How an Infinite Loop Vulnerability Happens in Go's Text Normalization and How to Fix It

How an Infinite Loop Vulnerability Happens in Go's Text Normalization and How to Fix It

The fe-tool module was quietly sitting on a time-bomb dependency

The fe-tool/go.mod file lists the dependencies for a Go utility that parses 7-Zip archives (via github.com/bodgit/sevenzip) and Electron ASAR bundles (via github.com/dcboy/go-asar). Both of those libraries lean on golang.org/x/text for Unicode normalization — the process of converting text into a canonical form before comparing, storing, or displaying it. That normalization step is completely invisible to most developers, which is exactly what makes this class of vulnerability so dangerous: the attack surface is buried several layers deep in the dependency tree.

Trivy's dependency scanner flagged golang.org/x/text v0.27.0 in fe-tool/go.mod as matching CVE-2026-56852, a high-severity Denial-of-Service vulnerability. The fix — upgrading to v0.39.0 — is a single line change, but understanding why it matters requires a closer look at what norm.Iter does and how it can be made to spin forever.

The Vulnerability Explained

What is norm.Iter and why does it loop?

golang.org/x/text/unicode/norm provides iterators for walking through Unicode text one normalization segment at a time. The norm.Iter type is designed to be called repeatedly in a loop like this:

// Typical usage pattern (simplified)
var iter norm.Iter
iter.InitString(norm.NFC, inputString)

for !iter.Done() {
    segment := iter.Next() // advance to next normalized segment
    process(segment)
}

The contract is simple: each call to iter.Next() advances an internal byte offset, and iter.Done() returns true when the offset reaches the end of the input. In affected versions (≤ v0.27.0), a specific class of Unicode byte sequences — containing certain combining characters or malformed code-point boundaries — causes iter.Next() to return without advancing the internal offset. The offset stays at the same position on the next iteration, iter.Done() never becomes true, and the loop runs forever.

The vulnerable dependency line

Before the fix, fe-tool/go.mod pinned:

// fe-tool/go.mod (before fix)
golang.org/x/text v0.27.0 // indirect

Because this is an indirect dependency (pulled in by bodgit/sevenzip and dcboy/go-asar), it would not normally appear on a developer's radar during a routine code review. Yet it is exercised whenever either of those libraries normalizes a filename or string extracted from an archive.

How an attacker could exploit this

fe-tool processes user-supplied archive files. Consider this attack path:

1. An attacker crafts a 7-Zip or ASAR archive whose internal filenames contain a malformed Unicode sequence — for example, a combining diacritic that is encoded in a way that prevents norm.Iter from advancing past it.
2. The victim runs fe-tool against the malicious archive (or a service wraps fe-tool and processes user-uploaded archives automatically).
3. bodgit/sevenzip calls into golang.org/x/text to normalize the filename for path comparison or output.
4. norm.Iter.Next() stalls at the malformed byte offset, the loop never exits, and the fe-tool process hangs at 100% CPU until it is killed.

In an automated pipeline — a CI artifact processor, a package registry scanner, a game-mod distribution platform — this single malicious archive could take down the worker indefinitely, constituting a full denial of service.

The Fix

What changed in fe-tool/go.mod

The patch upgrades golang.org/x/text from the vulnerable v0.27.0 to the patched v0.39.0:

-   golang.org/x/text v0.27.0 // indirect
+   golang.org/x/text v0.39.0 // indirect

v0.39.0 corrects the loop-exit logic inside norm.Iter so that Next() always advances the byte offset by at least one position, guaranteeing termination regardless of the input content.

Additional changes in the same PR

The diff also promotes github.com/bodgit/sevenzip and github.com/dcboy/go-asar from indirect to direct dependencies:

+require (
+   github.com/bodgit/sevenzip v1.6.1
+   github.com/dcboy/go-asar v0.1.0
+)

 require (
    github.com/andybalholm/brotli v1.2.0 // indirect
    github.com/bodgit/plumbing v1.3.0 // indirect
-   github.com/bodgit/sevenzip v1.6.1 // indirect
    github.com/bodgit/windows v1.0.1 // indirect
-   github.com/dcboy/go-asar v0.1.0 // indirect

This is a best-practice cleanup: if fe-tool's own code imports these packages directly, they belong in the top-level require block rather than the indirect block. Accurate dependency classification makes it easier for tools like Trivy, govulncheck, and Dependabot to reason about the actual attack surface.

The Go toolchain version was also bumped from go 1.24.5 to go 1.25.0, and fe-tool/go.sum was updated with the new hashes for golang.org/x/text v0.39.0 and the reclassified direct dependencies.

Before vs. after at a glance

Key Takeaways

• norm.Iter in golang.org/x/text ≤ v0.27.0 is not safe to use with untrusted input — archive filenames, user-uploaded text, or any externally sourced string can trigger the infinite loop.
• Indirect dependencies in go.mod carry real CVEs. The // indirect comment is a classification hint for the Go toolchain, not a security boundary.
• Upgrading from v0.27.0 to v0.39.0 is the only reliable fix — input validation alone cannot paper over a loop-exit bug inside the library.
• Promoting bodgit/sevenzip and go-asar to direct dependencies makes the dependency graph more transparent and ensures future scanners correctly attribute reachability to fe-tool's own code.
• A single-line go.mod change can eliminate a high-severity DoS vector that would otherwise be invisible during normal code review.

How Orbis AppSec Detected This

• Source: User-supplied archive files (7-Zip or ASAR format) processed by fe-tool, whose internal filenames or string fields may contain arbitrary Unicode byte sequences.
• Sink: The norm.Iter.Next() call path inside golang.org/x/text v0.27.0, reached transitively through github.com/bodgit/sevenzip and github.com/dcboy/go-asar when they normalize strings from archive metadata.
• Missing control: No upper bound on the version of golang.org/x/text was enforced, allowing the vulnerable v0.27.0 to remain pinned long after the patch was available; no automated dependency-vulnerability gate was present in the pipeline.
• CWE: CWE-835 — Loop with Unreachable Exit Condition
• Fix: golang.org/x/text was upgraded from v0.27.0 to v0.39.0 in fe-tool/go.mod, replacing the defective norm.Iter loop-exit logic with a version that always advances the byte offset.

Orbis AppSec automatically detected this vulnerability and opened a pull request with the fix. Try Orbis AppSec on your repositories to find and fix issues like this automatically.

Conclusion

CVE-2026-56852 is a textbook example of how a deeply buried, "invisible" dependency can introduce a high-severity vulnerability into an application that never explicitly calls the affected API. The fe-tool module processes real-world archive files — a classic source of adversarially crafted input — and its transitive dependency on a flawed version of golang.org/x/text meant that any malicious Unicode sequence in an archive filename could hang the process indefinitely.

The fix is surgical and low-risk: a single version bump in go.mod from v0.27.0 to v0.39.0, plus a dependency-graph cleanup that makes the true attack surface explicit. The broader lesson is that dependency hygiene — automated scanning, accurate direct/indirect classification, and CI gates on known CVEs — is not optional overhead. It is the layer of defense that catches the vulnerabilities your code review will never see.