Walk-Through: Checking Four Hosts for Rootkits and Bootkits
A worked rootkit check of four synthetic hosts: a refused BYOVD driver, a hidden kernel module with a preload library, a Gapz-like disk, a BlackLotus-like ESP.
TL;DR. "Try a sample" on the checker opens four synthetic hosts, 44 files in all, and each one raises findings:
- FIN-WKS-07, the Windows 11 workstation of the sister sites' svc_backup story. At 10:07:31 UTC, svc_backup installs the kernel driver service
IntelNetDiagfor an unsigned driver inC:\ProgramData\Intel, named one character away from Intel's vulnerableiqvw64e.sys. Two seconds later Windows refuses it: CodeIntegrity 3004 and System 7000. It is a failed BYOVD attempt, between Defender's detection ofm64.exeand the Defender exclusions. - fin-jump-01, the Ubuntu jump host. An unsigned module that no package owns sits among the kernel's own modules, carries syscall-table and hook-target strings, was loaded at 10:08:09, is set to load at every boot, and is missing from
/proc/moduleswhile/sys/modulestill lists it. A preloaded library replacesreaddirandfopen. - LAB-VM-01, a lab disk head. Unknown MBR code, high-entropy data before the first partition, and an NTFS boot sector whose hidden-sectors value points at that data: the Gapz layout.
- LAB-VM-02, a lab EFI system partition. A staging folder, an extra
grubx64.efiandwinload.efi, and a BCD that loads another store: the BlackLotus layout.
Read each finding with its evidence, decide the next step, and write down what files cannot show: memory, firmware and UEFI variables.
The sample
The four hosts are invented, and every file is synthetic and harmless. Placeholder executables are valid files that do nothing, certificates are generated, domains use .example and IP addresses come from the documentation ranges. All times are UTC.
| Host | What it is | Files | Modules with findings |
|---|---|---|---|
| FIN-WKS-07 | Windows 11 workstation, collected at 10:55:12 on 2026-09-14 | 7 | windrivers |
| fin-jump-01 | Ubuntu 24.04 jump host, collected with the Linux collect script at 11:04:52 the same day | 28 | linuxkernel |
| LAB-VM-01 | The first MiB of a lab VM's MBR disk | 1 | bootsector |
| LAB-VM-02 | A lab VM's EFI system partition | 8 | esp |
FIN-WKS-07 and fin-jump-01 belong to the intrusion that the sister sites follow on 2026-09-14. The lab hosts stand apart: they show two boot-chain layouts from public reports, built from placeholder bytes.
FIN-WKS-07: a driver Windows refused
What the other artifacts already show
| Time | Event | Where it is told |
|---|---|---|
| 09:57:41 | svc_backup created (Security 4720) | BITS Parser's walk-through |
| 09:59:28 | svc_backup logs on over RDP | Same |
| 10:06:51 | Defender detects m64.exe (1116) and quarantines it a second later | Defender Parser's walk-through |
| 10:07:36 | Tamper protection refuses a change to DisableRealtimeMonitoring (5013) | Same |
| 10:08:12 | Two 5007 events add the exclusions for C:\ProgramData\Intel and m64.exe | Same |
| 10:09:44 | m64.exe runs from the excluded folder | Same |
What the checker finds
The windrivers module reads the SYSTEM hive, the driver file and the System and CodeIntegrity logs:
| Time | Evidence | Finding (severity) |
|---|---|---|
| 10:07:29 | C:\ProgramData\Intel\iqvw64x.sys written: a 5 KiB driver image with no signature, whose version resource calls it a synthetic sample | "An unsigned kernel driver" (high) |
| 10:07:31 | Service key IntelNetDiag: Type 1 (kernel driver), Start 3 (demand), ImagePath \??\C:\ProgramData\Intel\iqvw64x.sys, key last written at 10:07:31 | "A driver service loads its image from C:\ProgramData" (high); "iqvw64x.sys looks like the name of another driver (iqvw64e.sys)" (medium); "A driver service written shortly before the latest activity on the host" (low) |
| 10:07:31 | System 7045: service IntelNetDiag, "kernel mode driver", "demand start", logged with the user SID S-1-5-21-3623811015-3361044348-30300820-1119 (svc_backup) | "A kernel driver service was installed: IntelNetDiag" (medium) |
| 10:07:33 | CodeIntegrity 3004 for \Device\HarddiskVolume3\ProgramData\Intel\iqvw64x.sys, requested by System | "Code Integrity refused a file whose signature could not be verified" (high) |
| 10:07:33 | System 7000 for IntelNetDiag with %%577: Windows cannot verify the digital signature for this file | "A driver failed to start: IntelNetDiag" (medium) |
The driver's hashes are not in the LOLDrivers snapshot the checker carries: the file is synthetic.
What it means
- Who. Installing a kernel driver service needs administrator rights, and the 7045 event carries svc_backup's SID. The account created at 09:57:41 is an administrator.
- What.
iqvw64e.sysis Intel's network adapter diagnostic driver, which LOLDrivers lists as vulnerable (CVE-2015-2291). A name one character away, in a writable folder, loaded on demand: the shape of "bring your own vulnerable driver", the approach of EDR killers such as POORTRY. Reading it as an attempt to blind Defender is an inference from the name and the timing; this file does nothing. - Why it failed. The file carries no signature. With test signing off and code integrity on, as the Measured Boot log of boot 42 records, 64-bit Windows loads only signed kernel drivers, so Code Integrity refused the image hash (3004) and the Service Control Manager logged the failed start with
ERROR_INVALID_IMAGE_HASH(7000). Nothing from this file ran in the kernel. - What came next. Three seconds later, tamper protection refused the attempt to turn off real-time protection (5013 at 10:07:36), and at 10:08:12 the attacker fell back to exclusions. The driver, the Defender setting and the exclusions look like three tries at the same goal.
- What a real attempt would look like. With the genuine, signed
iqvw64e.sys, the name finding and "An unsigned kernel driver" would not appear. If the file's hash matched a LOLDrivers sample, "A known vulnerable driver (LOLDrivers): iqvw64e.sys" would, and whether Windows loaded it would depend on Microsoft's vulnerable driver blocklist.
What else the host shows, with no findings
- ESP. Only
EFI\Microsoft\Boot\BCD(2026-09-14 07:52:10), whose boot entries carry default settings. The module notes that the partition holds no EFI binaries: Microsoft's files cannot be shipped, so their signatures were not checked. - Measured Boot. Boots 41 (log written 2026-09-11 07:58:52) and 42 (2026-09-14 07:52:31): Secure Boot on,
\EFI\Microsoft\Boot\bootmgfw.efias the only boot application, test signing off, code integrity on, the same 19 modules loaded by the boot loader. The digests in these logs are stand-ins. Boot 42 started before the intrusion andIntelNetDiagis a demand-start driver, so no Measured Boot log covers the attempt. - SYSTEM. 12 driver services: 11 Windows ones and
IntelNetDiag. The memory integrity value is absent. The hive also holdsIntelGfxHelper(key written 10:09:51), a service that runsC:\ProgramData\Intel\m64.exe -svc. It is not a driver, sowindriversdoes not list it; Windows Persistence covers it. - Amcache. None in the sample: whether the Compatibility Appraiser inventories a driver that never loaded is not documented.
- System log. 6005 and 6009 at 07:52:41, and 7036 for WinDefend at 07:53:02: the boot and Defender starting.
Next steps
- Collect what the checker does not read: the Security log (4624, 4672, 4697), Sysmon event 6 if Sysmon ran, the
VulnerableDriverBlocklistEnablevalue underControl\CI\Config, and Defender's own artifacts, as in Defender Parser's walk-through. - Search every host svc_backup reached for 7045 events, driver files in
ProgramDataor temporary folders, the service nameIntelNetDiagand the file nameiqvw64x.sys. Hunting BYOVD drivers offline has the commands. - Remove the service and the file after preserving them, and treat the host as compromised for the reasons the other sites give:
m64.exeran at 10:09:44.
fin-jump-01: a hidden module and a preload library
What the other artifacts already show
Linux Log Parser's walk-through follows svc_backup on this host: a password login at 10:02:11 after password guessing, and a sudo -i root shell from 10:04:02 to 10:09:15. Linux Persistence's sample adds the persistence: a changed password, a root SSH key, a sudoers entry, a crontab and a user service started at 10:12:30.
What the checker finds
The module tree is lib/modules/6.8.0-45-generic/, the release _live/uname reports. Its six in-tree modules are .ko.zst files dated 2024-08-30 12:02:04 and signed by "Sample Build Key"; modules.order lists them and the dpkg list of linux-modules-6.8.0-45-generic owns them. They raise nothing. What the linuxkernel module flags:
| Time | Evidence | Finding (severity) |
|---|---|---|
| 10:07:52.411 | kernel/drivers/powercap/intel_rapl_perf.ko: a plain .ko, no signature, no intree tag. Imports register_kprobe, prepare_creds, commit_creds; strings kallsyms_lookup_name, sys_call_table, __x64_sys_getdents64, __x64_sys_getdents, __x64_sys_kill | "Unsigned module among signed modules" (high); "Module unknown to the kernel build and to every package" (high); "Looks up the system call table" (high); "Names kernel functions that rootkits hook" (high); "Out-of-tree module inside the kernel's own folder" (medium); "Resolves unexported kernel symbols" (medium); "Packaged differently from the other modules" (low) |
| 10:08:03.127 | modules.dep rewritten, as depmod -a does; it now lists the module | None: the file is indexed |
| 10:08:09.612 | kern.log, and the same lines in _live/dmesg: "intel_rapl_perf: loading out-of-tree module taints kernel." and "intel_rapl_perf: module verification failed: signature and/or required key missing - tainting kernel" | "Kernel log: out-of-tree module loaded" (medium); "Kernel log: unsigned module loaded" (medium) |
| 10:08:21.906 | /etc/modules-load.d/intel-rapl.conf, line 2: intel_rapl_perf | "Boot configuration loads a flagged module" (high) |
| 10:08:40.233 | /usr/local/lib/libraplmon.so.1, owned by no package, exporting readdir, readdir64, fopen and fopen64 | "Library replaces libc directory and file functions" (high) |
| 10:08:47.518 | /etc/ld.so.preload names /usr/local/lib/libraplmon.so.1 | "Library preloaded into every program" (high); "Preloaded library outside the system library folders" (medium) |
| 11:04:52 | _live/: taint 12288 (O and E); /sys/module/intel_rapl_perf with initstate live and taint OE; /proc/modules lists the six other modules but not this one | "Kernel tainted by an out-of-tree or unsigned module" (medium); "Loaded module missing from /proc/modules" (high) |
What it means
- A module posing as a driver.
intel_rapl_commonandintel_rapl_msrare real in-tree modules in the same folder.intel_rapl_perfsits beside them but is unsigned, uncompressed, absent frommodules.orderand owned by no package: it arrived after the kernel package was installed. modules.deplists it.depmodindexes whatever is in the tree when it runs, and it ran eleven seconds after the file appeared. Presence inmodules.depproves nothing; the checker relies onmodules.orderand package ownership instead.- What it is built to do. A
sys_call_tablelookup,kallsyms_lookup_namefound through a kprobe (the usual way since Linux 5.7), the names of thegetdents,getdents64andkillhandlers, and the credential functions are the combination Diamorphine's public source uses to hide files and processes and to grant root on a signal. These are capabilities read from strings and imports, not observed behaviour. - It was loaded. The kernel named it in both taint lines at 10:08:09, and the saved taint value still carries O and E. The host booted on 2026-08-20 at 06:31:12 and has not rebooted since, so the boot entry written at 10:08:21 has not run yet; it would load the module at the next boot.
- It hides. At 11:04:52 the module has a live
/sys/moduleentry with taint OE but is missing from/proc/modules, whichlsmodreads. A module that unlinks itself from the kernel's module list leaves exactly this, as reported for Diamorphine. Other LKM rootkits hide the same way, and the checker attaches no family to these findings: call it Diamorphine-style, not Diamorphine. - A second layer in userland. Seven seconds after the library was written,
/etc/ld.so.preloadnamed it, so every dynamically linked program started from then on loads it. A library that replacesreaddirandfopencan filter whatls,psand other tools show, the shape of libprocesshider. Which names it hides needs its strings or a disassembler. - One root session. Every planted file falls between 10:07:52 and 10:08:47, inside the root shell Linux Log Parser shows from 10:04:02 to 10:09:15.
The _live/ outputs come from this kernel, with the module and the library active. On a real host, a collection taken this way is a lower bound: the rootkit decides what the collect script sees.
Next steps
- Capture memory before any reboot or cleaning: AVML, or LiME, which itself taints the kernel. Run Volatility 3's
linux.malware.hidden_modules,linux.malware.check_modules,linux.malware.check_syscallandlinux.tracing.ftraceto see what the module hooks. - Image the disk and read it on a clean machine; compare with the collection.
- Analyse
intel_rapl_perf.koandlibraplmon.so.1for hidden names and prefixes, then search other hosts for those, the file names andintel-rapl.conf. - Rebuild the host: code ran in its kernel. Rotate every credential used on it or from it. Checking Linux kernel modules offline and ld.so.preload and userland rootkits cover each check in depth.
LAB-VM-01: a Gapz-like disk head
disk0.img holds the first MiB of a 20 GiB MBR disk. Its file time is 2026-10-06 09:30:00; disk structures carry no times.
| Sectors | Content | Finding (severity) |
|---|---|---|
| LBA 0 | Placeholder boot code; one active NTFS partition at LBA 63 | "MBR boot code does not match a known loader" (medium) |
| LBA 33 to 62 | Pseudo-random data: 15,291 non-zero bytes from LBA 33, entropy 7.99 bits per byte | "Data before the first partition does not match a known loader" (medium) |
| LBA 63 | A standard NTFS boot sector with Windows' boot code, but a BPB hidden-sectors value of 32 for a partition that starts at 63 | "VBR hidden-sectors value differs from the partition start" (high), linked to Gapz; the value points into the sectors before the first partition, where data starts at LBA 33 |
| LBA 64 to 78 | Windows' NTFS loader, recognised | None |
What it means
The NTFS boot code reads its next sectors from the position the hidden-sectors field gives. Here that is LBA 33, the start of the high-entropy data, so a BIOS boot from this disk would run that data instead of Windows' loader, which is left untouched. ESET reported that Gapz changes only these 4 bytes and keeps its code before the first partition when there is room. The checker links only the hidden-sectors finding to Gapz. The unknown MBR code is a separate trait: MBR bootkits such as TDL4 replace that code, while Gapz's VBR variant does not need to; read each finding on its own.
There are innocent explanations for each finding alone: other boot managers, a partition moved by a tool that did not update its boot sector. Together, and pointing at each other, they need an image and an explanation.
What it cannot show
The image is a disk head: the module notes that the partition runs past the end of the file, so it skipped the end-of-disk checks, and Gapz can also keep its code after the last partition. The sectors hold noise and placeholders, not code to analyse. A disk read through the running system could have shown clean sectors, since these bootkits filter disk reads.
Next steps
Image the whole disk from a clean environment, extract LBA 33 to 62 for analysis, compare the MBR and boot sector with a clean disk of the same build, then rebuild as the Gapz page describes and move the machine to UEFI with Secure Boot.
LAB-VM-02: a BlackLotus-like ESP
Windows was installed on 2026-03-02 at 09:14:08, the last clean boot was on 2026-10-30, and an installer ran on 2026-11-02 at 14:21. The esp module finds:
| File | Content | Findings (severity) |
|---|---|---|
EFI\Microsoft\Boot\BCD (14:21:37) | {bootmgr} with bcdfilepath set to \system32\bcd and avoidlowmemory 0x10000000 | "The boot manager is told to load another BCD store" (high); "The BCD sets avoidlowmemory" (medium) |
EFI\Microsoft\Boot\BCDR (2026-10-30 08:02:11) | A backup of the original BCD | "BCDR does not belong in the Windows boot folder" (medium) |
EFI\Microsoft\Boot\bootmgfw.efi (14:21:36) | Placeholder signed by "Synthetic Shim Publisher (sample)", issued by "Synthetic Third-Party UEFI CA (sample)"; the digest matches | "The Windows Boot Manager is not signed by a Windows production CA" (high); "is much newer than the other files in its folder", by 124 days (medium) |
EFI\Microsoft\Boot\grubx64.efi (14:21:36) | Placeholder signed by "Synthetic Self-Signed CA (sample)" | "A self-signed signature on" the file (high); "grubx64.efi does not belong in the Windows boot folder" (high); newer than its folder (medium) |
EFI\Microsoft\Boot\winload.efi (2026-03-02) | Unsigned placeholder | "winload.efi does not belong in the Windows boot folder" (high); "An unsigned EFI binary" (low) |
system32\bcd (14:21:34) | A loader object with testsigning, nointegritychecks and truncatememory 0x10000000 | "A system32 folder at the root of the EFI system partition" (high); "The BCD enables test signing" (high); "The BCD disables integrity checks" (high); "The BCD sets truncatememory" (high) |
BOOTSTAT.DAT, boot.stl | Windows' own files | None |
Most of these findings link to the BlackLotus page.
What it means
The layout is the one ESET's report and Microsoft's guidance describe after the installer runs: a staging folder holding the attacker's BCD, whose settings drive the exploit of CVE-2022-21894; a replaced bootmgfw.efi; a self-signed grubx64.efi that shim starts through an enrolled Machine Owner Key; the original boot manager renamed winload.efi, which keeps its old time; and BCDR, a backup of the BCD from the last clean boot. Times on a FAT partition are local times the collecting machine can misread, so the 124 days are a lead; the sample's times are UTC.
Placeholders, not the real files
The signature results differ from a real infection. A real bootmgfw.efi would be a shim signed through Microsoft's third-party UEFI CA: the same finding, linked to BlackLotus as well. A real winload.efi would be Microsoft's own boot manager, signed: "An unsigned EFI binary" would not appear, but "winload.efi does not belong in the Windows boot folder" would. The installer deletes its staged files; one was left in system32\ so that the folder appears in a collection, which cannot hold an empty folder.
What it cannot show
This host has only its ESP. Its Measured Boot logs would show grubx64.efi and winload.efi started from the Windows boot folder, and its SYSTEM hive whether memory integrity was turned off. The MOK list, db and dbx live in NVRAM, and the checker does not check signatures against a trust store or revocation list, so a revoked but validly signed boot manager would pass.
Next steps
Collect C:\Windows\Logs\MeasuredBoot\ and the SYSTEM hive, read the MOK list with mokutil --list-enrolled and the revocation state on the live host, and image the partition to recover the deleted staging files. Then follow the BlackLotus page: rebuild the ESP and Windows from trusted media, remove the attacker's MOK, and apply Microsoft's revocations. Checking the ESP for a bootkit offline covers each step.
The timeline
| Time (UTC, 2026-09-14) | Host | Event | Source |
|---|---|---|---|
| 09:59:28 | FIN-WKS-07 | svc_backup logs on over RDP | Security log |
| 10:02:11 | fin-jump-01 | svc_backup logs in with a password over SSH | auth.log |
| 10:04:02 | fin-jump-01 | sudo -i root shell opens | auth.log |
| 10:06:51 | FIN-WKS-07 | Defender detects m64.exe | Defender (1116) |
| 10:07:29 | FIN-WKS-07 | iqvw64x.sys written | File time |
| 10:07:31 | FIN-WKS-07 | IntelNetDiag created | Service key, System 7045 |
| 10:07:33 | FIN-WKS-07 | Windows refuses the driver | CodeIntegrity 3004, System 7000 |
| 10:07:36 | FIN-WKS-07 | Tamper protection refuses a Defender change | Defender (5013) |
| 10:07:52 | fin-jump-01 | intel_rapl_perf.ko written | File time |
| 10:08:03 | fin-jump-01 | modules.dep rewritten | File time |
| 10:08:09 | fin-jump-01 | The module loads and taints the kernel | kern.log, dmesg |
| 10:08:12 | FIN-WKS-07 | Defender exclusions added | Defender (5007 ×2) |
| 10:08:21 | fin-jump-01 | Module set to load at boot | modules-load.d file time |
| 10:08:40 | fin-jump-01 | libraplmon.so.1 written | File time |
| 10:08:47 | fin-jump-01 | /etc/ld.so.preload names it | File time |
| 10:09:15 | fin-jump-01 | Root shell closes | auth.log |
| 10:09:44 | FIN-WKS-07 | m64.exe runs | PCA, Prefetch |
| 10:55:12 | FIN-WKS-07 | Collection | Responder's notes |
| 11:04:52 | fin-jump-01 | Collect script runs; the module hides from /proc/modules | _live/ |
What the sample cannot show
- Memory. Whether
intel_rapl_perfstill runs, what it hooks and what it hides; which processes map the preload library; on FIN-WKS-07, any code loaded without a file. - Firmware and NVRAM. No module reads the SPI flash, the MOK list,
dbordbx. - Real signatures and code. The lab binaries are placeholders with synthetic certificates, the Measured Boot digests are stand-ins, the disk sectors are noise, and the module and the library are inert: the kernel would refuse the module and the dynamic linker the library.
- Absence. A host with no findings means nothing was found in the files collected. Firmware implants such as MoonBounce leave no file, and a collection taken on a running compromised host can be filtered. What an offline rootkit check can and cannot prove goes layer by layer.
Writing it up
- State what the files show: "The SYSTEM hive of FIN-WKS-07 holds the kernel driver service
IntelNetDiag, written 2026-09-14 10:07:31 UTC, for the unsignedC:\ProgramData\Intel\iqvw64x.sys. System 7045 records its installation with svc_backup's SID; CodeIntegrity 3004 and System 7000 at 10:07:33 record that Windows refused to load it." Not "svc_backup disabled Defender with a driver". - For fin-jump-01: "An unsigned module that no package owns,
intel_rapl_perf, was loaded at 10:08:09 and at 11:04:52 was listed under/sys/modulebut not in/proc/modules, consistent with Diamorphine-style hiding." Not "Diamorphine was installed". - For the lab hosts, name the layout and the report it matches, and say the bytes were not analysed as code.
- For each host, list what was not collected and what only memory or firmware could show.
FAQ
Does the sample contain real malware?
No. Every file is synthetic and harmless: the driver, the kernel modules, the preload library and the EFI binaries are valid-format placeholders that do nothing, the certificates are generated, the disk sectors are pseudo-random filler, and no real Microsoft, vendor or malware binary or hash is included. The lab hosts' binaries are placeholders, so their signature results differ from a real infection.
Is the module on fin-jump-01 Diamorphine?
The files cannot say. The module looks up the system call table, finds kallsyms_lookup_name through a kprobe, names getdents, getdents64 and kill as hook targets, and is listed under /sys/module but missing from /proc/modules. That is Diamorphine-style hiding, and the checker attaches no family to Linux kernel findings. Naming the rootkit needs analysis of the module and of a memory image.
Why does a driver that never loaded matter?
It shows what svc_backup tried to do with administrator rights: install a kernel driver named like a known vulnerable Intel driver, seconds before turning to Defender's settings. Windows refused this file because it had no valid signature. The same account could bring a signed vulnerable driver next, here or on another host, so look for other 7045 events and driver files.
What can the checker not see on these hosts?
Memory, firmware and NVRAM. It cannot say whether the hidden module still runs or what it hooks, which processes the preload library filters, what is in the SPI flash, or which certificates are in db, dbx and the MOK list. It reads only the files in the collection, and a collection taken on a running compromised host may be filtered.
Related articles
Frequently asked questions
Does the sample contain real malware?
No. Every file is synthetic and harmless: the driver, the kernel modules, the preload library and the EFI binaries are valid-format placeholders that do nothing, the certificates are generated, the disk sectors are pseudo-random filler, and no real Microsoft, vendor or malware binary or hash is included. The lab hosts' binaries are placeholders, so their signature results differ from a real infection.
Is the module on fin-jump-01 Diamorphine?
The files cannot say. The module looks up the system call table, finds kallsyms_lookup_name through a kprobe, names getdents, getdents64 and kill as hook targets, and is listed under /sys/module but missing from /proc/modules. That is Diamorphine-style hiding, and the checker attaches no family to Linux kernel findings. Naming the rootkit needs analysis of the module and of a memory image.
Why does a driver that never loaded matter?
It shows what svc_backup tried to do with administrator rights: install a kernel driver named like a known vulnerable Intel driver, seconds before turning to Defender's settings. Windows refused this file because it had no valid signature. The same account could bring a signed vulnerable driver next, here or on another host, so look for other 7045 events and driver files.
What can the checker not see on these hosts?
Memory, firmware and NVRAM. It cannot say whether the hidden module still runs or what it hooks, which processes the preload library filters, what is in the SPI flash, or which certificates are in db, dbx and the MOK list. It reads only the files in the collection, and a collection taken on a running compromised host may be filtered.