Merge pull request #1822 from ipuustin/intel-rdt

Add support to Intel RDT.

crates/libcontainer/Cargo.toml

@@ -33,6 +33,7 @@ mio = { version = "0.8.6", features = ["os-ext", "os-poll"] }
 nix = "0.26.2"
 path-clean = "1.0.1"
 oci-spec = { version = "^0.6.0", features = ["runtime"] }
+once_cell = "1.17.1"
 procfs = "0.15.1"
 prctl = "1.0.0"
 libcgroups = { version = "0.0.5", path = "../libcgroups", default-features = false }
@@ -42,6 +43,7 @@ serde_json = "1.0"
 syscalls = "0.6.10"
 rust-criu = "0.4.0"
 clone3 = "0.2.3"
+regex = "1.7.3"
 
 [dev-dependencies]
 oci-spec = { version = "^0.6.0", features = ["proptests", "runtime"] }

crates/libcontainer/src/container/builder_impl.rs

@@ -5,6 +5,7 @@ use crate::{
     process::{
         self,
         args::{ContainerArgs, ContainerType},
+        intel_rdt::delete_resctrl_subdirectory,
     },
     rootless::Rootless,
     syscall::Syscall,
@@ -137,7 +138,8 @@ impl<'a> ContainerBuilderImpl<'a> {
             executor_manager: &self.executor_manager,
         };
 
-        let init_pid = process::container_main_process::container_main_process(&container_args)?;
+        let (init_pid, need_to_clean_up_intel_rdt_dir) =
+            process::container_main_process::container_main_process(&container_args)?;
 
         // if file to write the pid to is specified, write pid of the child
         if let Some(pid_file) = &self.pid_file {
@@ -150,6 +152,7 @@ impl<'a> ContainerBuilderImpl<'a> {
                 .set_status(ContainerStatus::Created)
                 .set_creator(nix::unistd::geteuid().as_raw())
                 .set_pid(init_pid.as_raw())
+                .set_clean_up_intel_rdt_directory(need_to_clean_up_intel_rdt_dir)
                 .save()
                 .context("Failed to save container state")?;
         }
@@ -171,11 +174,23 @@ impl<'a> ContainerBuilderImpl<'a> {
         )?;
 
         let mut errors = Vec::new();
+
         if let Err(e) = cmanager.remove().context("failed to remove cgroup") {
             errors.push(e.to_string());
         }
 
         if let Some(container) = &self.container {
+            if let Some(true) = container.clean_up_intel_rdt_subdirectory() {
+                if let Err(e) = delete_resctrl_subdirectory(container.id()).with_context(|| {
+                    format!(
+                        "failed to delete resctrl subdirectory: {:?}",
+                        container.id()
+                    )
+                }) {
+                    errors.push(e.to_string());
+                }
+            }
+
             if container.root.exists() {
                 if let Err(e) = fs::remove_dir_all(&container.root)
                     .with_context(|| format!("could not delete {:?}", container.root))

crates/libcontainer/src/container/container.rs

@@ -126,6 +126,15 @@ impl Container {
         self
     }
 
+    pub fn set_clean_up_intel_rdt_directory(&mut self, clean_up: bool) -> &mut Self {
+        self.state.clean_up_intel_rdt_subdirectory = Some(clean_up);
+        self
+    }
+
+    pub fn clean_up_intel_rdt_subdirectory(&self) -> Option<bool> {
+        self.state.clean_up_intel_rdt_subdirectory
+    }
+
     pub fn status(&self) -> ContainerStatus {
         self.state.status
     }

crates/libcontainer/src/container/container_delete.rs

@@ -1,6 +1,7 @@
 use super::{Container, ContainerStatus};
 use crate::config::YoukiConfig;
 use crate::hooks;
+use crate::process::intel_rdt::delete_resctrl_subdirectory;
 use anyhow::{bail, Context, Result};
 use libcgroups;
 use nix::sys::signal;
@@ -66,6 +67,14 @@ impl Container {
         // Once reached here, the container is verified that it can be deleted.
         debug_assert!(self.status().can_delete());
 
+        if let Some(true) = &self.clean_up_intel_rdt_subdirectory() {
+            if let Err(err) = delete_resctrl_subdirectory(self.id()) {
+                log::warn!(
+                    "failed to delete resctrl subdirectory due to: {err:?}, continue to delete"
+                );
+            }
+        }
+
         if self.root.exists() {
             match YoukiConfig::load(&self.root) {
                 Ok(config) => {

crates/libcontainer/src/container/state.rs

@@ -98,6 +98,8 @@ pub struct State {
     pub creator: Option<u32>,
     // Specifies if systemd should be used to manage cgroups
     pub use_systemd: Option<bool>,
+    // Specifies if the Intel RDT subdirectory needs be cleaned up.
+    pub clean_up_intel_rdt_subdirectory: Option<bool>,
 }
 
 impl State {
@@ -119,6 +121,7 @@ impl State {
             created: None,
             creator: None,
             use_systemd: None,
+            clean_up_intel_rdt_subdirectory: None,
         }
     }
 

crates/libcontainer/src/process/args.rs

@@ -26,7 +26,7 @@ pub struct ContainerArgs<'a> {
     pub console_socket: Option<RawFd>,
     /// The Unix Domain Socket to communicate container start
     pub notify_socket: NotifyListener,
-    /// File descriptos preserved/passed to the container init process.
+    /// File descriptors preserved/passed to the container init process.
     pub preserve_fds: i32,
     /// Container state
     pub container: &'a Option<Container>,

crates/libcontainer/src/process/container_main_process.rs

@@ -1,6 +1,9 @@
 use crate::{
     container::ContainerProcessState,
-    process::{args::ContainerArgs, channel, container_intermediate_process, fork},
+    process::{
+        args::ContainerArgs, channel, container_intermediate_process, fork,
+        intel_rdt::setup_intel_rdt,
+    },
     rootless::Rootless,
     utils,
 };
@@ -20,7 +23,7 @@ use oci_spec::runtime;
 #[cfg(feature = "libseccomp")]
 use std::{io::IoSlice, path::Path};
 
-pub fn container_main_process(container_args: &ContainerArgs) -> Result<Pid> {
+pub fn container_main_process(container_args: &ContainerArgs) -> Result<(Pid, bool)> {
     // We use a set of channels to communicate between parent and child process.
     // Each channel is uni-directional. Because we will pass these channel to
     // cloned process, we have to be deligent about closing any unused channel.
@@ -68,6 +71,7 @@ pub fn container_main_process(container_args: &ContainerArgs) -> Result<Pid> {
     // The intermediate process will send the init pid once it forks the init
     // process.  The intermediate process should exit after this point.
     let init_pid = main_receiver.wait_for_intermediate_ready()?;
+    let mut need_to_clean_up_intel_rdt_subdirectory = false;
 
     if let Some(linux) = container_args.spec.linux() {
         if let Some(seccomp) = linux.seccomp() {
@@ -89,6 +93,14 @@ pub fn container_main_process(container_args: &ContainerArgs) -> Result<Pid> {
             sync_seccomp(seccomp, &state, init_sender, main_receiver)
                 .context("failed to sync seccomp with init")?;
         }
+        if let Some(intel_rdt) = linux.intel_rdt() {
+            let container_id = container_args
+                .container
+                .as_ref()
+                .map(|container| container.id());
+            need_to_clean_up_intel_rdt_subdirectory =
+                setup_intel_rdt(container_id, &init_pid, intel_rdt)?;
+        }
     }
 
     // We don't need to send anything to the init process after this point, so
@@ -124,7 +136,7 @@ pub fn container_main_process(container_args: &ContainerArgs) -> Result<Pid> {
         Err(err) => bail!("failed to wait for intermediate process: {err}"),
     };
 
-    Ok(init_pid)
+    Ok((init_pid, need_to_clean_up_intel_rdt_subdirectory))
 }
 
 #[cfg(feature = "libseccomp")]