purple-ssh 3.15.18

Open-source terminal SSH manager that keeps ~/.ssh/config in sync with your cloud infra. Spin up a VM on AWS, GCP, Azure, Hetzner or 12 other cloud providers and it appears in your host list. Destroy it and the entry dims. Search hundreds of hosts, transfer files, manage Docker and Podman over SSH, sign Vault SSH certs. Rust TUI, MIT licensed.
Documentation
use std::fs;
use std::io;
use std::path::{Path, PathBuf};

use log::{debug, error};

/// Advisory file lock using a `.lock` file.
/// The lock is released when the `FileLock` is dropped.
pub struct FileLock {
    lock_path: PathBuf,
    #[cfg(unix)]
    _file: fs::File,
}

impl FileLock {
    /// Acquire an advisory lock for the given path.
    /// Creates a `.purple_lock` file alongside the target and holds an `flock` on it.
    /// Blocks until the lock is acquired (or returns an error on failure).
    pub fn acquire(path: &Path) -> io::Result<Self> {
        let mut lock_name = path.file_name().unwrap_or_default().to_os_string();
        lock_name.push(".purple_lock");
        let lock_path = path.with_file_name(lock_name);

        #[cfg(unix)]
        {
            use std::os::unix::fs::OpenOptionsExt;
            let file = fs::OpenOptions::new()
                .write(true)
                .create(true)
                .truncate(false)
                .mode(0o600)
                .open(&lock_path)?;

            // SAFETY: flock() is safe to call on any valid file descriptor.
            // The fd comes from a File we just opened and own. LOCK_EX
            // requests an exclusive advisory lock, blocking until acquired.
            let ret =
                unsafe { libc::flock(std::os::unix::io::AsRawFd::as_raw_fd(&file), libc::LOCK_EX) };
            if ret != 0 {
                return Err(io::Error::last_os_error());
            }

            Ok(FileLock {
                lock_path,
                _file: file,
            })
        }

        #[cfg(not(unix))]
        {
            // On non-Unix, use a simple lock file (best-effort)
            let file = fs::OpenOptions::new()
                .write(true)
                .create_new(true)
                .open(&lock_path)
                .or_else(|_| {
                    // If it already exists, wait briefly and retry
                    std::thread::sleep(std::time::Duration::from_millis(100));
                    fs::remove_file(&lock_path).ok();
                    fs::OpenOptions::new()
                        .write(true)
                        .create_new(true)
                        .open(&lock_path)
                })?;
            Ok(FileLock {
                lock_path,
                _file: file,
            })
        }
    }
}

impl Drop for FileLock {
    fn drop(&mut self) {
        // On Unix, flock is released when the file descriptor is closed (automatic).
        // The lockfile itself is intentionally left on disk: unlinking it here
        // creates a race where a second process opens a fresh inode at the
        // same path and obtains an independent flock, so both processes
        // think they hold the lock. Leaving the file (1 byte at chmod 600)
        // matches the standard advisory-lock pattern.
        // The `lock_path` field is kept for diagnostics (Debug, logging).
        let _ = &self.lock_path;
    }
}

/// Atomic write: write content to a PID-suffixed temp file with chmod 600, then rename.
/// Uses O_EXCL (create_new) to prevent symlink attacks on the temp file path.
/// Cleans up the temp file on failure.
///
/// When the target file already exists, its mode is preserved across the
/// rename — clamped to a minimum of 0o600 so a write never widens the
/// permission set of an SSH config file. A target with mode 0o644 stays
/// 0o644; a target with mode 0o400 is tightened from 0o600 (the temp file's
/// initial mode) up to 0o600 — i.e. the more restrictive of the two wins
/// only when it's still at least 0o600.
///
/// Logs a warning when the target is a hard link with more than one name:
/// `rename(2)` substitutes the inode atomically, so any sibling hard link
/// silently keeps the OLD content. Common dotfiles managers (chezmoi, stow)
/// use symlinks rather than hard links so this is rare, but worth surfacing.
pub fn atomic_write(path: &Path, content: &[u8]) -> io::Result<()> {
    debug!("Atomic write: {}", path.display());
    // Ensure parent directory exists
    if let Some(parent) = path.parent() {
        fs::create_dir_all(parent)?;
    }

    // Capture the target's existing mode (if it exists) so the rename does
    // not silently flip e.g. 0o644 to 0o600. `symlink_metadata` here would
    // miss the case where the target is a symlink we just resolved; use
    // `metadata` which follows.
    #[cfg(unix)]
    let target_mode: Option<u32> = {
        use std::os::unix::fs::MetadataExt;
        fs::metadata(path).ok().map(|m| m.mode() & 0o777)
    };

    // Detect hard-linked targets and emit a one-time warning so a user with
    // a dotfiles repo hard-linked into ~/.ssh sees why their other name
    // diverges after a save.
    #[cfg(unix)]
    if let Ok(meta) = fs::symlink_metadata(path) {
        use std::os::unix::fs::MetadataExt;
        if meta.nlink() > 1 {
            log::warn!(
                "[purple] {} has {} hard links; atomic write will keep this name's content but leave siblings pointing at the previous inode",
                path.display(),
                meta.nlink()
            );
        }
    }

    let mut tmp_name = path.file_name().unwrap_or_default().to_os_string();
    tmp_name.push(format!(".purple_tmp.{}", std::process::id()));
    let tmp_path = path.with_file_name(tmp_name);

    #[cfg(unix)]
    {
        use std::io::Write;
        use std::os::unix::fs::OpenOptionsExt;
        // Try O_EXCL first. If a stale tmp file exists from a crashed run, remove
        // it and retry once. This avoids a TOCTOU gap from removing before creating.
        let open = || {
            fs::OpenOptions::new()
                .write(true)
                .create_new(true)
                .mode(0o600)
                .open(&tmp_path)
        };
        let mut file = match open() {
            Ok(f) => f,
            Err(e) if e.kind() == io::ErrorKind::AlreadyExists => {
                let _ = fs::remove_file(&tmp_path);
                open().map_err(|e| {
                    io::Error::new(
                        e.kind(),
                        format!("Failed to create temp file {}: {}", tmp_path.display(), e),
                    )
                })?
            }
            Err(e) => {
                return Err(io::Error::new(
                    e.kind(),
                    format!("Failed to create temp file {}: {}", tmp_path.display(), e),
                ));
            }
        };
        if let Err(e) = file.write_all(content) {
            drop(file);
            let _ = fs::remove_file(&tmp_path);
            return Err(e);
        }
        if let Err(e) = file.sync_all() {
            drop(file);
            let _ = fs::remove_file(&tmp_path);
            return Err(e);
        }
        // Preserve the target's mode if it existed. Clamp to a minimum of
        // 0o600 so an SSH config file is never silently widened by this
        // write; 0o400 or 0o600 stays as-is, 0o644 stays 0o644, anything
        // wider keeps its original perms. Best-effort: a chmod failure
        // shouldn't abort the write (the temp file is already at 0o600).
        if let Some(mode) = target_mode {
            use std::os::unix::fs::PermissionsExt;
            let preserved = if mode < 0o600 { 0o600 } else { mode };
            if let Err(e) = fs::set_permissions(&tmp_path, fs::Permissions::from_mode(preserved)) {
                debug!(
                    "[purple] could not preserve target mode {:o} on {}: {e}",
                    preserved,
                    tmp_path.display()
                );
            }
        }
    }

    #[cfg(not(unix))]
    {
        if let Err(e) = fs::write(&tmp_path, content) {
            let _ = fs::remove_file(&tmp_path);
            return Err(e);
        }
        // sync_all via reopen since fs::write doesn't return a File handle
        match fs::File::open(&tmp_path) {
            Ok(f) => {
                if let Err(e) = f.sync_all() {
                    let _ = fs::remove_file(&tmp_path);
                    return Err(e);
                }
            }
            Err(e) => {
                let _ = fs::remove_file(&tmp_path);
                return Err(e);
            }
        }
    }

    let result = fs::rename(&tmp_path, path);
    if let Err(ref err) = result {
        let _ = fs::remove_file(&tmp_path);
        error!("[purple] Atomic write failed: {}: {err}", path.display());
        return result;
    }

    // Durable rename: the temp file's data was synced before rename, but the
    // directory entry change produced by `rename` lives in the page cache
    // until the parent directory itself is synced. Without this, a crash
    // within seconds of save can leave the directory pointing at the old
    // inode (= silently dropped edit). Best-effort: log but don't fail the
    // write if the parent sync itself fails — the rename already succeeded.
    #[cfg(unix)]
    if let Some(parent) = path.parent() {
        if let Err(err) = fs::File::open(parent).and_then(|d| d.sync_all()) {
            debug!(
                "[purple] parent dir sync after rename failed (rename succeeded): {}: {err}",
                parent.display()
            );
        }
    }

    result
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn file_lock_does_not_remove_lockfile_on_drop() {
        // Regression for the lockfile-unlink race: a second process can open
        // a new inode at the same path between fd-close and remove_file,
        // and then both processes hold flock on independent inodes.
        // Leaving the lockfile in place after drop prevents this entirely.
        let dir = tempfile::tempdir().expect("tempdir");
        let target = dir.path().join("config");
        let lockfile = dir.path().join("config.purple_lock");

        {
            let _lock = FileLock::acquire(&target).expect("acquire");
            assert!(lockfile.exists(), "lockfile must be created on acquire");
        }
        assert!(
            lockfile.exists(),
            "lockfile must remain after drop (not unlinked)"
        );
    }

    #[test]
    fn atomic_write_creates_file_with_content() {
        let dir = tempfile::tempdir().expect("tempdir");
        let target = dir.path().join("file");
        atomic_write(&target, b"hello\n").expect("write");
        let content = std::fs::read_to_string(&target).expect("read");
        assert_eq!(content, "hello\n");
    }

    #[test]
    fn atomic_write_replaces_existing_file() {
        let dir = tempfile::tempdir().expect("tempdir");
        let target = dir.path().join("file");
        std::fs::write(&target, b"old").expect("write old");
        atomic_write(&target, b"new").expect("write new");
        let content = std::fs::read_to_string(&target).expect("read");
        assert_eq!(content, "new");
    }

    #[test]
    fn atomic_write_leaves_no_temp_file() {
        let dir = tempfile::tempdir().expect("tempdir");
        let target = dir.path().join("file");
        atomic_write(&target, b"content").expect("write");
        let stem = target.file_name().unwrap().to_string_lossy().to_string();
        let leftovers: Vec<_> = std::fs::read_dir(dir.path())
            .unwrap()
            .filter_map(|e| e.ok())
            .filter(|e| {
                let n = e.file_name().to_string_lossy().to_string();
                n.starts_with(&format!("{}.purple_tmp.", stem))
            })
            .collect();
        assert!(
            leftovers.is_empty(),
            "temp file leaked after successful write: {:?}",
            leftovers.iter().map(|e| e.path()).collect::<Vec<_>>()
        );
    }
}