-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA256 NEFARIOUSPLAN-CANONICAL-V1 {"body_md":"## What the bug actually does\n\nCVE-2025-48804 is a flaw in how the Windows boot manager processes System Deployment Image files. SDI is the container WinRE uses to ship a Windows Imaging payload at boot. The PoC's own parser documents the structure verbatim:\n\n```python\nSDI_SIGNATURE = b\"$SDI0001\"\nHEADER_SIZE = 512\nBLOB_TABLE_OFFSET = 0x200\nBLOB_ENTRY_SIZE = 64\nBLOB_TABLE_ENTRIES = 64\n```\n\nA 512-byte header at offset 0, a 64-entry blob table at offset 0x200, payload after that. Each entry carries a 4-byte type tag (`PART`, `WIM\\x00`, `BOOT`), a 64-bit data offset at byte 0x10 of the entry, and a 64-bit data size at byte 0x18. The boot manager reads the `WIM\\x00` entry, takes its `data_offset` and `data_size`, and boots a Windows Imaging payload out of the ramdisk bytes at that location.\n\n`patch_sdi.py` appends an attacker-controlled WIM to the end of a stock `boot.sdi` and rewrites the `WIM\\x00` entry's `data_offset` field to point at the appended file. The hash check the boot manager performs runs over a different region of the same ramdisk. The script's own layout printout names the two regions next to each other:\n\n```\n0x000XXXXX Custom WIM (appended, N bytes) <-- WIM blob points here\n0x000YYYYY Trusted WIM (loaded separately by boot loader)\n ^-- hash verified on this one, but NOT booted\n```\n\nThe verifier reads one buffer. The loader reads another. The check and the action do not target the same bytes.\n\nMicrosoft patched this in July 2025. The new `bootmgfw.efi` parses the SDI in a way that does not let the blob table redirect what gets booted. That is a real fix. It is also the part of the chain Intrinsec's PoC skips.\n\n## The PoC's first file is a Microsoft-signed boot manager\n\nThe repository ships a single binary under three paths so the firmware never has to think about which copy to request:\n\n```\nTFTP-root/bootmgfw.efi 1,557,304 bytes\nTFTP-root/EFI/Microsoft/Boot/bootmgfw.efi 1,557,304 bytes\nUSB/EFI/Boot/bootx64.efi 1,557,304 bytes\n```\n\nByte-identical. MD5 `ae8af199ef80311f9cee9de104a15496`. The embedded PE certificate table chains to the issuer `Microsoft Windows Production PCA 2011`. It is a Microsoft retail boot manager, signed by Microsoft, dated before July 2025.\n\nA Windows 11 device booting this binary off USB or PXE goes through Secure Boot validation in firmware. Firmware reads the embedded signature, walks the certificate chain, and resolves the chain's root against the device's Secure Boot DB. The DB contains an entry for the production PCA 2011 issuer. The signature is valid. The firmware launches the binary. The binary then loads the attacker's modified BCD (which the README walks the operator through producing with three `bcdedit` commands run from inside WinRE), loads the modified SDI off the same USB or TFTP path, and reaches the SDI bug Microsoft patched ten months ago. Once the malicious WinRE comes up, the boot path that selected it still satisfies the PCR 7 and PCR 11 measurements the TPM expects, so the TPM seals the volume master key to the platform, the platform asks for it, the TPM hands it over, and the volume mounts. The README's section heading for this step is \"Profit.\"\n\nThe PoC does not build the pre-patch boot manager. It does not patch it. It does not re-sign it. It carries it.\n\n## Secure Boot's trust grain is the signer\n\nSecure Boot does not check what version of a binary it is being asked to load. The firmware-resident DB enumerates keys, hashes, and issuer certificates the firmware will accept. The DBX enumerates specific hashes and certificates the firmware should refuse. There is no third list that says \"load this version of bootmgfw.efi, refuse the others.\" The validation grain is the signer.\n\n*Microsoft Windows Production PCA 2011* has signed every retail Microsoft Windows boot manager since Windows 8. Every one of those binaries chains to the same root. Every one of those binaries passes DB validation on the same firmware. Replacing the binary on the operating system's disk does not change what the firmware will load if a different binary is presented to it. The disk is one storage location. The firmware accepts boot managers from any storage location the boot menu reaches, including external USB and PXE.\n\nNVD's CVE-2025-48804 entry describes the SDI bug. MSRC's update guide describes the boot manager that fixes it. Neither document mentions that the pre-patch boot manager is still loadable. Neither document mentions it because the trust model does not have a way to refuse it without separate, operator-initiated action. The CVE record cannot acknowledge this without acknowledging the architectural decision that drives it. The fix lives at one layer. The trust lives at another.\n\nThis is an [unpatchable-primitive](/patterns/unpatchable-primitive) at the firmware layer. Patches close instances. The primitive (signatures do not expire, validation has no version awareness, every binary the signer ever produced remains acceptable until the operator explicitly says otherwise) is the design.\n\n## KB5025885 exists. It has been opt-in for three years.\n\nMicrosoft has the mechanism. KB5025885, published in May 2023, walks an administrator through extending DBX with hashes of vulnerable pre-patch boot managers and, in its final stage, with the *Microsoft Windows Production PCA 2011* issuer itself. Once that final stage is applied, the firmware refuses to load any binary signed by that certificate. The PoC's `bootmgfw.efi` becomes a non-bootable blob on the next boot.\n\nKB5025885 was not originally written for CVE-2025-48804. It was written for CVE-2023-24932, the BlackLotus UEFI bootkit, which exploited a different bootmgfw bug to install a persistent boot-level rootkit. The remediation Microsoft engineered for BlackLotus is the same remediation that closes the BitUnlocker downgrade three years later, because both attacks depend on the same primitive: a pre-patch boot manager that the firmware will still load. The infrastructure for refusing those binaries existed before this CVE was assigned. It existed before the CVE family was reported to MSRC at all. CVE-2025-48804 inherits a defense that has been sitting in the toolbox since 2023.\n\nKB5025885 is not in the Patch Tuesday channel. It is documented as a multi-stage opt-in administrator process, gated at each stage on the operator verifying that the device has migrated to a Windows UEFI CA 2023 boot manager and that no recovery media or alternate installations still rely on PCA 2011 signatures. The compatibility risk is real. Older Microsoft-signed installation media, third-party recovery products, and preboot tools chain to PCA 2011, and they stop loading once that anchor is in DBX. Intrinsec's writeup names this *un véritable challenge opérationnel pour Microsoft*. The challenge is real. The state of the field, three years after KB5025885 published, is that the production PCA 2011 anchor is still trusted on the substantial majority of Windows 11 devices in production.\n\nMicrosoft has not declined to revoke it. Microsoft has declined to revoke it by default.\n\nThe July 2025 update did not stage any new DBX entries. The patched boot manager runs alongside the pre-patch one, both signed by the same issuer, both accepted by Secure Boot. Choose either at boot time.\n\n## Bring Your Own Vulnerable Boot Manager\n\nThe [LnvMSRIO post](/posts/lenovo-lnvmsrio-cve-2025-8061-driver-name-is-the-spec) named the kernel-driver version of this shape. A signed Lenovo kernel driver exposes a user-to-kernel R/W primitive. Lenovo patches the driver in version 3.2. The 3.0 and 3.1 binaries remain code-signed and WHQL-attested. An attacker with administrator on any Windows host (Lenovo or not, domain-joined or not) drops the old `.sys`, registers it as a service, and gets the kernel R/W against a system that has never run a Lenovo product. The technique has a name. Bring Your Own Vulnerable Driver.\n\nCVE-2025-48804 is the same primitive one layer below the kernel. The signed binary is `bootmgfw.efi`. The signer is Microsoft itself. The trust enforcement is Secure Boot, applied by the firmware, before the operating system exists. The operator needs physical access to the device or PXE control on the same network segment. Both are inside the threat model BitLocker was designed against (Evil Maid attacks, lost laptops, retired hardware in resale). The operator then carries the binary. Same primitive, lower layer. The chain runs through firmware instead of through `nt!ZwLoadDriver`. The LOLDrivers project tracks known-vulnerable signed Windows drivers; the equivalent registry for boot managers has fewer entries because the supply of pre-patch Microsoft-signed boot managers is smaller, but the entries it has produce more privileged attacks because the boot manager runs before the operating system gets a vote.\n\nDefenders inherit two consequences. First, the patched `bootmgfw.efi` does not make this device safe against this attack. The patched binary supersedes the pre-patch one in every way except the one Secure Boot checks. Second, no further Microsoft-side patch can close the primitive without invalidating the trust anchor every Windows 11 device shipped before late 2025 inherited at the factory. The two architectural alternatives (TPM + PIN, or full migration to Windows UEFI CA 2023 via KB5025885) are end-state migrations, not patches against this CVE. The first stops the TPM from releasing the volume master key when the boot path changes; it requires user training and breaks the single-sign-on flows some enterprises rely on. The second requires the multi-year DBX rollout Microsoft has not driven by default in the three years it has been available.\n\nIntrinsec's first commit on this PoC is dated April 30, 2026 (Cassius Garat, `cassius.garat@yahoo.fr`). Ten months after Microsoft fixed the boot manager. Three years after Microsoft published the mechanism that would have made the fix matter. The repository's working parts are a 300 MB patched SDI file, a Microsoft-signed pre-patch `bootmgfw.efi`, and operator instructions for triggering a USB or PXE boot from WinRE. None of those operator instructions involve a vulnerability discovery, an exploit primitive search, or a signature forgery. The work of the PoC is logistics.\n\nPoC: [garatc/BitUnlocker](https://github.com/garatc/BitUnlocker)","closing_line":"The patched bootmgfw.efi supersedes the pre-patch one in every way except the one Secure Boot checks.","hook_md":"The Intrinsec PoC for CVE-2025-48804 ships a pre-patch `bootmgfw.efi`. The binary is 1,557,304 bytes. Its certificate chain resolves to *Microsoft Windows Production PCA 2011*. Microsoft fixed the underlying SDI bug in July 2025 by replacing `bootmgfw.efi` with a version that does not let the SDI blob table redirect what gets booted. The replacement does not revoke the predecessor. The production PCA 2011 issuer is in the Secure Boot database of every Windows 11 device shipped before late 2025, and the database does not pin a version.\nThe patched bootmgfw.efi supersedes the pre-patch one in every way except the one Secure Boot checks.","post_id":350,"slug":"cve-2025-48804-pre-patch-bootmgr-still-signed","title":"CVE-2025-48804: The Pre-Patch bootmgfw.efi Is Still Signed","type":"initial","unreadable_sentence":"The patched bootmgfw.efi supersedes the pre-patch one in every way except the one Secure Boot checks."} -----BEGIN PGP SIGNATURE----- iHUEARYIAB0WIQRf0htP5+SjynlxywneZjl4jgkQJgUCajlqjQAKCRDeZjl4jgkQ Jl14AQDnaI5G+nGnngJqTtKRmo4uwk9QbbqMWs8/O6e2UUd8CQEAi6C8pn/xet9H w2YvUbQAKqnLQvO3NVf3MMEj5arYGgQ= =CLm1 -----END PGP SIGNATURE-----