llmenv 1.0.5

//! XDG paths and path helpers.

use std::path::{Path, PathBuf};

/// Expand a leading `~` or `~/` to `$HOME`. Other input is returned unchanged.
/// Returns the input unchanged when `HOME` is unset.
#[must_use]
pub fn expand_tilde(p: &str) -> String {
    let Ok(home) = std::env::var("HOME") else {
        return p.to_string();
    };
    if let Some(rest) = p.strip_prefix("~/") {
        format!("{home}/{rest}")
    } else if p == "~" {
        home
    } else {
        p.to_string()
    }
}

/// True if `path` contains any parent (`..`) component, parsed
/// component-wise rather than by substring. Catches traversal that string
/// matching misses: `foo/..`, mixed separators on the host OS, and a bare
/// `..` with no trailing slash. A leading `/` (root) is fine; only `..`
/// components are rejected.
///
/// Note: this does NOT check whether `path` is absolute. `Path::join` with
/// an absolute argument returns the argument unchanged, escaping the base
/// directory. When validating relative paths supplied by user-controlled
/// data, use [`is_unsafe_join_target`] instead.
#[must_use]
pub fn has_parent_component(path: &str) -> bool {
    use std::path::Component;
    Path::new(path)
        .components()
        .any(|c| matches!(c, Component::ParentDir))
}

/// True if joining `path` onto a base directory would escape it. Returns
/// true when `path` contains `..` components OR is absolute (since
/// `Path::join` with an absolute argument discards the base). Use this at
/// every site that does `base.join(user_controlled_rel)`.
#[must_use]
pub fn is_unsafe_join_target(path: &str) -> bool {
    let p = Path::new(path);
    p.is_absolute() || has_parent_component(path)
}

/// Return true if `cwd` is at or below `prefix`, treating both as filesystem
/// paths (component-wise) rather than raw strings. This avoids the
/// `/home/alice/git/xyz` matches prefix `/home/alice/git/x` bug.
#[must_use]
pub fn cwd_under_prefix(cwd: &str, prefix: &str) -> bool {
    let cwd_p = Path::new(cwd);
    let pre_p = PathBuf::from(prefix);
    cwd_p.starts_with(&pre_p)
}

pub fn config_dir() -> anyhow::Result<PathBuf> {
    if let Ok(dir) = std::env::var("LLMENV_CONFIG_DIR") {
        Ok(PathBuf::from(dir))
    } else {
        let home = std::env::var("HOME")?;
        Ok(PathBuf::from(home).join(".config/llmenv"))
    }
}

pub fn config_path() -> anyhow::Result<PathBuf> {
    Ok(config_dir()?.join("config.yaml"))
}

pub fn state_dir() -> anyhow::Result<PathBuf> {
    if let Ok(dir) = std::env::var("LLMENV_STATE_DIR") {
        Ok(PathBuf::from(dir))
    } else {
        let home = std::env::var("HOME")?;
        Ok(PathBuf::from(home).join(".local/state/llmenv"))
    }
}

/// Write `content` to `path` with owner-only permissions (mode 0o600) on Unix.
/// On Windows falls back to default permissions. Creates the file if absent,
/// truncates if present. Use for any file containing user state or
/// credentials (settings, sync state, MCP configs, ICM memory) where
/// world-readable defaults would leak data on shared systems.
pub fn write_owner_only(path: &Path, content: &[u8]) -> std::io::Result<()> {
    use std::io::Write;
    #[cfg(unix)]
    {
        use std::os::unix::fs::OpenOptionsExt;
        let mut file = std::fs::OpenOptions::new()
            .write(true)
            .create(true)
            .truncate(true)
            .mode(0o600)
            .open(path)?;
        file.write_all(content)?;
    }
    #[cfg(not(unix))]
    {
        std::fs::write(path, content)?;
    }
    Ok(())
}

/// Atomically write `content` to `path` with owner-only permissions.
///
/// Steps: write to a same-directory temp file `<path>.<pid>.<nanos>.tmp`,
/// `fsync` it for durability, then `rename` over the destination (POSIX
/// atomic replace). Readers observing `path` mid-write see either the prior
/// good contents or the new contents — never a torn document. On error the
/// temp file is removed.
///
/// Use for any structured/JSON state file where a half-written file would
/// break the next read: `icm.json`, `sync.json`, `settings.json`, `mcp.json`.
pub fn write_owner_only_atomic(path: &Path, content: &[u8]) -> std::io::Result<()> {
    let parent = path.parent().ok_or_else(|| {
        std::io::Error::new(
            std::io::ErrorKind::InvalidInput,
            format!("path has no parent: {}", path.display()),
        )
    })?;
    let file_name = path.file_name().ok_or_else(|| {
        std::io::Error::new(
            std::io::ErrorKind::InvalidInput,
            format!("path has no file name: {}", path.display()),
        )
    })?;
    if parent.as_os_str().is_empty() {
        // For paths like "foo.json" (no parent dir), use current dir.
        return write_owner_only_atomic_in_dir(Path::new("."), file_name, path, content);
    }
    std::fs::create_dir_all(parent)?;
    // Harden parent dir to 0o700 (owner-only). Without this, default umask
    // 0o022 leaves the state dir at 0o755 (world-listable), leaking the
    // existence and names of state files on shared systems. Failure is
    // non-fatal — on platforms that don't support it (Windows), or if the
    // dir was created by another process and we lack chmod rights, we still
    // proceed with the file-level 0o600 protection.
    #[cfg(unix)]
    {
        use std::os::unix::fs::PermissionsExt;
        let _ = std::fs::set_permissions(parent, std::fs::Permissions::from_mode(0o700));
    }
    write_owner_only_atomic_in_dir(parent, file_name, path, content)
}

/// Process-local counter used to disambiguate temp filenames when multiple
/// calls within the same process land in the same nanosecond. Combined with
/// `pid` and `nanos`, this guarantees uniqueness within a process and is
/// extremely unlikely to collide across processes (different pids).
static TMP_COUNTER: std::sync::atomic::AtomicU64 = std::sync::atomic::AtomicU64::new(0);

fn write_owner_only_atomic_in_dir(
    parent: &Path,
    file_name: &std::ffi::OsStr,
    final_path: &Path,
    content: &[u8],
) -> std::io::Result<()> {
    use std::io::Write;
    use std::time::{SystemTime, UNIX_EPOCH};

    let pid = std::process::id();
    let nanos = SystemTime::now()
        .duration_since(UNIX_EPOCH)
        .map(|d| d.as_nanos())
        .unwrap_or(0);

    // Retry on EEXIST up to a small number of times. A stale temp file (from
    // a prior crashed process with the same pid+nanos slice) or in-process
    // race could collide; the per-process counter and retry loop together
    // guarantee progress without unbounded blocking.
    let mut last_err: Option<std::io::Error> = None;
    for _ in 0..8 {
        let counter = TMP_COUNTER.fetch_add(1, std::sync::atomic::Ordering::Relaxed);
        let mut tmp_name = file_name.to_os_string();
        tmp_name.push(format!(".{pid}.{nanos}.{counter}.tmp"));
        let tmp_path = parent.join(&tmp_name);

        let result = (|| -> std::io::Result<()> {
            #[cfg(unix)]
            {
                use std::os::unix::fs::OpenOptionsExt;
                let mut file = std::fs::OpenOptions::new()
                    .write(true)
                    .create_new(true)
                    .mode(0o600)
                    .open(&tmp_path)?;
                file.write_all(content)?;
                file.sync_all()?;
            }
            #[cfg(not(unix))]
            {
                let mut file = std::fs::OpenOptions::new()
                    .write(true)
                    .create_new(true)
                    .open(&tmp_path)?;
                file.write_all(content)?;
                file.sync_all()?;
            }
            std::fs::rename(&tmp_path, final_path)?;
            Ok(())
        })();

        match result {
            Ok(()) => return Ok(()),
            Err(e) if e.kind() == std::io::ErrorKind::AlreadyExists => {
                last_err = Some(e);
                continue;
            }
            Err(e) => {
                let _ = std::fs::remove_file(&tmp_path);
                return Err(e);
            }
        }
    }
    Err(last_err.unwrap_or_else(|| {
        std::io::Error::new(
            std::io::ErrorKind::AlreadyExists,
            "exhausted temp-file collision retries",
        )
    }))
}

#[cfg(test)]
#[allow(clippy::unwrap_used, clippy::expect_used, clippy::panic)]
mod tests {
    use super::*;

    #[test]
    fn cwd_under_prefix_respects_component_boundary() {
        assert!(cwd_under_prefix("/home/alice/git/x", "/home/alice/git/x"));
        assert!(cwd_under_prefix(
            "/home/alice/git/x/sub",
            "/home/alice/git/x"
        ));
        assert!(!cwd_under_prefix(
            "/home/alice/git/xyz",
            "/home/alice/git/x"
        ));
        assert!(!cwd_under_prefix("/home/alice", "/home/alice/git"));
    }

    #[test]
    fn has_parent_component_detects_traversal_substring_misses() {
        // Trailing `..` with no slash — substring check for "../" misses this.
        assert!(has_parent_component("foo/.."));
        assert!(has_parent_component(".."));
        assert!(has_parent_component("/foo/../bar"));
        assert!(has_parent_component("a/b/../c"));
    }

    #[test]
    fn has_parent_component_allows_safe_paths() {
        assert!(!has_parent_component("/home/alice/.cache/llmenv"));
        assert!(!has_parent_component("relative/path"));
        assert!(!has_parent_component("~/.cache/llmenv"));
        // A `..` embedded in a name is not a parent component.
        assert!(!has_parent_component("/foo/..bar/baz"));
        assert!(!has_parent_component("file..txt"));
        assert!(!has_parent_component(""));
    }

    #[test]
    fn has_parent_component_does_not_check_absolute_paths() {
        // Documents that has_parent_component alone is INSUFFICIENT for
        // safe-join validation. Callers must use is_unsafe_join_target.
        assert!(!has_parent_component("/etc/passwd"));
        assert!(!has_parent_component("/abs/secret"));
    }

    #[test]
    fn is_unsafe_join_target_rejects_traversal_and_absolute() {
        // Parent components — same as has_parent_component.
        assert!(is_unsafe_join_target(".."));
        assert!(is_unsafe_join_target("foo/.."));
        assert!(is_unsafe_join_target("a/b/../c"));
        // Absolute paths — would escape via Path::join semantics.
        assert!(is_unsafe_join_target("/etc/passwd"));
        assert!(is_unsafe_join_target("/abs"));
        // Safe: plain relative paths.
        assert!(!is_unsafe_join_target("rel/path"));
        assert!(!is_unsafe_join_target("file.txt"));
        assert!(!is_unsafe_join_target("a/b/c"));
        // Embedded `..` in a name is not a parent component.
        assert!(!is_unsafe_join_target("file..txt"));
    }

    #[cfg(unix)]
    #[test]
    fn write_owner_only_sets_mode_0o600() {
        use std::os::unix::fs::PermissionsExt;
        let tmp = tempfile::TempDir::new().expect("tempdir");
        let path = tmp.path().join("secret");
        write_owner_only(&path, b"sensitive").expect("write");
        let mode = std::fs::metadata(&path)
            .expect("metadata")
            .permissions()
            .mode();
        // Group/other bits must be clear — file is owner-only.
        assert_eq!(mode & 0o077, 0, "group/other bits set: {mode:o}");
        let body = std::fs::read(&path).expect("read");
        assert_eq!(body, b"sensitive");
    }

    #[cfg(unix)]
    #[test]
    fn write_owner_only_truncates_existing_file() {
        let tmp = tempfile::TempDir::new().expect("tempdir");
        let path = tmp.path().join("file");
        write_owner_only(&path, b"longer content").expect("write1");
        write_owner_only(&path, b"short").expect("write2");
        let body = std::fs::read(&path).expect("read");
        assert_eq!(body, b"short");
    }

    #[cfg(unix)]
    #[test]
    fn write_owner_only_atomic_creates_file_with_mode_0o600() {
        use std::os::unix::fs::PermissionsExt;
        let tmp = tempfile::TempDir::new().expect("tempdir");
        let path = tmp.path().join("atomic");
        write_owner_only_atomic(&path, b"payload").expect("atomic write");
        let mode = std::fs::metadata(&path)
            .expect("metadata")
            .permissions()
            .mode();
        assert_eq!(mode & 0o077, 0, "group/other bits set: {mode:o}");
        assert_eq!(std::fs::read(&path).expect("read"), b"payload");
    }

    #[test]
    fn write_owner_only_atomic_replaces_existing_file() {
        let tmp = tempfile::TempDir::new().expect("tempdir");
        let path = tmp.path().join("file");
        write_owner_only_atomic(&path, b"v1").expect("v1");
        write_owner_only_atomic(&path, b"v2-longer").expect("v2");
        assert_eq!(std::fs::read(&path).expect("read"), b"v2-longer");
    }

    #[test]
    fn write_owner_only_atomic_leaves_no_temp_files() {
        let tmp = tempfile::TempDir::new().expect("tempdir");
        let path = tmp.path().join("file");
        write_owner_only_atomic(&path, b"x").expect("write");
        write_owner_only_atomic(&path, b"y").expect("write");
        let entries: Vec<_> = std::fs::read_dir(tmp.path())
            .expect("read_dir")
            .filter_map(Result::ok)
            .map(|e| e.file_name())
            .collect();
        assert_eq!(entries.len(), 1, "found stray files: {entries:?}");
    }

    #[test]
    fn write_owner_only_atomic_creates_parent_dir() {
        let tmp = tempfile::TempDir::new().expect("tempdir");
        let path = tmp.path().join("a/b/c/file.json");
        write_owner_only_atomic(&path, b"nested").expect("write");
        assert_eq!(std::fs::read(&path).expect("read"), b"nested");
    }

    #[test]
    fn write_owner_only_atomic_concurrent_writers_no_torn_reads() {
        // Spawn N threads writing distinct fixed-size payloads to the same
        // path. Every reader sees one of the written payloads — never a
        // partial document, never an empty file.
        let tmp = tempfile::TempDir::new().expect("tempdir");
        let path = tmp.path().join("contended.json");
        write_owner_only_atomic(&path, b"initial").expect("seed");

        let payloads: Vec<Vec<u8>> = (0..8)
            .map(|i| format!("{{\"writer\":{i},\"data\":\"{}\"}}", "x".repeat(256)).into_bytes())
            .collect();
        let valid: std::collections::HashSet<Vec<u8>> = std::iter::once(b"initial".to_vec())
            .chain(payloads.iter().cloned())
            .collect();

        let writers: Vec<_> = payloads
            .into_iter()
            .map(|payload| {
                let p = path.clone();
                std::thread::spawn(move || {
                    for _ in 0..20 {
                        write_owner_only_atomic(&p, &payload).expect("concurrent write");
                    }
                })
            })
            .collect();

        let reader_path = path.clone();
        let reader_valid = valid.clone();
        let reader = std::thread::spawn(move || {
            for _ in 0..200 {
                let body = std::fs::read(&reader_path).expect("concurrent read");
                assert!(
                    reader_valid.contains(&body),
                    "reader observed torn write: {body:?}"
                );
            }
        });

        for w in writers {
            w.join().expect("writer join");
        }
        reader.join().expect("reader join");
    }

    #[test]
    fn tilde_passthrough_for_absolute_and_relative() {
        // Tests the non-HOME-dependent branches.
        assert_eq!(expand_tilde("/abs/path"), "/abs/path");
        assert_eq!(expand_tilde("rel/path"), "rel/path");
        assert_eq!(expand_tilde(""), "");
    }

    // ===== Property tests for atomic-write byte roundtrip (#156 / #157) =====

    use proptest::prelude::*;

    proptest! {
        // Arbitrary byte payloads written through write_owner_only_atomic must
        // round-trip exactly via fs::read. Catches truncation, encoding, or
        // mid-write corruption regressions across the full u8 range including
        // NUL bytes and high-bit values.
        #[test]
        fn atomic_write_byte_roundtrip(payload in proptest::collection::vec(any::<u8>(), 0..8192)) {
            let dir = tempfile::TempDir::new().expect("tempdir");
            let path = dir.path().join("payload.bin");
            write_owner_only_atomic(&path, &payload).expect("atomic write");
            let read = std::fs::read(&path).expect("read");
            prop_assert_eq!(payload, read);
        }

        // Repeated overwrites must end with the final payload exactly — no
        // residual bytes from prior writes, no torn state, no permission
        // escalation.
        #[test]
        fn atomic_write_overwrite_idempotent(
            first in proptest::collection::vec(any::<u8>(), 0..4096),
            second in proptest::collection::vec(any::<u8>(), 0..4096),
        ) {
            let dir = tempfile::TempDir::new().expect("tempdir");
            let path = dir.path().join("payload.bin");
            write_owner_only_atomic(&path, &first).expect("write 1");
            write_owner_only_atomic(&path, &second).expect("write 2");
            let read = std::fs::read(&path).expect("read");
            prop_assert_eq!(second, read);

            #[cfg(unix)]
            {
                use std::os::unix::fs::PermissionsExt;
                let mode = std::fs::metadata(&path).expect("meta").permissions().mode();
                prop_assert_eq!(mode & 0o077, 0, "group/other bits set after overwrite: {:o}", mode);
            }
        }
    }
}