[protonnegativo]

Since d78db4b ("security fixes") the vnc module fails against every RFB 3.7/3.8 server (i.e. most current VNC servers): each attempt ends with [ERROR] VNC server connection failed (short read) and no password is found.

Cause

start_vnc() now requires the first recv() to return 20 bytes. That is only true for RFB 3.3 with VNC auth (4-byte type + 16-byte challenge). With 3.7/3.8 the server sends just the security type list (1 + n bytes) and then waits for the client to choose a type, so the read is always short. The same commit also changed the type loop from i <= buf2[0] to i < buf2[0], which skips the last offered type (a server offering only type 2 never matches).

Fix

• Store the recv() result and require only what the protocol version sends at that point: 1 byte for 3.7+, 4 for 3.3 (an RFB 3.3 server with no auth sends only those 4 bytes and was broken too).
• Fix the type loop back to i <= buf2[0], and read the rest of the type list if it arrives in more than one segment (fail only if the connection closes). The buffer gets one more byte so the largest accepted count (20) fits.
• In the RFB 3.3 VNC auth branch, read the rest of the 16-byte challenge if it arrives in a separate segment (fail only if the connection closes).

Testing

Mock RFB server (Python, pycryptodome for the DES check), password secret, wordlist a b secret c:

python3 vnc_server.py 47301 3.8 secret 2      # RFB 3.8, offers type 2
python3 vnc_server.py 47302 3.3 secret        # RFB 3.3
python3 vnc_server.py 47303 3.8 secret 16,2   # RFB 3.8, offers 16 and 2
./hydra -P pw.txt -t 1 -s <port> 127.0.0.1 vnc

Before (master):

[ERROR] VNC server connection failed (short read)
[ERROR] VNC server connection failed (short read)
[ERROR] VNC server connection failed (short read)
[ERROR] VNC server connection failed (short read)
1 of 1 target completed, 0 valid password found

(3.3 still worked; with hydra-vnc.c from d78db4b~1 all three were found.)

After:

[47301][vnc] host: 127.0.0.1   password: secret
1 of 1 target successfully completed, 1 valid password found

Same for 47302 and 47303. With a wordlist without the password: 0 valid password found on all three. A 3.8 server offering only type 1 still reports "VNC server does not require authentication."

With SPLIT=1 the mock sends the first field in its own write, 50 ms before the rest (RFB 3.8: the type count, then the types; RFB 3.3: the type, then the challenge). Both failed on master ("short read"); the password is found after, also for 3.8 with 16,2. A 3.8 server that announces 3 types, sends one and closes gives "short read" instead of hanging.

vnc_server.py

#!/usr/bin/env python3
"""Tiny RFB (VNC) server for testing hydra's vnc module.
usage: vnc_server.py PORT VERSION(3.3|3.8) PASSWORD [SECTYPES e.g. 2 or 16,2]
SPLIT=1 sends the first field (type count / security type) in its own segment."""
import os, socket, struct, sys, threading, time
from Crypto.Cipher import DES

port, ver, password = int(sys.argv[1]), sys.argv[2], sys.argv[3].encode()
sectypes = [int(x) for x in (sys.argv[4] if len(sys.argv) > 4 else "2").split(",")]
split = os.environ.get("SPLIT") == "1"


def send(c, head, rest):
    if split:
        c.sendall(head)
        time.sleep(0.05)
        c.sendall(rest)
    else:
        c.sendall(head + rest)


def vnc_key(pw):
    k = (pw + b"\0" * 8)[:8]
    return bytes(int(f"{b:08b}"[::-1], 2) for b in k)


def handle(c):
    try:
        c.sendall(b"RFB 003.008\n" if ver == "3.8" else b"RFB 003.003\n")
        c.recv(12)
        chal = os.urandom(16)
        if ver == "3.8":
            send(c, bytes([len(sectypes)]), bytes(sectypes))
            if c.recv(1) != b"\x02":
                return
            c.sendall(chal)
        else:
            send(c, struct.pack(">I", 2), chal)
        resp = b""
        while len(resp) < 16:
            d = c.recv(16 - len(resp))
            if not d:
                return
            resp += d
        ok = resp == DES.new(vnc_key(password), DES.MODE_ECB).encrypt(chal)
        if ok:
            c.sendall(struct.pack(">I", 0))
        else:
            reason = b"Authentication failed"
            c.sendall(struct.pack(">I", 1) + (struct.pack(">I", len(reason)) + reason if ver == "3.8" else b""))
    finally:
        c.close()


s = socket.socket()
s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
s.bind(("127.0.0.1", port))
s.listen(16)
while True:
    c, _ = s.accept()
    threading.Thread(target=handle, args=(c,), daemon=True).start()

[Copilot]

🟡 Changes recommended

Fragmented RFB 3.7/3.8 security-type lists still cause valid handshakes to fail.

1 open issue

• Read complete fragmented security list before selecting a type · New

What changed in this PR

This PR addresses short-read regressions in the VNC module’s RFB security negotiation, restoring support for common RFB 3.7/3.8 handshakes and split RFB 3.3 challenges.

Changes:

• Apply protocol-specific minimum lengths to the initial security read.
• Include the last received security type while bounding the scan by received bytes.
• Accumulate the remaining RFB 3.3 authentication challenge before encryption.

File	Description
hydra-vnc.c	Updates security negotiation checks, type scanning, and RFB 3.3 challenge reads.

_{Review effort: Balanced}

[Copilot]

🟡 Changes recommended

Blocking receives in both new handshake loops can leave workers waiting beyond the configured timeout when a server sends incomplete data and keeps the connection open.

1 open issue

• Use timeout-aware receives for incomplete VNC handshakes · New

1 resolved since last review

• Read complete fragmented security list before selecting a type

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

There are no active review issues for the VNC authentication handshake changes.

No open issues 🎉

1 resolved since last review

• Use timeout-aware receives for incomplete VNC handshakes

_{Review effort: Balanced}

[vanhauser-thc]

thanks