AnyBot/src/vnc.rs

// People watching and helping on the bots' computer: VNC, carried to them as a stream.
//
// The box already runs a VNC server (x11vnc, inside the container, with a password). A person's
// viewer (noVNC in a browser: the phone, or any computer through the account server) speaks VNC
// to this proxy, which reaches the box's server through one `docker exec` (a few lines of
// Python in the box joining its stdin/stdout to 127.0.0.1:5900). VNC sends only what changed on
// the screen and every mouse move and key as it happens, so watching is live and helping is
// like sitting at it — unlike pictures sent one by one.
//
// The proxy signs in to the box with its password itself and offers the viewer "no password",
// so the password never leaves this AnyBot; whoever reaches the proxy was already let in (an
// account session or one-time ticket at the account server, or the phone's password here).
// While the person clicks or types, the bot's own steps wait (HELD_BY_PERSON), and go on a
// little after they stop (person_active).

use std::io::{Read, Write};
use std::process::{Child, ChildStdin};
use std::sync::atomic::{AtomicUsize, Ordering};
use std::sync::Mutex;
use std::time::{Duration, Instant};

/// The bot waits this long after the person's last click or key before it goes on.
const HELD_AFTER_INPUT: Duration = Duration::from_secs(8);

/// Joins the box's VNC server to stdin/stdout, and ends when either side does (a loop left in
/// the box would hold the connection open).
const BRIDGE: &str = r#"import os,select,socket
s=socket.create_connection(('127.0.0.1',5900))
while True:
 r=select.select([s,0],[],[])[0]
 if s in r:
  d=s.recv(65536)
  if not d: break
  os.write(1,d)
 if 0 in r:
  d=os.read(0,65536)
  if not d: break
  s.sendall(d)
"#;

/// Viewers connected now (the box stays awake while there is one).
static OPEN: AtomicUsize = AtomicUsize::new(0);
/// When the person last clicked or typed through a viewer.
static LAST_INPUT: Mutex<Option<Instant>> = Mutex::new(None);

/// VNC authentication: the 16-byte challenge encrypted with DES, the key being the password's
/// first 8 bytes with each byte's bits reversed (RFC 6143 §7.2.2, as every VNC server does it).
pub fn vnc_response(password: &str, challenge: &[u8; 16]) -> [u8; 16] {
    use des::cipher::{BlockEncrypt, KeyInit};
    let mut key = [0u8; 8];
    for (k, b) in key.iter_mut().zip(password.bytes()) {
        *k = b.reverse_bits();
    }
    let c = des::Des::new_from_slice(&key).expect("8-byte key");
    let mut out = *challenge;
    for block in out.chunks_mut(8) {
        c.encrypt_block(des::cipher::generic_array::GenericArray::from_mut_slice(block));
    }
    out
}

/// This proxy's first words to a viewer, as a VNC server: version 3.8 (then "no password"; see
/// Session::feed).
pub const GREETING: &[u8] = b"RFB 003.008\n";

/// One person's viewer joined to the box.
pub struct Session {
    child: Child,
    stdin: ChildStdin,
    /// The viewer's side of the handshake still to read: its version (12 bytes) and its choice
    /// of security type (1 byte).
    greeting: usize,
    /// Viewer → box messages not yet complete (to notice clicks and keys).
    pending: Vec<u8>,
    sniffing: bool,
    /// The viewer's ClientInit (one byte) is still to come.
    init: bool,
}

/// Open a viewer's way to the box: sign in to its VNC server, and start sending what the server
/// says to `out` (from a thread of its own; `out` returns false when the viewer is gone, and
/// gets an empty slice when the box's side ended). The caller then sends the viewer GREETING
/// (once it is ready for the viewer's answer).
pub fn open(out: impl Fn(&[u8]) -> bool + Send + 'static) -> Result<Session, String> {
    crate::cloudbox::ensure_running()?;
    let mut child = crate::cloudbox::exec_piped(&["python3", "-c", BRIDGE])?;
    let mut stdin = child.stdin.take().ok_or("連不上這台電腦的畫面")?;
    let mut stdout = child.stdout.take().ok_or("連不上這台電腦的畫面")?;
    let fail = |child: &mut Child, why: &str| -> String {
        let _ = child.kill();
        let _ = child.wait();
        format!("連不上這台電腦的畫面：{why}")
    };
    let mut read = |n: usize| -> Result<Vec<u8>, String> {
        let mut b = vec![0u8; n];
        stdout.read_exact(&mut b).map(|_| b).map_err(|_| "畫面伺服器沒有回應".to_string())
    };
    // The box's side, as a VNC client: version 3.8, then its password.
    let signed_in = (|| -> Result<(), String> {
        let v = read(12)?;
        if !v.starts_with(b"RFB 003.") {
            return Err("不是 VNC 伺服器".into());
        }
        stdin.write_all(b"RFB 003.008\n").map_err(|e| e.to_string())?;
        let n = read(1)?[0] as usize;
        if n == 0 {
            return Err("畫面伺服器拒絕連線".into());
        }
        let types = read(n)?;
        if types.contains(&2) {
            stdin.write_all(&[2]).map_err(|e| e.to_string())?;
            let c: [u8; 16] = read(16)?.try_into().map_err(|_| "認證失敗".to_string())?;
            stdin.write_all(&vnc_response(&crate::cloudbox::box_vnc_password(), &c)).map_err(|e| e.to_string())?;
        } else if types.contains(&1) {
            stdin.write_all(&[1]).map_err(|e| e.to_string())?;
        } else {
            return Err("畫面伺服器要的登入方式不支援".into());
        }
        if read(4)? != [0, 0, 0, 0] {
            return Err("畫面伺服器的密碼不對".into());
        }
        Ok(())
    })();
    if let Err(e) = signed_in {
        return Err(fail(&mut child, &e));
    }
    OPEN.fetch_add(1, Ordering::SeqCst);
    crate::cloudbox::set_viewer_open(true);
    std::thread::spawn(move || {
        let mut buf = vec![0u8; 64 * 1024];
        loop {
            match stdout.read(&mut buf) {
                Ok(0) | Err(_) => break,
                Ok(n) => {
                    if !out(&buf[..n]) {
                        return;
                    }
                }
            }
        }
        // The box's side ended: an empty send says so.
        out(&[]);
    });
    Ok(Session { child, stdin, greeting: 13, pending: Vec::new(), sniffing: true, init: true })
}

impl Session {
    /// Bytes from the viewer. False when the box's side is gone.
    pub fn feed(&mut self, mut b: &[u8], reply: impl Fn(&[u8]) -> bool) -> bool {
        if self.greeting > 0 {
            let take = self.greeting.min(b.len());
            let before = self.greeting;
            self.greeting -= take;
            b = &b[take..];
            // After the version (12 bytes): the one security type, "None".
            if before > 1 && self.greeting <= 1 && !reply(&[1, 1]) {
                return false;
            }
            // After the choice: OK.
            if self.greeting == 0 && !reply(&[0, 0, 0, 0]) {
                return false;
            }
            if b.is_empty() {
                return true;
            }
        }
        self.notice_input(b);
        self.stdin.write_all(b).and_then(|_| self.stdin.flush()).is_ok()
    }

    /// Follow the viewer's messages (RFC 6143 §7.5) to see clicks and keys. Anything it doesn't
    /// know stops the watching, never the forwarding.
    fn notice_input(&mut self, b: &[u8]) {
        if !self.sniffing {
            return;
        }
        self.pending.extend_from_slice(b);
        // The first byte after the handshake is ClientInit (the shared flag).
        if self.init && !self.pending.is_empty() {
            self.pending.remove(0);
            self.init = false;
        }
        let mut at = 0;
        loop {
            let rest = &self.pending[at..];
            let Some(&kind) = rest.first() else { break };
            let len = match kind {
                0 => 20,
                2 if rest.len() >= 4 => 4 + 4 * u16::from_be_bytes([rest[2], rest[3]]) as usize,
                3 => 10,
                4 => 8,
                5 => 6,
                6 if rest.len() >= 8 => 8 + u32::from_be_bytes([rest[4], rest[5], rest[6], rest[7]]) as usize,
                2 | 6 => break,
                _ => {
                    self.sniffing = false;
                    self.pending.clear();
                    return;
                }
            };
            if rest.len() < len {
                break;
            }
            match kind {
                4 if rest[1] == 1 => person_acted(),
                5 if rest[1] != 0 => person_acted(),
                _ => {}
            }
            at += len;
        }
        self.pending.drain(..at);
    }
}

impl Drop for Session {
    fn drop(&mut self) {
        let _ = self.child.kill();
        let _ = self.child.wait();
        if OPEN.fetch_sub(1, Ordering::SeqCst) == 1 {
            crate::cloudbox::set_viewer_open(false);
        }
    }
}

/// The person clicked or typed: the bot's steps wait until a little after they stop.
pub fn person_acted() {
    *LAST_INPUT.lock().unwrap_or_else(|e| e.into_inner()) = Some(Instant::now());
}

/// The person is using the box through a viewer right now (input in the last few seconds).
pub fn person_active() -> bool {
    LAST_INPUT.lock().unwrap_or_else(|e| e.into_inner()).map(|t| t.elapsed() < HELD_AFTER_INPUT).unwrap_or(false)
}

/// Text the person typed in the viewer's own box (any language): pasted into the box.
pub fn person_typed(text: &str) {
    if !text.is_empty() {
        person_acted();
        crate::cloudbox::viewer_text(text);
    }
}

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

    #[test]
    fn the_bridge_keeps_its_indentation() {
        // Python needs its indentation (a Rust "\" line continuation would strip it).
        assert!(BRIDGE.contains("\n r=select") && BRIDGE.contains("\n  if not d: break"));
    }

    #[test]
    fn vnc_password_answer_matches_the_standard() {
        // Password "password", challenge 00..0f: DES-ECB with the bit-reversed key 0e86ceceeef64e26
        // (worked out with `openssl enc -des-ecb -nopad`).
        let c: [u8; 16] = core::array::from_fn(|i| i as u8);
        let want = [0xb8, 0x66, 0x92, 0x41, 0x25, 0xc8, 0xee, 0xbb, 0x9d, 0xeb, 0xc1, 0xdb, 0x61, 0xc5, 0x38, 0xe2];
        assert_eq!(vnc_response("password", &c), want);
        // Only the first 8 bytes of a password count.
        assert_eq!(vnc_response("password-and-more", &c), want);
    }
}