# Reachability analysis runs in two passes: a code pass for a path, a runtime pass for a call

Guide, 19 Nov 2026, read 19 Nov 2026

**The short answer.** Reachability analysis checks whether the vulnerable code element is called by your software, from the code (static) or from a running workload (runtime). The worked run below takes CVE-2026-111497, a Brotli decoder flaw in Node.js, through both passes on 19 Nov 2026: the application code had no Brotli reference, a dependency did, and a probe saw the call.

## The method asks whether the vulnerable element is used

Snyk describes the method as "identifying whether your application is calling a code element (for example, functions, classes, modules, and more) related to the vulnerability", and reports one of three statuses: REACHABLE ("A direct or indirect path was found from your application to the vulnerable code"), NO PATH FOUND, and NOT APPLICABLE ("Reachability is not supported for this vulnerability"). Semgrep Supply Chain pairs the affected dependency versions with "the pattern for the vulnerable code" and scans the codebase for it. Endor Labs builds static call graphs over the project and its dependencies, which it says often comprise over 100 packages.

The runtime variant watches the workload instead. Endor Labs describes instrumented container reachability as embedding "a runtime sensor in your container image to record how your application uses the image in a real environment".

The two passes need different inputs and miss different things. Static analysis needs source or build artifacts and lockfiles. Its limit, in Snyk's wording, is that "just because there is no evidence of this does not mean that the element cannot be reached": reflection and dynamic behavior can hide a call. Runtime analysis needs a sensor or probe on a live process and an observation window. Its limit is the window: code that does not run during the window leaves no trace. Neither pass tells you whether an attacker can reach the input, which is a separate exposure question.

## Step 1 is to name the vulnerable element from the advisory

CVE-2026-111497 is a flaw in the Brotli decoder. NVD lists Google Brotli 1.0.3 through 1.2.0, fixed in 1.0.10, 1.1.1 and 1.2.1. The Node.js security release of 21 Oct 2026 bundles the fix in v22.23.4, v24.22.1 and v26.12.1 and says an attacker "who can supply Brotli data that an application decompresses with node:zlib can corrupt process memory". CISA listed it in KEV on 27 Oct 2026.

That sentence names the element: a Brotli decompression stream created through node:zlib. The reachability question for any service on an affected Node.js line is whether anything in the process creates one. Whoever owns the service runs it, and the record it leaves is the answer, the evidence and the date.

## Step 2, the static pass: the application is clean and a dependency is not

The sample service, report-api, is an 8-line Node.js program written for this page. It calls a partner feed with axios 1.21.1 and returns the length of the body. Its source never mentions Brotli or node:zlib. A search of the application code is the first static check:

grep -rniE "brotli|node:zlib|require('zlib')" app.js printed nothing. Read alone, that is a NO PATH FOUND result, and the usual answer is wrong here. The advisory says "an application, or a library it calls", so the search has to cover the dependency tree:

grep -rlE "createBrotliDecompress|brotliDecompress" node_modules printed one file, node_modules/axios/lib/adapters/http.js. Reading it shows two relevant lines. Line 57 builds the request header as 'gzip, compress, deflate' plus ', br' when zlib.createBrotliDecompress exists. Line 1328 calls zlib.createBrotliDecompress when the response header says content-encoding br.

The path is app.js, axios.get, the http adapter, zlib.createBrotliDecompress. In Snyk's vocabulary that is REACHABLE, with a call path. A text search is a crude call graph: it finds the call site but not whether the branch can execute, and it would miss a call made through a computed name. Tools in this category build the graph across the whole dependency tree for that reason.

## Step 3, the runtime pass: a probe records the call and the window decides the result

The probe is a file loaded before the application with node --require ./probe.js app.js. It replaces zlib.createBrotliDecompress with a wrapper that prints the caller each time a decoder is created.

The first version assigned the wrapper with a plain assignment, zlib.createBrotliDecompress = function ... . It ran silently and recorded zero calls, because the assignment had no effect on the zlib export. That is a false NOT LOADED from a bad probe, the failure to watch for in any runtime sensor. Replacing the assignment with Object.defineProperty fixed it.

Window one: the partner upstream replied with content-encoding br, after seeing axios send Accept-Encoding: gzip, compress, deflate, br. The probe printed: probe: createBrotliDecompress call 1 from at RedirectableRequest.handleResponse (/tmp/demo/node_modules/axios/dist/node/axios.cjs:4951:33). The request returned 100 bytes.

Window two: the same service, the same request, but the upstream ignored the br offer and sent the body uncompressed. The probe printed nothing. Whether the decoder runs is decided by what the other side sends, so a quiet window proves little. A runtime result is true for the window that was watched.

## Step 4 is to record each answer with its limit

The machine used for this run reported Node.js v22.23.5, at or above the fix in v22.23.4, so this machine is not affected. The table is what the record would say for an image of the same service on a 22.x build before v22.23.4.

| Question | Method | Answer | Evidence | Limit |
| --- | --- | --- | --- | --- |
| Is the affected build present? | Version check | Depends on the image: below v22.23.4 is affected | node --version in each image | Says nothing about use |
| Does the application code call Brotli? | Static, application code | No hits | grep of app.js, 19 Nov 2026 | Misses libraries |
| Does a dependency call it? | Static, dependency tree | Yes, axios http adapter, lines 57 and 1328 | grep of node_modules plus reading the file, 19 Nov 2026 | Call site found, branch not proven |
| Was a decoder created at runtime? | Runtime probe | Yes in window one, no in window two | Probe output with call stack, 19 Nov 2026 | True for the window watched |
| Can an attacker control the compressed bytes? | Exposure check, outside both passes | Open: depends on who controls the partner feed and the network path | Not tested in this run | Needs its own evidence |

## Where the usual answer is wrong

A clean search of your own code does not clear a finding when the advisory names a library path, as the axios result shows. A status of NO PATH FOUND is, in Snyk's words, not a statement that the vulnerability "is completely unreachable or unexploitable", and Snyk notes that results labeled this way "can change to Reachable over time" as the analysis improves. Re-run after a dependency update.

Some findings are not reachability questions at all. CVE-2026-75604 in Next.js turns on the host being Windows and the router type, and CVE-2026-123158 in OpenSSH turns on musl, PAM and account state. A call graph has nothing to say about either. Read the advisory's preconditions first and choose the method after.

## Documented offerings, alphabetically

The vendors below each document at least one of the two passes.

- Endor Labs: static call graphs across project and dependencies; instrumented container reachability with a runtime sensor.
- Semgrep: Supply Chain rules that pair affected versions with a code pattern.
- Snyk: call-graph analysis of application against dependencies, with REACHABLE, NO PATH FOUND and NOT APPLICABLE statuses.

## Related pages

- [Vulnerability triage with evidence](/guides/vulnerability-triage-with-evidence/)
- [How to prove a finding is not exploitable](/guides/prove-a-finding-is-not-exploitable/)
- [Reading a vendor advisory's preconditions](/guides/reading-a-vendor-advisory/)

## Sources

- Snyk reachability analysis: https://docs.snyk.io/manage-risk/prioritize-issues-for-fixing/reachability-analysis (read 19 Nov 2026)
- Semgrep Supply Chain overview: https://semgrep.dev/docs/semgrep-supply-chain/overview (read 19 Nov 2026)
- Endor Labs call graphs: https://docs.endorlabs.com/scan/sca/call-graphs/index (read 19 Nov 2026)
- Endor Labs instrumented container reachability: https://docs.endorlabs.com/scan/containers/instrumented-reachability/index (read 19 Nov 2026)
- NVD record for CVE-2026-111497: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-111497 (read 19 Nov 2026)
- Node.js security releases, 21 Oct 2026: https://nodejs.org/en/blog/vulnerability/october-2026-security-releases (read 19 Nov 2026)
- CISA KEV catalog JSON feed (release 2026.11.18): https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json (read 19 Nov 2026)
- axios 1.21.1 package, lib/adapters/http.js, from the npm registry: https://registry.npmjs.org/axios/1.21.1 (read 19 Nov 2026)
- NVD record for CVE-2026-75604: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-75604 (read 19 Nov 2026)
- NVD record for CVE-2026-123158: https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2026-123158 (read 19 Nov 2026)

Last reviewed 19 Nov 2026. Canonical: https://reachabilityanalysis.com/guides/reachability-analysis-in-practice/
