I also use the verified-boot features for several boards. The behavior you're describing is my understanding too.
But I think the expectation: "do not enforce signature checks if there are no public keys" is OK. The expectation "enforce a signature check if there's a signature" doesn't seem complete to me.
On Fri, May 25, 2018 at 7:46 AM, Nelson, Mark Mark.Nelson@commscope.com wrote:
Let me preface this by saying I'm using the xilinx-u-boot repo based on the 2017.4 tag.
I've been studying how to sign FIT images, using mkimage. I successfully signed an image and verified the signature as follows:
${UBTOOLS4}/mkimage -T multi -A arm64 -O linux -f images/linux/fitimage.its -K images/linux/public.dtb -k ${SEC_KEYSTORE} -r images/linux/image_secure.ub
${UBTOOLS4}/fit_check_sign -f images/linux/image_secure.ub -k images/linux/public.dtb
I'm signing the configuration in the fitimage.its as follows: conf@1 { description = "Boot Linux kernel with FDT blob + ramdisk"; kernel = "kernel@0"; fdt = "fdt@0"; ramdisk = "ramdisk@0"; signature@1 { algo = "sha256,rsa2048"; key-name-hint = "linux_rsa_priv"; sign-images = "kernel", "fdt", "ramdisk"; }; };
However, when I tested booting the FIT image without providing the public key in Uboots control fdt, it booted successfully. It obviously should not boot a signed image without the key information. When I looked into this, I found that there are two routines called to verify required signatures when booting: fit_image_verify_required_sigs() and fit_config_verify_required_sigs(). The latter is for verifying the signatures in configuration nodes. Each of these routines begins with this logic: /* Work out what we need to verify */ *no_sigsp = 1; sig_node = fdt_subnode_offset(sig_blob, 0, FIT_SIG_NODENAME); if (sig_node < 0) { debug("%s: No signature node found: %s\n", __func__, fdt_strerror(sig_node)); return 0; }
The problem is that if the fdt signature node(with the public key info) is not provided, this simply returns 0 for success, and the boot process continues successfully after it checks the hashes of the images. To fix the problem, I simply changed fit_config_verify_required_sigs() to return sig_node instead. If the signature is not there, the returned value is FDT_ERR_NOTFOUND, and the boot processing ends with: Bad Data Hash ERROR: can't get kernel image!
Notice that I didn't change the logic in fit_image_verify_required_sigs(), because I'm not including signatures on the images, just hashes. In that case, I want it to ignore the fact that it didn't find a signature node.
But surely, this solution is not the correct way. I know that IMAGE_ENABLE_VERIFY is used to gate access to fit_image_verify_required_sigs(), but there is nothing gating access to fit_config_verify_required_sigs() that I can see. I don't see any way to tell u-boot which signature method it should be using to verify: images or configs. It would seem that an additional preprocessor directive is needed. Am I missing something?
The suggestion of a build-time option to enforce a signature check regardless of the availability of public keys is interesting. A complementary build option to choose image or config could work too.
This could enforce a failure if (A) there are no public keys available and also (B) there are no signatures available for verifying. I am interested in others' thoughts!
[cid:image002.png@01D357D5.908FF380]http://www.commscope.com/ Mark Nelson Principal Engineer CommScope 250 Apollo Drive, Suite 100 Chelmsford, MA 01824 Mobile: 352-444-0150
U-Boot mailing list U-Boot@lists.denx.de https://lists.denx.de/listinfo/u-boot