runner: hold lock on task state dir while task is running

Do not allow concurrent runs of the same task.

2 files changed, 12 insertions, 2 deletions

diff --git a/crates/runner/src/paths.rs b/crates/runner/src/paths.rs
index 9539782..5bfef25 100644
--- a/crates/runner/src/paths.rs
+++ b/crates/runner/src/paths.rs
@@ -21,7 +21,9 @@
 //! │       │   ├── layer/            # overlayfs layer dir (moved to layer/ after build)
 //! │       │   ├── work/             # overlayfs work dir (discarded after build)
 //! │       │   ├── task.json.tmp     # during write
-//! │       │   └── task.json         # after write
+//! │       │   ├── task.json         # after write
+//! │       │   ├── task.log          # stdout/stderr output of the task
+//! │       │   └── task.lock         # task lockfile
 //! │       └── ...
 //! └── tmp/                          # tmpfs (mounted on start of rebel)
 //!     ├── rootfs/                   # unpacked rootfs.tar
@@ -95,6 +97,10 @@ pub fn task_log_filename(hash: &InputHash) -> String {
 	join(&[TASK_STATE_DIR, &hash.to_string(), "task.log"])
 }
 
+pub fn task_lock_filename(hash: &InputHash) -> String {
+	join(&[TASK_STATE_DIR, &hash.to_string(), "task.lock"])
+}
+
 pub fn layer_dir(hash: &LayerHash) -> String {
 	join(&[LAYER_STATE_DIR, &hash.to_string()])
 }
diff --git a/crates/runner/src/task.rs b/crates/runner/src/task.rs
index 659b04d..ff27399 100644
--- a/crates/runner/src/task.rs
+++ b/crates/runner/src/task.rs
@@ -24,7 +24,7 @@ use super::{
 	ns, tar,
 	util::{cjson, fs, Checkable},
 };
-use crate::paths;
+use crate::{paths, util::unix};
 
 const BUILD_UID: Uid = Uid::from_raw(1000);
 const BUILD_GID: Gid = Gid::from_raw(1000);
@@ -433,6 +433,10 @@ fn save_cached(input_hash: &InputHash, output: &TaskOutput) -> Result<()> {
 pub fn handle(task: Task, jobserver: Jobserver) -> Result<TaskOutput> {
 	let input_hash = input_hash(&task);
 
+	// TODO: We should not hold a job token while waiting on the lock
+	let _lock = unix::lock(paths::task_lock_filename(&input_hash), true, true)
+		.context("Failed to get task lock")?;
+
 	if let Ok(task_output) = load_cached(&input_hash) {
 		return Ok(task_output);
 	}