agent: Add SELinux support for containers

The kata-agent supports SELinux for containers inside the guest
to comply with the OCI runtime specification.

Fixes: #4812

Signed-off-by: Manabu Sugimoto <Manabu.Sugimoto@sony.com>

src/agent/Cargo.lock

@@ -1764,6 +1764,7 @@ dependencies = [
  "tempfile",
  "test-utils",
  "tokio",
+ "xattr",
  "zbus",
 ]
 
@@ -2552,6 +2553,15 @@ version = "0.36.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c811ca4a8c853ef420abd8592ba53ddbbac90410fab6903b3e79972a631f7680"
 
+[[package]]
+name = "xattr"
+version = "0.2.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6d1526bbe5aaeb5eb06885f4d987bcdfa5e23187055de9b83fe00156a821fabc"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "zbus"
 version = "2.3.2"

src/agent/rustjail/Cargo.toml

@@ -35,6 +35,7 @@ inotify = "0.9.2"
 libseccomp = { version = "0.3.0", optional = true }
 zbus = "2.3.0"
 bit-vec= "0.6.3"
+xattr = "0.2.3"
 
 [dev-dependencies]
 serial_test = "0.5.0"

src/agent/rustjail/src/container.rs

@@ -30,6 +30,7 @@ use crate::log_child;
 use crate::process::Process;
 #[cfg(feature = "seccomp")]
 use crate::seccomp;
+use crate::selinux;
 use crate::specconv::CreateOpts;
 use crate::{mount, validator};
 
@@ -537,6 +538,8 @@ fn do_init_child(cwfd: RawFd) -> Result<()> {
         }
     }
 
+    let selinux_enabled = selinux::is_enabled()?;
+
     sched::unshare(to_new & !CloneFlags::CLONE_NEWUSER)?;
 
     if userns {
@@ -638,6 +641,18 @@ fn do_init_child(cwfd: RawFd) -> Result<()> {
         capctl::prctl::set_no_new_privs().map_err(|_| anyhow!("cannot set no new privileges"))?;
     }
 
+    // Set SELinux label
+    if !oci_process.selinux_label.is_empty() {
+        if !selinux_enabled {
+            return Err(anyhow!(
+                "SELinux label for the process is provided but SELinux is not enabled on the running kernel"
+            ));
+        }
+
+        log_child!(cfd_log, "Set SELinux label to the container process");
+        selinux::set_exec_label(&oci_process.selinux_label)?;
+    }
+
     // Log unknown seccomp system calls in advance before the log file descriptor closes.
     #[cfg(feature = "seccomp")]
     if let Some(ref scmp) = linux.seccomp {

src/agent/rustjail/src/lib.rs

@@ -38,6 +38,7 @@ pub mod pipestream;
 pub mod process;
 #[cfg(feature = "seccomp")]
 pub mod seccomp;
+pub mod selinux;
 pub mod specconv;
 pub mod sync;
 pub mod sync_with_async;

src/agent/rustjail/src/mount.rs

@@ -25,6 +25,7 @@ use std::fs::File;
 use std::io::{BufRead, BufReader};
 
 use crate::container::DEFAULT_DEVICES;
+use crate::selinux;
 use crate::sync::write_count;
 use std::string::ToString;
 
@@ -181,6 +182,8 @@ pub fn init_rootfs(
         None => flags |= MsFlags::MS_SLAVE,
     }
 
+    let label = &linux.mount_label;
+
     let root = spec
         .root
         .as_ref()
@@ -244,7 +247,7 @@ pub fn init_rootfs(
                 }
             }
 
-            mount_from(cfd_log, m, rootfs, flags, &data, "")?;
+            mount_from(cfd_log, m, rootfs, flags, &data, label)?;
             // bind mount won't change mount options, we need remount to make mount options
             // effective.
             // first check that we have non-default options required before attempting a
@@ -524,7 +527,6 @@ pub fn pivot_rootfs<P: ?Sized + NixPath + std::fmt::Debug>(path: &P) -> Result<(
 
 fn rootfs_parent_mount_private(path: &str) -> Result<()> {
     let mount_infos = parse_mount_table(MOUNTINFO_PATH)?;
-
     let mut max_len = 0;
     let mut mount_point = String::from("");
     let mut options = String::from("");
@@ -767,9 +769,9 @@ fn mount_from(
     rootfs: &str,
     flags: MsFlags,
     data: &str,
-    _label: &str,
+    label: &str,
 ) -> Result<()> {
-    let d = String::from(data);
+    let mut d = String::from(data);
     let dest = secure_join(rootfs, &m.destination);
 
     let src = if m.r#type.as_str() == "bind" {
@@ -822,6 +824,37 @@ fn mount_from(
         e
     })?;
 
+    // Set the SELinux context for the mounts
+    let mut use_xattr = false;
+    if !label.is_empty() {
+        if selinux::is_enabled()? {
+            let device = Path::new(&m.source)
+                .file_name()
+                .ok_or_else(|| anyhow!("invalid device source path: {}", &m.source))?
+                .to_str()
+                .ok_or_else(|| anyhow!("failed to convert device source path: {}", &m.source))?;
+
+            match device {
+                // SELinux does not support labeling of /proc or /sys
+                "proc" | "sysfs" => (),
+                // SELinux does not support mount labeling against /dev/mqueue,
+                // so we use setxattr instead
+                "mqueue" => {
+                    use_xattr = true;
+                }
+                _ => {
+                    log_child!(cfd_log, "add SELinux mount label to {}", dest.as_str());
+                    selinux::add_mount_label(&mut d, label);
+                }
+            }
+        } else {
+            log_child!(
+                cfd_log,
+                "SELinux label for the mount is provided but SELinux is not enabled on the running kernel"
+            );
+        }
+    }
+
     mount(
         Some(src.as_str()),
         dest.as_str(),
@@ -834,6 +867,10 @@ fn mount_from(
         e
     })?;
 
+    if !label.is_empty() && selinux::is_enabled()? && use_xattr {
+        xattr::set(dest.as_str(), "security.selinux", label.as_bytes())?;
+    }
+
     if flags.contains(MsFlags::MS_BIND)
         && flags.intersects(
             !(MsFlags::MS_REC

src/agent/rustjail/src/selinux.rs

@@ -0,0 +1,80 @@
+// Copyright 2022 Sony Group Corporation
+//
+// SPDX-License-Identifier: Apache-2.0
+//
+
+use anyhow::{Context, Result};
+use nix::unistd::gettid;
+use std::fs::{self, OpenOptions};
+use std::io::prelude::*;
+use std::path::Path;
+
+pub fn is_enabled() -> Result<bool> {
+    let buf = fs::read_to_string("/proc/mounts")?;
+    let enabled = buf.contains("selinuxfs");
+
+    Ok(enabled)
+}
+
+pub fn add_mount_label(data: &mut String, label: &str) {
+    if data.is_empty() {
+        let context = format!("context=\"{}\"", label);
+        data.push_str(&context);
+    } else {
+        let context = format!(",context=\"{}\"", label);
+        data.push_str(&context);
+    }
+}
+
+pub fn set_exec_label(label: &str) -> Result<()> {
+    let mut attr_path = Path::new("/proc/thread-self/attr/exec").to_path_buf();
+    if !attr_path.exists() {
+        // Fall back to the old convention
+        attr_path = Path::new("/proc/self/task")
+            .join(gettid().to_string())
+            .join("attr/exec")
+    }
+
+    let mut file = OpenOptions::new()
+        .write(true)
+        .truncate(true)
+        .open(attr_path)?;
+    file.write_all(label.as_bytes())
+        .with_context(|| "failed to apply SELinux label")?;
+
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    const TEST_LABEL: &str = "system_u:system_r:unconfined_t:s0";
+
+    #[test]
+    fn test_is_enabled() {
+        let ret = is_enabled();
+        assert!(ret.is_ok(), "Expecting Ok, Got {:?}", ret);
+    }
+
+    #[test]
+    fn test_add_mount_label() {
+        let mut data = String::new();
+        add_mount_label(&mut data, TEST_LABEL);
+        assert_eq!(data, format!("context=\"{}\"", TEST_LABEL));
+
+        let mut data = String::from("defaults");
+        add_mount_label(&mut data, TEST_LABEL);
+        assert_eq!(data, format!("defaults,context=\"{}\"", TEST_LABEL));
+    }
+
+    #[test]
+    fn test_set_exec_label() {
+        let ret = set_exec_label(TEST_LABEL);
+        if is_enabled().unwrap() {
+            assert!(ret.is_ok(), "Expecting Ok, Got {:?}", ret);
+        } else {
+            assert!(ret.is_err(), "Expecting error, Got {:?}", ret);
+        }
+    }
+}

src/agent/rustjail/src/validator.rs

@@ -6,6 +6,7 @@
 use crate::container::Config;
 use anyhow::{anyhow, Context, Result};
 use oci::{Linux, LinuxIdMapping, LinuxNamespace, Spec};
+use regex::Regex;
 use std::collections::HashMap;
 use std::path::{Component, PathBuf};
 
@@ -86,6 +87,23 @@ fn hostname(oci: &Spec) -> Result<()> {
 
 fn security(oci: &Spec) -> Result<()> {
     let linux = get_linux(oci)?;
+    let label_pattern = r".*_u:.*_r:.*_t:s[0-9]|1[0-5].*";
+    let label_regex = Regex::new(label_pattern)?;
+
+    if let Some(ref process) = oci.process {
+        if !process.selinux_label.is_empty() && !label_regex.is_match(&process.selinux_label) {
+            return Err(anyhow!(
+                "SELinux label for the process is invalid format: {}",
+                &process.selinux_label
+            ));
+        }
+    }
+    if !linux.mount_label.is_empty() && !label_regex.is_match(&linux.mount_label) {
+        return Err(anyhow!(
+            "SELinux label for the mount is invalid format: {}",
+            &linux.mount_label
+        ));
+    }
 
     if linux.masked_paths.is_empty() && linux.readonly_paths.is_empty() {
         return Ok(());
@@ -95,8 +113,6 @@ fn security(oci: &Spec) -> Result<()> {
         return Err(anyhow!("Linux namespace does not contain mount"));
     }
 
-    // don't care about selinux at present
-
     Ok(())
 }
 
@@ -285,7 +301,7 @@ pub fn validate(conf: &Config) -> Result<()> {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use oci::Mount;
+    use oci::{Mount, Process};
 
     #[test]
     fn test_namespace() {
@@ -388,6 +404,29 @@ mod tests {
         ];
         spec.linux = Some(linux);
         security(&spec).unwrap();
+
+        // SELinux
+        let valid_label = "system_u:system_r:container_t:s0:c123,c456";
+        let mut process = Process::default();
+        process.selinux_label = valid_label.to_string();
+        spec.process = Some(process);
+        security(&spec).unwrap();
+
+        let mut linux = Linux::default();
+        linux.mount_label = valid_label.to_string();
+        spec.linux = Some(linux);
+        security(&spec).unwrap();
+
+        let invalid_label = "system_u:system_r:container_t";
+        let mut process = Process::default();
+        process.selinux_label = invalid_label.to_string();
+        spec.process = Some(process);
+        security(&spec).unwrap_err();
+
+        let mut linux = Linux::default();
+        linux.mount_label = invalid_label.to_string();
+        spec.linux = Some(linux);
+        security(&spec).unwrap_err();
     }
 
     #[test]