Archive extraction is a trust decision. A file inside an archive can contain a relative path, an absolute path, a symbolic link, or a hardlink. Defensive code often checks the visible filename and treats hardlink metadata as an implementation detail. BusyBox's cpio extractor made that mistake.
CVE-2026-124744 describes a flaw in BusyBox cpio extraction when FEATURE_PATH_TRAVERSAL_PROTECTION is enabled. The extractor sanitizes the member path but does not properly sanitize a hardlink target. An archive can therefore create a hardlink that resolves outside the intended extraction directory.
The visible path is not the write target
A hardlink is another name for an existing file. The security-relevant operation is not only where the archive member appears in the extraction tree. It is which existing path the link is allowed to reference.
If a privileged extraction process follows a malicious hardlink target, it can write to a location that the archive never legitimately owned. The practical outcome depends on the caller, the filesystem, existing target files, and whether the process has permission to the referenced path. The Red Hat record rates the issue 6.3 medium with a local attack vector, high complexity, no privileges, and required user interaction.
That score should not be read as proof that every cpio extraction is harmless. Build systems, firmware pipelines, container image assembly, backup restore jobs, package installation, and appliance update scripts often run extraction as a service account with broad filesystem access. A malicious archive can turn a path bug into overwritten configuration, corrupted application files, or persistence.
Patch the extraction utility
Use the patched BusyBox package from the operating system or appliance vendor. Embedded systems may carry a private BusyBox build, so the distribution package version is not enough. Inventory routers, cameras, industrial gateways, NAS devices, and recovery images that include cpio.
If a device vendor has not released a fix, avoid extracting untrusted archives with a privileged account. Perform extraction in a disposable directory owned by an unprivileged user and on a filesystem that does not contain valuable host paths. Disable hardlink creation when the archive format does not require it, or use a parser that resolves and validates the target before creating any link.
Do not place an extraction job inside a directory that is writable by the same untrusted user who supplies the archive. A hardlink target can be outside the extraction root even when the visible member name looks harmless. Use a dedicated mount namespace or container with a read-only host view and an empty writable workspace.
Test archives, not just firmware versions
Create test archives with a normal member, a member with ../, an absolute path, a hardlink to an outside path, and a hardlink chain. Run the test as the exact service account used in production. Confirm that every rejected link fails closed and that the process does not create a file outside the workspace.
Log archive name, submitter, extraction root, and rejected link target. If the extractor returns a generic failure, preserve the archive and the command line for investigation. Review file ownership and modification times after a failed extraction because a partial write may precede the error.
For appliances, include BusyBox in the firmware bill of materials even when the user interface hides it. A version string from a web page or update bundle may not represent the recovery partition or the cpio used during manufacturing.
Recovery and containment
If a vulnerable host extracted an untrusted archive, identify the account and root used for extraction. Search for unexpected hardlinks, changed configuration files, and files owned by the extraction service outside the intended tree. Compare hashes against a known-good package or image.
Restore affected files from trusted media, rotate credentials stored in overwritten configuration, and rebuild the host if the extraction process ran with broad privileges and integrity is uncertain. The patch closes the reported target-validation gap; the post-incident review should also reduce unnecessary extraction authority.