CVE-2021-27138 is a node-confusion flaw in U-Boot's FIT image handling before 2021.04-rc2: because libfdt ignores the @ unit address when matching node names, a crafted image can make signature verification check a different node from the one U-Boot then uses. That matters on devices with U-Boot verified boot. NVD scores it 7.8; U-Boot 2021.04 fixes it.
Vulnerability summary
CVSS base score
7.8
High severity
EPSS score (FIRST)
1.1%
Probability of exploitation in the wild in the next 30 days
CISA KEV catalog
No
Not listed
Published
Vulnerability ID: CVE-2021-27138 · Source: NVD
Affected components
| Component | CPE | Version range |
|---|---|---|
| Das U-Boot bootloader (FIT image handling) | cpe:2.3:a:denx:u-boot:*:*:*:*:*:*:*:* | Up to and including 2021.01, and 2021.04-rc1 (NVD); before 2021.04-rc2 (CVE record); first fixed release 2021.04 |
| Debian u-boot package | pkg:deb/debian/u-boot | bullseye fixed in 2021.01+dfsg-5+deb11u2 (DLA-4320-1, 2025-10-01); unstable fixed in 2021.07+dfsg-2; buster marked no-dsa (minor issue); bookworm and later fixed (Debian security tracker, as of 2026-11-20) |
| Buildroot boot/uboot (default 'latest version') | — | 2021.01 in Buildroot 2021.02 and the 2021.02.x branch, with no patches; 2021.04 from Buildroot 2021.05 (commit bd62da59). Boards that pin their own U-Boot version need a separate check |
| Yocto Project poky meta/recipes-bsp/u-boot | — | hardknott (3.3) builds 2021.01 with backported CVE-2021-27097 and CVE-2021-27138 patches; dunfell (3.1) builds 2020.01 with no patch for either (as of 2026-11-20) |
What does this mean for affected devices?
NVD's description: "The boot loader in Das U-Boot before 2021.04-rc2 mishandles use of unit addresses in a FIT." A unit address is the @... suffix on a devicetree node name, as in kernel@1. FIT (Flattened Image Tree) is U-Boot's devicetree-based image format, which the FIT_SIGNATURE option verifies for signed boot. The record was published on 2021-02-17 (NVD, CVE record).
Why are unit addresses a problem for signed images?
Commit 79af75f7 ("fit: Don't allow verification of images with @ nodes") explains it: "When searching for a node called 'fred', any unit address appended to the name is ignored by libfdt, meaning that 'fred' can match 'fred@1'. This means that we cannot be sure that the node originally intended is the one that is used." A crafted FIT could therefore carry a validly signed node and a second, same-named node with a unit address, and the code that verifies the signature might not look at the same node as the code that boots. Commit 3f04db89 ("image: Check for unit addresses in FITs") makes fit_check_format() reject any node name containing @ when FIT_SIGNATURE is enabled; the code comment notes that "U-Boot stopped using unit addressed in 2017." Both commits are by Simon Glass and credit Bruce Monroe, Arie Haenel and Julien Lenoir of Intel. Tom Rini's merge commit b6f4c757 fixes this CVE together with CVE-2021-27097; the merge is in tag v2021.04-rc2, so U-Boot 2021.04 is the first release with the fix.
How severe is CVE-2021-27138?
NVD's Primary score is CVSS v3.1 7.8 (CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H), the same as the MITRE CNA score; v2 is 6.8. NVD's weakness entry is NVD-CWE-noinfo. An attacker has to get a crafted FIT onto the boot path, through an update package or writable boot storage. The impact is on devices that rely on signed FIT images.
Will the fix break existing images?
It can. After the update, a U-Boot with FIT_SIGNATURE enabled refuses any FIT that uses @ in node names, printing "Signature checking prevents use of unit addresses (@) in nodes". Image source files (.its) written in the old kernel@1 / fdt@1 style must be renamed (for example kernel-1, fdt-1) and re-signed before the new bootloader ships. Test the rebuilt images on the target before rolling the bootloader out.
Which firmware builds carry it?
Any product with FIT signature verification on U-Boot older than 2021.04 without a backport. NVD's CPE covers U-Boot up to and including 2021.01 plus 2021.04-rc1. No device vendor or CISA advisory names products for this CVE as of 2026-11-20; downstream status comes from the distributions and build systems:
- Debian fixed bullseye in 2021.01+dfsg-5+deb11u2 with DLA-4320-1 (dated 2025-10-01, covering both CVEs) and unstable in 2021.07+dfsg-2. Buster was marked no-dsa as a minor issue; bookworm, trixie and later ship fixed packages (Debian security tracker).
- Buildroot's default U-Boot ("latest version") was 2021.01 in Buildroot 2021.02, and the 2021.02.x branch kept it with no patches in
boot/uboot. The bump to 2021.04 of 2021-04-20 shipped in Buildroot 2021.05. Many board configs pin their own U-Boot version or a vendor fork, so check the configured version, not the Buildroot release. - Yocto Project: the poky hardknott branch (3.3) builds U-Boot 2021.01 with backported patches for both CVEs. The dunfell branch (3.1) builds 2020.01 and, as of 2026-11-20, carries patches for CVE-2020-8432 and CVE-2020-10648 but none for these two. Vendor BSP layers often replace the poky recipe, so check the U-Boot your BSP actually builds.
How do you fix CVE-2021-27138?
Rebuild on U-Boot 2021.04 or later, or apply the merge b6f4c757 commits to your fork, rename @ nodes in your image sources and re-sign. Our U-Boot security guide lists this CVE with its sibling and covers verified boot hardening; the secure boot and hardware root of trust guide covers the wider chain. An earlier FIT signature bypass is CVE-2020-10648; for a U-Boot shell bug, see CVE-2022-34835.
Is CVE-2021-27138 exploited?
It is not in the CISA KEV catalog as of 2026-11-20 (catalog version 2026.11.18, 1,786 entries). FIRST EPSS gives it 0.01118, the 66.8th percentile, in the 2026-11-19 dataset (CISA KEV, FIRST EPSS).
Evidence
- NVD (read 2026-11-20): status Modified; published 2021-02-17; NVD Primary CVSS v3.1 7.8 CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H, v2 6.8; MITRE Secondary 7.8; CWE NVD-CWE-noinfo; CPE denx:u-boot up to and including 2021.01 and 2021.04-rc1; references commits 3f04db89, 79af75f7, b6f4c757 and DLA-4320-1. — NVD
- U-Boot commits 79af75f7 and 3f04db89 (Simon Glass, 2021-02-15/16; reported by Bruce Monroe, Arie Haenel, Julien Lenoir of Intel), read 2026-11-20: libfdt ignores unit addresses so 'fred' can match 'fred@1'; fit_check_format() rejects node names containing '@' when FIT_SIGNATURE is enabled, with the message 'Signature checking prevents use of unit addresses (@) in nodes'; comment 'U-Boot stopped using unit addressed in 2017'. Merge b6f4c757 fixes CVE-2021-27097 and CVE-2021-27138 and is contained in tag v2021.04-rc2. — U-Boot repository
- Debian security tracker and DLA-4320-1 (read 2026-11-20): bullseye fixed in 2021.01+dfsg-5+deb11u2 (DLA-4320-1, 2025-10-01); unstable fixed in 2021.07+dfsg-2; buster no-dsa (minor issue); bookworm and trixie fixed. — Debian
- Buildroot (read 2026-11-20): boot/uboot latest version 2021.01 in tag 2021.02 and branch 2021.02.x with no patches; commit bd62da59 bumps to 2021.04 (2021-04-20); tag 2021.05 latest version 2021.04. — Buildroot
- Yocto poky (read 2026-11-20): hardknott u-boot 2021.01 with CVE-2021-27138-1 and -2 patches; dunfell u-boot 2020.01 without them. — Yocto Project poky
- CVE record (checked 2026-11-20): state PUBLISHED, CNA MITRE, published 2021-02-17; CNA CVSS v3.1 7.8; references the three upstream commits. — CVE.org
- Not in the CISA KEV catalog, version 2026.11.18 (1,786 entries), checked 2026-11-20. — CISA KEV
- EPSS 0.01118, percentile 0.66811, dataset 2026-11-19, read 2026-11-20. — FIRST EPSS

Related advisories
Sources
- NVD
- U-Boot merge commit b6f4c757: fix CVE-2021-27097 and CVE-2021-27138
- U-Boot commit 79af75f7: fit: Don't allow verification of images with @ nodes
- U-Boot commit 3f04db89: image: Check for unit addresses in FITs
- Debian security tracker: CVE-2021-27138
- Debian LTS Advisory DLA-4320-1: u-boot security update
- Buildroot commit bd62da59: boot/uboot: bump to version 2021.04
- Yocto Project poky (hardknott): u-boot-common.inc with CVE patches
- CVE.org record CVE-2021-27138 (CNA: MITRE)
- CISA KEV
- FIRST EPSS
Frequently asked questions
Which U-Boot version fixes CVE-2021-27138?
U-Boot 2021.04 is the first release with the fix; the merge commit b6f4c757 is in tag v2021.04-rc2. Debian backported it to bullseye in 2021.01+dfsg-5+deb11u2 (DLA-4320-1), and Yocto hardknott carries backported patches on 2021.01.
Why will my old signed FIT images stop booting after the update?
With FIT_SIGNATURE enabled, the fixed U-Boot rejects any node name containing @ and prints 'Signature checking prevents use of unit addresses (@) in nodes'. Rename nodes such as kernel@1 to kernel-1 in the image source and re-sign the image.
What can an attacker do with CVE-2021-27138?
Build a FIT in which the signature check and the boot code look at different same-named nodes, so a validly signed node vouches for content that is not the one booted. It needs the ability to place an image on the device's boot path, which is why NVD scores it as local with user interaction (7.8).
Is CVE-2021-27138 always fixed together with CVE-2021-27097?
Upstream, yes: one merge (b6f4c757) fixes both, and Debian DLA-4320-1 and the Yocto hardknott recipe patch both. A vendor fork that cherry-picked only some commits may have one fix without the other, so check for both.
Questions about this page?