Merge pull request #26 from Tim-Zhang/memory-apply-customized-attrs

Apply customized attributes in MemController.apply

.cargo/config

@@ -0,0 +1,2 @@
+[target.x86_64-unknown-linux-gnu]
+runner = 'sudo -E'

.github/workflows/PR-wip-checks.yaml

@@ -0,0 +1,21 @@
+name: Pull request WIP checks
+on:
+  pull_request:
+    types:
+      - opened
+      - synchronize
+      - reopened
+      - edited
+      - labeled
+      - unlabeled
+
+jobs:
+  pr_wip_check:
+    runs-on: ubuntu-latest
+    name: WIP Check
+    steps:
+    - name: WIP Check
+      uses: tim-actions/wip-check@1c2a1ca6c110026b3e2297bb2ef39e1747b5a755
+      with:
+        labels: '["do-not-merge", "wip", "rfc"]'
+        keywords: '["WIP", "wip", "RFC", "rfc", "dnm", "DNM", "do-not-merge"]'

.github/workflows/commit-message-check.yaml

@@ -0,0 +1,53 @@
+name: Commit Message Check
+on:
+  pull_request:
+    types:
+      - opened
+      - reopened
+      - synchronize
+
+env:
+  error_msg: |+
+    See the document below for help on formatting commits for the project.
+
+    https://github.com/kata-containers/community/blob/master/CONTRIBUTING.md#patch-forma
+
+jobs:
+  commit-message-check:
+    runs-on: ubuntu-latest
+    name: Commit Message Check
+    steps:
+    - name: Get PR Commits
+      id: 'get-pr-commits'
+      uses: tim-actions/get-pr-commits@v1.0.0
+      with:
+        token: ${{ secrets.GITHUB_TOKEN }}
+
+    - name: DCO Check
+      uses: tim-actions/dco@2fd0504dc0d27b33f542867c300c60840c6dcb20
+      with:
+        commits: ${{ steps.get-pr-commits.outputs.commits }}
+
+    - name: Commit Body Missing Check
+      if: ${{ success() || failure() }}
+      uses: tim-actions/commit-body-check@v1.0.2
+      with:
+        commits: ${{ steps.get-pr-commits.outputs.commits }}
+
+    - name: Check Subject Line Length
+      if: ${{ success() || failure() }}
+      uses: tim-actions/commit-message-checker-with-regex@v0.3.1
+      with:
+        commits: ${{ steps.get-pr-commits.outputs.commits }}
+        pattern: '^.{0,75}(\n.*)*$'
+        error: 'Subject too long (max 75)'
+        post_error: ${{ env.error_msg }}
+
+    - name: Check Body Line Length
+      if: ${{ success() || failure() }}
+      uses: tim-actions/commit-message-checker-with-regex@v0.3.1
+      with:
+        commits: ${{ steps.get-pr-commits.outputs.commits }}
+        pattern: '^.+(\n.{0,72})*$|^.+\n\s*[^a-zA-Z\s\n]|^.+\n\S+$'
+        error: 'Body line too long (max 72)'
+        post_error: ${{ env.error_msg }}

.travis.yml

@@ -1,11 +1,31 @@
+# Copyright (c) 2018 Levente Kurusa
+# Copyright (c) 2020 Ant Group
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+
+dist: bionic
+os: linux
 language: rust
+cache: cargo
+
+arch:
+  - amd64
+  - arm64
+
+install:
+  - rustup component add rustfmt
+
+script:
+  - RUSTFLAGS="--deny warnings" cargo build
+  - if [ "$TRAVIS_CPU_ARCH" == "amd64" ]; then cargo test -- --color always --nocapture ; fi
+  - cargo fmt -- --check
+
 rust:
-  - stable
-  - beta
+  - 1.44.1
   - nightly
-matrix:
+
+jobs:
   allow_failures:
     - rust: nightly
   fast_finish: true
-script:
-  - cargo build --verbose --all

Cargo.toml

@@ -1,15 +1,20 @@
 [package]
-name = "cgroups"
+name = "cgroups-rs"
 description = "Native Rust crate for managing control groups on Linux"
-repository = "https://github.com/levex/cgroups-rs"
+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 = "0.0.2"
-authors = ["Levente Kurusa <lkurusa@acm.org>"]
+version = "0.2.2"
+authors = ["The Kata Containers community <kata-dev@lists.katacontainers.io>", "Levente Kurusa <lkurusa@acm.org>", "Sam Wilson <tecywiz121@hotmail.com>"]
+edition = "2018"
 
 [dependencies]
+log = "0.4"
+regex = "1.1"
+nix = "0.18.0"
+libc = "0.2"
+procinfo = "0.4.2"
 
 [dev-dependencies]
-nix = "0.11.0"
-libc = "0.2.43"
+libc = "0.2.76"

README.md

@@ -1,25 +1,47 @@
-# cgroups-rs ![Build](https://travis-ci.org/levex/cgroups-rs.svg?branch=master)
+# cgroups-rs ![Build](https://travis-ci.org/kata-containers/cgroups-rs.svg?branch=master)
 Native Rust library for managing control groups under Linux
 
-# Example
+Both v1 and v2 of cgroups are supported.
 
-## Create a control group, and limit the pid resource
+# Examples
+
+## Create a control group using the builder pattern
 
 ``` rust
-// Acquire a handle for the V1 cgroup hierarchy.
-let hier = ::hierarchies::V1::new();
-// Create a control group named "example" in the hierarchy.
-let cg = Cgroup::new(&hier, String::from("example"), 0);
-{
-    // Get a handle to the pids controller of the control group.
-    let pids: &PidController = cg.controller_of().expect("No pids controller in V1 hierarchy!");
-    // Set the maximum amount of processes in the cgroup.
-    pids.set_pid_max(PidMax::Value(10));
-    // Check that this has had the desired effect by reading the value back from the kernel.
-    assert_eq!(pids.get_pid_max(), Some(PidMax::Value(10)));
-}
-// Once done, delete the control group (and its associated controllers).
+
+
+use cgroups_rs::*;
+use cgroups_rs::cgroup_builder::*;
+
+// Acquire a handle for the cgroup hierarchy.
+let hier = cgroups_rs::hierarchies::auto();
+
+// Use the builder pattern (see the documentation to create the control group)
+//
+// This creates a control group named "example" in the V1 hierarchy.
+    let cg: Cgroup = CgroupBuilder::new("example")
+        .cpu()
+        .shares(85)
+        .done()
+        .build(hier);
+
+// Now `cg` is a control group that gets 85% of the CPU time in relative to
+// other control groups.
+
+// Get a handle to the CPU controller.
+let cpus: &cgroups_rs::cpu::CpuController = cg.controller_of().unwrap();
+cpus.add_task(&CgroupPid::from(1234u64));
+
+// [...]
+
+// Finally, clean up and delete the control group.
 cg.delete();
+
+// Note that `Cgroup` does not implement `Drop` and therefore when the
+// structure is dropped, the Cgroup will stay around. This is because, later
+// you can then re-create the `Cgroup` using `load()`. We aren't too set on
+// this behavior, so it might change in the feature. Rest assured, it will be a
+// major version change.
 ```
 
 # Disclaimer
@@ -27,7 +49,7 @@ cg.delete();
 This crate is licensed under:
 
 - MIT License (see LICENSE-MIT); or
-- Apache 2.0 LIcense (see LICENSE-Apache-2.0),
+- Apache 2.0 License (see LICENSE-Apache-2.0),
 
 at your option.
 

src/cgroup.rs

@@ -1,74 +1,173 @@
+// 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 {CgroupError, CgroupPid, Resources, ControllIdentifier, Controller, Hierarchy, Subsystem};
+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: 
+/// 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.
-pub struct Cgroup<'b> {
+#[derive(Debug)]
+pub struct Cgroup {
     /// The list of subsystems that control this cgroup
     subsystems: Vec<Subsystem>,
 
     /// The hierarchy.
-    hier: &'b Hierarchy,
+    hier: Box<dyn Hierarchy>,
+    path: String,
 }
 
-impl<'b> Cgroup<'b> {
-
-    /// Create this control group.
-    fn create(self: &Self) {
-        for subsystem in &self.subsystems {
-            subsystem.to_controller().create();
+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().clone(), &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.
-    ///
-    /// Note that if the handle goes out of scope and is dropped, the control group is _not_
-    /// destroyed.
-    pub fn new(hier: &Hierarchy, path: String) -> Cgroup {
+    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.
-    ///
-    /// Note that if the handle goes out of scope and is dropped, the control group is _not_
-    /// destroyed.
-    pub fn load(hier: &Hierarchy, path: String) -> 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 != ""  {
-            subsystems = subsystems.into_iter().map(|x| x.enter(&path)).collect::<Vec<_>>();
+        if path.as_os_str() != "" {
+            subsystems = subsystems
+                .into_iter()
+                .map(|x| x.enter(path))
+                .collect::<Vec<_>>();
+        }
+
+        let cg = Cgroup {
+            path: path.to_str().unwrap().to_string(),
+            subsystems: subsystems,
+            hier,
+        };
+
+        cg
+    }
+
+    /// 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<_>>();
         }
 
         let cg = Cgroup {
             subsystems: subsystems,
-            hier: hier,
+            hier,
+            path: path.to_str().unwrap().to_string(),
         };
 
         cg
     }
 
     /// The list of subsystems that this control group supports.
-    pub fn subsystems(self: &Self) -> &Vec<Subsystem> {
+    pub fn subsystems(&self) -> &Vec<Subsystem> {
         &self.subsystems
     }
 
@@ -78,31 +177,39 @@ impl<'b> Cgroup<'b> {
     /// 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: Self) {
-        self.subsystems.into_iter().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(),
+    pub fn delete(&self) -> Result<()> {
+        if self.v2() {
+            if self.path != "" {
+                let mut p = self.hier.root().clone();
+                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: &Self, res: &Resources) {
-        for subsystem in &self.subsystems {
-            subsystem.to_controller().apply(res);
-        }
+    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.
@@ -116,15 +223,14 @@ impl<'b> Cgroup<'b> {
     ///                             .expect("No cpu controller attached!");
     /// ```
     pub fn controller_of<'a, T>(self: &'a Self) -> Option<&'a T>
-        where &'a T: From<&'a Subsystem>,
-                  T: Controller + ControllIdentifier,
+    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
-                 */
+                // N.B.:
+                // https://play.rust-lang.org/?gist=978b2846bacebdaa00be62374f4f4334&version=stable&mode=debug&edition=2015
                 return Some(i.into());
             }
         }
@@ -135,25 +241,143 @@ impl<'b> Cgroup<'b> {
     ///
     /// 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: &Self, pid: CgroupPid) {
+    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: &Self, pid: CgroupPid) -> Result<(), CgroupError> {
-        self.subsystems().iter().try_for_each(|sub| sub.to_controller().add_task(&pid))
+    pub fn add_task(&self, pid: CgroupPid) -> Result<()> {
+        if self.v2() {
+            let subsystems = self.subsystems();
+            if subsystems.len() > 0 {
+                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: &Self) -> Vec<CgroupPid> {
-        /* Collect the tasks from all subsystems */
-        let mut v = self.subsystems().iter()
-            .map(|x| x.to_controller().tasks())
-            .fold(vec![], |mut acc, mut x| { acc.append(&mut x); acc });
+    pub fn tasks(&self) -> Vec<CgroupPid> {
+        // Collect the tasks from all subsystems
+        let mut v = if self.v2() {
+            let subsystems = self.subsystems();
+            if subsystems.len() > 0 {
+                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: &'static str = "/sys/fs/cgroup";
+
+fn enable_controllers(controllers: &Vec<String>, path: &PathBuf) {
+    let mut f = path.clone();
+    f.push("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(String::new())
+        .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() {
+            match ::std::fs::create_dir(fp.clone()) {
+                Err(e) => return Err(Error::with_cause(ErrorKind::FsError, e)),
+                Ok(_) => {}
+            }
+        }
+
+        if i < last_index {
+            // enable controllers for substree
+            enable_controllers(&controllers, &fp);
+        }
+    }
+
+    Ok(())
+}
+
+pub fn get_cgroups_relative_paths() -> Result<HashMap<String, String>> {
+    let mut m = HashMap::new();
+    let content =
+        fs::read_to_string("/proc/self/cgroup").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/cgroup_builder.rs

@@ -0,0 +1,390 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+//! This module allows the user to create a control group using the Builder pattern.
+//! # Example
+//!
+//! The following example demonstrates how the control group builder looks like.  The user
+//! specifies the name of the control group (here: "hello") and the hierarchy it belongs to (here:
+//! a V1 hierarchy). Next, the user selects a subsystem by calling functions like `memory()`,
+//! `cpu()` and `devices()`. The user can then add restrictions and details via subsystem-specific
+//! calls. To finalize a subsystem, the user may call `done()`. Finally, if the control group build
+//! is done and all requirements/restrictions have been specified, the control group can be created
+//! 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();
+//! let cgroup: Cgroup = CgroupBuilder::new("hello")
+//!      .memory()
+//!          .kernel_memory_limit(1024 * 1024)
+//!          .memory_hard_limit(1024 * 1024)
+//!          .done()
+//!      .cpu()
+//!          .shares(100)
+//!          .done()
+//!      .devices()
+//!          .device(1000, 10, DeviceType::Block, true,
+//!             vec![DevicePermissions::Read,
+//!                  DevicePermissions::Write,
+//!                  DevicePermissions::MkNod])
+//!          .device(6, 1, DeviceType::Char, false, vec![])
+//!          .done()
+//!      .network()
+//!          .class_id(1337)
+//!          .priority("eth0".to_string(), 100)
+//!          .priority("wl0".to_string(), 200)
+//!          .done()
+//!      .hugepages()
+//!          .limit("2M".to_string(), 0)
+//!          .limit("4M".to_string(), 4 * 1024 * 1024 * 100)
+//!          .limit("2G".to_string(), 2 * 1024 * 1024 * 1024)
+//!          .done()
+//!      .blkio()
+//!          .weight(123)
+//!          .leaf_weight(99)
+//!          .weight_device(6, 1, Some(100), Some(55))
+//!          .weight_device(6, 1, Some(100), Some(55))
+//!          .throttle_iops()
+//!              .read(6, 1, 10)
+//!              .write(11, 1, 100)
+//!          .throttle_bps()
+//!              .read(6, 1, 10)
+//!              .write(11, 1, 100)
+//!          .done()
+//!      .build(h);
+//! ```
+
+use crate::{
+    BlkIoDeviceResource, BlkIoDeviceThrottleResource, Cgroup, DeviceResource, Hierarchy,
+    HugePageResource, MaxValue, NetworkPriority, Resources,
+};
+
+macro_rules! gen_setter {
+    ($res:ident, $cont:ident, $func:ident, $name:ident, $ty:ty) => {
+        /// See the similarly named function in the respective controller.
+        pub fn $name(mut self, $name: $ty) -> Self {
+            self.cgroup.resources.$res.$name = Some($name);
+            self
+        }
+    };
+}
+
+/// A control group builder instance
+pub struct CgroupBuilder {
+    name: String,
+    /// Internal, unsupported field: use the associated builders instead.
+    resources: Resources,
+}
+
+impl CgroupBuilder {
+    /// Start building a control group with the supplied hierarchy and name pair.
+    ///
+    /// Note that this does not actually create the control group until `build()` is called.
+    pub fn new(name: &str) -> CgroupBuilder {
+        CgroupBuilder {
+            name: name.to_owned(),
+            resources: Resources::default(),
+        }
+    }
+
+    /// Builds the memory resources of the control group.
+    pub fn memory(self) -> MemoryResourceBuilder {
+        MemoryResourceBuilder { cgroup: self }
+    }
+
+    /// Builds the pid resources of the control group.
+    pub fn pid(self) -> PidResourceBuilder {
+        PidResourceBuilder { cgroup: self }
+    }
+
+    /// Builds the cpu resources of the control group.
+    pub fn cpu(self) -> CpuResourceBuilder {
+        CpuResourceBuilder { cgroup: self }
+    }
+
+    /// Builds the devices resources of the control group, disallowing or
+    /// allowing access to certain devices in the system.
+    pub fn devices(self) -> DeviceResourceBuilder {
+        DeviceResourceBuilder { cgroup: self }
+    }
+
+    /// Builds the network resources of the control group, setting class id, or
+    /// various priorities on networking interfaces.
+    pub fn network(self) -> NetworkResourceBuilder {
+        NetworkResourceBuilder { cgroup: self }
+    }
+
+    /// Builds the hugepage/hugetlb resources available to the control group.
+    pub fn hugepages(self) -> HugepagesResourceBuilder {
+        HugepagesResourceBuilder { cgroup: self }
+    }
+
+    /// Builds the block I/O resources available for the control group.
+    pub fn blkio(self) -> BlkIoResourcesBuilder {
+        BlkIoResourcesBuilder {
+            cgroup: self,
+            throttling_iops: false,
+        }
+    }
+
+    /// 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
+    }
+}
+
+/// A builder that configures the memory controller of a control group.
+pub struct MemoryResourceBuilder {
+    cgroup: CgroupBuilder,
+}
+
+impl MemoryResourceBuilder {
+    gen_setter!(
+        memory,
+        MemController,
+        set_kmem_limit,
+        kernel_memory_limit,
+        i64
+    );
+    gen_setter!(memory, MemController, set_limit, memory_hard_limit, i64);
+    gen_setter!(
+        memory,
+        MemController,
+        set_soft_limit,
+        memory_soft_limit,
+        i64
+    );
+    gen_setter!(
+        memory,
+        MemController,
+        set_tcp_limit,
+        kernel_tcp_memory_limit,
+        i64
+    );
+    gen_setter!(
+        memory,
+        MemController,
+        set_memswap_limit,
+        memory_swap_limit,
+        i64
+    );
+    gen_setter!(memory, MemController, set_swappiness, swappiness, u64);
+
+    /// Finish the construction of the memory resources of a control group.
+    pub fn done(self) -> CgroupBuilder {
+        self.cgroup
+    }
+}
+
+/// A builder that configures the pid controller of a control group.
+pub struct PidResourceBuilder {
+    cgroup: CgroupBuilder,
+}
+
+impl PidResourceBuilder {
+    gen_setter!(
+        pid,
+        PidController,
+        set_pid_max,
+        maximum_number_of_processes,
+        MaxValue
+    );
+
+    /// Finish the construction of the pid resources of a control group.
+    pub fn done(self) -> CgroupBuilder {
+        self.cgroup
+    }
+}
+
+/// A builder that configures the cpuset & cpu controllers of a control group.
+pub struct CpuResourceBuilder {
+    cgroup: CgroupBuilder,
+}
+
+impl CpuResourceBuilder {
+    gen_setter!(cpu, CpuSetController, set_cpus, cpus, String);
+    gen_setter!(cpu, CpuSetController, set_mems, mems, String);
+    gen_setter!(cpu, CpuController, set_shares, shares, u64);
+    gen_setter!(cpu, CpuController, set_cfs_quota, quota, i64);
+    gen_setter!(cpu, CpuController, set_cfs_period, period, u64);
+    gen_setter!(cpu, CpuController, set_rt_runtime, realtime_runtime, i64);
+    gen_setter!(cpu, CpuController, set_rt_period, realtime_period, u64);
+
+    /// Finish the construction of the cpu resources of a control group.
+    pub fn done(self) -> CgroupBuilder {
+        self.cgroup
+    }
+}
+
+/// A builder that configures the devices controller of a control group.
+pub struct DeviceResourceBuilder {
+    cgroup: CgroupBuilder,
+}
+
+impl DeviceResourceBuilder {
+    /// Restrict (or allow) a device to the tasks inside the control group.
+    pub fn device(
+        mut self,
+        major: i64,
+        minor: i64,
+        devtype: crate::devices::DeviceType,
+        allow: bool,
+        access: Vec<crate::devices::DevicePermissions>,
+    ) -> DeviceResourceBuilder {
+        self.cgroup.resources.devices.devices.push(DeviceResource {
+            major,
+            minor,
+            devtype,
+            allow,
+            access,
+        });
+        self
+    }
+
+    /// Finish the construction of the devices resources of a control group.
+    pub fn done(self) -> CgroupBuilder {
+        self.cgroup
+    }
+}
+
+/// A builder that configures the net_cls & net_prio controllers of a control group.
+pub struct NetworkResourceBuilder {
+    cgroup: CgroupBuilder,
+}
+
+impl NetworkResourceBuilder {
+    gen_setter!(network, NetclsController, set_class, class_id, u64);
+
+    /// Set the priority of the tasks when operating on a networking device defined by `name` to be
+    /// `priority`.
+    pub fn priority(mut self, name: String, priority: u64) -> NetworkResourceBuilder {
+        self.cgroup
+            .resources
+            .network
+            .priorities
+            .push(NetworkPriority { name, priority });
+        self
+    }
+
+    /// Finish the construction of the network resources of a control group.
+    pub fn done(self) -> CgroupBuilder {
+        self.cgroup
+    }
+}
+
+/// A builder that configures the hugepages controller of a control group.
+pub struct HugepagesResourceBuilder {
+    cgroup: CgroupBuilder,
+}
+
+impl HugepagesResourceBuilder {
+    /// Limit the usage of certain hugepages (determined by `size`) to be at most `limit` bytes.
+    pub fn limit(mut self, size: String, limit: u64) -> HugepagesResourceBuilder {
+        self.cgroup
+            .resources
+            .hugepages
+            .limits
+            .push(HugePageResource { size, limit });
+        self
+    }
+
+    /// Finish the construction of the network resources of a control group.
+    pub fn done(self) -> CgroupBuilder {
+        self.cgroup
+    }
+}
+
+/// A builder that configures the blkio controller of a control group.
+pub struct BlkIoResourcesBuilder {
+    cgroup: CgroupBuilder,
+    throttling_iops: bool,
+}
+
+impl BlkIoResourcesBuilder {
+    gen_setter!(blkio, BlkIoController, set_weight, weight, u16);
+    gen_setter!(blkio, BlkIoController, set_leaf_weight, leaf_weight, u16);
+
+    /// Set the weight of a certain device.
+    pub fn weight_device(
+        mut self,
+        major: u64,
+        minor: u64,
+        weight: Option<u16>,
+        leaf_weight: Option<u16>,
+    ) -> BlkIoResourcesBuilder {
+        self.cgroup
+            .resources
+            .blkio
+            .weight_device
+            .push(BlkIoDeviceResource {
+                major,
+                minor,
+                weight,
+                leaf_weight,
+            });
+        self
+    }
+
+    /// Start configuring the I/O operations per second metric.
+    pub fn throttle_iops(mut self) -> BlkIoResourcesBuilder {
+        self.throttling_iops = true;
+        self
+    }
+
+    /// Start configuring the bytes per second metric.
+    pub fn throttle_bps(mut self) -> BlkIoResourcesBuilder {
+        self.throttling_iops = false;
+        self
+    }
+
+    /// Limit the read rate of the current metric for a certain device.
+    pub fn read(mut self, major: u64, minor: u64, rate: u64) -> BlkIoResourcesBuilder {
+        let throttle = BlkIoDeviceThrottleResource { major, minor, rate };
+        if self.throttling_iops {
+            self.cgroup
+                .resources
+                .blkio
+                .throttle_read_iops_device
+                .push(throttle);
+        } else {
+            self.cgroup
+                .resources
+                .blkio
+                .throttle_read_bps_device
+                .push(throttle);
+        }
+        self
+    }
+
+    /// Limit the write rate of the current metric for a certain device.
+    pub fn write(mut self, major: u64, minor: u64, rate: u64) -> BlkIoResourcesBuilder {
+        let throttle = BlkIoDeviceThrottleResource { major, minor, rate };
+        if self.throttling_iops {
+            self.cgroup
+                .resources
+                .blkio
+                .throttle_write_iops_device
+                .push(throttle);
+        } else {
+            self.cgroup
+                .resources
+                .blkio
+                .throttle_write_bps_device
+                .push(throttle);
+        }
+        self
+    }
+
+    /// Finish the construction of the blkio resources of a control group.
+    pub fn done(self) -> CgroupBuilder {
+        self.cgroup
+    }
+}

src/cpu.rs

@@ -1,22 +1,37 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `cpu` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/scheduler/sched-design-CFS.txt](https://www.kernel.org/doc/Documentation/scheduler/sched-design-CFS.txt)
 //!  paragraph 7 ("GROUP SCHEDULER EXTENSIONS TO CFS").
-use std::path::PathBuf;
+use std::fs::File;
 use std::io::{Read, Write};
+use std::path::PathBuf;
 
-use {CgroupError, CpuResources, Controllers, Controller, Resources, ControllIdentifier, Subsystem};
+use crate::error::ErrorKind::*;
+use crate::error::*;
+use crate::{parse_max_value, read_i64_from, read_u64_from};
+
+use crate::{
+    ControllIdentifier, ControllerInternal, Controllers, CpuResources, CustomizedAttribute,
+    MaxValue, Resources, Subsystem,
+};
 
 /// A controller that allows controlling the `cpu` subsystem of a Cgroup.
-/// 
+///
 /// In essence, it allows gathering information about how much the tasks inside the control group
 /// are using the CPU and creating rules that limit their usage. Note that this crate does not yet
 /// support managing realtime tasks.
 #[derive(Debug, Clone)]
-pub struct CpuController{
+pub struct CpuController {
     base: PathBuf,
     path: PathBuf,
+    v2: bool,
 }
 
 /// The current state of the control group and its processes.
@@ -28,23 +43,49 @@ pub struct Cpu {
     pub stat: String,
 }
 
-impl Controller for CpuController {
-    fn control_type(self: &Self) -> Controllers { Controllers::Cpu}
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+/// The current state of the control group and its processes.
+#[derive(Debug)]
+struct CFSQuotaAndPeriod {
+    quota: MaxValue,
+    period: u64,
+}
 
-    fn apply(self: &Self, res: &Resources) {
-        /* get the resources that apply to this controller */
+impl ControllerInternal for CpuController {
+    fn control_type(&self) -> Controllers {
+        Controllers::Cpu
+    }
+
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
+
+    fn is_v2(&self) -> bool {
+        self.v2
+    }
+
+    fn apply(&self, res: &Resources) -> Result<()> {
+        // get the resources that apply to this controller
         let res: &CpuResources = &res.cpu;
 
-        if res.update_values {
-            /* apply pid_max */
-            let _ = self.set_shares(res.shares);
-            let _ = self.set_cfs_period(res.period);
-            let _ = self.set_cfs_quota(res.quota as u64);
-            /* TODO: rt properties (CONFIG_RT_GROUP_SCHED) are not yet supported */
-        }
+        update_and_test!(self, set_shares, res.shares, shares);
+        update_and_test!(self, set_cfs_period, res.period, cfs_period);
+        update_and_test!(self, set_cfs_quota, res.quota, cfs_quota);
+
+        res.attrs.iter().for_each(|(k, v)| {
+            let _ = self.set(k, v);
+        });
+
+        // TODO: rt properties (CONFIG_RT_GROUP_SCHED) are not yet supported
+
+        Ok(())
     }
 }
 
@@ -61,63 +102,205 @@ impl<'a> From<&'a Subsystem> for &'a CpuController {
                 Subsystem::Cpu(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
 impl CpuController {
-    /// Contructs a new `CpuController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Contructs a new `CpuController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf, v2: bool) -> Self {
         Self {
             base: root.clone(),
             path: root,
+            v2: v2,
         }
     }
 
     /// Returns CPU time statistics based on the processes in the control group.
-    pub fn cpu(self: &Self) -> Cpu {
+    pub fn cpu(&self) -> Cpu {
         Cpu {
-            stat: self.open_path("cpu.stat", false).and_then(|mut file| {
-                let mut s = String::new();
-                let res = file.read_to_string(&mut s);
-                match res {
-                    Ok(_) => Ok(s),
-                    Err(e) => Err(CgroupError::ReadError(e)),
-                }
-            }).unwrap_or("".to_string()),
+            stat: self
+                .open_path("cpu.stat", false)
+                .and_then(|mut file| {
+                    let mut s = String::new();
+                    let res = file.read_to_string(&mut s);
+                    match res {
+                        Ok(_) => Ok(s),
+                        Err(e) => Err(Error::with_cause(ReadFailed, e)),
+                    }
+                })
+                .unwrap_or("".to_string()),
         }
     }
 
     /// Configures the CPU bandwidth (in relative relation to other control groups and this control
     /// group's parent).
-    /// 
+    ///
     /// For example, setting control group `A`'s `shares` to `100`, and control group `B`'s
     /// `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: &Self, shares: u64) -> Result<(), CgroupError> {
-        self.open_path("cpu.shares", true).and_then(|mut file| {
-            file.write_all(shares.to_string().as_ref()).map_err(CgroupError::WriteError)
+    pub fn set_shares(&self, shares: u64) -> Result<()> {
+        let mut file = "cpu.shares";
+        if self.v2 {
+            file = "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))
         })
     }
 
+    /// Retrieve the CPU bandwidth that this control group (relative to other control groups and
+    /// this control group's parent) can use.
+    pub fn shares(&self) -> Result<u64> {
+        let mut file = "cpu.shares";
+        if self.v2 {
+            file = "cpu.weight";
+        }
+        self.open_path(file, false).and_then(read_u64_from)
+    }
+
     /// Specify a period (when using the CFS scheduler) of time in microseconds for how often this
     /// control group's access to the CPU should be reallocated.
-    pub fn set_cfs_period(self: &Self, us: u64) -> Result<(), CgroupError> {
-        self.open_path("cpu.cfs_period_us", true).and_then(|mut file| {
-            file.write_all(us.to_string().as_ref()).map_err(CgroupError::WriteError)
-        })
+    pub fn set_cfs_period(&self, us: u64) -> Result<()> {
+        if self.v2 {
+            return self.set_cfs_quota_and_period(None, Some(us));
+        }
+        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))
+            })
+    }
+
+    /// Retrieve the period of time of how often this cgroup's access to the CPU should be
+    /// reallocated in microseconds.
+    pub fn cfs_period(&self) -> Result<u64> {
+        if self.v2 {
+            let current_value = self
+                .open_path("cpu.max", false)
+                .and_then(parse_cfs_quota_and_period)?;
+            return Ok(current_value.period);
+        }
+        self.open_path("cpu.cfs_period_us", false)
+            .and_then(read_u64_from)
     }
 
     /// Specify a quota (when using the CFS scheduler) of time in microseconds for which all tasks
     /// in this control group can run during one period (see: `set_cfs_period()`).
-    pub fn set_cfs_quota(self: &Self, us: u64) -> Result<(), CgroupError> {
-        self.open_path("cpu.cfs_quota_us", true).and_then(|mut file| {
-            file.write_all(us.to_string().as_ref()).map_err(CgroupError::WriteError)
+    pub fn set_cfs_quota(&self, us: i64) -> Result<()> {
+        if self.v2 {
+            return self.set_cfs_quota_and_period(Some(us), None);
+        }
+        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))
+            })
+    }
+
+    /// Retrieve the quota of time for which all tasks in this cgroup can run during one period, in
+    /// microseconds.
+    pub fn cfs_quota(&self) -> Result<i64> {
+        if self.v2 {
+            let current_value = self
+                .open_path("cpu.max", false)
+                .and_then(parse_cfs_quota_and_period)?;
+            return Ok(current_value.quota.to_i64());
+        }
+
+        self.open_path("cpu.cfs_quota_us", false)
+            .and_then(read_i64_from)
+    }
+
+    pub fn set_cfs_quota_and_period(&self, quota: Option<i64>, period: Option<u64>) -> Result<()> {
+        if !self.v2 {
+            if let Some(q) = quota {
+                self.set_cfs_quota(q)?;
+            }
+            if let Some(p) = period {
+                self.set_cfs_period(p)?;
+            }
+            return Ok(());
+        }
+
+        // https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html
+
+        // cpu.max
+        // A read-write two value file which exists on non-root cgroups. The default is “max 100000”.
+        // The maximum bandwidth limit. It’s in the following format:
+        // $MAX $PERIOD
+        // which indicates that the group may consume upto $MAX in each $PERIOD duration.
+        // “max” for $MAX indicates no limit. If only one number is written, $MAX is updated.
+
+        let current_value = self
+            .open_path("cpu.max", false)
+            .and_then(parse_cfs_quota_and_period)?;
+
+        let new_quota = if let Some(q) = quota {
+            if q > 0 {
+                q.to_string()
+            } else {
+                "max".to_string()
+            }
+        } else {
+            current_value.quota.to_string()
+        };
+
+        let new_period = if let Some(p) = period {
+            p.to_string()
+        } else {
+            current_value.period.to_string()
+        };
+
+        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))
         })
     }
+
+    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))
+            })
+    }
+
+    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))
+            })
+    }
+}
+
+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))?;
+
+    let fields = content.trim().split(' ').collect::<Vec<&str>>();
+    if fields.len() != 2 {
+        return Err(Error::from_string(format!("invaild format: {}", content)));
+    }
+
+    let quota = parse_max_value(&fields[0].to_string())?;
+    let period = fields[1]
+        .parse::<u64>()
+        .map_err(|e| Error::with_cause(ParseError, e))?;
+
+    Ok(CFSQuotaAndPeriod {
+        quota: quota,
+        period: period,
+    })
 }

src/cpuacct.rs

@@ -1,12 +1,20 @@
+// Copyright (c) 2018 Levente Kurusa
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `cpuacct` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/cgroup-v1/cpuacct.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/cpuacct.txt)
+use std::io::Write;
 use std::path::PathBuf;
-use std::io::{Read, Write};
-use std::fs::File;
 
-use {CgroupError, Controllers, Resources, Subsystem, ControllIdentifier, Controller};
+use crate::error::ErrorKind::*;
+use crate::error::*;
+
+use crate::{read_string_from, read_u64_from};
+use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
 
 /// A controller that allows controlling the `cpuacct` subsystem of a Cgroup.
 ///
@@ -49,13 +57,22 @@ pub struct CpuAcct {
     pub usage_user: u64,
 }
 
-impl Controller for CpuAcctController {
-    fn control_type(self: &Self) -> Controllers { Controllers::CpuAcct }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+impl ControllerInternal for CpuAcctController {
+    fn control_type(&self) -> Controllers {
+        Controllers::CpuAcct
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
 
-    fn apply(self: &Self, _res: &Resources) {
+    fn apply(&self, _res: &Resources) -> Result<()> {
+        Ok(())
     }
 }
 
@@ -72,39 +89,17 @@ impl<'a> From<&'a Subsystem> for &'a CpuAcctController {
                 Subsystem::CpuAcct(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
-fn read_u64_from(mut file: File) -> Result<u64, CgroupError> {
-    let mut string = String::new();
-    let res = file.read_to_string(&mut string);
-    match res {
-        Ok(_) => match string.trim().parse() {
-            Ok(e) => Ok(e),
-            Err(_) => Err(CgroupError::ParseError),
-        },
-        Err(e) => Err(CgroupError::ReadError(e)),
-    }
-}
-
-fn read_string_from(mut file: File) -> Result<String, CgroupError> {
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => Ok(string.trim().to_string()),
-        Err(e) => Err(CgroupError::ReadError(e)),
-    }
-}
-
 impl CpuAcctController {
-
-    /// Contructs a new `CpuAcctController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Contructs a new `CpuAcctController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf) -> Self {
         Self {
             base: root.clone(),
             path: root,
@@ -112,34 +107,48 @@ impl CpuAcctController {
     }
 
     /// Gathers the statistics that are available in the control group into a `CpuAcct` structure.
-    pub fn cpuacct(self: &Self) -> CpuAcct {
+    pub fn cpuacct(&self) -> CpuAcct {
         CpuAcct {
-            stat: self.open_path("cpuacct.stat", false)
-                    .and_then(|file| read_string_from(file)).unwrap_or("".to_string()),
-            usage: self.open_path("cpuacct.usage", false)
-                    .and_then(|file| read_u64_from(file))
-                    .unwrap_or(0),
-            usage_all: self.open_path("cpuacct.usage_all", false)
-                    .and_then(|file| read_string_from(file)).unwrap_or("".to_string()),
-            usage_percpu: self.open_path("cpuacct.usage_percpu", false)
-                    .and_then(|file| read_string_from(file)).unwrap_or("".to_string()),
-            usage_percpu_sys: self.open_path("cpuacct.usage_percpu_sys", false)
-                    .and_then(|file| read_string_from(file)).unwrap_or("".to_string()),
-            usage_percpu_user: self.open_path("cpuacct.usage_percpu_user", false)
-                    .and_then(|file| read_string_from(file)).unwrap_or("".to_string()),
-            usage_sys: self.open_path("cpuacct.usage_sys", false)
-                    .and_then(|file| read_u64_from(file))
-                    .unwrap_or(0),
-            usage_user: self.open_path("cpuacct.usage_user", false)
-                    .and_then(|file| read_u64_from(file))
-                    .unwrap_or(0),
+            stat: self
+                .open_path("cpuacct.stat", false)
+                .and_then(|file| read_string_from(file))
+                .unwrap_or("".to_string()),
+            usage: self
+                .open_path("cpuacct.usage", false)
+                .and_then(|file| read_u64_from(file))
+                .unwrap_or(0),
+            usage_all: self
+                .open_path("cpuacct.usage_all", false)
+                .and_then(|file| read_string_from(file))
+                .unwrap_or("".to_string()),
+            usage_percpu: self
+                .open_path("cpuacct.usage_percpu", false)
+                .and_then(|file| read_string_from(file))
+                .unwrap_or("".to_string()),
+            usage_percpu_sys: self
+                .open_path("cpuacct.usage_percpu_sys", false)
+                .and_then(|file| read_string_from(file))
+                .unwrap_or("".to_string()),
+            usage_percpu_user: self
+                .open_path("cpuacct.usage_percpu_user", false)
+                .and_then(|file| read_string_from(file))
+                .unwrap_or("".to_string()),
+            usage_sys: self
+                .open_path("cpuacct.usage_sys", false)
+                .and_then(|file| read_u64_from(file))
+                .unwrap_or(0),
+            usage_user: self
+                .open_path("cpuacct.usage_user", false)
+                .and_then(|file| read_u64_from(file))
+                .unwrap_or(0),
         }
     }
 
     /// Reset the statistics the kernel has gathered about the control group.
-    pub fn reset(self: &Self) -> Result<(), CgroupError> {
+    pub fn reset(&self) -> Result<()> {
         self.open_path("cpuacct.usage", true).and_then(|mut file| {
-            file.write_all(b"0").map_err(CgroupError::WriteError)
+            file.write_all(b"0")
+                .map_err(|e| Error::with_cause(WriteFailed, e))
         })
     }
 }

src/cpuset.rs

@@ -1,39 +1,54 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `cpuset` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/cgroup-v1/cpusets.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/cpusets.txt)
-use std::path::PathBuf;
-use std::io::{Read, Write};
-use std::fs::File;
 
-use {CgroupError, CpuResources, Resources, Controller, ControllIdentifier, Subsystem, Controllers};
-use CgroupError::*;
+use log::*;
+use std::io::Write;
+use std::path::PathBuf;
+
+use crate::error::ErrorKind::*;
+use crate::error::*;
+
+use crate::{read_string_from, read_u64_from};
+use crate::{
+    ControllIdentifier, ControllerInternal, Controllers, CpuResources, Resources, Subsystem,
+};
 
 /// A controller that allows controlling the `cpuset` subsystem of a Cgroup.
-/// 
+///
 /// In essence, this controller is responsible for restricting the tasks in the control group to a
 /// set of CPUs and/or memory nodes.
 #[derive(Debug, Clone)]
 pub struct CpuSetController {
     base: PathBuf,
     path: PathBuf,
+    v2: bool,
 }
 
 /// The current state of the `cpuset` controller for this control group.
 pub struct CpuSet {
     /// If true, no other control groups can share the CPUs listed in the `cpus` field.
     pub cpu_exclusive: bool,
-    /// The list of CPUs the tasks of the control group can run on. This is a comma-separated list
-    /// with dashes between numbers representing ranges.
-    pub cpus: String,
+    /// The list of CPUs the tasks of the control group can run on.
+    ///
+    /// This is a vector of `(start, end)` tuples, where each tuple is a range of CPUs where the
+    /// control group is allowed to run on. Both sides of the range are inclusive.
+    pub cpus: Vec<(u64, u64)>,
     /// The list of CPUs that the tasks can effectively run on. This removes the list of CPUs that
     /// the parent (and all of its parents) cannot run on from the `cpus` field of this control
     /// group.
-    pub effective_cpus: String,
+    pub effective_cpus: Vec<(u64, u64)>,
     /// The list of memory nodes that the tasks can effectively use. This removes the list of nodes that
     /// the parent (and all of its parents) cannot use from the `mems` field of this control
     /// group.
-    pub effective_mems: String,
+    pub effective_mems: Vec<(u64, u64)>,
     /// If true, no other control groups can share the memory nodes listed in the `mems` field.
     pub mem_exclusive: bool,
     /// If true, the control group is 'hardwalled'. Kernel memory allocations (except for a few
@@ -48,13 +63,14 @@ pub struct CpuSet {
     /// the memory pressure for control groups or not.
     pub memory_pressure_enabled: Option<bool>,
     /// If true, filesystem buffers are spread across evenly between the nodes specified in `mems`.
-    pub memory_spread_page: bool, 
+    pub memory_spread_page: bool,
     /// If true, kernel slab caches for file I/O are spread across evenly between the nodes
     /// specified in `mems`.
-    pub memory_spread_slab: bool, 
-    /// The list of memory nodes the tasks of the control group can use. This is a comma-separated list
-    /// with dashes between numbers representing ranges.
-    pub mems: String,
+    pub memory_spread_slab: bool,
+    /// The list of memory nodes the tasks of the control group can use.
+    ///
+    /// The format is the same as the `cpus`, `effective_cpus` and `effective_mems` fields.
+    pub mems: Vec<(u64, u64)>,
     /// If true, the kernel will attempt to rebalance the load between the CPUs specified in the
     /// `cpus` field of this control group.
     pub sched_load_balance: bool,
@@ -70,27 +86,103 @@ pub struct CpuSet {
     /// |           5          | Immediately balance the load between CPUs even if the system is NUMA |
     /// |           6          | Immediately balance the load between all CPUs |
     pub sched_relax_domain_level: u64,
-
 }
 
-impl Controller for CpuSetController {
-    fn control_type(self: &Self) -> Controllers { Controllers::CpuSet }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+impl ControllerInternal for CpuSetController {
+    fn control_type(&self) -> Controllers {
+        Controllers::CpuSet
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
 
-    fn apply(self: &Self, res: &Resources) {
-        /* get the resources that apply to this controller */
+    fn is_v2(&self) -> bool {
+        self.v2
+    }
+
+    fn apply(&self, res: &Resources) -> Result<()> {
+        // get the resources that apply to this controller
         let res: &CpuResources = &res.cpu;
 
-        if res.update_values {
-            /* apply pid_max */
-            let _ = self.set_cpus(&res.cpus);
-            let _ = self.set_mems(&res.mems);
+        update!(self, set_cpus, res.cpus.as_ref());
+        update!(self, set_mems, res.mems.as_ref());
+
+        Ok(())
+    }
+
+    fn post_create(&self) {
+        if self.is_v2() {
+            return;
+        }
+        let current = self.get_path();
+
+        if current != self.get_base() {
+            match copy_from_parent(current.to_str().unwrap(), "cpuset.cpus") {
+                Ok(_) => (),
+                Err(err) => error!("error create_dir for cpuset.cpus {:?}", err),
+            }
+            match copy_from_parent(current.to_str().unwrap(), "cpuset.mems") {
+                Ok(_) => (),
+                Err(err) => error!("error create_dir for cpuset.mems {:?}", err),
+            }
         }
     }
 }
 
+fn find_no_empty_parent(from: &str, file: &str) -> Result<(String, Vec<PathBuf>)> {
+    let mut current_path = ::std::path::Path::new(from).to_path_buf();
+    let mut v = vec![];
+
+    loop {
+        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)),
+            };
+
+        if current_value != "" {
+            return Ok((current_value, v));
+        }
+        v.push(current_path.clone());
+
+        let parent = match current_path.parent() {
+            Some(p) => p,
+            None => return Ok(("".to_string(), v)),
+        };
+
+        // next loop, find parent
+        current_path = parent.to_path_buf();
+    }
+}
+
+/// copy_from_parent copy the cpuset.cpus and cpuset.mems from the parent
+/// directory to the current directory if the file's contents are 0
+fn copy_from_parent(current: &str, file: &str) -> Result<()> {
+    // find not empty cpus/memes from current directory.
+    let (value, parents) = find_no_empty_parent(current, file)?;
+
+    if value == "" || parents.len() == 0 {
+        return Ok(());
+    }
+
+    for p in parents.iter().rev() {
+        let mut pb = p.clone();
+        pb.push(file);
+        match ::std::fs::write(pb.to_str().unwrap(), value.as_bytes()) {
+            Ok(_) => (),
+            Err(e) => return Err(Error::with_cause(WriteFailed, e)),
+        }
+    }
+
+    Ok(())
+}
+
 impl ControllIdentifier for CpuSetController {
     fn controller_type() -> Controllers {
         Controllers::CpuSet
@@ -104,94 +196,145 @@ impl<'a> From<&'a Subsystem> for &'a CpuSetController {
                 Subsystem::CpuSet(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
-fn read_string_from(mut file: File) -> Result<String, CgroupError> {
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => Ok(string.trim().to_string()),
-        Err(e) => Err(CgroupError::ReadError(e)),
-    }
-}
+/// Parse a string like "1,2,4-5,8" into a list of (start, end) tuples.
+fn parse_range(s: String) -> Result<Vec<(u64, u64)>> {
+    let mut fin = Vec::new();
 
-fn read_u64_from(mut file: File) -> Result<u64, CgroupError> {
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => string.trim().parse().map_err(|_| ParseError),
-        Err(e) => Err(CgroupError::ReadError(e)),
+    if s == "".to_string() {
+        return Ok(fin);
     }
+
+    // first split by commas
+    let comma_split = s.split(",");
+
+    for sp in comma_split {
+        if sp.contains("-") {
+            // this is a true range
+            let dash_split = sp.split("-").collect::<Vec<_>>();
+            if dash_split.len() != 2 {
+                return Err(Error::new(ParseError));
+            }
+            let first = dash_split[0].parse::<u64>();
+            let second = dash_split[1].parse::<u64>();
+            if first.is_err() || second.is_err() {
+                return Err(Error::new(ParseError));
+            }
+            fin.push((first.unwrap(), second.unwrap()));
+        } else {
+            // this is just a single number
+            let num = sp.parse::<u64>();
+            if num.is_err() {
+                return Err(Error::new(ParseError));
+            }
+            fin.push((num.clone().unwrap(), num.clone().unwrap()));
+        }
+    }
+
+    Ok(fin)
 }
 
 impl CpuSetController {
-    /// Contructs a new `CpuSetController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Contructs a new `CpuSetController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf, v2: bool) -> Self {
         Self {
             base: root.clone(),
             path: root,
+            v2: v2,
         }
     }
 
     /// Returns the statistics gathered by the kernel for this control group. See the struct for
     /// more information on what information this entails.
-    pub fn cpuset(self: &Self) -> CpuSet {
+    pub fn cpuset(&self) -> CpuSet {
         CpuSet {
             cpu_exclusive: {
-                self.open_path("cpuset.cpu_exclusive", false).and_then(|file| {
-                    read_u64_from(file)
-                }).map(|x| x == 1).unwrap_or(false)
+                self.open_path("cpuset.cpu_exclusive", false)
+                    .and_then(|file| read_u64_from(file))
+                    .map(|x| x == 1)
+                    .unwrap_or(false)
             },
             cpus: {
-                self.open_path("cpuset.cpus", false).and_then(read_string_from).unwrap_or("".to_string())
+                self.open_path("cpuset.cpus", false)
+                    .and_then(read_string_from)
+                    .and_then(parse_range)
+                    .unwrap_or(Vec::new())
             },
             effective_cpus: {
-                self.open_path("cpuset.effective_cpus", false).and_then(read_string_from).unwrap_or("".to_string())
+                self.open_path("cpuset.effective_cpus", false)
+                    .and_then(read_string_from)
+                    .and_then(parse_range)
+                    .unwrap_or(Vec::new())
             },
             effective_mems: {
-                self.open_path("cpuset.effective_mems", false).and_then(read_string_from).unwrap_or("".to_string())
+                self.open_path("cpuset.effective_mems", false)
+                    .and_then(read_string_from)
+                    .and_then(parse_range)
+                    .unwrap_or(Vec::new())
             },
             mem_exclusive: {
-                self.open_path("cpuset.mem_exclusive", false).and_then(read_u64_from)
-                    .map(|x| x == 1).unwrap_or(false)
+                self.open_path("cpuset.mem_exclusive", false)
+                    .and_then(read_u64_from)
+                    .map(|x| x == 1)
+                    .unwrap_or(false)
             },
             mem_hardwall: {
-                self.open_path("cpuset.mem_hardwall", false).and_then(read_u64_from)
-                    .map(|x| x == 1).unwrap_or(false)
+                self.open_path("cpuset.mem_hardwall", false)
+                    .and_then(read_u64_from)
+                    .map(|x| x == 1)
+                    .unwrap_or(false)
             },
             memory_migrate: {
-                self.open_path("cpuset.memory_migrate", false).and_then(read_u64_from)
-                    .map(|x| x == 1).unwrap_or(false)
+                self.open_path("cpuset.memory_migrate", false)
+                    .and_then(read_u64_from)
+                    .map(|x| x == 1)
+                    .unwrap_or(false)
             },
             memory_pressure: {
-                self.open_path("cpuset.memory_pressure", false).and_then(read_u64_from).unwrap_or(0)
+                self.open_path("cpuset.memory_pressure", false)
+                    .and_then(read_u64_from)
+                    .unwrap_or(0)
             },
             memory_pressure_enabled: {
-                self.open_path("cpuset.memory_pressure_enabled", false).and_then(read_u64_from)
-                    .map(|x| x == 1).ok()
+                self.open_path("cpuset.memory_pressure_enabled", false)
+                    .and_then(read_u64_from)
+                    .map(|x| x == 1)
+                    .ok()
             },
             memory_spread_page: {
-                self.open_path("cpuset.memory_spread_page", false).and_then(read_u64_from)
-                    .map(|x| x == 1).unwrap_or(false)
+                self.open_path("cpuset.memory_spread_page", false)
+                    .and_then(read_u64_from)
+                    .map(|x| x == 1)
+                    .unwrap_or(false)
             },
             memory_spread_slab: {
-                self.open_path("cpuset.memory_spread_slab", false).and_then(read_u64_from)
-                    .map(|x| x == 1).unwrap_or(false)
+                self.open_path("cpuset.memory_spread_slab", false)
+                    .and_then(read_u64_from)
+                    .map(|x| x == 1)
+                    .unwrap_or(false)
             },
             mems: {
-                self.open_path("cpuset.mems", false).and_then(read_string_from).unwrap_or("".to_string())
+                self.open_path("cpuset.mems", false)
+                    .and_then(read_string_from)
+                    .and_then(parse_range)
+                    .unwrap_or(Vec::new())
             },
             sched_load_balance: {
-                self.open_path("cpuset.sched_load_balance", false).and_then(read_u64_from)
-                    .map(|x| x == 1).unwrap_or(false)
+                self.open_path("cpuset.sched_load_balance", false)
+                    .and_then(read_u64_from)
+                    .map(|x| x == 1)
+                    .unwrap_or(false)
             },
             sched_relax_domain_level: {
-                self.open_path("cpuset.sched_relax_domain_level", false).and_then(read_u64_from)
+                self.open_path("cpuset.sched_relax_domain_level", false)
+                    .and_then(read_u64_from)
                     .unwrap_or(0)
             },
         }
@@ -199,44 +342,52 @@ impl CpuSetController {
 
     /// Control whether the CPUs selected via `set_cpus()` should be exclusive to this control
     /// group or not.
-    pub fn set_cpu_exclusive(self: &Self, b: bool) -> Result<(), CgroupError> {
-        self.open_path("cpuset.cpu_exclusive", true).and_then(|mut file| {
-            if b {
-                file.write_all(b"1").map_err(CgroupError::WriteError)
-            } else {
-                file.write_all(b"0").map_err(CgroupError::WriteError)
-            }
-        })
+    pub fn set_cpu_exclusive(&self, b: bool) -> Result<()> {
+        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))
+                } else {
+                    file.write_all(b"0")
+                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                }
+            })
     }
 
     /// Control whether the memory nodes selected via `set_memss()` should be exclusive to this control
     /// group or not.
-    pub fn set_mem_exclusive(self: &Self, b: bool) -> Result<(), CgroupError> {
-        self.open_path("cpuset.mem_exclusive", true).and_then(|mut file| {
-            if b {
-                file.write_all(b"1").map_err(CgroupError::WriteError)
-            } else {
-                file.write_all(b"0").map_err(CgroupError::WriteError)
-            }
-        })
+    pub fn set_mem_exclusive(&self, b: bool) -> Result<()> {
+        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))
+                } else {
+                    file.write_all(b"0")
+                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                }
+            })
     }
 
     /// Set the CPUs that the tasks in this control group can run on.
     ///
     /// Syntax is a comma separated list of CPUs, with an additional extension that ranges can
     /// be represented via dashes.
-    pub fn set_cpus(self: &Self, cpus: &String) -> Result<(), CgroupError> {
+    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(CgroupError::WriteError)
+            file.write_all(cpus.as_ref())
+                .map_err(|e| Error::with_cause(WriteFailed, e))
         })
     }
 
     /// Set the memory nodes that the tasks in this control group can use.
     ///
     /// Syntax is the same as with `set_cpus()`.
-    pub fn set_mems(self: &Self, mems: &String) -> Result<(), CgroupError> {
+    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(CgroupError::WriteError)
+            file.write_all(mems.as_ref())
+                .map_err(|e| Error::with_cause(WriteFailed, e))
         })
     }
 
@@ -245,86 +396,139 @@ impl CpuSetController {
     ///
     /// Note that some kernel allocations, most notably those that are made in interrupt handlers
     /// may disregard this.
-    pub fn set_hardwall(self: &Self, b: bool) -> Result<(), CgroupError> {
-        self.open_path("cpuset.mem_hardwall", true).and_then(|mut file| {
-            if b {
-                file.write_all(b"1").map_err(CgroupError::WriteError)
-            } else {
-                file.write_all(b"0").map_err(CgroupError::WriteError)
-            }
-        })
+    pub fn set_hardwall(&self, b: bool) -> Result<()> {
+        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))
+                } else {
+                    file.write_all(b"0")
+                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                }
+            })
     }
 
     /// Controls whether the kernel should attempt to rebalance the load between the CPUs specified in the
     /// `cpus` field of this control group.
-    pub fn set_load_balancing(self: &Self, b: bool) -> Result<(), CgroupError> {
-        self.open_path("cpuset.sched_load_balance", true).and_then(|mut file| {
-            if b {
-                file.write_all(b"1").map_err(CgroupError::WriteError)
-            } else {
-                file.write_all(b"0").map_err(CgroupError::WriteError)
-            }
-        })
+    pub fn set_load_balancing(&self, b: bool) -> Result<()> {
+        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))
+                } else {
+                    file.write_all(b"0")
+                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                }
+            })
     }
 
     /// Contorl how much effort the kernel should invest in rebalacing the control group.
     ///
     /// See @CpuSet 's similar field for more information.
-    pub fn set_rebalance_relax_domain_level(self: &Self, i: i64) -> Result<(), CgroupError> {
-        self.open_path("cpuset.sched_relax_domain_level", true).and_then(|mut file| {
-            file.write_all(i.to_string().as_ref()).map_err(CgroupError::WriteError)
-        })
+    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))
+            })
     }
 
     /// Control whether when using `set_mems()` the existing memory used by the tasks should be
     /// migrated over to the now-selected nodes.
-    pub fn set_memory_migration(self: &Self, b: bool) -> Result<(), CgroupError> {
-        self.open_path("cpuset.memory_migrate", true).and_then(|mut file| {
-            if b {
-                file.write_all(b"1").map_err(CgroupError::WriteError)
-            } else {
-                file.write_all(b"0").map_err(CgroupError::WriteError)
-            }
-        })
+    pub fn set_memory_migration(&self, b: bool) -> Result<()> {
+        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))
+                } else {
+                    file.write_all(b"0")
+                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                }
+            })
     }
 
     /// Control whether filesystem buffers should be evenly split across the nodes selected via
     /// `set_mems()`.
-    pub fn set_memory_spread_page(self: &Self, b: bool) -> Result<(), CgroupError> {
-        self.open_path("cpuset.memory_spread_page", true).and_then(|mut file| {
-            if b {
-                file.write_all(b"1").map_err(CgroupError::WriteError)
-            } else {
-                file.write_all(b"0").map_err(CgroupError::WriteError)
-            }
-        })
+    pub fn set_memory_spread_page(&self, b: bool) -> Result<()> {
+        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))
+                } else {
+                    file.write_all(b"0")
+                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                }
+            })
     }
 
     /// Control whether the kernel's slab cache for file I/O should be evenly split across the
     /// nodes selected via `set_mems()`.
-    pub fn set_memory_spread_slab(self: &Self, b: bool) -> Result<(), CgroupError> {
-        self.open_path("cpuset.memory_spread_slab", true).and_then(|mut file| {
-            if b {
-                file.write_all(b"1").map_err(CgroupError::WriteError)
-            } else {
-                file.write_all(b"0").map_err(CgroupError::WriteError)
-            }
-        })
+    pub fn set_memory_spread_slab(&self, b: bool) -> Result<()> {
+        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))
+                } else {
+                    file.write_all(b"0")
+                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                }
+            })
     }
 
     /// Control whether the kernel should collect information to calculate memory pressure for
     /// control groups.
     ///
-    /// Note: This is a no-operation if the control group referred by `self` is not the root
+    /// Note: This will fail with `InvalidOperation` if the current congrol group is not the root
     /// control group.
-    pub fn set_enable_memory_pressure(self: &Self, b: bool) -> Result<(), CgroupError> {
-        /* XXX: this file should only be present in the root cpuset cg */
-        self.open_path("cpuset.memory_pressure_enabled", true).and_then(|mut file| {
-            if b {
-                file.write_all(b"1").map_err(CgroupError::WriteError)
-            } else {
-                file.write_all(b"0").map_err(CgroupError::WriteError)
-            }
-        })
+    pub fn set_enable_memory_pressure(&self, b: bool) -> Result<()> {
+        if !self.path_exists("cpuset.memory_pressure_enabled") {
+            return Err(Error::new(InvalidOperation));
+        }
+        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))
+                } else {
+                    file.write_all(b"0")
+                        .map_err(|e| Error::with_cause(WriteFailed, e))
+                }
+            })
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::cpuset;
+    #[test]
+    fn test_parse_range() {
+        let test_cases = vec![
+            "1,2,4-6,9".to_string(),
+            "".to_string(),
+            "1".to_string(),
+            "1-111".to_string(),
+            "1,2,3,4".to_string(),
+            "1-5,6-7,8-9".to_string(),
+        ];
+        let expecteds = vec![
+            vec![(1, 1), (2, 2), (4, 6), (9, 9)],
+            vec![],
+            vec![(1, 1)],
+            vec![(1, 111)],
+            vec![(1, 1), (2, 2), (3, 3), (4, 4)],
+            vec![(1, 5), (6, 7), (8, 9)],
+        ];
+
+        for (i, case) in test_cases.into_iter().enumerate() {
+            let range = cpuset::parse_range(case.clone());
+            println!("{:?} => {:?}", case, range);
+            assert!(range.is_ok());
+            assert_eq!(range.unwrap(), expecteds[i]);
+        }
     }
 }

src/devices.rs

@@ -1,43 +1,169 @@
+// Copyright (c) 2018 Levente Kurusa
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `devices` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/cgroup-v1/devices.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/devices.txt)
-use std::path::PathBuf;
 use std::io::{Read, Write};
+use std::path::PathBuf;
 
-use {CgroupError, DeviceResources, Controllers, Controller, Resources, ControllIdentifier, Subsystem};
+use log::*;
+
+use crate::error::ErrorKind::*;
+use crate::error::*;
+
+use crate::{
+    ControllIdentifier, ControllerInternal, Controllers, DeviceResource, DeviceResources,
+    Resources, Subsystem,
+};
 
 /// A controller that allows controlling the `devices` subsystem of a Cgroup.
 ///
 /// In essence, using the devices controller, it is possible to allow or disallow sets of devices to
 /// be used by the control group's tasks.
 #[derive(Debug, Clone)]
-pub struct DevicesController{
+pub struct DevicesController {
     base: PathBuf,
     path: PathBuf,
 }
 
-impl Controller for DevicesController {
-    fn control_type(self: &Self) -> Controllers { Controllers::Devices }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+/// An enum holding the different types of devices that can be manipulated using this controller.
+#[derive(Debug, Copy, Clone, PartialEq, Eq)]
+pub enum DeviceType {
+    /// The rule applies to all devices.
+    All,
+    /// The rule only applies to character devices.
+    Char,
+    /// The rule only applies to block devices.
+    Block,
+}
 
-    fn apply(self: &Self, res: &Resources) {
-        /* get the resources that apply to this controller */
-        let res: &DeviceResources = &res.devices;
+impl Default for DeviceType {
+    fn default() -> Self {
+        DeviceType::All
+    }
+}
 
-        if res.update_values {
-            for i in &res.devices {
-                let wstr = format!("{} {}:{} {}",
-                                   i.devtype, i.major, i.minor, i.access);
-                if i.allow {
-                    let _ = self.allow_device(&wstr);
-                } else {
-                    let _ = self.deny_device(&wstr);
-                }
+impl DeviceType {
+    /// Convert a DeviceType into the character that the kernel recognizes.
+    pub fn to_char(&self) -> char {
+        match self {
+            DeviceType::All => 'a',
+            DeviceType::Char => 'c',
+            DeviceType::Block => 'b',
+        }
+    }
+
+    /// Convert the kenrel's representation into the DeviceType type.
+    pub fn from_char(c: Option<char>) -> Option<DeviceType> {
+        match c {
+            Some('a') => Some(DeviceType::All),
+            Some('c') => Some(DeviceType::Char),
+            Some('b') => Some(DeviceType::Block),
+            _ => None,
+        }
+    }
+}
+
+/// An enum with the permissions that can be allowed/denied to the control group.
+#[derive(Debug, Copy, Clone, PartialEq, Eq)]
+pub enum DevicePermissions {
+    /// Permission to read from the device.
+    Read,
+    /// Permission to write to the device.
+    Write,
+    /// Permission to execute the `mknod(2)` system call with the device's major and minor numbers.
+    /// That is, the permission to create a special file that refers to the device node.
+    MkNod,
+}
+
+impl DevicePermissions {
+    /// Convert a DevicePermissions into the character that the kernel recognizes.
+    pub fn to_char(&self) -> char {
+        match self {
+            DevicePermissions::Read => 'r',
+            DevicePermissions::Write => 'w',
+            DevicePermissions::MkNod => 'm',
+        }
+    }
+
+    /// Convert a char to a DevicePermission if there is such a mapping.
+    pub fn from_char(c: char) -> Option<DevicePermissions> {
+        match c {
+            'r' => Some(DevicePermissions::Read),
+            'w' => Some(DevicePermissions::Write),
+            'm' => Some(DevicePermissions::MkNod),
+            _ => None,
+        }
+    }
+
+    /// Checks whether the string is a valid descriptor of DevicePermissions.
+    pub fn is_valid(s: &str) -> bool {
+        if s == "" {
+            return false;
+        }
+        for i in s.chars() {
+            if i != 'r' && i != 'w' && i != 'm' {
+                return false;
             }
         }
+        return true;
+    }
+
+    /// Returns a Vec will all the permissions that a device can have.
+    pub fn all() -> Vec<DevicePermissions> {
+        vec![
+            DevicePermissions::Read,
+            DevicePermissions::Write,
+            DevicePermissions::MkNod,
+        ]
+    }
+
+    /// Convert a string into DevicePermissions.
+    pub fn from_str(s: &str) -> Result<Vec<DevicePermissions>> {
+        let mut v = Vec::new();
+        if s == "" {
+            return Ok(v);
+        }
+        for e in s.chars() {
+            let perm = DevicePermissions::from_char(e).ok_or_else(|| Error::new(ParseError))?;
+            v.push(perm);
+        }
+
+        Ok(v)
+    }
+}
+
+impl ControllerInternal for DevicesController {
+    fn control_type(&self) -> Controllers {
+        Controllers::Devices
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
+
+    fn apply(&self, res: &Resources) -> Result<()> {
+        // get the resources that apply to this controller
+        let res: &DeviceResources = &res.devices;
+
+        for i in &res.devices {
+            if i.allow {
+                let _ = self.allow_device(i.devtype, i.major, i.minor, &i.access);
+            } else {
+                let _ = self.deny_device(i.devtype, i.major, i.minor, &i.access);
+            }
+        }
+
+        Ok(())
     }
 }
 
@@ -54,18 +180,17 @@ impl<'a> From<&'a Subsystem> for &'a DevicesController {
                 Subsystem::Devices(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
 impl DevicesController {
-    /// Constructs a new `DevicesController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Constructs a new `DevicesController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf) -> Self {
         Self {
             base: root.clone(),
             path: root,
@@ -74,42 +199,110 @@ impl DevicesController {
 
     /// Allow a (possibly, set of) device(s) to be used by the tasks in the control group.
     ///
-    /// The format of `dev` is rather simple:
-    /// `$type $major:$minor $rwm`
-    /// where `$rwm` is a combination of the characters `r`, `w`, `m`, each standing for read,
-    /// write, mknod permissions.
-    ///
-    /// Note that `dev` can be "regex"-like: both `$major` and `$minor` can be `*` which implies
-    /// that their value does not matter.
-    pub fn allow_device(self: &Self, dev: &String) -> Result<(), CgroupError> {
+    /// When `-1` is passed as `major` or `minor`, the kernel interprets that value as "any",
+    /// meaning that it will match any device.
+    pub fn allow_device(
+        &self,
+        devtype: DeviceType,
+        major: i64,
+        minor: i64,
+        perm: &Vec<DevicePermissions>,
+    ) -> Result<()> {
+        let perms = perm
+            .iter()
+            .map(DevicePermissions::to_char)
+            .collect::<String>();
+        let minor = if minor == -1 {
+            "*".to_string()
+        } else {
+            format!("{}", minor)
+        };
+        let major = if major == -1 {
+            "*".to_string()
+        } else {
+            format!("{}", major)
+        };
+        let final_str = format!("{} {}:{} {}", devtype.to_char(), major, minor, perms);
         self.open_path("devices.allow", true).and_then(|mut file| {
-            file.write_all(dev.as_ref()).map_err(CgroupError::WriteError)
+            file.write_all(final_str.as_ref())
+                .map_err(|e| Error::with_cause(WriteFailed, e))
         })
     }
 
     /// Deny the control group's tasks access to the devices covered by `dev`.
     ///
-    /// The format of `dev` is rather simple:
-    /// `$type $major:$minor $rwm`
-    /// where `$rwm` is a combination of the characters `r`, `w`, `m`, each standing for read,
-    /// write, mknod permissions.
-    ///
-    /// Note that `dev` can be "regex"-like: both `$major` and `$minor` can be `*` which implies
-    /// that their value does not matter.
-    pub fn deny_device(self: &Self, dev: &String) -> Result<(), CgroupError> {
+    /// When `-1` is passed as `major` or `minor`, the kernel interprets that value as "any",
+    /// meaning that it will match any device.
+    pub fn deny_device(
+        &self,
+        devtype: DeviceType,
+        major: i64,
+        minor: i64,
+        perm: &Vec<DevicePermissions>,
+    ) -> Result<()> {
+        let perms = perm
+            .iter()
+            .map(DevicePermissions::to_char)
+            .collect::<String>();
+        let minor = if minor == -1 {
+            "*".to_string()
+        } else {
+            format!("{}", minor)
+        };
+        let major = if major == -1 {
+            "*".to_string()
+        } else {
+            format!("{}", major)
+        };
+        let final_str = format!("{} {}:{} {}", devtype.to_char(), major, minor, perms);
         self.open_path("devices.deny", true).and_then(|mut file| {
-            file.write_all(dev.as_ref()).map_err(CgroupError::WriteError)
+            file.write_all(final_str.as_ref())
+                .map_err(|e| Error::with_cause(WriteFailed, e))
         })
     }
 
     /// Get the current list of allowed devices.
-    pub fn allowed_devices(self: &Self) -> Result<String, CgroupError> {
+    pub fn allowed_devices(&self) -> Result<Vec<DeviceResource>> {
         self.open_path("devices.list", false).and_then(|mut file| {
             let mut s = String::new();
             let res = file.read_to_string(&mut s);
             match res {
-                Ok(_) => Ok(s),
-                Err(e) => Err(CgroupError::ReadError(e)),
+                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().nth(0));
+                            let mut major = ls[1].parse::<i64>();
+                            let mut minor = ls[2].parse::<i64>();
+                            if major.is_err() && ls[1] == "*".to_string() {
+                                major = Ok(-1);
+                            }
+                            if minor.is_err() && ls[2] == "*".to_string() {
+                                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: access,
+                                });
+                                Ok(acc)
+                            }
+                        }
+                    })
+                },
+                Err(e) => Err(Error::with_cause(ReadFailed, e)),
             }
         })
     }

src/error.rs

@@ -0,0 +1,112 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+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)]
+pub enum ErrorKind {
+    FsError,
+    Common(String),
+
+    /// An error occured while writing to a control group file.
+    WriteFailed,
+
+    /// An error occured while trying to read from a control group file.
+    ReadFailed,
+
+    /// An error occured while trying 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.
+    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.
+    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.
+    InvalidPath,
+
+    InvalidBytesSize,
+
+    /// An unknown error has occured.
+    Other,
+}
+
+#[derive(Debug)]
+pub struct Error {
+    kind: ErrorKind,
+    cause: Option<Box<dyn StdError + Send + Sync>>,
+}
+
+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)
+        } else {
+            write!(f, "{}", msg)
+        }
+    }
+}
+
+impl StdError for Error {
+    fn cause(&self) -> Option<&dyn StdError> {
+        match self.cause {
+            Some(ref x) => Some(&**x),
+            None => None,
+        }
+    }
+}
+
+impl Error {
+    pub(crate) fn from_string(s: String) -> Self {
+        Self {
+            kind: ErrorKind::Common(s),
+            cause: None,
+        }
+    }
+    pub(crate) fn new(kind: ErrorKind) -> Self {
+        Self { kind, cause: None }
+    }
+
+    pub(crate) fn with_cause<E>(kind: ErrorKind, cause: E) -> Self
+    where
+        E: 'static + Send + Sync + StdError,
+    {
+        Self {
+            kind,
+            cause: Some(Box::new(cause)),
+        }
+    }
+
+    pub fn kind(&self) -> &ErrorKind {
+        &self.kind
+    }
+}
+
+pub type Result<T> = ::std::result::Result<T, Error>;

src/events.rs

@@ -0,0 +1,90 @@
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use eventfd::{eventfd, EfdFlags};
+use nix::sys::eventfd;
+use std::fs::{self, File};
+use std::io::Read;
+use std::os::unix::io::{AsRawFd, FromRawFd};
+use std::path::{Path, PathBuf};
+use std::sync::mpsc::{self, Receiver};
+use std::thread;
+
+use crate::error::ErrorKind::*;
+use crate::error::*;
+
+// notify_on_oom returns channel on which you can expect event about OOM,
+// if process died without OOM this channel will be closed.
+pub fn notify_on_oom_v2(key: &str, dir: &PathBuf) -> Result<Receiver<String>> {
+    register_memory_event(key, dir, "memory.oom_control", "")
+}
+
+// notify_on_oom returns channel on which you can expect event about OOM,
+// if process died without OOM this channel will be closed.
+pub fn notify_on_oom_v1(key: &str, dir: &PathBuf) -> Result<Receiver<String>> {
+    register_memory_event(key, dir, "memory.oom_control", "")
+}
+
+// level is one of "low", "medium", or "critical"
+pub fn notify_memory_pressure(key: &str, dir: &PathBuf, level: &str) -> Result<Receiver<String>> {
+    if level != "low" && level != "medium" && level != "critical" {
+        return Err(Error::from_string(format!(
+            "invalid pressure level {}",
+            level
+        )));
+    }
+
+    register_memory_event(key, dir, "memory.pressure_level", level)
+}
+
+fn register_memory_event(
+    key: &str,
+    cg_dir: &PathBuf,
+    event_name: &str,
+    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 eventfd =
+        eventfd(0, EfdFlags::EFD_CLOEXEC).map_err(|e| Error::with_cause(ReadFailed, e))?;
+
+    let event_control_path = cg_dir.join("cgroup.event_control");
+    let data;
+    if arg == "" {
+        data = format!("{} {}", eventfd, event_file.as_raw_fd());
+    } else {
+        data = format!("{} {} {}", eventfd, event_file.as_raw_fd(), arg);
+    }
+
+    // write to file and set mode to 0700(FIXME)
+    fs::write(&event_control_path, data).map_err(|e| Error::with_cause(WriteFailed, e))?;
+
+    let mut eventfd_file = unsafe { File::from_raw_fd(eventfd) };
+
+    let (sender, receiver) = mpsc::channel();
+    let key = key.to_string();
+
+    thread::spawn(move || {
+        loop {
+            let mut buf = [0; 8];
+            match eventfd_file.read(&mut buf) {
+                Err(_err) => {
+                    return;
+                }
+                Ok(_) => {}
+            }
+
+            // When a cgroup is destroyed, an event is sent to eventfd.
+            // So if the control path is gone, return instead of notifying.
+            if !Path::new(&event_control_path).exists() {
+                return;
+            }
+            sender.send(key.clone()).unwrap();
+        }
+    });
+
+    Ok(receiver)
+}

src/freezer.rs

@@ -1,11 +1,20 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `freezer` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/cgroup-v1/freezer-subsystem.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/freezer-subsystem.txt)
-use std::path::PathBuf;
 use std::io::{Read, Write};
+use std::path::PathBuf;
 
-use {CgroupError, Controllers, Controller, Resources, ControllIdentifier, Subsystem};
+use crate::error::ErrorKind::*;
+use crate::error::*;
+
+use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
 
 /// A controller that allows controlling the `freezer` subsystem of a Cgroup.
 ///
@@ -16,9 +25,10 @@ use {CgroupError, Controllers, Controller, Resources, ControllIdentifier, Subsys
 /// Note that if the control group is currently in the `Frozen` or `Freezing` state, then no
 /// processes can be added to it.
 #[derive(Debug, Clone)]
-pub struct FreezerController{
+pub struct FreezerController {
     base: PathBuf,
     path: PathBuf,
+    v2: bool,
 }
 
 /// The current state of the control group
@@ -31,13 +41,22 @@ pub enum FreezerState {
     Frozen,
 }
 
-impl Controller for FreezerController {
-    fn control_type(self: &Self) -> Controllers { Controllers::Freezer }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+impl ControllerInternal for FreezerController {
+    fn control_type(&self) -> Controllers {
+        Controllers::Freezer
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
 
-    fn apply(self: &Self, _res: &Resources) {
+    fn apply(&self, _res: &Resources) -> Result<()> {
+        Ok(())
     }
 }
 
@@ -54,51 +73,72 @@ impl<'a> From<&'a Subsystem> for &'a FreezerController {
                 Subsystem::Freezer(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
 impl FreezerController {
-    /// Contructs a new `FreezerController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Contructs a new `FreezerController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf, v2: bool) -> Self {
         Self {
             base: root.clone(),
             path: root,
+            v2: v2,
         }
     }
 
     /// Freezes the processes in the control group.
-    pub fn freeze(self: &Self) -> Result<(), CgroupError> {
-        self.open_path("freezer.state", true).and_then(|mut file| {
-            file.write_all("FROZEN".to_string().as_ref()).map_err(CgroupError::WriteError)
+    pub fn freeze(&self) -> Result<()> {
+        let mut file = "freezer.state";
+        let mut content = "FROZEN".to_string();
+        if self.v2 {
+            file = "cgroup.freeze";
+            content = "1".to_string();
+        }
+
+        self.open_path(file, true).and_then(|mut file| {
+            file.write_all(content.as_ref())
+                .map_err(|e| Error::with_cause(WriteFailed, e))
         })
     }
 
     /// Thaws, that is, unfreezes the processes in the control group.
-    pub fn thaw(self: &Self) -> Result<(), CgroupError> {
-        self.open_path("freezer.state", true).and_then(|mut file| {
-            file.write_all("THAWED".to_string().as_ref()).map_err(CgroupError::WriteError)
+    pub fn thaw(&self) -> Result<()> {
+        let mut file = "freezer.state";
+        let mut content = "THAWED".to_string();
+        if self.v2 {
+            file = "cgroup.freeze";
+            content = "0".to_string();
+        }
+        self.open_path(file, true).and_then(|mut file| {
+            file.write_all(content.as_ref())
+                .map_err(|e| Error::with_cause(WriteFailed, e))
         })
     }
 
     /// Retrieve the state of processes in the control group.
-    pub fn state(self: &Self) -> Result<FreezerState, CgroupError> {
-        self.open_path("freezer.state", false).and_then(|mut file| {
+    pub fn state(&self) -> Result<FreezerState> {
+        let mut file = "freezer.state";
+        if self.v2 {
+            file = "cgroup.freeze";
+        }
+        self.open_path(file, false).and_then(|mut file| {
             let mut s = String::new();
             let res = file.read_to_string(&mut s);
             match res {
                 Ok(_) => match s.as_ref() {
                     "FROZEN" => Ok(FreezerState::Frozen),
                     "THAWED" => Ok(FreezerState::Thawed),
+                    "1" => Ok(FreezerState::Frozen),
+                    "0" => Ok(FreezerState::Thawed),
                     "FREEZING" => Ok(FreezerState::Freezing),
-                    _ => Err(CgroupError::ParseError),
+                    _ => Err(Error::new(ParseError)),
                 },
-                Err(e) => Err(CgroupError::ReadError(e)),
+                Err(e) => Err(Error::with_cause(ReadFailed, e)),
             }
         })
     }

src/hierarchies.rs

@@ -1,137 +1,255 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! 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::io::BufRead;
-use std::io::BufReader;
-use std::fs::File;
-use std::path::{Path, PathBuf};
+use procinfo::pid::{mountinfo_self, Mountinfo};
+use std::fs;
+use std::path::PathBuf;
 
-use {Controllers, Hierarchy, Subsystem};
-use ::pid::PidController;
-use ::memory::MemController;
-use ::cpuset::CpuSetController;
-use ::cpuacct::CpuAcctController;
-use ::cpu::CpuController;
-use ::freezer::FreezerController;
-use ::devices::DevicesController;
-use ::net_cls::NetClsController;
-use ::blkio::BlkIoController;
-use ::perf_event::PerfEventController;
-use ::net_prio::NetPrioController;
-use ::hugetlb::HugeTlbController;
-use ::rdma::RdmaController;
-
-use ::cgroup::Cgroup;
+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::cgroup::Cgroup;
 
 /// The standard, original cgroup implementation. Often referred to as "cgroupv1".
+#[derive(Debug)]
 pub struct V1 {
-    mount_point: String,
+    mountinfo: Vec<Mountinfo>,
+}
+
+#[derive(Debug)]
+pub struct V2 {
+    root: String,
 }
 
 impl Hierarchy for V1 {
-    fn subsystems(self: &Self) -> Vec<Subsystem> {
+    fn v2(&self) -> bool {
+        false
+    }
+
+    fn subsystems(&self) -> Vec<Subsystem> {
         let mut subs = vec![];
-        if self.check_support(Controllers::Pids) {
-            subs.push(Subsystem::Pid(PidController::new(self.root())));
+
+        // 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 self.check_support(Controllers::Mem) {
-            subs.push(Subsystem::Mem(MemController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::Mem) {
+            subs.push(Subsystem::Mem(MemController::new(root, false)));
         }
-        if self.check_support(Controllers::CpuSet) {
-            subs.push(Subsystem::CpuSet(CpuSetController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::Pids) {
+            subs.push(Subsystem::Pid(PidController::new(root, false)));
         }
-        if self.check_support(Controllers::CpuAcct) {
-            subs.push(Subsystem::CpuAcct(CpuAcctController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::CpuSet) {
+            subs.push(Subsystem::CpuSet(CpuSetController::new(root, false)));
         }
-        if self.check_support(Controllers::Cpu) {
-            subs.push(Subsystem::Cpu(CpuController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::CpuAcct) {
+            subs.push(Subsystem::CpuAcct(CpuAcctController::new(root)));
         }
-        if self.check_support(Controllers::Devices) {
-            subs.push(Subsystem::Devices(DevicesController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::Cpu) {
+            subs.push(Subsystem::Cpu(CpuController::new(root, false)));
         }
-        if self.check_support(Controllers::Freezer) {
-            subs.push(Subsystem::Freezer(FreezerController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::Devices) {
+            subs.push(Subsystem::Devices(DevicesController::new(root)));
         }
-        if self.check_support(Controllers::NetCls) {
-            subs.push(Subsystem::NetCls(NetClsController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::Freezer) {
+            subs.push(Subsystem::Freezer(FreezerController::new(root, false)));
         }
-        if self.check_support(Controllers::BlkIo) {
-            subs.push(Subsystem::BlkIo(BlkIoController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::NetCls) {
+            subs.push(Subsystem::NetCls(NetClsController::new(root)));
         }
-        if self.check_support(Controllers::PerfEvent) {
-            subs.push(Subsystem::PerfEvent(PerfEventController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::PerfEvent) {
+            subs.push(Subsystem::PerfEvent(PerfEventController::new(root)));
         }
-        if self.check_support(Controllers::NetPrio) {
-            subs.push(Subsystem::NetPrio(NetPrioController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::NetPrio) {
+            subs.push(Subsystem::NetPrio(NetPrioController::new(root)));
         }
-        if self.check_support(Controllers::HugeTlb) {
-            subs.push(Subsystem::HugeTlb(HugeTlbController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::HugeTlb) {
+            subs.push(Subsystem::HugeTlb(HugeTlbController::new(root, false)));
         }
-        if self.check_support(Controllers::Rdma) {
-            subs.push(Subsystem::Rdma(RdmaController::new(self.root())));
+        if let Some(root) = self.get_mount_point(Controllers::Rdma) {
+            subs.push(Subsystem::Rdma(RdmaController::new(root)));
+        }
+        if let Some(root) = self.get_mount_point(Controllers::Systemd) {
+            subs.push(Subsystem::Systemd(SystemdController::new(root, false)));
         }
 
         subs
     }
 
-    fn root_control_group(self: &Self) -> Cgroup {
-        Cgroup::load(self, "".to_string())
+    fn root_control_group(&self) -> Cgroup {
+        Cgroup::load(auto(), "".to_string())
     }
 
-    fn check_support(self: &Self, sub: Controllers) -> bool {
-        let root = self.root().read_dir().unwrap();
-        for entry in root {
-            if let Ok(entry) = entry {
-                if entry.file_name().into_string().unwrap() == sub.to_string() {
-                    return true;
+    fn root(&self) -> PathBuf {
+        self.mountinfo
+            .iter()
+            .find_map(|m| {
+                if m.fs_type.0 == "cgroup" {
+                    return Some(m.mount_point.parent().unwrap());
                 }
+                None
+            })
+            .unwrap()
+            .to_path_buf()
+    }
+}
+
+impl Hierarchy for V2 {
+    fn v2(&self) -> bool {
+        true
+    }
+
+    fn subsystems(&self) -> Vec<Subsystem> {
+        let p = format!("{}/{}", UNIFIED_MOUNTPOINT, "cgroup.controllers");
+        let ret = fs::read_to_string(p.as_str());
+        if ret.is_err() {
+            return vec![];
+        }
+
+        let mut subs = vec![];
+
+        let controllers = ret.unwrap().trim().to_string();
+        let controller_list: Vec<&str> = controllers.split(' ').collect();
+
+        for s in controller_list {
+            match s {
+                "cpu" => {
+                    subs.push(Subsystem::Cpu(CpuController::new(self.root(), true)));
+                }
+                "io" => {
+                    subs.push(Subsystem::BlkIo(BlkIoController::new(self.root(), true)));
+                }
+                "cpuset" => {
+                    subs.push(Subsystem::CpuSet(CpuSetController::new(self.root(), true)));
+                }
+                "memory" => {
+                    subs.push(Subsystem::Mem(MemController::new(self.root(), true)));
+                }
+                "pids" => {
+                    subs.push(Subsystem::Pid(PidController::new(self.root(), true)));
+                }
+                "freezer" => {
+                    subs.push(Subsystem::Freezer(FreezerController::new(
+                        self.root(),
+                        true,
+                    )));
+                }
+                "hugetlb" => {
+                    subs.push(Subsystem::HugeTlb(HugeTlbController::new(
+                        self.root(),
+                        true,
+                    )));
+                }
+                _ => {}
             }
         }
-        return false;
+
+        subs
     }
 
-    fn root(self: &Self) -> PathBuf {
-        PathBuf::from(self.mount_point.clone())
+    fn root_control_group(&self) -> Cgroup {
+        Cgroup::load(auto(), "".to_string())
+    }
+
+    fn root(&self) -> PathBuf {
+        PathBuf::from(self.root.clone())
     }
 }
 
 impl V1 {
     /// Finds where control groups are mounted to and returns a hierarchy in which control groups
     /// can be created.
-    pub fn new() -> Self {
-        let mount_point = find_v1_mount().unwrap();
+    pub fn new() -> V1 {
         V1 {
-            mount_point: mount_point,
+            mountinfo: mountinfo_self().unwrap(),
+        }
+    }
+
+    pub fn get_mount_point(&self, controller: Controllers) -> Option<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());
+            }
+            None
+        })
+    }
+}
+
+impl V2 {
+    /// Finds where control groups are mounted to and returns a hierarchy in which control groups
+    /// can be created.
+    pub fn new() -> V2 {
+        V2 {
+            root: String::from(UNIFIED_MOUNTPOINT),
         }
     }
 }
 
-fn find_v1_mount() -> Option<String> {
-    /* Open mountinfo so we can get a parseable mount list */
-    let mountinfo_path = Path::new("/proc/self/mountinfo");
+pub const UNIFIED_MOUNTPOINT: &'static str = "/sys/fs/cgroup";
 
-    /* If /proc isn't mounted, or something else happens, then bail out */
-    if mountinfo_path.exists() == false {
-        return None;
+#[cfg(all(target_os = "linux", not(target_env = "musl")))]
+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 mountinfo_file = File::open(mountinfo_path).unwrap();
-    let mountinfo_reader = BufReader::new(&mountinfo_file);
-    for _line in mountinfo_reader.lines() {
-        let line = _line.unwrap();
-        let mut fields = line.split_whitespace();
-        let index = line.find(" - ").unwrap();
-        let mut more_fields = line[index + 3..].split_whitespace().collect::<Vec<_>>();
-        let fstype = more_fields[0];
-        if fstype == "tmpfs" && more_fields[2].contains("ro") {
-            let cgroups_mount = fields.nth(4).unwrap();
-            println!("found cgroups at {:?}", cgroups_mount);
-            return Some(cgroups_mount.to_string());
+    // 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: &'static 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;
         }
     }
 
-    None
+    true
+}
+
+pub fn auto() -> Box<dyn Hierarchy> {
+    if is_cgroup2_unified_mode() {
+        Box::new(V2::new())
+    } else {
+        Box::new(V1::new())
+    }
 }

src/hugetlb.rs

@@ -1,14 +1,23 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `hugetlb` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/cgroup-v1/hugetlb.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/hugetlb.txt)
+use std::io::Write;
 use std::path::PathBuf;
-use std::fs::File;
-use std::io::{Write, Read};
 
-use {CgroupError, HugePageResources, Controllers, Controller, Resources, ControllIdentifier, Subsystem};
-use CgroupError::*;
+use crate::error::ErrorKind::*;
+use crate::error::*;
+use crate::{flat_keyed_to_vec, read_u64_from};
 
+use crate::{
+    ControllIdentifier, ControllerInternal, Controllers, HugePageResources, Resources, Subsystem,
+};
 
 /// A controller that allows controlling the `hugetlb` subsystem of a Cgroup.
 ///
@@ -18,23 +27,40 @@ use CgroupError::*;
 pub struct HugeTlbController {
     base: PathBuf,
     path: PathBuf,
+    sizes: Vec<String>,
+    v2: bool,
 }
 
-impl Controller for HugeTlbController {
-    fn control_type(self: &Self) -> Controllers { Controllers::HugeTlb }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+impl ControllerInternal for HugeTlbController {
+    fn control_type(&self) -> Controllers {
+        Controllers::HugeTlb
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
 
-    fn apply(self: &Self, res: &Resources) {
-        /* get the resources that apply to this controller */
+    fn is_v2(&self) -> bool {
+        self.v2
+    }
+
+    fn apply(&self, res: &Resources) -> Result<()> {
+        // get the resources that apply to this controller
         let res: &HugePageResources = &res.hugepages;
 
-        if res.update_values {
-            for i in &res.limits {
-                let _ = self.set_limit_in_bytes(&i.size, i.limit);
+        for i in &res.limits {
+            let _ = self.set_limit_in_bytes(&i.size, i.limit);
+            if self.limit_in_bytes(&i.size)? != i.limit {
+                return Err(Error::new(Other));
             }
         }
+
+        Ok(())
     }
 }
 
@@ -51,71 +77,224 @@ impl<'a> From<&'a Subsystem> for &'a HugeTlbController {
                 Subsystem::HugeTlb(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
-fn read_u64_from(mut file: File) -> Result<u64, CgroupError> {
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => string.trim().parse().map_err(|_| ParseError),
-        Err(e) => Err(CgroupError::ReadError(e)),
-    }
-}
-
 impl HugeTlbController {
-    /// Constructs a new `HugeTlbController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Constructs a new `HugeTlbController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf, v2: bool) -> Self {
+        let sizes = get_hugepage_sizes().unwrap();
         Self {
             base: root.clone(),
             path: root,
+            sizes: sizes,
+            v2: v2,
         }
     }
 
     /// Whether the system supports `hugetlb_size` hugepages.
-    pub fn size_supported(self: &Self, _hugetlb_size: String) -> bool {
-        /* TODO */
-        true
+    pub fn size_supported(&self, hugetlb_size: &str) -> bool {
+        for s in &self.sizes {
+            if s == hugetlb_size {
+                return true;
+            }
+        }
+        false
+    }
+
+    pub fn get_sizes(&self) -> Vec<String> {
+        self.sizes.clone()
+    }
+
+    fn failcnt_v2(&self, hugetlb_size: &str) -> Result<u64> {
+        self.open_path(&format!("hugetlb.{}.events", hugetlb_size), false)
+            .and_then(flat_keyed_to_vec)
+            .and_then(|x| {
+                if x.len() == 0 {
+                    return Err(Error::from_string(format!(
+                        "get empty from hugetlb.{}.events",
+                        hugetlb_size
+                    )));
+                }
+                Ok(x[0].1 as u64)
+            })
     }
 
     /// Check how many times has the limit of `hugetlb_size` hugepages been hit.
-    pub fn failcnt(self: &Self, hugetlb_size: &String) -> Result<u64, CgroupError> {
+    pub fn failcnt(&self, hugetlb_size: &str) -> Result<u64> {
+        if self.v2 {
+            return self.failcnt_v2(hugetlb_size);
+        }
         self.open_path(&format!("hugetlb.{}.failcnt", hugetlb_size), false)
             .and_then(read_u64_from)
     }
 
     /// 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: &Self, hugetlb_size: &String) -> Result<u64, CgroupError> {
+    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)
     }
 
     /// Get the current usage of memory that is backed by hugepages of a certain size
     /// (`hugetlb_size`).
-    pub fn usage_in_bytes(self: &Self, hugetlb_size: &String) -> Result<u64, CgroupError> {
-        self.open_path(&format!("hugetlb.{}.usage_in_bytes", hugetlb_size), false)
-            .and_then(read_u64_from)
+    pub fn usage_in_bytes(&self, hugetlb_size: &str) -> Result<u64> {
+        let mut file = format!("hugetlb.{}.usage_in_bytes", hugetlb_size);
+        if self.v2 {
+            file = format!("hugetlb.{}.current", hugetlb_size);
+        }
+        self.open_path(&file, false).and_then(read_u64_from)
     }
 
     /// Get the maximum observed usage of memory that is backed by hugepages of a certain size
     /// (`hugetlb_size`).
-    pub fn max_usage_in_bytes(self: &Self, hugetlb_size: &String) -> Result<u64, CgroupError> {
-        self.open_path(&format!("hugetlb.{}.max_usage_in_bytes", hugetlb_size), false)
-            .and_then(read_u64_from)
+    pub fn max_usage_in_bytes(&self, hugetlb_size: &str) -> Result<u64> {
+        self.open_path(
+            &format!("hugetlb.{}.max_usage_in_bytes", hugetlb_size),
+            false,
+        )
+        .and_then(read_u64_from)
     }
 
     /// 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: &Self, hugetlb_size: &String, limit: u64) -> Result<(), CgroupError> {
-        self.open_path(&format!("hugetlb.{}.limit_in_bytes", hugetlb_size), false)
-            .and_then(|mut file| {
-                file.write_all(limit.to_string().as_ref()).map_err(CgroupError::WriteError)
-            })
+    pub fn set_limit_in_bytes(&self, hugetlb_size: &str, limit: u64) -> Result<()> {
+        let mut file = format!("hugetlb.{}.limit_in_bytes", hugetlb_size);
+        if self.v2 {
+            file = 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))
+        })
     }
 }
+
+pub const HUGEPAGESIZE_DIR: &'static str = "/sys/kernel/mm/hugepages";
+use regex::Regex;
+use std::collections::HashMap;
+use std::fs;
+
+fn get_hugepage_sizes() -> Result<Vec<String>> {
+    let mut m = Vec::new();
+    let dirs = fs::read_dir(HUGEPAGESIZE_DIR);
+    if dirs.is_err() {
+        return Ok(m);
+    }
+
+    for e in dirs.unwrap() {
+        let entry = e.unwrap();
+        let name = entry.file_name().into_string().unwrap();
+        let parts: Vec<&str> = name.split('-').collect();
+        if parts.len() != 2 {
+            continue;
+        }
+        let bmap = get_binary_size_map();
+        let size = parse_size(parts[1], &bmap)?;
+        let dabbrs = get_decimal_abbrs();
+        m.push(custom_size(size as f64, 1024.0, &dabbrs));
+    }
+
+    Ok(m)
+}
+
+pub const KB: u128 = 1000;
+pub const MB: u128 = 1000 * KB;
+pub const GB: u128 = 1000 * MB;
+pub const TB: u128 = 1000 * GB;
+pub const PB: u128 = 1000 * TB;
+
+#[allow(non_upper_case_globals)]
+pub const KiB: u128 = 1024;
+#[allow(non_upper_case_globals)]
+pub const MiB: u128 = 1024 * KiB;
+#[allow(non_upper_case_globals)]
+pub const GiB: u128 = 1024 * MiB;
+#[allow(non_upper_case_globals)]
+pub const TiB: u128 = 1024 * GiB;
+#[allow(non_upper_case_globals)]
+pub const PiB: u128 = 1024 * TiB;
+
+pub fn get_binary_size_map() -> HashMap<String, u128> {
+    let mut m = HashMap::new();
+    m.insert("k".to_string(), KiB);
+    m.insert("m".to_string(), MiB);
+    m.insert("g".to_string(), GiB);
+    m.insert("t".to_string(), TiB);
+    m.insert("p".to_string(), PiB);
+    m
+}
+
+pub fn get_decimal_size_map() -> HashMap<String, u128> {
+    let mut m = HashMap::new();
+    m.insert("k".to_string(), KB);
+    m.insert("m".to_string(), MB);
+    m.insert("g".to_string(), GB);
+    m.insert("t".to_string(), TB);
+    m.insert("p".to_string(), PB);
+    m
+}
+
+pub fn get_decimal_abbrs() -> Vec<String> {
+    let m = vec![
+        "B".to_string(),
+        "KB".to_string(),
+        "MB".to_string(),
+        "GB".to_string(),
+        "TB".to_string(),
+        "PB".to_string(),
+        "EB".to_string(),
+        "ZB".to_string(),
+        "YB".to_string(),
+    ];
+    m
+}
+
+fn parse_size(s: &str, m: &HashMap<String, u128>) -> Result<u128> {
+    let re = Regex::new(r"(?P<num>\d+)(?P<mul>[kKmMgGtTpP]?)[bB]?$");
+
+    if re.is_err() {
+        return Err(Error::new(InvalidBytesSize));
+    }
+    let caps = re.unwrap().captures(s).unwrap();
+
+    let num = caps.name("num");
+    let size: u128 = if num.is_some() {
+        let n = num.unwrap().as_str().trim().parse::<u128>();
+        if n.is_err() {
+            return Err(Error::new(InvalidBytesSize));
+        }
+        n.unwrap()
+    } else {
+        return Err(Error::new(InvalidBytesSize));
+    };
+
+    let q = caps.name("mul");
+    let mul: u128 = if q.is_some() {
+        let t = m.get(q.unwrap().as_str());
+        if t.is_some() {
+            *t.unwrap()
+        } else {
+            return Err(Error::new(InvalidBytesSize));
+        }
+    } else {
+        return Err(Error::new(InvalidBytesSize));
+    };
+
+    Ok(size * mul)
+}
+
+fn custom_size(mut size: f64, base: f64, m: &Vec<String>) -> String {
+    let mut i = 0;
+    while size >= base && i < m.len() - 1 {
+        size /= base;
+        i += 1;
+    }
+
+    format!("{}{}", size, m[i].as_str())
+}

src/lib.rs

@@ -1,41 +1,78 @@
-use std::path::PathBuf;
-use std::fs::File;
-use std::io::{BufRead, BufReader, Write};
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+use log::*;
+
+use std::collections::HashMap;
+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 hierarchies;
-pub mod pid;
-pub mod memory;
-pub mod cpuset;
-pub mod cpuacct;
-pub mod cpu;
-pub mod devices;
-pub mod cgroup;
-pub mod freezer;
-pub mod net_cls;
 pub mod blkio;
-pub mod perf_event;
-pub mod net_prio;
+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 pid::PidController;
-use memory::MemController;
-use cpuset::CpuSetController;
-use cpuacct::CpuAcctController;
-use cpu::CpuController;
-use freezer::FreezerController;
-use devices::DevicesController;
-use net_cls::NetClsController;
-use blkio::BlkIoController;
-use perf_event::PerfEventController;
-use net_prio::NetPrioController;
-use hugetlb::HugeTlbController;
-use rdma::RdmaController;
+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;
 
-pub use cgroup::Cgroup;
+#[doc(inline)]
+pub use crate::cgroup::Cgroup;
 
 /// Contains all the subsystems that are available in this crate.
-#[derive(Debug)]
+#[derive(Debug, Clone)]
 pub enum Subsystem {
     /// Controller for the `Pid` subsystem, see `PidController` for more information.
     Pid(PidController),
@@ -63,33 +100,12 @@ pub enum Subsystem {
     HugeTlb(HugeTlbController),
     /// Controller for the `Rdma` subsystem, see `RdmaController` for more information.
     Rdma(RdmaController),
-}
-
-/// The different types of errors that can occur while manipulating control groups.
-#[derive(Debug)]
-pub enum CgroupError {
-    /// An error occured while writing to a control group file.
-    WriteError(std::io::Error),
-    /// An error occured while trying to read from a control group file.
-    ReadError(std::io::Error),
-    /// 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.
-    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.
-    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.
-    InvalidPath,
+    /// Controller for the `Systemd` subsystem, see `SystemdController` for more information.
+    Systemd(SystemdController),
 }
 
 #[doc(hidden)]
-#[derive(Eq, PartialEq, Debug)]
+#[derive(Eq, PartialEq, Debug, Clone)]
 pub enum Controllers {
     Pids,
     Mem,
@@ -104,10 +120,11 @@ pub enum Controllers {
     NetPrio,
     HugeTlb,
     Rdma,
+    Systemd,
 }
 
 impl Controllers {
-    pub fn to_string(self: &Self) -> String {
+    pub fn to_string(&self) -> String {
         match self {
             Controllers::Pids => return "pids".to_string(),
             Controllers::Mem => return "memory".to_string(),
@@ -122,97 +139,243 @@ impl Controllers {
             Controllers::NetPrio => return "net_prio".to_string(),
             Controllers::HugeTlb => return "hugetlb".to_string(),
             Controllers::Rdma => return "rdma".to_string(),
+            Controllers::Systemd => return "name=systemd".to_string(),
         }
     }
 }
 
+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) => return Err(Error::with_cause(ErrorKind::WriteFailed, e)),
+                    Ok(file) => return Ok(file),
+                }
+            } else {
+                match File::open(&path) {
+                    Err(e) => return Err(Error::with_cause(ErrorKind::ReadFailed, e)),
+                    Ok(file) => return 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 let Err(_) = self.verify_path() {
+                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: &Self, res: &Resources);
+    fn apply(&self, res: &Resources) -> Result<()>;
 
-    /* meta stuff */
-    #[doc(hidden)]
-    fn control_type(self: &Self) -> Controllers;
-    #[doc(hidden)]
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf;
-    #[doc(hidden)]
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf;
-    #[doc(hidden)]
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf;
+    /// Create this controller
+    fn create(&self);
 
-    #[doc(hidden)]
-    fn verify_path(self: &Self) -> bool {
-        self.get_path().starts_with(self.get_base())
+    /// 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) {
-        if self.verify_path() {
-            match ::std::fs::create_dir(self.get_path()) {
-                Ok(_) => (),
-                Err(e) => println!("error create_dir {:?}", e),
-            }
+    fn create(&self) {
+        self.verify_path()
+            .expect(format!("path should be valid: {:?}", self.path()).as_str());
+
+        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: &Self) -> bool {
+    fn exists(&self) -> bool {
         self.get_path().exists()
     }
 
     /// Delete the controller.
-    fn delete(self: &Self) {
-        if self.get_path().exists() {
-            let _ = ::std::fs::remove_dir(self.get_path());
-        }
-    }
-
-    #[doc(hidden)]
-    fn open_path(self: &Self, p: &str, w: bool) -> Result<File, CgroupError> {
-        let mut path = self.get_path().clone();
-        path.push(p);
-
-        if !self.verify_path() {
-            return Err(CgroupError::InvalidPath);
+    fn delete(&self) -> Result<()> {
+        if !self.get_path().exists() {
+            return Ok(());
         }
 
-        if w {
-            match File::create(&path) {
-                Err(e) => return Err(CgroupError::WriteError(e)),
-                Ok(file) => return Ok(file),
-            }
-        } else {
-            match File::open(&path) {
-                Err(e) => return Err(CgroupError::ReadError(e)),
-                Ok(file) => return Ok(file),
-            }
-        }
+        fs::remove_dir(self.get_path()).map_err(|e| Error::with_cause(ErrorKind::RemoveFailed, e))
     }
 
     /// Attach a task to this controller.
-    fn add_task(self: &Self, pid: &CgroupPid) -> Result<(), CgroupError> {
-        self.open_path("tasks", true).and_then(|mut file| {
-            file.write_all(pid.pid.to_string().as_ref()).map_err(CgroupError::WriteError)
+    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: &Self) -> Vec<CgroupPid> {
-        self.open_path("tasks", false).and_then(|file| {
-            let bf = BufReader::new(file);
-            let mut v = Vec::new();
-            for line in bf.lines() {
-                if let Ok(line) = line {
-                    let n = line.trim().parse().unwrap_or(0u64);
-                    v.push(n);
+    fn tasks(&self) -> Vec<CgroupPid> {
+        let mut file = "tasks";
+        if self.is_v2() {
+            file = "cgroup.procs";
+        }
+        self.open_path(file, false)
+            .and_then(|file| {
+                let bf = BufReader::new(file);
+                let mut v = Vec::new();
+                for line in bf.lines() {
+                    if let Ok(line) = line {
+                        let n = line.trim().parse().unwrap_or(0u64);
+                        v.push(n);
+                    }
                 }
-            }
-            Ok(v.into_iter().map(CgroupPid::from).collect())
-        }).unwrap_or(vec![])
+                Ok(v.into_iter().map(CgroupPid::from).collect())
+            })
+            .unwrap_or(vec![])
+    }
+
+    fn v2(&self) -> bool {
+        self.is_v2()
     }
 }
 
@@ -223,84 +386,90 @@ pub trait ControllIdentifier {
 
 /// Control group hierarchy (right now, only V1 is supported, but in the future Unified will be
 /// implemented as well).
-pub trait Hierarchy {
+pub trait Hierarchy: std::fmt::Debug + Send + Sync {
     /// Returns what subsystems are supported by the hierarchy.
-    fn subsystems(self: &Self) -> Vec<Subsystem>;
+    fn subsystems(&self) -> Vec<Subsystem>;
 
     /// Returns the root directory of the hierarchy.
-    fn root(self: &Self) -> PathBuf;
+    fn root(&self) -> PathBuf;
 
     /// Return a handle to the root control group in the hierarchy.
-    fn root_control_group(self: &Self) -> Cgroup;
+    fn root_control_group(&self) -> Cgroup;
 
-    /// Checks whether a certain subsystem is supported in the hierarchy.
-    ///
-    /// This is an internal function and should not be used.
-    #[doc(hidden)]
-    fn check_support(self: &Self, sub: Controllers) -> bool;
+    fn v2(&self) -> bool;
 }
 
 /// Resource limits for the memory subsystem.
 #[derive(Debug, Clone, Eq, PartialEq, Default)]
 pub struct MemoryResources {
-    /// Whether values should be applied to the controller.
-    pub update_values: bool,
     /// How much memory (in bytes) can the kernel consume.
-    pub kernel_memory_limit: u64,
+    pub kernel_memory_limit: Option<i64>,
     /// Upper limit of memory usage of the control group's tasks.
-    pub memory_hard_limit: u64,
+    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: u64,
+    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: u64,
+    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: u64,
+    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: u64,
+    pub swappiness: Option<u64>,
+    /// Customized key-value attributes
+    ///
+    /// # Usage:
+    /// ```
+    /// let resource = &mut cgroups_rs::Resources::default();
+    /// resource.memory.attrs.insert("memory.numa_balancing", "true".to_string());
+    /// // apply here
+    pub attrs: std::collections::HashMap<&'static str, String>,
 }
 
 /// Resources limits on the number of processes.
 #[derive(Debug, Clone, Eq, PartialEq, Default)]
 pub struct PidResources {
-    /// Whether values should be applied to the controller.
-    pub update_values: bool,
     /// 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: pid::PidMax,
+    pub maximum_number_of_processes: Option<MaxValue>,
 }
 
 /// Resources limits about how the tasks can use the CPU.
 #[derive(Debug, Clone, Eq, PartialEq, Default)]
 pub struct CpuResources {
-    /// Whether values should be applied to the controller.
-    pub update_values: bool,
-    /* cpuset */
+    // cpuset
     /// A comma-separated list of CPU IDs where the task in the control group can run. Dashes
     /// between numbers indicate ranges.
-    pub cpus: String,
+    pub cpus: Option<String>,
     /// Same syntax as the `cpus` field of this structure, but applies to memory nodes instead of
     /// processors.
-    pub mems: String,
-    /* cpu */
+    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: u64,
+    pub shares: Option<u64>,
     /// In one `period`, how much can the tasks run in nanoseconds.
-    pub quota: i64,
+    pub quota: Option<i64>,
     /// Period of time in nanoseconds.
-    pub period: u64,
+    pub period: Option<u64>,
     /// This is currently a no-operation.
-    pub realtime_runtime: i64,
+    pub realtime_runtime: Option<i64>,
     /// This is currently a no-operation.
-    pub realtime_period: u64,
+    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", "10".to_string());
+    /// // apply here
+    /// ```
+    pub attrs: std::collections::HashMap<&'static str, String>,
 }
 
 /// A device resource that can be allowed or denied access to.
@@ -309,20 +478,18 @@ 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: String,
+    pub devtype: crate::devices::DeviceType,
     /// The major number of the device.
-    pub major: u64,
+    pub major: i64,
     /// The minor number of the device.
-    pub minor: u64,
+    pub minor: i64,
     /// Sequence of `'r'`, `'w'` or `'m'`, each denoting read, write or mknod permissions.
-    pub access: String,
+    pub access: Vec<crate::devices::DevicePermissions>,
 }
 
 /// Limit the usage of devices for the control group's tasks.
 #[derive(Debug, Clone, Eq, PartialEq, Default)]
 pub struct DeviceResources {
-    /// Whether values should be applied to the controller.
-    pub update_values: bool,
     /// For each device in the list, the limits in the structure are applied.
     pub devices: Vec<DeviceResource>,
 }
@@ -340,12 +507,10 @@ pub struct NetworkPriority {
 /// control group.
 #[derive(Debug, Clone, Eq, PartialEq, Default)]
 pub struct NetworkResources {
-    /// Whether values should be applied to the controller.
-    pub update_values: bool,
     /// 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: u64,
+    pub class_id: Option<u64>,
     /// Priority of the egress traffic for each interface.
     pub priorities: Vec<NetworkPriority>,
 }
@@ -363,8 +528,6 @@ pub struct HugePageResource {
 /// Provides the ability to set consumption limit on each type of hugepages.
 #[derive(Debug, Clone, Eq, PartialEq, Default)]
 pub struct HugePageResources {
-    /// Whether values should be applied to the controller.
-    pub update_values: bool,
     /// Set a limit of consumption for each hugepages type.
     pub limits: Vec<HugePageResource>,
 }
@@ -377,9 +540,9 @@ pub struct BlkIoDeviceResource {
     /// The minor number of the device.
     pub minor: u64,
     /// The weight of the device against the descendant nodes.
-    pub weight: u16,
+    pub weight: Option<u16>,
     /// The weight of the device against the sibling nodes.
-    pub leaf_weight: u16,
+    pub leaf_weight: Option<u16>,
 }
 
 /// Provides the ability to throttle a device (both byte/sec, and IO op/s)
@@ -396,12 +559,10 @@ pub struct BlkIoDeviceThrottleResource {
 /// General block I/O resource limits.
 #[derive(Debug, Clone, Eq, PartialEq, Default)]
 pub struct BlkIoResources {
-    /// Whether values should be applied to the controller.
-    pub update_values: bool,
     /// The weight of the control group against descendant nodes.
-    pub weight: u16,
+    pub weight: Option<u16>,
     /// The weight of the control group against sibling nodes.
-    pub leaf_weight: u16,
+    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.
@@ -443,23 +604,18 @@ pub struct CgroupPid {
 
 impl From<u64> for CgroupPid {
     fn from(u: u64) -> CgroupPid {
-        CgroupPid {
-            pid: u,
-        }
+        CgroupPid { pid: u }
     }
 }
 
 impl<'a> From<&'a std::process::Child> for CgroupPid {
     fn from(u: &std::process::Child) -> CgroupPid {
-        CgroupPid {
-            pid: u.id() as u64,
-        }
+        CgroupPid { pid: u.id() as u64 }
     }
 }
 
-
 impl Subsystem {
-    fn enter(self: Self, path: &String) -> Self {
+    fn enter(self, path: &Path) -> Self {
         match self {
             Subsystem::Pid(cont) => Subsystem::Pid({
                 let mut c = cont.clone();
@@ -526,10 +682,15 @@ impl Subsystem {
                 c.get_path_mut().push(path);
                 c
             }),
+            Subsystem::Systemd(cont) => Subsystem::Systemd({
+                let mut c = cont.clone();
+                c.get_path_mut().push(path);
+                c
+            }),
         }
     }
 
-    fn to_controller(self: &Self) -> &dyn Controller {
+    pub fn to_controller(&self) -> &dyn Controller {
         match self {
             Subsystem::Pid(cont) => cont,
             Subsystem::Mem(cont) => cont,
@@ -544,6 +705,163 @@ impl Subsystem {
             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)]
+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 {
+    fn to_i64(&self) -> i64 {
+        match self {
+            MaxValue::Max => -1,
+            MaxValue::Value(num) => *num,
+        }
+    }
+
+    fn to_string(&self) -> String {
+        match self {
+            MaxValue::Max => "max".to_string(),
+            MaxValue::Value(num) => num.to_string(),
         }
     }
 }
+
+pub fn parse_max_value(s: &String) -> 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 {
+            match parts[1].parse::<i64>() {
+                Ok(i) => {
+                    v.push((parts[0].to_string(), i));
+                }
+                Err(_) => {}
+            }
+        }
+    }
+    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 {
+            match parts[1].parse::<i64>() {
+                Ok(i) => {
+                    h.insert(parts[0].to_string(), i);
+                }
+                Err(_) => {}
+            }
+        }
+    }
+    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.len() == 0 {
+            continue;
+        }
+        let mut th = HashMap::new();
+        for item in parts[1..].into_iter() {
+            let fields: Vec<&str> = item.split('=').collect();
+            if fields.len() == 2 {
+                match fields[1].parse::<i64>() {
+                    Ok(i) => {
+                        th.insert(fields[0].to_string(), i);
+                    }
+                    Err(_) => {}
+                }
+            }
+        }
+        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/net_cls.rs

@@ -1,13 +1,22 @@
+// Copyright (c) 2018 Levente Kurusa
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `net_cls` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/cgroup-v1/net_cls.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/net_cls.txt)
+use std::io::Write;
 use std::path::PathBuf;
-use std::io::{Read, Write};
-use std::fs::File;
 
-use {CgroupError, NetworkResources, Controllers, Controller, Resources, ControllIdentifier, Subsystem};
-use CgroupError::*;
+use crate::error::ErrorKind::*;
+use crate::error::*;
+
+use crate::read_u64_from;
+use crate::{
+    ControllIdentifier, ControllerInternal, Controllers, NetworkResources, Resources, Subsystem,
+};
 
 /// A controller that allows controlling the `net_cls` subsystem of a Cgroup.
 ///
@@ -20,19 +29,27 @@ pub struct NetClsController {
     path: PathBuf,
 }
 
-impl Controller for NetClsController {
-    fn control_type(self: &Self) -> Controllers { Controllers::NetCls }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+impl ControllerInternal for NetClsController {
+    fn control_type(&self) -> Controllers {
+        Controllers::NetCls
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
 
-    fn apply(self: &Self, res: &Resources) {
-        /* get the resources that apply to this controller */
+    fn apply(&self, res: &Resources) -> Result<()> {
+        // get the resources that apply to this controller
         let res: &NetworkResources = &res.network;
 
-        if res.update_values {
-            let _ = self.set_class(res.class_id);
-        }
+        update_and_test!(self, set_class, res.class_id, get_class);
+
+        return Ok(());
     }
 }
 
@@ -49,44 +66,36 @@ impl<'a> From<&'a Subsystem> for &'a NetClsController {
                 Subsystem::NetCls(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
-fn read_u64_from(mut file: File) -> Result<u64, CgroupError> {
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => string.trim().parse().map_err(|_| ParseError),
-        Err(e) => Err(CgroupError::ReadError(e)),
-    }
-}
-
 impl NetClsController {
-    /// Constructs a new `NetClsController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Constructs a new `NetClsController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf) -> Self {
         Self {
             base: root.clone(),
             path: root,
         }
     }
-    
+
     /// Set the network class id of the outgoing packets of the control group's tasks.
-    pub fn set_class(self: &Self, class: u64) -> Result<(), CgroupError> {
-        self.open_path("net_cls.classid", true).and_then(|mut file| {
-            let s = format!("{:#08X}", class);
-            file.write_all(s.as_ref()).map_err(CgroupError::WriteError)
-        })
+    pub fn set_class(&self, class: u64) -> Result<()> {
+        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))
+            })
     }
 
     /// Get the network class id of the outgoing packets of the control group's tasks.
-    pub fn get_class(self: &Self) -> Result<u64, CgroupError> {
-        self.open_path("net_cls.classid", false).and_then(|file| {
-            read_u64_from(file)
-        })
+    pub fn get_class(&self) -> Result<u64> {
+        self.open_path("net_cls.classid", false)
+            .and_then(|file| read_u64_from(file))
     }
 }

src/net_prio.rs

@@ -1,14 +1,23 @@
+// Copyright (c) 2018 Levente Kurusa
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `net_prio` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/cgroup-v1/net_prio.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/net_prio.txt)
-use std::path::PathBuf;
-use std::io::{BufReader, BufRead, Write, Read};
-use std::fs::File;
 use std::collections::HashMap;
+use std::io::{BufRead, BufReader, Write};
+use std::path::PathBuf;
 
-use {CgroupError, NetworkResources, Controllers, Controller, Resources, ControllIdentifier, Subsystem};
-use CgroupError::*;
+use crate::error::ErrorKind::*;
+use crate::error::*;
+
+use crate::read_u64_from;
+use crate::{
+    ControllIdentifier, ControllerInternal, Controllers, NetworkResources, Resources, Subsystem,
+};
 
 /// A controller that allows controlling the `net_prio` subsystem of a Cgroup.
 ///
@@ -21,21 +30,29 @@ pub struct NetPrioController {
     path: PathBuf,
 }
 
-impl Controller for NetPrioController {
-    fn control_type(self: &Self) -> Controllers { Controllers::NetPrio }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+impl ControllerInternal for NetPrioController {
+    fn control_type(&self) -> Controllers {
+        Controllers::NetPrio
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
 
-    fn apply(self: &Self, res: &Resources) {
-        /* get the resources that apply to this controller */
+    fn apply(&self, res: &Resources) -> Result<()> {
+        // get the resources that apply to this controller
         let res: &NetworkResources = &res.network;
 
-        if res.update_values {
-            for i in &res.priorities {
-                let _ = self.set_if_prio(&i.name, i.priority);
-            }
+        for i in &res.priorities {
+            let _ = self.set_if_prio(&i.name, i.priority);
         }
+
+        Ok(())
     }
 }
 
@@ -52,26 +69,17 @@ impl<'a> From<&'a Subsystem> for &'a NetPrioController {
                 Subsystem::NetPrio(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
-fn read_u64_from(mut file: File) -> Result<u64, CgroupError> {
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => string.trim().parse().map_err(|_| ParseError),
-        Err(e) => Err(CgroupError::ReadError(e)),
-    }
-}
-
 impl NetPrioController {
-    /// Constructs a new `NetPrioController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Constructs a new `NetPrioController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf) -> Self {
         Self {
             base: root.clone(),
             path: root,
@@ -79,46 +87,50 @@ impl NetPrioController {
     }
 
     /// Retrieves the current priority of the emitted packets.
-    pub fn prio_idx(self: &Self) -> u64 {
+    pub fn prio_idx(&self) -> u64 {
         self.open_path("net_prio.prioidx", false)
             .and_then(read_u64_from)
             .unwrap_or(0)
     }
 
     /// A map of priorities for each network interface.
-    pub fn ifpriomap(self: &Self) -> Result<HashMap<String, u64>, CgroupError> {
-        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();
-                    let ifname = sp.nth(0);
-                    let ifprio = sp.nth(1);
-                    if ifname.is_none() || ifprio.is_none() {
-                        Err(CgroupError::ParseError)
+    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 ifname = ifname.unwrap();
-                        let ifprio = ifprio.unwrap().trim().parse();
-                        if ifprio.is_err() {
-                            Err(CgroupError::ParseError)
+                        let mut acc = acc.unwrap();
+                        let l = line.unwrap();
+                        let mut sp = l.split_whitespace();
+                        let ifname = sp.nth(0);
+                        let ifprio = sp.nth(1);
+                        if ifname.is_none() || ifprio.is_none() {
+                            Err(Error::new(ParseError))
                         } else {
-                            acc.insert(ifname.to_string(), ifprio.unwrap());
-                            Ok(acc)
+                            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)
+                                }
+                            }
                         }
                     }
-                }
+                })
             })
-        })
     }
 
     /// Set the priority of the network traffic on `eif` to be `prio`.
-    pub fn set_if_prio(self: &Self, eif: &String, prio: u64) -> Result<(), CgroupError> {
-        self.open_path("net_prio.ifpriomap", true).and_then(|mut file| {
-            file.write_all(format!("{} {}", eif, prio).as_ref()).map_err(CgroupError::WriteError)
-        })
+    pub fn set_if_prio(&self, eif: &str, prio: u64) -> Result<()> {
+        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))
+            })
     }
 }

src/perf_event.rs

@@ -1,10 +1,17 @@
+// Copyright (c) 2018 Levente Kurusa
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `perf_event` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [tools/perf/Documentation/perf-record.txt](https://raw.githubusercontent.com/torvalds/linux/master/tools/perf/Documentation/perf-record.txt)
 use std::path::PathBuf;
 
-use {Controllers, Controller, Resources, ControllIdentifier, Subsystem};
+use crate::error::*;
+
+use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
 
 /// A controller that allows controlling the `perf_event` subsystem of a Cgroup.
 ///
@@ -16,13 +23,22 @@ pub struct PerfEventController {
     path: PathBuf,
 }
 
-impl Controller for PerfEventController {
-    fn control_type(self: &Self) -> Controllers { Controllers::PerfEvent }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+impl ControllerInternal for PerfEventController {
+    fn control_type(&self) -> Controllers {
+        Controllers::PerfEvent
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
 
-    fn apply(self: &Self, _res: &Resources) {
+    fn apply(&self, _res: &Resources) -> Result<()> {
+        Ok(())
     }
 }
 
@@ -39,18 +55,17 @@ impl<'a> From<&'a Subsystem> for &'a PerfEventController {
                 Subsystem::PerfEvent(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
 impl PerfEventController {
-    /// Constructs a new `PerfEventController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Constructs a new `PerfEventController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf) -> Self {
         Self {
             base: root.clone(),
             path: root,

src/pid.rs

@@ -1,59 +1,72 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `pids` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/cgroups-v1/pids.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/pids.txt)
+use std::io::{Read, Write};
 use std::path::PathBuf;
-use std::io::{Write, Read};
-use std::fs::File;
 
-use {CgroupError, Resources, PidResources, Controller, ControllIdentifier, Subsystem, Controllers};
-use CgroupError::*;
+use crate::error::ErrorKind::*;
+use crate::error::*;
+
+use crate::read_u64_from;
+use crate::{
+    parse_max_value, ControllIdentifier, ControllerInternal, Controllers, MaxValue, PidResources,
+    Resources, Subsystem,
+};
 
 /// A controller that allows controlling the `pids` subsystem of a Cgroup.
 #[derive(Debug, Clone)]
 pub struct PidController {
     base: PathBuf,
     path: PathBuf,
+    v2: bool,
 }
 
-/// The values found in the `pids.max` file in a Cgroup's `pids` subsystem.
-#[derive(Eq, PartialEq, Copy, Clone, Debug)]
-pub enum PidMax {
-    /// This value is returned when the text found `pids.max` is `"max"`.
-    Max,
-    /// When the value in `pids.max` is a numerical value, they are returned via this enum field.
-    Value(i64),
-}
-
-impl Default for PidMax {
-    /// By default, (as per the kernel) `pids.max` should contain `"max"`.
-    fn default() -> Self {
-        PidMax::Max
-    }
-}
-
-impl Controller for PidController {
-    fn control_type(self: &Self) -> Controllers { Controllers::Pids }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
-
-    fn apply(self: &Self, res: &Resources) {
-        /* get the resources that apply to this controller */
-        let pidres: &PidResources = &res.pid;
-
-        if pidres.update_values {
-            /* apply pid_max */
-            let _ = self.set_pid_max(pidres.maximum_number_of_processes);
-        }
-    }
-}
-
-/*impl<'a> ControllIdentifier for &'a PidController {
-    fn controller_type() -> Controllers {
+impl ControllerInternal for PidController {
+    fn control_type(&self) -> Controllers {
         Controllers::Pids
     }
-}*/
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
+
+    fn is_v2(&self) -> bool {
+        self.v2
+    }
+
+    fn apply(&self, res: &Resources) -> Result<()> {
+        // get the resources that apply to this controller
+        let pidres: &PidResources = &res.pid;
+
+        // apply pid_max
+        update_and_test!(
+            self,
+            set_pid_max,
+            pidres.maximum_number_of_processes,
+            get_pid_max
+        );
+
+        Ok(())
+    }
+}
+
+// impl<'a> ControllIdentifier for &'a PidController {
+//     fn controller_type() -> Controllers {
+//         Controllers::Pids
+//     }
+// }
 
 impl ControllIdentifier for PidController {
     fn controller_type() -> Controllers {
@@ -68,72 +81,56 @@ impl<'a> From<&'a Subsystem> for &'a PidController {
                 Subsystem::Pid(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
-fn read_u64_from(mut file: File) -> Result<u64, CgroupError> {
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => string.trim().parse().map_err(|_| ParseError),
-        Err(e) => Err(CgroupError::ReadError(e)),
-    }
-}
-
 impl PidController {
-    /// Constructors a new `PidController` instance, with `oroot` serving as the controller's root
+    /// Constructors a new `PidController` instance, with `root` serving as the controller's root
     /// directory.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    pub fn new(root: PathBuf, v2: bool) -> Self {
         Self {
             base: root.clone(),
             path: root,
+            v2: v2,
         }
     }
 
     /// The number of times `fork` failed because the limit was hit.
-    pub fn get_pid_events(self: &Self) -> Result<u64, CgroupError> {
+    pub fn get_pid_events(&self) -> Result<u64> {
         self.open_path("pids.events", false).and_then(|mut file| {
             let mut string = String::new();
             match file.read_to_string(&mut string) {
-                Ok(_) => {
-                    match string.split_whitespace().nth(1) {
-                        Some(elem) => match elem.parse() {
-                            Ok(val) => Ok(val),
-                            Err(_) => Err(CgroupError::ParseError),
-                        },
-                        None => Err(CgroupError::ParseError),
-                    }
+                Ok(_) => match string.split_whitespace().nth(1) {
+                    Some(elem) => match elem.parse() {
+                        Ok(val) => Ok(val),
+                        Err(e) => Err(Error::with_cause(ParseError, e)),
+                    },
+                    None => Err(Error::new(ParseError)),
                 },
-                Err(e) => Err(CgroupError::ReadError(e)),
+                Err(e) => Err(Error::with_cause(ReadFailed, e)),
             }
         })
     }
 
     /// The number of processes currently.
-    pub fn get_pid_current(self: &Self) -> Result<u64, CgroupError> {
-        self.open_path("pids.current", false).and_then(read_u64_from)
+    pub fn get_pid_current(&self) -> Result<u64> {
+        self.open_path("pids.current", false)
+            .and_then(read_u64_from)
     }
 
     /// The maximum number of processes that can exist at one time in the control group.
-    pub fn get_pid_max(self: &Self) -> Result<PidMax, CgroupError> {
+    pub fn get_pid_max(&self) -> Result<MaxValue> {
         self.open_path("pids.max", false).and_then(|mut file| {
             let mut string = String::new();
             let res = file.read_to_string(&mut string);
             match res {
-                Ok(_) => if string.trim() == "max" {
-                    Ok(PidMax::Max)
-                } else {
-                    match string.trim().parse() {
-                        Ok(val) => Ok(PidMax::Value(val)),
-                        Err(_) => Err(CgroupError::ParseError),
-                    }
-                },
-                Err(e) => Err(CgroupError::ReadError(e)),
+                Ok(_) => parse_max_value(&string),
+                Err(e) => Err(Error::with_cause(ReadFailed, e)),
             }
         })
     }
@@ -143,15 +140,12 @@ impl PidController {
     /// Note that if `get_pid_current()` returns a higher number than what you
     /// are about to set (`max_pid`), then no processess will be killed. Additonally, attaching
     /// extra processes to a control group disregards the limit.
-    pub fn set_pid_max(self: &Self, max_pid: PidMax) -> Result<(), CgroupError> {
+    pub fn set_pid_max(&self, max_pid: MaxValue) -> Result<()> {
         self.open_path("pids.max", true).and_then(|mut file| {
-            let string_to_write = match max_pid {
-                PidMax::Max => "max".to_string(),
-                PidMax::Value(num) => num.to_string(),
-            };
+            let string_to_write = max_pid.to_string();
             match file.write_all(string_to_write.as_ref()) {
                 Ok(_) => Ok(()),
-                Err(e) => Err(CgroupError::WriteError(e)),
+                Err(e) => Err(Error::with_cause(WriteFailed, e)),
             }
         })
     }

src/rdma.rs

@@ -1,12 +1,20 @@
+// Copyright (c) 2018 Levente Kurusa
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! This module contains the implementation of the `rdma` cgroup subsystem.
-//! 
+//!
 //! See the Kernel's documentation for more information about this subsystem, found at:
 //!  [Documentation/cgroup-v1/rdma.txt](https://www.kernel.org/doc/Documentation/cgroup-v1/rdma.txt)
+use std::io::Write;
 use std::path::PathBuf;
-use std::io::{Write, Read};
-use std::fs::File;
 
-use {CgroupError, Controllers, Controller, Resources, ControllIdentifier, Subsystem};
+use crate::error::ErrorKind::*;
+use crate::error::*;
+
+use crate::read_string_from;
+use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
 
 /// A controller that allows controlling the `rdma` subsystem of a Cgroup.
 ///
@@ -18,13 +26,22 @@ pub struct RdmaController {
     path: PathBuf,
 }
 
-impl Controller for RdmaController {
-    fn control_type(self: &Self) -> Controllers { Controllers::Rdma }
-    fn get_path<'a>(self: &'a Self) -> &'a PathBuf { &self.path }
-    fn get_path_mut<'a>(self: &'a mut Self) -> &'a mut PathBuf { &mut self.path }
-    fn get_base<'a>(self: &'a Self) -> &'a PathBuf { &self.base }
+impl ControllerInternal for RdmaController {
+    fn control_type(&self) -> Controllers {
+        Controllers::Rdma
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
 
-    fn apply(self: &Self, _res: &Resources) {
+    fn apply(&self, _res: &Resources) -> Result<()> {
+        Ok(())
     }
 }
 
@@ -41,26 +58,17 @@ impl<'a> From<&'a Subsystem> for &'a RdmaController {
                 Subsystem::Rdma(c) => c,
                 _ => {
                     assert_eq!(1, 0);
-                    ::std::mem::uninitialized()
-                },
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
             }
         }
     }
 }
 
-fn read_string_from(mut file: File) -> Result<String, CgroupError> {
-    let mut string = String::new();
-    match file.read_to_string(&mut string) {
-        Ok(_) => Ok(string.trim().to_string()),
-        Err(e) => Err(CgroupError::ReadError(e)),
-    }
-}
-
 impl RdmaController {
-    /// Constructs a new `RdmaController` with `oroot` serving as the root of the control group.
-    pub fn new(oroot: PathBuf) -> Self {
-        let mut root = oroot;
-        root.push(Self::controller_type().to_string());
+    /// Constructs a new `RdmaController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf) -> Self {
         Self {
             base: root.clone(),
             path: root,
@@ -68,15 +76,16 @@ impl RdmaController {
     }
 
     /// Returns the current usage of RDMA/IB specific resources.
-    pub fn current(self: &Self) -> Result<String, CgroupError> {
+    pub fn current(&self) -> Result<String> {
         self.open_path("rdma.current", false)
             .and_then(read_string_from)
     }
 
     /// Set a maximum usage for each RDMA/IB resource.
-    pub fn set_max(self: &Self, max: &String) -> Result<(), CgroupError> {
+    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(CgroupError::WriteError)
+            file.write_all(max.as_ref())
+                .map_err(|e| Error::with_cause(WriteFailed, e))
         })
     }
 }

src/systemd.rs

@@ -0,0 +1,72 @@
+// Copyright (c) 2020 Ant Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+//! This module contains the implementation of the `systemd` cgroup subsystem.
+//!
+use std::path::PathBuf;
+
+use crate::error::*;
+
+use crate::{ControllIdentifier, ControllerInternal, Controllers, Resources, Subsystem};
+
+/// A controller that allows controlling the `systemd` subsystem of a Cgroup.
+///
+#[derive(Debug, Clone)]
+pub struct SystemdController {
+    base: PathBuf,
+    path: PathBuf,
+    v2: bool,
+}
+
+impl ControllerInternal for SystemdController {
+    fn control_type(&self) -> Controllers {
+        Controllers::Systemd
+    }
+    fn get_path(&self) -> &PathBuf {
+        &self.path
+    }
+    fn get_path_mut(&mut self) -> &mut PathBuf {
+        &mut self.path
+    }
+    fn get_base(&self) -> &PathBuf {
+        &self.base
+    }
+
+    fn apply(&self, _res: &Resources) -> Result<()> {
+        Ok(())
+    }
+}
+
+impl ControllIdentifier for SystemdController {
+    fn controller_type() -> Controllers {
+        Controllers::Systemd
+    }
+}
+
+impl<'a> From<&'a Subsystem> for &'a SystemdController {
+    fn from(sub: &'a Subsystem) -> &'a SystemdController {
+        unsafe {
+            match sub {
+                Subsystem::Systemd(c) => c,
+                _ => {
+                    assert_eq!(1, 0);
+                    let v = std::mem::MaybeUninit::uninit();
+                    v.assume_init()
+                }
+            }
+        }
+    }
+}
+
+impl SystemdController {
+    /// Constructs a new `SystemdController` with `root` serving as the root of the control group.
+    pub fn new(root: PathBuf, v2: bool) -> Self {
+        Self {
+            base: root.clone(),
+            path: root,
+            v2: v2,
+        }
+    }
+}

tests/builder.rs

@@ -0,0 +1,164 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 And Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+//! 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::*;
+
+#[test]
+pub fn test_cpu_res_build() {
+    let h = cgroups_rs::hierarchies::auto();
+    let cg: Cgroup = CgroupBuilder::new("test_cpu_res_build")
+        .cpu()
+        .shares(85)
+        .done()
+        .build(h);
+
+    {
+        let cpu: &CpuController = cg.controller_of().unwrap();
+        assert!(cpu.shares().is_ok());
+        assert_eq!(cpu.shares().unwrap(), 85);
+    }
+
+    cg.delete().unwrap();
+}
+
+#[test]
+pub fn test_memory_res_build() {
+    let h = cgroups_rs::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);
+
+    {
+        let c: &MemController = cg.controller_of().unwrap();
+        if !c.v2() {
+            assert_eq!(c.kmem_stat().limit_in_bytes, 128 * 1024 * 1024);
+            assert_eq!(c.memory_stat().swappiness, 70);
+        }
+        assert_eq!(c.memory_stat().limit_in_bytes, 1024 * 1024 * 1024);
+    }
+
+    cg.delete().unwrap();
+}
+
+#[test]
+pub fn test_pid_res_build() {
+    let h = cgroups_rs::hierarchies::auto();
+    let cg: Cgroup = CgroupBuilder::new("test_pid_res_build")
+        .pid()
+        .maximum_number_of_processes(MaxValue::Value(123))
+        .done()
+        .build(h);
+
+    {
+        let c: &PidController = cg.controller_of().unwrap();
+        assert!(c.get_pid_max().is_ok());
+        assert_eq!(c.get_pid_max().unwrap(), MaxValue::Value(123));
+    }
+
+    cg.delete().unwrap();
+}
+
+#[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 cg: Cgroup = CgroupBuilder::new("test_devices_res_build")
+        .devices()
+        .device(1, 6, DeviceType::Char, true, vec![DevicePermissions::Read])
+        .done()
+        .build(h);
+
+    {
+        let c: &DevicesController = cg.controller_of().unwrap();
+        assert!(c.allowed_devices().is_ok());
+        assert_eq!(
+            c.allowed_devices().unwrap(),
+            vec![DeviceResource {
+                allow: true,
+                devtype: DeviceType::Char,
+                major: 1,
+                minor: 6,
+                access: vec![DevicePermissions::Read],
+            }]
+        );
+    }
+    cg.delete().unwrap();
+}
+
+#[test]
+pub fn test_network_res_build() {
+    let h = cgroups_rs::hierarchies::auto();
+    if h.v2() {
+        // FIXME add cases for v2
+        return;
+    }
+    let cg: Cgroup = CgroupBuilder::new("test_network_res_build")
+        .network()
+        .class_id(1337)
+        .done()
+        .build(h);
+
+    {
+        let c: &NetClsController = cg.controller_of().unwrap();
+        assert!(c.get_class().is_ok());
+        assert_eq!(c.get_class().unwrap(), 1337);
+    }
+    cg.delete().unwrap();
+}
+
+#[test]
+pub fn test_hugepages_res_build() {
+    let h = cgroups_rs::hierarchies::auto();
+    if h.v2() {
+        // FIXME add cases for v2
+        return;
+    }
+    let cg: Cgroup = CgroupBuilder::new("test_hugepages_res_build")
+        .hugepages()
+        .limit("2MB".to_string(), 4 * 2 * 1024 * 1024)
+        .done()
+        .build(h);
+
+    {
+        let c: &HugeTlbController = cg.controller_of().unwrap();
+        assert!(c.limit_in_bytes(&"2MB".to_string()).is_ok());
+        assert_eq!(
+            c.limit_in_bytes(&"2MB".to_string()).unwrap(),
+            4 * 2 * 1024 * 1024
+        );
+    }
+    cg.delete().unwrap();
+}
+
+#[test]
+#[ignore] // high version kernel not support `blkio.weight`
+pub fn test_blkio_res_build() {
+    let h = cgroups_rs::hierarchies::auto();
+    let cg: Cgroup = CgroupBuilder::new("test_blkio_res_build")
+        .blkio()
+        .weight(100)
+        .done()
+        .build(h);
+
+    {
+        let c: &BlkIoController = cg.controller_of().unwrap();
+        assert_eq!(c.blkio().weight, 100);
+    }
+    cg.delete().unwrap();
+}

tests/cgroup.rs

@@ -1,23 +1,31 @@
-//! Simple unit tests about the control groups system.
-extern crate cgroups;
-use cgroups::{Cgroup, CgroupPid};
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 And Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
 
-extern crate nix;
-extern crate libc;
+//! Simple unit tests about the control groups system.
+use cgroups_rs::memory::MemController;
+use cgroups_rs::Controller;
+use cgroups_rs::{Cgroup, CgroupPid, Subsystem};
 
 #[test]
 fn test_tasks_iterator() {
-    let hier = cgroups::hierarchies::V1::new();
+    let h = cgroups_rs::hierarchies::auto();
     let pid = libc::pid_t::from(nix::unistd::getpid()) as u64;
-    let cg = Cgroup::new(&hier, String::from("test_tasks_iterator"));
+    let cg = Cgroup::new(h, String::from("test_tasks_iterator"));
     {
         // Add a task to the control group.
-        cg.add_task(CgroupPid::from(pid));
+        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
         assert_eq!(tasks.next(), Some(CgroupPid::from(pid)));
         assert_eq!(tasks.next(), None);
-        
+
         // Now, try removing it.
         cg.remove_task(CgroupPid::from(pid));
         tasks = cg.tasks().into_iter();
@@ -25,5 +33,73 @@ fn test_tasks_iterator() {
         // Verify that it was indeed removed.
         assert_eq!(tasks.next(), None);
     }
-    cg.delete();
+    cg.delete().unwrap();
+}
+
+#[test]
+fn test_cgroup_with_relative_paths() {
+    if cgroups_rs::hierarchies::is_cgroup2_unified_mode() {
+        return;
+    }
+    let h = cgroups_rs::hierarchies::auto();
+    let cgroup_root = h.root();
+    let cgroup_name = "test_cgroup_with_relative_paths";
+
+    let cg = Cgroup::load(h, String::from(cgroup_name));
+    {
+        let subsystems = cg.subsystems();
+        subsystems.into_iter().for_each(|sub| match sub {
+            Subsystem::Pid(c) => {
+                let cgroup_path = c.path().to_str().unwrap();
+                let relative_path = "/pids/";
+                // cgroup_path = cgroup_root + relative_path + cgroup_name
+                assert_eq!(
+                    cgroup_path,
+                    format!(
+                        "{}{}{}",
+                        cgroup_root.to_str().unwrap(),
+                        relative_path,
+                        cgroup_name
+                    )
+                );
+            }
+            Subsystem::Mem(c) => {
+                let cgroup_path = c.path().to_str().unwrap();
+                // cgroup_path = cgroup_root + relative_path + cgroup_name
+                assert_eq!(
+                    cgroup_path,
+                    format!("{}/memory/{}", cgroup_root.to_str().unwrap(), cgroup_name)
+                );
+            }
+            _ => {}
+        });
+    }
+    cg.delete().unwrap();
+}
+
+#[test]
+fn test_cgroup_v2() {
+    if !cgroups_rs::hierarchies::is_cgroup2_unified_mode() {
+        return;
+    }
+    let h = cgroups_rs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_v2"));
+
+    let mem_controller: &MemController = cg.controller_of().unwrap();
+    let (mem, swp, rev) = (4 * 1024 * 1000, 2 * 1024 * 1000, 1024 * 1000);
+
+    mem_controller.set_limit(mem).unwrap();
+    mem_controller.set_memswap_limit(swp).unwrap();
+    mem_controller.set_soft_limit(rev).unwrap();
+
+    let memory_stat = mem_controller.memory_stat();
+    println!("memory_stat {:?}", memory_stat);
+    assert_eq!(mem, memory_stat.limit_in_bytes);
+    assert_eq!(rev, memory_stat.soft_limit_in_bytes);
+
+    let memswap = mem_controller.memswap();
+    println!("memswap {:?}", memswap);
+    assert_eq!(swp, memswap.limit_in_bytes);
+
+    cg.delete().unwrap();
 }

tests/cpu.rs

@@ -0,0 +1,61 @@
+// Copyright (c) 2020 And Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+//! Simple unit tests about the CPU control groups system.
+use cgroups_rs::cpu::CpuController;
+use cgroups_rs::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 cpu_controller: &CpuController = cg.controller_of().unwrap();
+
+    let current_quota = cpu_controller.cfs_quota().unwrap();
+    let current_peroid = cpu_controller.cfs_period().unwrap();
+
+    // verify default value
+    // The default is “max 100000”.
+    assert_eq!(-1, current_quota);
+    assert_eq!(100000, current_peroid);
+
+    // case 1 set quota
+    let _ = cpu_controller.set_cfs_quota(2000);
+
+    let current_quota = cpu_controller.cfs_quota().unwrap();
+    let current_peroid = cpu_controller.cfs_period().unwrap();
+    assert_eq!(2000, current_quota);
+    assert_eq!(100000, current_peroid);
+
+    // case 2 set period
+    cpu_controller.set_cfs_period(1000000).unwrap();
+    let current_quota = cpu_controller.cfs_quota().unwrap();
+    let current_peroid = cpu_controller.cfs_period().unwrap();
+    assert_eq!(2000, current_quota);
+    assert_eq!(1000000, current_peroid);
+
+    // case 3 set both quota and period
+    cpu_controller
+        .set_cfs_quota_and_period(Some(5000), Some(100000))
+        .unwrap();
+
+    let current_quota = cpu_controller.cfs_quota().unwrap();
+    let current_peroid = cpu_controller.cfs_period().unwrap();
+    assert_eq!(5000, current_quota);
+    assert_eq!(100000, current_peroid);
+
+    // case 4 set both quota and period, set quota to -1
+    cpu_controller
+        .set_cfs_quota_and_period(Some(-1), None)
+        .unwrap();
+
+    let current_quota = cpu_controller.cfs_quota().unwrap();
+    let current_peroid = cpu_controller.cfs_period().unwrap();
+    assert_eq!(-1, current_quota);
+    assert_eq!(100000, current_peroid);
+
+    cg.delete().unwrap();
+}

tests/cpuset.rs

@@ -0,0 +1,93 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 And Group
+//
+// 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;
+
+#[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 cpuset: &CpuSetController = cg.controller_of().unwrap();
+
+        // This is not a root control group, so it should fail via InvalidOperation.
+        let res = cpuset.set_enable_memory_pressure(true);
+        assert_eq!(res.unwrap_err().kind(), &ErrorKind::InvalidOperation);
+    }
+    cg.delete().unwrap();
+}
+
+#[test]
+fn test_cpuset_set_cpus() {
+    let h = cgroups_rs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_cpuset_set_cpus"));
+    {
+        let cpuset: &CpuSetController = cg.controller_of().unwrap();
+
+        let set = cpuset.cpuset();
+        if cg.v2() {
+            assert_eq!(0, set.cpus.len());
+        } else {
+            // for cgroup v1, cpuset is copied from parent.
+            assert_eq!(true, set.cpus.len() > 0);
+        }
+
+        // 0
+        let r = cpuset.set_cpus("0");
+        assert_eq!(true, r.is_ok());
+
+        let set = cpuset.cpuset();
+        assert_eq!(1, set.cpus.len());
+        assert_eq!((0, 0), set.cpus[0]);
+
+        // all cpus in system
+        let cpus =
+            fs::read_to_string("/sys/fs/cgroup/cpuset.cpus.effective").unwrap_or("".to_string());
+        let cpus = cpus.trim();
+        if cpus != "" {
+            let r = cpuset.set_cpus(&cpus);
+            assert_eq!(true, r.is_ok());
+            let set = cpuset.cpuset();
+            assert_eq!(1, set.cpus.len());
+            assert_eq!(format!("{}-{}", set.cpus[0].0, set.cpus[0].1), cpus);
+        }
+    }
+    cg.delete().unwrap();
+}
+
+#[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 cpuset: &CpuSetController = cg.controller_of().unwrap();
+    let set = cpuset.cpuset();
+    if cg.v2() {
+        assert_eq!(0, set.cpus.len());
+    } else {
+        // for cgroup v1, cpuset is copied from parent.
+        assert_eq!(true, set.cpus.len() > 0);
+    }
+
+    // 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 tasks = cg.tasks();
+    assert_eq!(true, tasks.len() > 0);
+    println!("tasks after added: {:?}", tasks);
+
+    // remove task
+    let _ = cg.remove_task(CgroupPid::from(pid_i));
+    let tasks = cg.tasks();
+    println!("tasks after deleted: {:?}", tasks);
+    assert_eq!(0, tasks.len());
+
+    cg.delete().unwrap();
+}

tests/devices.rs

@@ -0,0 +1,73 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 And Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+//! Integration tests about the devices subsystem
+
+use cgroups_rs::devices::{DevicePermissions, DeviceType, DevicesController};
+use cgroups_rs::{Cgroup, DeviceResource};
+
+#[test]
+fn test_devices_parsing() {
+    // now only v2
+    if cgroups_rs::hierarchies::is_cgroup2_unified_mode() {
+        return;
+    }
+
+    let h = cgroups_rs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_devices_parsing"));
+    {
+        let devices: &DevicesController = cg.controller_of().unwrap();
+
+        // Deny access to all devices first
+        devices
+            .deny_device(
+                DeviceType::All,
+                -1,
+                -1,
+                &vec![
+                    DevicePermissions::Read,
+                    DevicePermissions::Write,
+                    DevicePermissions::MkNod,
+                ],
+            )
+            .unwrap();
+        // Acquire the list of allowed devices after we denied all
+        let allowed_devices = devices.allowed_devices();
+        // Verify that there are no devices that we can access.
+        assert!(allowed_devices.is_ok());
+        assert_eq!(allowed_devices.unwrap(), Vec::new());
+
+        // Now add mknod access to /dev/null device
+        devices
+            .allow_device(DeviceType::Char, 1, 3, &vec![DevicePermissions::MkNod])
+            .unwrap();
+        let allowed_devices = devices.allowed_devices();
+        assert!(allowed_devices.is_ok());
+        let allowed_devices = allowed_devices.unwrap();
+        assert_eq!(allowed_devices.len(), 1);
+        assert_eq!(
+            allowed_devices[0],
+            DeviceResource {
+                allow: true,
+                devtype: DeviceType::Char,
+                major: 1,
+                minor: 3,
+                access: vec![DevicePermissions::MkNod],
+            }
+        );
+
+        // Now deny, this device explicitly.
+        devices
+            .deny_device(DeviceType::Char, 1, 3, &DevicePermissions::all())
+            .unwrap();
+        // Finally, check that.
+        let allowed_devices = devices.allowed_devices();
+        // Verify that there are no devices that we can access.
+        assert!(allowed_devices.is_ok());
+        assert_eq!(allowed_devices.unwrap(), Vec::new());
+    }
+    cg.delete().unwrap();
+}

tests/hugetlb.rs

@@ -0,0 +1,44 @@
+// Copyright (c) 2020 And Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+//! Integration tests about the hugetlb subsystem
+use cgroups_rs::error::*;
+use cgroups_rs::hugetlb::{self, HugeTlbController};
+use cgroups_rs::Cgroup;
+use std::fs;
+
+#[test]
+fn test_hugetlb_sizes() {
+    // now only v2
+    if cgroups_rs::hierarchies::is_cgroup2_unified_mode() {
+        return;
+    }
+
+    let h = cgroups_rs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_hugetlb_sizes"));
+    {
+        let hugetlb_controller: &HugeTlbController = cg.controller_of().unwrap();
+        let _ = hugetlb_controller.get_sizes();
+
+        // test sizes count
+        let sizes = hugetlb_controller.get_sizes();
+        let sizes_count = fs::read_dir(hugetlb::HUGEPAGESIZE_DIR).unwrap().count();
+        assert_eq!(sizes.len(), sizes_count);
+
+        for size in sizes {
+            let supported = hugetlb_controller.size_supported(&size);
+            assert_eq!(supported, true);
+            assert_no_error(hugetlb_controller.failcnt(&size));
+            assert_no_error(hugetlb_controller.limit_in_bytes(&size));
+            assert_no_error(hugetlb_controller.usage_in_bytes(&size));
+            assert_no_error(hugetlb_controller.max_usage_in_bytes(&size));
+        }
+    }
+    cg.delete().unwrap();
+}
+
+fn assert_no_error(r: Result<u64>) {
+    assert_eq!(!r.is_err(), true)
+}

tests/memory.rs

@@ -0,0 +1,91 @@
+// Copyright (c) 2020 And Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
+//! Integration tests about the hugetlb subsystem
+use cgroups_rs::memory::{MemController, SetMemory};
+use cgroups_rs::Controller;
+use cgroups_rs::{Cgroup, 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 mem_controller: &MemController = cg.controller_of().unwrap();
+
+        // before disable
+        let m = mem_controller.memory_stat();
+        assert_eq!(m.oom_control.oom_kill_disable, false);
+
+        // now only v1
+        if !mem_controller.v2() {
+            // disable oom killer
+            let r = mem_controller.disable_oom_killer();
+            assert_eq!(r.is_err(), false);
+
+            // after disable
+            let m = mem_controller.memory_stat();
+            assert_eq!(m.oom_control.oom_kill_disable, true);
+        }
+    }
+    cg.delete().unwrap();
+}
+
+#[test]
+fn set_mem_v2() {
+    let h = cgroups_rs::hierarchies::auto();
+    if !h.v2() {
+        return;
+    }
+
+    let cg = Cgroup::new(h, String::from("set_mem_v2"));
+    {
+        let mem_controller: &MemController = cg.controller_of().unwrap();
+
+        // before disable
+        let m = mem_controller.get_mem().unwrap();
+        // case 1: get default value
+        assert_eq!(m.low, Some(MaxValue::Value(0)));
+        assert_eq!(m.min, Some(MaxValue::Value(0)));
+        assert_eq!(m.high, Some(MaxValue::Max));
+        assert_eq!(m.max, Some(MaxValue::Max));
+
+        // case 2: set parts
+        let m = SetMemory {
+            low: Some(MaxValue::Value(1024 * 1024 * 2)),
+            high: Some(MaxValue::Value(1024 * 1024 * 1024 * 2)),
+            min: Some(MaxValue::Value(1024 * 1024 * 3)),
+            max: None,
+        };
+        let r = mem_controller.set_mem(m);
+        assert_eq!(true, r.is_ok());
+
+        let m = mem_controller.get_mem().unwrap();
+        // get
+        assert_eq!(m.low, Some(MaxValue::Value(1024 * 1024 * 2)));
+        assert_eq!(m.min, Some(MaxValue::Value(1024 * 1024 * 3)));
+        assert_eq!(m.high, Some(MaxValue::Value(1024 * 1024 * 1024 * 2)));
+        assert_eq!(m.max, Some(MaxValue::Max));
+
+        // case 3: set parts
+        let m = SetMemory {
+            max: Some(MaxValue::Value(1024 * 1024 * 1024 * 2)),
+            min: Some(MaxValue::Value(1024 * 1024 * 4)),
+            high: Some(MaxValue::Max),
+            low: None,
+        };
+        let r = mem_controller.set_mem(m);
+        assert_eq!(true, r.is_ok());
+
+        let m = mem_controller.get_mem().unwrap();
+        // get
+        assert_eq!(m.low, Some(MaxValue::Value(1024 * 1024 * 2)));
+        assert_eq!(m.min, Some(MaxValue::Value(1024 * 1024 * 4)));
+        assert_eq!(m.max, Some(MaxValue::Value(1024 * 1024 * 1024 * 2)));
+        assert_eq!(m.high, Some(MaxValue::Max));
+    }
+
+    cg.delete().unwrap();
+}

tests/pids.rs

@@ -1,70 +1,77 @@
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 And Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
+
 //! Integration tests about the pids subsystem
-extern crate cgroups;
-use cgroups::{CgroupError, CgroupPid, Cgroup, Resources, PidResources};
-use cgroups::pid::{PidController, PidMax};
-use cgroups::Controller;
+use cgroups_rs::pid::PidController;
+use cgroups_rs::Controller;
+use cgroups_rs::{Cgroup, MaxValue};
 
-extern crate nix;
-use nix::unistd::{Pid, fork, ForkResult};
 use nix::sys::wait::{waitpid, WaitStatus};
+use nix::unistd::{fork, ForkResult};
 
-extern crate libc;
 use libc::pid_t;
 
-use std::thread;
-
 #[test]
 fn create_and_delete_cgroup() {
-    let hier = cgroups::hierarchies::V1::new();
-    let cg = Cgroup::new(&hier, String::from("create_and_delete_cgroup"));
+    let h = cgroups_rs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("create_and_delete_cgroup"));
     {
         let pidcontroller: &PidController = cg.controller_of().unwrap();
-        pidcontroller.set_pid_max(PidMax::Value(1337));
-        assert_eq!(pidcontroller.get_pid_max(), Some(PidMax::Value(1337)));
+        pidcontroller.set_pid_max(MaxValue::Value(1337)).unwrap();
+        let max = pidcontroller.get_pid_max();
+        assert!(max.is_ok());
+        assert_eq!(max.unwrap(), MaxValue::Value(1337));
     }
-    cg.delete();
+    cg.delete().unwrap();
 }
 
 #[test]
 fn test_pids_current_is_zero() {
-    let hier = cgroups::hierarchies::V1::new();
-    let cg = Cgroup::new(&hier, String::from("test_pids_current_is_zero"));
+    let h = cgroups_rs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_pids_current_is_zero"));
     {
         let pidcontroller: &PidController = cg.controller_of().unwrap();
-        assert_eq!(pidcontroller.get_pid_current(), 0);
+        let current = pidcontroller.get_pid_current();
+        assert_eq!(current.unwrap(), 0);
     }
-    cg.delete();
+    cg.delete().unwrap();
 }
 
 #[test]
 fn test_pids_events_is_zero() {
-    let hier = cgroups::hierarchies::V1::new();
-    let cg = Cgroup::new(&hier, String::from("test_pids_events_is_zero"));
+    let h = cgroups_rs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("test_pids_events_is_zero"));
     {
         let pidcontroller: &PidController = cg.controller_of().unwrap();
-        assert_eq!(pidcontroller.get_pid_events(), 0);
+        let events = pidcontroller.get_pid_events();
+        assert!(events.is_ok());
+        assert_eq!(events.unwrap(), 0);
     }
-    cg.delete();
+    cg.delete().unwrap();
 }
 
 #[test]
 fn test_pid_events_is_not_zero() {
-    let hier = cgroups::hierarchies::V1::new();
-    let cg = Cgroup::new(&hier, String::from("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 pids: &PidController = cg.controller_of().unwrap();
         let before = pids.get_pid_events();
+        let before = before.unwrap();
 
         match fork() {
             Ok(ForkResult::Parent { child, .. }) => {
                 // move the process into the control group
-                pids.add_task(&(pid_t::from(child) as u64).into());
+                let _ = pids.add_task(&(pid_t::from(child) as u64).into());
 
                 println!("added task to cg: {:?}", child);
 
                 // Set limit to one
-                pids.set_pid_max(PidMax::Value(1));
-                println!("err = {:?}", pids.get_pid_max());
+                let _ = pids.set_pid_max(MaxValue::Value(1));
+                println!("current pid.max = {:?}", pids.get_pid_max());
 
                 // wait on the child
                 let res = waitpid(child, None);
@@ -75,21 +82,22 @@ fn test_pid_events_is_not_zero() {
                 }
 
                 // Check pids.events
-                assert_eq!(pids.get_pid_events(), before + 1);
-            },
-            Ok(ForkResult::Child) => {
-                loop {
-                    if pids.get_pid_max() == Some(PidMax::Value(1)) {
-                        if let Err(_) = fork() {
-                            unsafe { libc::exit(0) };
-                        } else {
-                            unsafe { libc::exit(1) };
-                        }
+                let events = pids.get_pid_events();
+                assert!(events.is_ok());
+                assert_eq!(events.unwrap(), before + 1);
+            }
+            Ok(ForkResult::Child) => loop {
+                let pids_max = pids.get_pid_max();
+                if pids_max.is_ok() && pids_max.unwrap() == MaxValue::Value(1) {
+                    if let Err(_) = fork() {
+                        unsafe { libc::exit(0) };
+                    } else {
+                        unsafe { libc::exit(1) };
                     }
                 }
             },
             Err(_) => panic!("failed to fork"),
         }
     }
-    cg.delete();
+    cg.delete().unwrap();
 }

tests/resources.rs

@@ -1,26 +1,31 @@
-//! Integration test about setting resources using `apply()`
-extern crate cgroups;
+// Copyright (c) 2018 Levente Kurusa
+// Copyright (c) 2020 And Group
+//
+// SPDX-License-Identifier: Apache-2.0 or MIT
+//
 
-use cgroups::{Cgroup, Resources, PidResources};
-use cgroups::pid::{PidController, PidMax};
+//! Integration test about setting resources using `apply()`
+use cgroups_rs::pid::PidController;
+use cgroups_rs::{Cgroup, MaxValue, PidResources, Resources};
 
 #[test]
 fn pid_resources() {
-    let hier = cgroups::hierarchies::V1::new();
-    let cg = Cgroup::new(&hier, String::from("pid_resources"));
+    let h = cgroups_rs::hierarchies::auto();
+    let cg = Cgroup::new(h, String::from("pid_resources"));
     {
         let res = Resources {
             pid: PidResources {
-                update_values: true,
-                maximum_number_of_processes: PidMax::Value(512),
+                maximum_number_of_processes: Some(MaxValue::Value(512)),
             },
             ..Default::default()
         };
-        cg.apply(&res);
+        cg.apply(&res).unwrap();
 
-        /* verify */
+        // verify
         let pidcontroller: &PidController = cg.controller_of().unwrap();
-        assert_eq!(pidcontroller.get_pid_max(), Some(PidMax::Value(512)));
+        let pid_max = pidcontroller.get_pid_max();
+        assert_eq!(pid_max.is_ok(), true);
+        assert_eq!(pid_max.unwrap(), MaxValue::Value(512));
     }
-    cg.delete();
+    cg.delete().unwrap();
 }

tools/create_cgroup.sh

@@ -1,4 +1,9 @@
 #!/bin/sh
+#
+# Copyright (c) 2018 Levente Kurusa
+#
+# SPDX-License-Identifier: Apache-2.0 or MIT
+#
 
 CONTROL_GROUPS=`cargo test -- --list 2>/dev/null | egrep 'test$' | egrep -v '^src' | cut -d':' -f1`
 

tools/delete_cgroup.sh

@@ -1,4 +1,9 @@
 #!/bin/sh
+#
+# Copyright (c) 2018 Levente Kurusa
+#
+# SPDX-License-Identifier: Apache-2.0 or MIT
+#
 
 CONTROL_GROUPS=`cargo test -- --list 2>/dev/null | egrep 'test$' | egrep -v '^src' | cut -d':' -f1`