GhostMap Linux eBPF rootkit is Aqua Nautilus’s name for a concealment component found in examined IronWorm and Atomic Arch payloads. Its purpose is to change what selected Linux programs see when they inspect processes, files, and sockets. Aqua’s November 17, 2026, research measured both the concealment and its limits.
For defenders, the striking result is a mismatch between views of the same host. Aqua could still reach a hidden process by its known identifier. A listener absent from one kernel-provided listing remained visible to the tested ss utility, while Tracee alerted on the rootkit’s eBPF attachments.
How the GhostMap Linux eBPF Rootkit Enters the Host
In Aqua’s tested IronWorm specimen, a compromised software package started an installer when lifecycle scripts were enabled. The implant created a persistent systemd service, then its successor process loaded GhostMap. Watching only the installer’s immediate children could therefore miss the process that attached the eBPF programs.
Aqua also examined a payload associated with Atomic Arch, delivered through compromised Arch User Repository build scripts. The two delivery channels matter because a package name alone is a weak boundary for this investigation. The researchers’ detailed concealment tests, however, were performed on the IronWorm specimen.
A Cyber Threat Intelligence workflow can connect package-delivery evidence with runtime behavior across campaigns. For a separate Linux persistence mechanism, see our GSocket Linux persistence analysis. The two cases call for distinct detection logic.
- 1Package install Persistent serviceInstaller establishes the service
- 2Persistent service Linux kernelSuccessor process attaches GhostMap programs
- 3Linux kernel InvestigatorSelected listings omit entries
- 4Investigator Linux kernelCompare independent observations
The sequence separates delivery from hiding. GhostMap is the rootkit component, while IronWorm is the broader implant that installs it and runs local collection. Aqua’s laboratory execution used root privileges, so that test does not demonstrate an unprivileged loading path.
Why Process and File Listings Can Be Misleading
Linux exposes process identifiers as entries under /proc. Directory-reading system calls give applications a series of records to interpret. Aqua found that GhostMap changes the record chain returned to a reader, causing the reader to step past selected entries.
In the controlled experiments, benign test processes selected for hiding disappeared from directory enumeration. Their known process paths remained directly accessible, and the entries returned after the rootkit stopped. Control processes outside the hiding selection remained visible throughout the comparison.
Aqua matched before-and-after buffers from the same calls and found a small record-length change behind this effect. The hidden records were still present in the buffer; the reader skipped them because the preceding record directed it farther ahead. The researchers reproduced the altered-buffer result six times, with one byte changed in each observed buffer.
Selection also depended on execution behavior. Aqua observed a fork-only child remain visible, while a child that executed a matching program became hidden. A child shown beneath a missing parent is therefore an investigation lead, not a standalone finding of GhostMap.
Why the Two Socket Views Disagreed
GhostMap targeted both the text listing at /proc/net/tcp and diagnostic messages consumed by network tools. In Aqua’s listener experiment, the selected listener vanished from the text listing in three observations. A benign comparison listener stayed visible.
The tested ss utility still displayed the selected listener in all three queries. Aqua observed that GhostMap modified a diagnostic message but left useful socket information in its payload. The process-owner annotation disappeared, yet the listening socket itself remained visible to that reader.
That result is specific to the tested ss build, iproute2 6.1.0-1ubuntu6.4. Other consumers may interpret the modified diagnostic message differently. The practical check is to compare independent socket views and retained runtime telemetry, rather than treating one listing as a complete inventory.
What the Implant Collected Locally
Aqua verified that the IronWorm specimen accepted synthetic local records and stored them as an extended attribute on its installed executable. The researchers confirmed seven accepted records through storage and readback. Static analysis connected that stored data to the construction of outbound command-and-control messages.
The experiment established local collection and storage. Aqua did not observe successful external delivery or authenticated remote command execution in this investigation. GhostMap helped conceal parts of the implant’s activity; it was not the component that parsed incoming records.
When preserving evidence, retain file metadata as well as executable content. A copy that includes only normal file bytes may omit the extended attribute Aqua observed. Teams can use an incident response plan template to document that evidence requirement before containment.
What Connects IronWorm and Atomic Arch
Aqua compared the embedded rootkit objects and their recorded behavior across the two payloads. Both contained 13 named eBPF programs, ten matching map names, and matching helper-use patterns for every compared program. Their surrounding implants also showed closely matching service and local-listener behavior.
These similarities strongly support reuse of a shared rootkit implementation. They do not establish that the same operator ran both delivery campaigns. The detailed process, socket, and Tracee measurements belong to the tested IronWorm specimen, so defenders should test any new variant separately.
This distinction also keeps the Linux research separate from Brandefense’s earlier GhostMap Android flaw report. The names overlap, but the Android crash-loop issue and Aqua’s Linux eBPF component are different subjects. Search and monitoring rules should include platform and behavior, not the name alone.
How Runtime Monitoring Exposed the Hiding Behavior
Aqua ran Tracee before starting IronWorm. Tracee then alerted as GhostMap attached programs with the capability to write into application memory. Three highlighted findings corresponded to directory and network concealment paths.
Those attachment alerts identified suspicious capabilities and active programs. They did not report every individual buffer mutation. Aqua’s separate instrumentation established the precise record changes, while Tracee retained all 48 tested directory-reader calls during concealment.
Aqua demonstrated detection in its lab, not prevention. A program’s memory-writing capability alone is also not proof of maliciousness. Investigators need to connect the attachment to its owner, timing, observed behavior, and the host’s other evidence.
Record the attachment as a reusable analytic, including program capability, owner, and timing. A package-name match alone misses behavior that crosses delivery channels. Keep the detection tied to observable host events.
A GhostMap Investigation Workflow for Defenders
Start with the suspected host’s retained installation and service history. Look for an unexpected service that appeared after package installation, then connect its executable identity to the process that loaded eBPF programs. Preserve that chain before removing a service or detaching programs.
Next, compare process enumeration with direct access to known process identifiers and historical execution telemetry. Check attached eBPF programs, maps, and links with ownership and timing in mind. A discrepancy should trigger further investigation, not an automatic family attribution.
- Preserve installation and service recordsKeep the delivery and restart sequence before containment.
- Compare process viewsMatch enumeration against known identifiers and retained execution events.
- Inspect eBPF attachment activityCorrelate program capabilities, owner, and attachment time.
- Compare socket viewsCheck text listings against diagnostic tools and runtime telemetry.
- Preserve file metadataInclude extended attributes when collecting the installed executable.
Use care when inspecting a protected process interactively. Aqua’s controlled debugger-attachment attempt terminated the attaching task. Plan collection so a failed inspection does not erase evidence or interrupt another response action.
Aqua also observed a service stop remove programs and links while three pinned maps remained. Those maps alone did not keep hiding active and did not prove survival across reboot. Check current attachments instead of assuming any retained map represents active concealment.
What the Chatter Count Can and Cannot Say
Brandefense Blackhole recorded 20 dark-chat posts matching GhostMap from August 19 through November 17, 2026. The measured unit is posts matching the name, not Linux infections, distinct campaigns, or confirmed victims. The count can include discussion of the unrelated Android subject, so it must not be used to estimate IronWorm deployment.
For this investigation, the priority comes from Aqua’s observed host behavior: persistent loading, altered reader views, and detectable attachment activity. Treat the Blackhole count as a prompt to disambiguate incoming mentions. Keep platform, delivery channel, and observed mechanism attached to each case before escalation.
Aqua’s evidence supports a practical response rule: a missing process or socket entry can reflect a modified view, while the resource continues running. Preserve independent runtime observations, confirm the implant’s installation chain, and test a suspected variant against its own behavior.
Frequently Asked Questions
Does a missing process listing prove IronWorm stopped?
No. Aqua accessed selected hidden processes through their known process paths while the rootkit was active.
Did GhostMap hide the listener from every network tool?
No. Aqua saw it disappear from a text listing, but the tested ss build still displayed it.
Are IronWorm and Atomic Arch run by the same actor?
Aqua found strong evidence of a shared rootkit implementation. Its comparison did not independently establish a common operator.
