Summary

AppImage targets built by app-builder-lib could use an empty path component when setting the LDLIBRARYPATH environment variable at runtime. This causes the current working directory to be added to the dynamic linker search path, which may allow an attacker to execute arbitrary code by placing a malicious shared library in the directory from which the AppImage is launched.

This vulnerability is the same class as CVE-2024-41817.

Details

The vulnerability existed in two independent code paths within app-builder-lib (toolset 1.0.0) and through upstream dependency app-builder-bin (toolset 0.0.0).

Path 1 — Modern static runtime (AppRun generated by TypeScript)

The AppRun script generated by app-builder-lib contained this line:

bash export LDLIBRARYPATH="${APPDIR}/usr/lib:${LDLIBRARYPATH}"

When LDLIBRARYPATH is not set in the environment at launch time, this evaluates to:

/path/to/app.AppDir/usr/lib:

The trailing : is treated by the dynamic linker as an empty path component, which resolves to the current working directory. If an attacker can place a malicious shared library (e.g., libfoo.so) in the directory from which the AppImage is executed, that library will be loaded in place of the legitimate one, resulting in arbitrary code execution.

The same issue affected PATH, XDGDATADIRS, and GSETTINGSSCHEMADIR in the same script.

bash export LDLIBRARYPATH="${APPDIR}/usr/lib${LDLIBRARYPATH:+:${LDLIBRARYPATH}}"

Path 2 — Legacy FUSE2 toolset (app-builder-bin)

AppImage targets built using the legacy FUSE2 toolset (toolsets.appimage = "0.0.0") delegated AppRun script generation to the app-builder-bin Go binary, which contained the same vulnerable template:

https://github.com/develar/app-builder/blob/7004925f95d8f034fc88d7e782c9aa7583debb8e/pkg/package-format/appimage/templates/AppRun.sh#L27

Impact

An attacker with the ability to write files to the directory from which a vulnerable AppImage is executed can cause arbitrary shared libraries to be loaded into the application process, resulting in arbitrary code execution with the privileges of the user running the AppImage.

Affected Versions

This was fully resolved in app-builder-lib@26.15.0 (commit 01b8ba979, PR #9829) when app-builder-bin was removed from the dependency tree entirely and all AppImage construction was migrated to the TypeScript implementation.

Workarounds

Set LDLIBRARYPATH to a non-empty value before launching the AppImage, so that the concatenation does not produce an empty path component. Alternatively, avoid running AppImage files from world-writable directories such as /tmp.

Summary

In electron-builder's builder-util-runtime package, the HTTP redirect handler (HttpExecutor.prepareRedirectUrlOptions) only stripped a credential header whose key string matched exactly lowercase "authorization". Other credential-bearing headers — most notably PRIVATE-TOKEN (used by GitLab's personal access token flow) and mixed-case Authorization (used by GitLab's Bearer/OAuth flow) — were not stripped and could be forwarded to an attacker-controlled cross-origin redirect destination.

---

Details

Root cause

HttpExecutor.prepareRedirectUrlOptions (introduced in builder-util-runtime via PR #9211, first released in v26.0.20) performed its cross-origin credential strip with a single case-sensitive property check:

typescript // vulnerable code (electron-builder v26.0.20 – v26.14.x) [via builder-util-runtime <9.7.0] if (headers?.authorization) { if (HttpExecutor.isCrossOriginRedirect(originalUrl, parsedRedirectUrl)) { delete headers.authorization // only removes the exact key "authorization" } }

JavaScript object property access is case-sensitive. The guard headers?.authorization evaluates to undefined (falsy) when the key is "Authorization", "AUTHORIZATION", or any other casing, so the branch is never entered and no header is deleted for those cases.

Affected updater flows

The clearest reproduced path is the private GitLab updater flow.

packages/electron-updater/src/providers/GitLabProvider.ts sets one of two credential headers depending on the token type:

typescript // GitLabProvider.setAuthHeaderForToken (affected versions) if (token.startsWith("Bearer")) { headers.authorization = token // Bearer / OAuth token → key is lowercase } else { headers["PRIVATE-TOKEN"] = token // personal access token → key is "PRIVATE-TOKEN" }

During a private release update check, the updater requests a release asset through GitLab's directasseturl. GitLab commonly redirects asset downloads to an external object-storage origin (e.g., S3, GCS). Because the redirect crosses origins:

1. A personal access token in PRIVATE-TOKEN is never inspected by the vulnerable strip guard — it is forwarded intact. 2. A Bearer token set as headers.authorization (lowercase) is stripped correctly. 3. A Bearer token set as headers.Authorization (capital A) or any other mixed-case variant bypasses the guard and is forwarded intact.

GitLab is the concrete reproduced case; any other provider or custom updater configuration that places credentials in a non-lowercase-authorization header is equally affected.

Before v26.0.20

Versions prior to v26.0.20 did not contain prepareRedirectUrlOptions at all. All credential headers were forwarded unchanged on every redirect, regardless of origin. This represents a broader, pre-existing version of the same class of vulnerability.

---

Proof of concept (reproduction shape)

1. Configure the updater with an authenticated GitLab provider, supplying a personal access token (non-Bearer). The provider will set PRIVATE-TOKEN: <token> on requests. 2. Trigger an update check. The updater fetches release metadata and then requests a release asset URL. 3. The trusted GitLab origin returns a 3xx cross-origin redirect (e.g., to S3 object storage). 4. HttpExecutor.prepareRedirectUrlOptions is called. The guard headers?.authorization is falsy (the key is "PRIVATE-TOKEN"). No header is deleted. 5. The request to the redirect destination is issued with PRIVATE-TOKEN: <token> present in the headers.

Observed result: The personal access token is forwarded to the redirect destination. An attacker who controls or can observe the redirect destination receives the token.

---

Impact

This is a credential disclosure vulnerability. An automatic update check can leak:

- GitLab personal access tokens (PRIVATE-TOKEN) - Bearer/OAuth tokens sent under a mixed-case Authorization key - Any other credential header not named exactly "authorization" in lowercase

Disclosure of a GitLab PAT grants the attacker whatever repository and API permissions the token carries, enabling unauthorized access to private source code, packages, or release artifacts.

---

Patches

Fixed in electron-builder v26.15.0 [included via v9.7.0 builder-util-runtime] via PR #9834 (commit 22a7532bd).

The incomplete property-access guard was replaced with a separator-agnostic, case-insensitive lookup against a registry of known sensitive header names:

typescript // fixed code (v9.7.0+) const normalizeName = (name: string): string => name.toLowerCase().replace(/[-]/g, "")

const SENSITIVEREDIRECTHEADERS = new Set([ "authorization", "proxyauthorization", "privatetoken", "xapikey", "xauthtoken", "xaccesstoken", "xgitlabtoken", "cookie", "xcsrftoken", ])

// In prepareRedirectUrlOptions, on cross-origin redirect: for (const key of Object.keys(headers)) { if (SENSITIVEREDIRECTHEADERS.has(normalizeName(key))) { delete (headers as Record<string, unknown>)[key] } }

normalizeName converts to lowercase and strips - and separators, so PRIVATE-TOKEN, Private-Token, Authorization, AUTHORIZATION, X-Api-Key, etc. are all matched. The fix also exports addSensitiveRedirectHeader() to allow custom publishers to register additional headers.

Upgrade path: Update builder-util-runtime to >= 9.7.0.

---

Workarounds

There is no configuration-level workaround that prevents header forwarding in affected versions. The only mitigation short of upgrading is to avoid authenticated GitLab updater flows on versions < 9.7.0.

If operating in a network environment where you control all possible redirect destinations, you may be able to prevent the token from reaching an attacker-controlled host through network-layer controls, but this is not a reliable mitigation.