## Description `WrappedRE2::Replace` builds the replacement result and hands it to V8 with `.ToLocalChecked()` **without checking for the empty `MaybeLocal`** that V8 returns when the string/buffer exceeds its maximum length: `lib/replace.cc` (v1.24.1): ```cpp // L553 — Buffer return path info.GetReturnValue().Set(Nan::CopyBuffer(result.data(), result.size()).ToLocalChecked()); // L556 — String return path info.GetReturnValue().Set(Nan::New(result).ToLocalChecked()); ``` When a global replace uses an output-amplifying template — `$'` (text after the match) or `` $` `` (text before the match) — the result grows to **O(input²)**. For an input of ~40,000+ identical single-char matches the result exceeds V8's `String::kMaxLength` (~536,870,888 chars on 64-bit). `Nan::New(result)` then returns an **empty `MaybeLocal`**, and the unchecked `.ToLocalChecked()` calls `v8::Utils::ReportApiFailure` → **`FATAL ERROR: v8::ToLocalChecked Empty MaybeLocal`** → `abort()` (SIGABRT). This is an **uncatchable** crash: it is not a JavaScript exception, so a surrounding `try/catch` cannot stop it — the entire Node process (or worker) dies. **The built-in regex engine handles the identical case correctly** by throwing a *catchable* `RangeError: Invalid string length`. node-re2 diverges from that contract and aborts instead. ## Proof of concept ``` npm i re2 node poc.js ``` ```js const RE2 = require('re2'); // Built-in engine: same case -> CATCHABLE RangeError (correct) try { 'a'.repeat(50000).replace(/a/g, "$'"); } catch (e) { console.log('native:', e.constructor.name, e.message); } // RangeError: Invalid string length // re2: ABORTS the whole process (uncatchable; try/catch does not help) 'a'.repeat(50000).replace(new RE2('a', 'g'), "$'"); // -> FATAL ERROR: v8::ToLocalChecked Empty MaybeLocal (process exits 134 / SIGABRT) ``` Observed (Node v24, clean `npm i re2` → [email protected]): native branch prints `RangeError: Invalid string length`; the re2 branch aborts with `FATAL ERROR: v8::ToLocalChecked Empty MaybeLocal`, stack top `WrappedRE2::Replace`, process exit code **134**. Threshold matches the mechanism precisely: input of 30,000 chars completes; 40,000 aborts (30000²/2 ≈ 4.5e8 < 5.37e8 max; 40000²/2 ≈ 8e8 > max). `$&`/constant templates and non-global replaces do not amplify and do not crash. ## Impact A remote, unauthenticated denial of service against any service that runs `String.prototype.replace` / the re2 `[Symbol.replace]` path where either the **replacement template** (containing `$'` or `` $` ``) or the **input size** is attacker-influenced. Because the failure is a native `abort()`, it cannot be contained by `try/catch` or domains — one request takes down the whole process/worker. This is especially impactful for re2's core audience, who adopt it specifically to process untrusted patterns/inputs safely. ## Suggested fix Check the `MaybeLocal` before `ToLocalChecked` on both return paths (and the intermediate group-string builds), and throw a catchable `RangeError` to match the built-in engine: ```cpp auto maybe = Nan::New(result); if (maybe.IsEmpty()) { Nan::ThrowRangeError("Invalid string length"); return; } info.GetReturnValue().Set(maybe.ToLocalChecked()); ``` (Apply equivalently to the `Nan::CopyBuffer(...)` buffer path at L553 and to the per-group `Nan::New(data, size).ToLocalChecked()` sites used by the replacer-function path.) ## Resolution Resolved in `re2` `1.25.1`. `WrappedRE2::Replace` now checks the returned `MaybeLocal` on every result path and throws a catchable `RangeError: Invalid string length` (matching the built-in engine) instead of aborting the process with an uncatchable `SIGABRT`. No API changes --- upgrade to `re2` >= `1.25.1` via a plain `npm upgrade` to receive the fix.
### Impact Affected versions of `league/commonmark` can have quadratic time complexity when parsing specially crafted Markdown lines. In practical terms, doubling the length of an affected line can make the parser perform roughly four times as much work. The parser identifies locations using character positions, but regular-expression matches report byte positions. These positions differ when a UTF-8 character uses more than one byte. Several parsing paths repeatedly rescan growing portions of the line to translate between the two positions. The Autolink extension can also copy and validate the remaining line at every URL-like prefix. In current 2.x releases, a single non-ASCII character anywhere on a line can place that whole line on the slower multibyte path. An attacker can combine it with a long run of leading whitespace or repeated Markdown punctuation, causing increasingly large rescans. When the Autolink extension is enabled, repeated URL-like prefixes provide another trigger, even on ASCII-only lines. Each trigger fits within one long line, so complex Markdown structure is unnecessary. An attacker who can submit Markdown for conversion can use a comparatively small request to consume disproportionate CPU time and allocation activity. Repeated or concurrent requests can occupy all available PHP workers and prevent legitimate requests from completing. The core paths affect `CommonMarkConverter`, `GithubFlavoredMarkdownConverter`, and custom environments. The autolink-specific path affects applications using `AutolinkExtension` or `GithubFlavoredMarkdownExtension`. Applications that process only trusted Markdown are not remotely exploitable. The impact is limited to availability: it does not disclose data, change rendered output, or bypass rendering restrictions. Settings such as `html_input` and `allow_unsafe_links` do not mitigate the issue because the expensive work occurs before rendering. ### Patches The issue is patched in `2.9.0` and later. Starting in that release, the parser records UTF-8 character-to-byte positions incrementally, converts ordered regular-expression match positions without restarting from the beginning of the line, and matches autolinks against the original line instead of copying every remaining suffix. The affected work then grows in direct proportion to the input size while preserving existing Markdown output and configuration behavior. Versions from `0.6.0` through `2.8.3` are affected. The 0.x and 1.x release lines are no longer supported, so their users must upgrade to `2.9.0` or later. ### Workarounds If you cannot upgrade immediately, reject or truncate inputs with excessively long individual lines before passing them to the converter. A total request-size limit is also useful, but a per-line limit is important because every demonstrated trigger fits on one line. Choose limits appropriate for the application and enforce them before Markdown parsing begins. Restricting conversion to trusted users, applying strict execution-time limits, rate-limiting requests, and limiting concurrent conversions can further reduce exposure, but these measures are not complete substitutes for upgrading. Disabling `AutolinkExtension` and avoiding `GithubFlavoredMarkdownExtension` removes the autolink-specific trigger, but the core multibyte parsing paths remain reachable in the standard parser. Existing nesting, delimiter, raw-HTML, and unsafe-link configuration options do not eliminate all affected paths. Applications that must continue processing untrusted Markdown should therefore enforce input limits even when autolinking is disabled.