//! PID 1 of a job's microVM (docs/m0.md §2). It mounts the read-only
//! rootfs and the workspace, runs the job's argv as an unprivileged user
//! with exactly the job's environment, frames the output on the serial
//! console for the host, and powers off. It never decides anything the
//! host has not: the job comes from `/job.json` in the initramfs the host
//! built, and the host enforces the time limit.
//!
//! Frames, one per line on the console: `\x1eO ` stdout, `\x1eE ` stderr,
//! `\x1eX <code>` the exit code (128 + signal when killed), `\x1eF <text>`
//! a failure of init itself, `\x1eC <json>` a build cache pruned before
//! the guest powers off (oxygenflow#72): `{"target": ..., "pruned":
//! flow_init::prune::Pruned}`, sent before the exit code.

use flow_init::prune;
use serde::Deserialize;
use std::ffi::CString;
use std::io::{Read, Write};
use std::os::fd::FromRawFd;
use std::sync::{Arc, Mutex};

mod egress;
mod job;
use job::{dev_links, job_env, loopback_up, resolve_program, PATH};

#[derive(Deserialize)]
struct JobSpec {
    argv: Vec<String>,
    env: Vec<(String, String)>,
    #[serde(default)]
    mounts: Vec<MountSpec>,
    /// `network "loopback"`: bring up `lo` (the guest has no other device).
    #[serde(default)]
    loopback: bool,
    /// `network "egress"` (oxygenflow#55): `lo` comes up and a proxy on
    /// 127.0.0.1:3128 relays over vsock to the runner.
    #[serde(default)]
    egress: bool,
    /// Secrets to write before the job starts (spec §9): each at `path`
    /// under `/run/secrets`, 0400 and the job's user's, on a tmpfs.
    #[serde(default)]
    secrets: Vec<SecretSpec>,
}

#[derive(serde::Deserialize)]
struct SecretSpec {
    path: String,
    value: String,
}

/// An ephemeral mount from the host: a tmpfs of `size_mib`, or the disk at
/// `device`, mounted at `target` under the new root.
#[derive(serde::Deserialize)]
struct MountSpec {
    kind: String,
    target: String,
    size_mib: u32,
    #[serde(default)]
    device: Option<String>,
    /// A build cache's disk (oxygenflow#72): mounted `strictatime`, so
    /// what the job reads is seen, and after a passing job pruned to this
    /// percentage of its size before the guest powers off.
    #[serde(default)]
    prune: Option<u8>,
    /// Paths on the disk, relative to it, that pruning leaves: the job's
    /// outputs, which the runner reads after the guest stops.
    #[serde(default)]
    keep: Vec<String>,
}

fn main() {
    let console = Arc::new(Mutex::new(std::io::stdout()));
    let code = match run(&console) {
        Ok(code) => code,
        Err(e) => {
            frame(&console, b'F', e.as_bytes());
            125
        }
    };
    frame(&console, b'X', code.to_string().as_bytes());
    unsafe {
        libc::sync();
        // Firecracker ends the VM when the guest reboots (reboot=k).
        libc::reboot(libc::LINUX_REBOOT_CMD_RESTART);
    }
}

fn frame(console: &Mutex<std::io::Stdout>, kind: u8, text: &[u8]) {
    let mut out = console.lock().unwrap_or_else(|p| p.into_inner());
    let mut line = Vec::with_capacity(text.len() + 4);
    line.extend_from_slice(&[0x1e, kind, b' ']);
    line.extend_from_slice(text);
    line.push(b'\n');
    let _ = out.write_all(&line);
    let _ = out.flush();
}

fn cstr(s: &str) -> Result<CString, String> {
    CString::new(s).map_err(|_| format!("{s:?} contains NUL"))
}

fn chown(path: &str, uid: libc::uid_t, gid: libc::gid_t) -> Result<(), String> {
    let p = cstr(path)?;
    if unsafe { libc::chown(p.as_ptr(), uid, gid) } != 0 {
        return Err(format!("chown {path}: {}", std::io::Error::last_os_error()));
    }
    Ok(())
}

/// Create `path` and its missing parents, each owned by the job's user.
fn make_dirs_owned(path: &str, uid: libc::uid_t, gid: libc::gid_t) -> Result<(), String> {
    let mut missing = Vec::new();
    let mut p = std::path::Path::new(path);
    while !p.exists() {
        missing.push(p.to_path_buf());
        p = match p.parent() {
            Some(parent) => parent,
            None => break,
        };
    }
    for d in missing.iter().rev() {
        std::fs::create_dir(d).map_err(|e| format!("mkdir {}: {e}", d.display()))?;
        chown(&d.to_string_lossy(), uid, gid)?;
    }
    Ok(())
}

fn mount(src: &str, target: &str, fstype: &str, flags: libc::c_ulong, data: &str) -> Result<(), String> {
    let (s, t, f, d) = (cstr(src)?, cstr(target)?, cstr(fstype)?, cstr(data)?);
    let rc = unsafe { libc::mount(s.as_ptr(), t.as_ptr(), f.as_ptr(), flags, d.as_ptr() as *const libc::c_void) };
    if rc != 0 {
        return Err(format!("mount {src} on {target}: {}", std::io::Error::last_os_error()));
    }
    Ok(())
}

fn run(console: &Arc<Mutex<std::io::Stdout>>) -> Result<i32, String> {
    let job: JobSpec = serde_json::from_slice(&std::fs::read("/job.json").map_err(|e| format!("read /job.json: {e}"))?)
        .map_err(|e| format!("parse /job.json: {e}"))?;
    if job.argv.is_empty() {
        return Err("empty argv".into());
    }
    let nosuid = libc::MS_NOSUID | libc::MS_NODEV;
    std::fs::create_dir_all("/dev").ok();
    mount("devtmpfs", "/dev", "devtmpfs", libc::MS_NOSUID, "")?;
    std::fs::create_dir_all("/newroot").map_err(|e| e.to_string())?;
    mount("/dev/vda", "/newroot", "ext4", libc::MS_RDONLY | libc::MS_NODEV, "")?;
    mount("/dev/vdb", "/newroot/src", "ext4", nosuid, "")?;
    mount("tmpfs", "/newroot/tmp", "tmpfs", nosuid, "mode=1777")?;
    mount("proc", "/newroot/proc", "proc", nosuid | libc::MS_NOEXEC, "")?;
    mount("devtmpfs", "/newroot/dev", "devtmpfs", libc::MS_NOSUID, "")?;
    dev_links("/newroot/dev")?;
    if std::path::Path::new("/newroot/run").is_dir() {
        mount("tmpfs", "/newroot/run", "tmpfs", nosuid, "mode=755")?;
    }
    // The job runs as whoever owns the workspace (the host's builder), never root.
    let (uid, gid) = {
        let mut st: libc::stat = unsafe { std::mem::zeroed() };
        let p = cstr("/newroot/src")?;
        if unsafe { libc::stat(p.as_ptr(), &mut st) } != 0 {
            return Err("stat /src".into());
        }
        if st.st_uid == 0 {
            (1000, 1000)
        } else {
            (st.st_uid, st.st_gid)
        }
    };
    // Ephemeral mounts: mount points are made (and owned by the job's user)
    // under the writable /src or /tmp, then each is mounted and given to it.
    for m in &job.mounts {
        let target = format!("/newroot{}", m.target);
        make_dirs_owned(&target, uid, gid)?;
        match (m.kind.as_str(), &m.device) {
            ("tmpfs", _) => mount("tmpfs", &target, "tmpfs", nosuid, &format!("size={}m,uid={uid},gid={gid},mode=755", m.size_mib))?,
            ("disk", Some(dev)) => {
                let atime = if m.prune.is_some() { libc::MS_STRICTATIME | libc::MS_LAZYTIME } else { 0 };
                mount(dev, &target, "ext4", nosuid | atime, "")?;
                chown(&target, uid, gid)?;
            }
            (kind, _) => return Err(format!("unknown mount kind {kind}")),
        }
    }
    // Secrets: an in-memory mount of their own, only the job's user's,
    // each file 0400; the job file they came from is outside the new root.
    if !job.secrets.is_empty() {
        if !std::path::Path::new("/newroot/run").is_dir() {
            return Err("the image has no /run, where secrets are mounted".into());
        }
        let dir = "/newroot/run/secrets";
        std::fs::create_dir_all(dir).map_err(|e| format!("mkdir {dir}: {e}"))?;
        mount("tmpfs", dir, "tmpfs", nosuid | libc::MS_NOEXEC, &format!("size=1m,uid={uid},gid={gid},mode=700"))?;
        for s in &job.secrets {
            let Some(rel) = s.path.strip_prefix("/run/secrets/") else {
                return Err(format!("secret path {:?} is not under /run/secrets/", s.path));
            };
            let full = format!("{dir}/{rel}");
            if let Some(parent) = std::path::Path::new(&full).parent().and_then(|p| p.to_str()) {
                make_dirs_owned(parent, uid, gid)?;
            }
            std::fs::write(&full, s.value.as_bytes()).map_err(|e| format!("write {}: {e}", s.path))?;
            chown(&full, uid, gid)?;
            std::fs::set_permissions(&full, std::os::unix::fs::PermissionsExt::from_mode(0o400))
                .map_err(|e| format!("chmod {}: {e}", s.path))?;
        }
    }
    if job.loopback || job.egress {
        loopback_up()?;
    }
    if job.egress {
        egress::start()?;
    }
    let root = cstr("/newroot")?;
    if unsafe { libc::chroot(root.as_ptr()) } != 0 || std::env::set_current_dir("/src").is_err() {
        return Err(format!("enter the rootfs: {}", std::io::Error::last_os_error()));
    }

    // What the job reads or writes on a cache disk from now on is what
    // this run used.
    let started = job.mounts.iter().any(|m| m.prune.is_some()).then(prune::start_clock);
    let argv: Vec<CString> = job.argv.iter().map(|a| cstr(a)).collect::<Result<_, _>>()?;
    // glibc's execvpe searches the caller's PATH, and PID 1 has none of its
    // own, so the program is found here, in the job's PATH, before forking.
    let vars = job_env(std::fs::read_to_string("/etc/flow/env").ok().as_deref(), &job.env);
    let path = vars.get("PATH").map(String::as_str).unwrap_or(PATH);
    let program = cstr(&resolve_program(&job.argv[0], path).ok_or_else(|| format!("cannot run {:?}: not found in {path}", job.argv[0]))?)?;
    let env: Vec<CString> = vars.iter().map(|(k, v)| cstr(&format!("{k}={v}"))).collect::<Result<_, _>>()?;
    let mut out = [0; 2];
    let mut err = [0; 2];
    if unsafe { libc::pipe2(out.as_mut_ptr(), libc::O_CLOEXEC) } != 0 || unsafe { libc::pipe2(err.as_mut_ptr(), libc::O_CLOEXEC) } != 0 {
        return Err("pipe".into());
    }
    let pid = unsafe { libc::fork() };
    if pid < 0 {
        return Err("fork".into());
    }
    if pid == 0 {
        // Child: no way back to root, then the program itself (execvpe: no shell).
        unsafe {
            let null = libc::open(c"/dev/null".as_ptr(), libc::O_RDONLY);
            libc::dup2(null, 0);
            libc::dup2(out[1], 1);
            libc::dup2(err[1], 2);
            libc::setsid();
            if libc::setgroups(0, std::ptr::null()) != 0 || libc::setgid(gid) != 0 || libc::setuid(uid) != 0 {
                libc::_exit(126);
            }
            let mut a: Vec<*const libc::c_char> = argv.iter().map(|c| c.as_ptr()).collect();
            a.push(std::ptr::null());
            let mut e: Vec<*const libc::c_char> = env.iter().map(|c| c.as_ptr()).collect();
            e.push(std::ptr::null());
            libc::execve(program.as_ptr(), a.as_ptr(), e.as_ptr());
            let msg = format!("flow: cannot run {:?}: {}\n", job.argv[0], std::io::Error::last_os_error());
            libc::write(2, msg.as_ptr() as *const libc::c_void, msg.len());
            libc::_exit(127);
        }
    }
    unsafe {
        libc::close(out[1]);
        libc::close(err[1]);
    }
    let readers: Vec<_> = [(out[0], b'O'), (err[0], b'E')]
        .into_iter()
        .map(|(fd, kind)| {
            let console = console.clone();
            std::thread::spawn(move || pump(unsafe { std::fs::File::from_raw_fd(fd) }, kind, &console))
        })
        .collect();
    let mut status = 0;
    // Reap everything (we are PID 1) until the job itself exits.
    loop {
        let r = unsafe { libc::waitpid(-1, &mut status, 0) };
        if r == pid || r < 0 {
            break;
        }
    }
    for r in readers {
        let _ = r.join();
    }
    let code = if libc::WIFEXITED(status) { libc::WEXITSTATUS(status) } else { 128 + libc::WTERMSIG(status) };
    // A failing job's disks are thrown away: only a passing one's are
    // worth pruning.
    if let (Some(started), 0) = (started, code) {
        prune_caches(console, &job.mounts, started);
    }
    Ok(code)
}

/// Prune each cache disk and say so to the host. Whatever the job left
/// running stops first, so nothing changes the disks meanwhile.
fn prune_caches(console: &Mutex<std::io::Stdout>, mounts: &[MountSpec], started: std::time::SystemTime) {
    unsafe {
        libc::kill(-1, libc::SIGKILL);
        let mut status = 0;
        while libc::waitpid(-1, &mut status, 0) > 0 {}
    }
    for m in mounts {
        let Some(percent) = m.prune else { continue };
        let root = std::path::Path::new(&m.target);
        let pruned = match prune::usage(root) {
            Ok((used, size)) => {
                let mut p = prune::prune(root, started, &m.keep, size, used, percent).unwrap_or_else(|e| prune::Pruned {
                    size,
                    percent,
                    before: used,
                    after: used,
                    unpruned: Some(format!("pruning failed: {e}")),
                    ..Default::default()
                });
                // What is left as the file system counts it.
                if p.freed > 0 {
                    unsafe { libc::sync() };
                    if let Ok((now, _)) = prune::usage(root) {
                        p.after = now;
                    }
                }
                p
            }
            Err(e) => prune::Pruned { percent, unpruned: Some(format!("statvfs {}: {e}", m.target)), ..Default::default() },
        };
        let msg = serde_json::json!({ "target": m.target, "pruned": pruned });
        frame(console, b'C', msg.to_string().as_bytes());
    }
}

/// Frame a stream line by line; a long line is cut at 16 KiB.
fn pump(mut f: std::fs::File, kind: u8, console: &Mutex<std::io::Stdout>) {
    let mut buf = [0u8; 8192];
    let mut line: Vec<u8> = Vec::new();
    loop {
        let n = match f.read(&mut buf) {
            Ok(0) | Err(_) => break,
            Ok(n) => n,
        };
        for &b in &buf[..n] {
            if b == b'\n' || line.len() >= 16 * 1024 {
                frame(console, kind, &line);
                line.clear();
                if b == b'\n' {
                    continue;
                }
            }
            line.push(b);
        }
    }
    if !line.is_empty() {
        frame(console, kind, &line);
    }
}