mkimage regression when building ARCH=mips defconfig Linux kernel

[Simon Glass]

+Tom Rini

Hi Nathan,

On Fri, 9 Apr 2021 at 06:23, Nathan Chancellor <nathan at kernel.org> wrote:
>
> Hi Simon,
>
> Apologies if this is not the proper way to report a regression, this is my first
> time interacting with the U-Boot community.
>
> My distribution updated the uboot-tools package to 2021.04, which broke my
> Linux kernel builds for ARCH=mips:
>
> $ make -skj"$(nproc)" ARCH=mips CROSS_COMPILE=mips-linux- defconfig all
> ...
> /usr/bin/mkimage: verify_header failed for FIT Image support with exit code 1
> make[2]: *** [arch/mips/boot/Makefile:173: arch/mips/boot/vmlinux.gz.itb] Error 1
> ...
>
> I bisected this down to your commit:
>
> 3f04db891a353f4b127ed57279279f851c6b4917 is the first bad commit
> commit 3f04db891a353f4b127ed57279279f851c6b4917
> Author: Simon Glass <sjg at chromium.org>
> Date:   Mon Feb 15 17:08:12 2021 -0700
>
>     image: Check for unit addresses in FITs
>
>     Using unit addresses in a FIT is a security risk. Add a check for this
>     and disallow it.
>
>     CVE-2021-27138
>
>     Signed-off-by: Simon Glass <sjg at chromium.org>
>     Reported-by: Bruce Monroe <bruce.monroe at intel.com>
>     Reported-by: Arie Haenel <arie.haenel at intel.com>
>     Reported-by: Julien Lenoir <julien.lenoir at intel.com>
>
>  common/image-fit.c          | 56 +++++++++++++++++++++++++++++++++++++++++----
>  test/py/tests/test_vboot.py |  9 ++++----
>  2 files changed, 57 insertions(+), 8 deletions(-)
> bisect run success
>
> $ git bisect log
> # bad: [e9c99db7787e3b5c2ef05701177c43ed1c023c27] Merge branch '2021-04-07-CI-improvements'
> # good: [c4fddedc48f336eabc4ce3f74940e6aa372de18c] Prepare v2021.01
> git bisect start 'e9c99db7787e3b5c2ef05701177c43ed1c023c27' 'v2021.01'
> # good: [b2c86f596cfb1ea9f7f5138f72f1c5c49e3ae3f1] arm: dts: r8a774a1: Import DTS queued for Linux 5.12-rc1
> git bisect good b2c86f596cfb1ea9f7f5138f72f1c5c49e3ae3f1
> # bad: [74f4929c2c73beb595faf7d5d9bb6a78d710c2fd] ddr: marvell: axp: fix array types have different bounds warning
> git bisect bad 74f4929c2c73beb595faf7d5d9bb6a78d710c2fd
> # bad: [cbe607b920bc0827d8fe379ed4f5ae4e2058513e] Merge tag 'xilinx-for-v2021.04-rc3' of https://gitlab.denx.de/u-boot/custodians/u-boot-microblaze
> git bisect bad cbe607b920bc0827d8fe379ed4f5ae4e2058513e
> # good: [d5f3aadacbc63df3b690d6fd9f0aa3f575b43356] test: Add tests for the 'evil' vboot attacks
> git bisect good d5f3aadacbc63df3b690d6fd9f0aa3f575b43356
> # bad: [a1a652e8016426e2d67148cab225cd5ec45189fb] Merge tag 'mmc-2021-2-19' of https://gitlab.denx.de/u-boot/custodians/u-boot-mmc
> git bisect bad a1a652e8016426e2d67148cab225cd5ec45189fb
> # bad: [aeedeae40733131467de72c68e639cf9d795e059] spl: fit: Replace #ifdef blocks with more readable constructs
> git bisect bad aeedeae40733131467de72c68e639cf9d795e059
> # bad: [eb5fd9e46c11ea41430d9c5bcc81d4583424216e] usb: kbd: destroy device after console is stopped
> git bisect bad eb5fd9e46c11ea41430d9c5bcc81d4583424216e
> # bad: [99cb2b996bd649d98069a95941beaaade0a4447a] stdio: Split out nulldev_register() and move it under #if
> git bisect bad 99cb2b996bd649d98069a95941beaaade0a4447a
> # bad: [3f04db891a353f4b127ed57279279f851c6b4917] image: Check for unit addresses in FITs
> git bisect bad 3f04db891a353f4b127ed57279279f851c6b4917
> # good: [6f3c2d8aa5e6cbd80b5e869bbbddecb66c329d01] image: Add an option to do a full check of the FIT
> git bisect good 6f3c2d8aa5e6cbd80b5e869bbbddecb66c329d01
> # good: [124c255731c76a2b09587378b2bcce561bcd3f2d] libfdt: Check for multiple/invalid root nodes
> git bisect good 124c255731c76a2b09587378b2bcce561bcd3f2d
> # first bad commit: [3f04db891a353f4b127ed57279279f851c6b4917] image: Check for unit addresses in FITs
>
> Is this an actual regression or is this now the expected behavior? I have added
> Thomas and the linux-mips mailing list to take a look and see if the Linux
> kernel needs to have its sources updated.

It is expected. See the code in that commit:

      /*
       * U-Boot stopped using unit addressed in 2017. Since libfdt
       * can match nodes ignoring any unit address, signature
       * verification can see the wrong node if one is inserted with
       * the same name as a valid node but with a unit address
       * attached. Protect against this by disallowing unit addresses.
       */
      if (!ret && CONFIG_IS_ENABLED(FIT_SIGNATURE)) {
         ret = fdt_check_no_at(fit, 0);

         if (ret) {
            log_debug("FIT check error %d\n", ret);
            return ret;
         }
      }

Possibly you are using @ nodes in your FIT files in the kernel. Is it
possible to use a hyphen instead?

Regards,
Simon