builtin_scan() reads a file's full contents into memory before checking the 5MB size cap

[kimdzhekhon]

Summary

In src/scanners.rs::builtin_scan():

let Ok(bytes) = std::fs::read(entry.path()) else {
    continue;
};
if bytes.len() > 5_000_000 || is_probably_binary(&bytes) {
    continue;
}

The 5MB size cap is meant to bound memory/CPU use per file, but std::fs::read unconditionally reads the entire file into memory first, and only checks the length (and binary-ness) afterward. A large file anywhere under the scan target (a stray VM image, database dump, log file, or an adversarially-placed huge file) is read fully into memory regardless of the cap, defeating its purpose. Scanning a directory containing a multi-GB file can cause excessive memory use / slowdown, i.e. a resource-exhaustion / DoS vector for a tool whose whole job is to be run against arbitrary repos before a push.

Suggested fix

Check the file size via entry.metadata() (or std::fs::metadata) before calling std::fs::read, and skip files over the cap without reading them at all:

let Ok(meta) = entry.metadata() else { continue; };
if meta.len() > 5_000_000 {
    continue;
}
let Ok(bytes) = std::fs::read(entry.path()) else { continue; };
if is_probably_binary(&bytes) { continue; }

Filing before fixing per repo convention.