blockdev: Fix loopback device resource leak on signal interruption

blockdev/Cargo.toml

@@ -0,0 +1,25 @@
+[package]
+description = "Internal blockdev code"
+# Should never be published to crates.io
+publish = false
+edition = "2021"
+license = "MIT OR Apache-2.0"
+name = "bootc-blockdev"
+repository = "https://github.com/bootc-dev/bootc"
+version = "0.0.0"
+
+[dependencies]
+anyhow = { workspace = true }
+bootc-utils = { path = "../utils" }
+camino = { workspace = true, features = ["serde1"] }
+fn-error-context = { workspace = true }
+regex = "1.10.4"
+serde = { workspace = true, features = ["derive"] }
+serde_json = { workspace = true }
+tracing = { workspace = true }
+
+[dev-dependencies]
+indoc = "2.0.5"
+
+[lib]
+path = "src/blockdev.rs"

crates/blockdev/Cargo.toml

@@ -11,9 +11,13 @@ anyhow = { workspace = true }
 bootc-utils = { package = "bootc-internal-utils", path = "../utils", version = "0.0.0" }
 camino = { workspace = true, features = ["serde1"] }
 fn-error-context = { workspace = true }
+libc = { workspace = true }
 regex = "1.10.4"
+rustix = { workspace = true }
 serde = { workspace = true, features = ["derive"] }
 serde_json = { workspace = true }
+tempfile = { workspace = true }
+tokio = { workspace = true, features = ["signal"] }
 tracing = { workspace = true }
 
 [dev-dependencies]

crates/blockdev/src/blockdev.rs

@@ -181,6 +181,14 @@ pub fn partitions_of(dev: &Utf8Path) -> Result<PartitionTable> {
 
 pub struct LoopbackDevice {
     pub dev: Option<Utf8PathBuf>,
+    // Handle to the cleanup helper process
+    cleanup_handle: Option<LoopbackCleanupHandle>,
+}
+
+/// Handle to manage the cleanup helper process for loopback devices
+struct LoopbackCleanupHandle {
+    /// Child process handle
+    child: std::process::Child,
 }
 
 impl LoopbackDevice {
@@ -208,7 +216,15 @@ impl LoopbackDevice {
             .run_get_string()?;
         let dev = Utf8PathBuf::from(dev.trim());
         tracing::debug!("Allocated loopback {dev}");
-        Ok(Self { dev: Some(dev) })
+
+        // Try to spawn cleanup helper process - if it fails, make it fatal
+        let cleanup_handle = Self::spawn_cleanup_helper(dev.as_str())
+            .context("Failed to spawn loopback cleanup helper")?;
+
+        Ok(Self {
+            dev: Some(dev),
+            cleanup_handle: Some(cleanup_handle),
+        })
     }
 
     // Access the path to the loopback block device.
@@ -217,13 +233,49 @@ impl LoopbackDevice {
         self.dev.as_deref().unwrap()
     }
 
+    /// Spawn a cleanup helper process that will clean up the loopback device
+    /// if the parent process dies unexpectedly
+    fn spawn_cleanup_helper(device_path: &str) -> Result<LoopbackCleanupHandle> {
+        use std::process::{Command, Stdio};
+
+        // Get the path to our own executable
+        let self_exe =
+            std::fs::read_link("/proc/self/exe").context("Failed to read /proc/self/exe")?;
+
+        // Create the helper process
+        let mut cmd = Command::new(self_exe);
+        cmd.args(["loopback-cleanup-helper", "--device", device_path]);
+
+        // Set environment variable to indicate this is a cleanup helper
+        cmd.env("BOOTC_LOOPBACK_CLEANUP_HELPER", "1");
+
+        // Set up stdio to redirect to /dev/null
+        cmd.stdin(Stdio::null());
+        cmd.stdout(Stdio::null());
+        cmd.stderr(Stdio::null());
+
+        // Spawn the process
+        let child = cmd
+            .spawn()
+            .context("Failed to spawn loopback cleanup helper")?;
+
+        Ok(LoopbackCleanupHandle { child })
+    }
+
     // Shared backend for our `close` and `drop` implementations.
     fn impl_close(&mut self) -> Result<()> {
         // SAFETY: This is the only place we take the option
         let Some(dev) = self.dev.take() else {
             tracing::trace!("loopback device already deallocated");
             return Ok(());
         };
+
+        // Kill the cleanup helper since we're cleaning up normally
+        if let Some(mut cleanup_handle) = self.cleanup_handle.take() {
+            // Send SIGTERM to the child process
+            let _ = cleanup_handle.child.kill();
+        }
+
         Command::new("losetup").args(["-d", dev.as_str()]).run()
     }
 
@@ -240,6 +292,46 @@ impl Drop for LoopbackDevice {
     }
 }
 
+/// Main function for the loopback cleanup helper process
+/// This function does not return - it either exits normally or via signal
+pub async fn run_loopback_cleanup_helper(device_path: &str) -> Result<()> {
+    // Check if we're running as a cleanup helper
+    if std::env::var("BOOTC_LOOPBACK_CLEANUP_HELPER").is_err() {
+        anyhow::bail!("This function should only be called as a cleanup helper");
+    }
+
+    // Set up death signal notification - we want to be notified when parent dies
+    rustix::process::set_parent_process_death_signal(Some(rustix::process::Signal::TERM))
+        .context("Failed to set parent death signal")?;
+
+    // Wait for SIGTERM (either from parent death or normal cleanup)
+    tokio::signal::unix::signal(tokio::signal::unix::SignalKind::terminate())
+        .expect("Failed to create signal stream")
+        .recv()
+        .await;
+
+    // Clean up the loopback device
+    let status = std::process::Command::new("losetup")
+        .args(["-d", device_path])
+        .status();
+
+    match status {
+        Ok(exit_status) if exit_status.success() => {
+            // Log to systemd journal instead of stderr
+            tracing::info!("Cleaned up leaked loopback device {}", device_path);
+            std::process::exit(0);
+        }
+        Ok(_) => {
+            tracing::error!("Failed to clean up loopback device {}", device_path);
+            std::process::exit(1);
+        }
+        Err(e) => {
+            tracing::error!("Error cleaning up loopback device {}: {}", device_path, e);
+            std::process::exit(1);
+        }
+    }
+}
+
 /// Parse key-value pairs from lsblk --pairs.
 /// Newer versions of lsblk support JSON but the one in CentOS 7 doesn't.
 fn split_lsblk_line(line: &str) -> HashMap<String, String> {

crates/lib/src/cli.rs

@@ -464,6 +464,12 @@ pub(crate) enum InternalsOpts {
         #[clap(allow_hyphen_values = true)]
         args: Vec<OsString>,
     },
+    /// Loopback device cleanup helper (internal use only)
+    LoopbackCleanupHelper {
+        /// Device path to clean up
+        #[clap(long)]
+        device: String,
+    },
     /// Invoked from ostree-ext to complete an installation.
     BootcInstallCompletion {
         /// Path to the sysroot
@@ -1261,6 +1267,9 @@ async fn run_from_opt(opt: Opt) -> Result<()> {
                 let rootfs = &Dir::open_ambient_dir("/", cap_std::ambient_authority())?;
                 crate::install::completion::run_from_ostree(rootfs, &sysroot, &stateroot).await
             }
+            InternalsOpts::LoopbackCleanupHelper { device } => {
+                crate::blockdev::run_loopback_cleanup_helper(&device).await
+            }
             #[cfg(feature = "rhsm")]
             InternalsOpts::PublishRhsmFacts => crate::rhsm::publish_facts(&root).await,
         },

crates/lib/src/lib.rs

@@ -38,3 +38,6 @@ mod kernel;
 
 #[cfg(feature = "rhsm")]
 mod rhsm;
+
+// Re-export blockdev crate for internal use
+pub(crate) use bootc_blockdev as blockdev;