From 0a8aaa2c249f0aa647ed7e738fbc7a2eb5de7813 Mon Sep 17 00:00:00 2001
From: Patrick Insinger <patrickinsinger@gmail.com>
Date: Wed, 29 Sep 2021 16:07:03 -0700
Subject: [PATCH] zenith_utils - add crashsafe_dir

Utility for creating directories and directory trees in a crash safe
manor.

Minimizes calls to fsync for trees.
---
 Cargo.lock                        |   1 +
 zenith_utils/Cargo.toml           |   1 +
 zenith_utils/src/crashsafe_dir.rs | 125 ++++++++++++++++++++++++++++++
 zenith_utils/src/lib.rs           |   3 +
 4 files changed, 130 insertions(+)
 create mode 100644 zenith_utils/src/crashsafe_dir.rs

diff --git a/Cargo.lock b/Cargo.lock
index 47493c6efd..b168bf8333 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -2606,6 +2606,7 @@ dependencies = [
  "slog-scope",
  "slog-stdlog",
  "slog-term",
+ "tempfile",
  "thiserror",
  "tokio",
  "webpki",
diff --git a/zenith_utils/Cargo.toml b/zenith_utils/Cargo.toml
index f315a473af..22c1c9bab6 100644
--- a/zenith_utils/Cargo.toml
+++ b/zenith_utils/Cargo.toml
@@ -38,3 +38,4 @@ rustls-split = "0.2.1"
 hex-literal = "0.3"
 bytes = "1.0"
 webpki = "0.21"
+tempfile = "3.2"
diff --git a/zenith_utils/src/crashsafe_dir.rs b/zenith_utils/src/crashsafe_dir.rs
new file mode 100644
index 0000000000..fcb4ddcce3
--- /dev/null
+++ b/zenith_utils/src/crashsafe_dir.rs
@@ -0,0 +1,125 @@
+use std::{
+    fs::{self, File},
+    io,
+    path::Path,
+};
+
+/// Similar to [`std::fs::create_dir`], except we fsync the
+/// created directory and its parent.
+pub fn create_dir(path: impl AsRef<Path>) -> io::Result<()> {
+    let path = path.as_ref();
+
+    fs::create_dir(path)?;
+    File::open(path)?.sync_all()?;
+
+    if let Some(parent) = path.parent() {
+        File::open(parent)?.sync_all()
+    } else {
+        Err(io::Error::new(
+            io::ErrorKind::InvalidInput,
+            "can't find parent",
+        ))
+    }
+}
+
+/// Similar to [`std::fs::create_dir_all`], except we fsync all
+/// newly created directories and the pre-existing parent.
+pub fn create_dir_all(path: impl AsRef<Path>) -> io::Result<()> {
+    let mut path = path.as_ref();
+
+    let mut dirs_to_create = Vec::new();
+
+    // Figure out which directories we need to create.
+    loop {
+        match path.metadata() {
+            Ok(metadata) if metadata.is_dir() => break,
+            Ok(_) => {
+                return Err(io::Error::new(
+                    io::ErrorKind::AlreadyExists,
+                    format!("non-directory found in path: {:?}", path),
+                ));
+            }
+            Err(ref e) if e.kind() == io::ErrorKind::NotFound => {}
+            Err(e) => return Err(e),
+        }
+
+        dirs_to_create.push(path);
+
+        match path.parent() {
+            Some(parent) => path = parent,
+            None => {
+                return Err(io::Error::new(
+                    io::ErrorKind::InvalidInput,
+                    "can't find parent",
+                ))
+            }
+        }
+    }
+
+    // Create directories from parent to child.
+    for &path in dirs_to_create.iter().rev() {
+        fs::create_dir(path)?;
+    }
+
+    // Fsync the created directories from child to parent.
+    for &path in dirs_to_create.iter() {
+        File::open(path)?.sync_all()?;
+    }
+
+    // If we created any new directories, fsync the parent.
+    if !dirs_to_create.is_empty() {
+        File::open(path)?.sync_all()?;
+    }
+
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use tempfile::tempdir;
+
+    use super::*;
+
+    #[test]
+    fn test_create_dir_fsyncd() {
+        let dir = tempdir().unwrap();
+
+        let existing_dir_path = dir.path();
+        let err = create_dir(existing_dir_path).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::AlreadyExists);
+
+        let child_dir = existing_dir_path.join("child");
+        create_dir(child_dir).unwrap();
+
+        let nested_child_dir = existing_dir_path.join("child1").join("child2");
+        let err = create_dir(nested_child_dir).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::NotFound);
+    }
+
+    #[test]
+    fn test_create_dir_all_fsyncd() {
+        let dir = tempdir().unwrap();
+
+        let existing_dir_path = dir.path();
+        create_dir_all(existing_dir_path).unwrap();
+
+        let child_dir = existing_dir_path.join("child");
+        assert!(!child_dir.exists());
+        create_dir_all(&child_dir).unwrap();
+        assert!(child_dir.exists());
+
+        let nested_child_dir = existing_dir_path.join("child1").join("child2");
+        assert!(!nested_child_dir.exists());
+        create_dir_all(&nested_child_dir).unwrap();
+        assert!(nested_child_dir.exists());
+
+        let file_path = existing_dir_path.join("file");
+        std::fs::write(&file_path, b"").unwrap();
+
+        let err = create_dir_all(&file_path).unwrap_err();
+        assert_eq!(err.kind(), io::ErrorKind::AlreadyExists);
+
+        let invalid_dir_path = file_path.join("folder");
+        create_dir_all(&invalid_dir_path).unwrap_err();
+    }
+}
diff --git a/zenith_utils/src/lib.rs b/zenith_utils/src/lib.rs
index 142c65e6f1..ca26be5df2 100644
--- a/zenith_utils/src/lib.rs
+++ b/zenith_utils/src/lib.rs
@@ -18,6 +18,9 @@ pub mod pq_proto;
 // dealing with connstring parsing and handy access to it's parts
 pub mod connstring;
 
+// helper functions for creating and fsyncing directories/trees
+pub mod crashsafe_dir;
+
 // common authentication routines
 pub mod auth;