PureEdDSA and support for calculating SHA directly on device #2080

de-nordic · 2024-10-02T07:59:43Z

Note on latest change: I have divided commits into bootutil and Zephyr themed so that any problem with Zephyr would not immediately break the feature PR.

Six commits:

Adds support for sha512 when TinyCrypt is used
adds PureEdDSA where image signature is calculated over image directly, rather than hash
adds support for calculating sha directly on storage, rather than loading chunks to memory
Zephyr Kconfig change that enables SHA512 with tinycrypt
Zephyr Kconfig change that enables SHA directly on storage
Zephyr Kconfig change that enables PureEdDSA

Note that mentioned below Kconfig options have proper MCUBOOT_ identifiers in code, but it is easier to communicate change by using them in examples.

Both commits provide support only on devices that are within address space.
There is possibility to extend both commits to work with external devices via features like XIP, where external device can be mapped to internal address space, but this is not approached here.

The PureEdDSA change adds TLV that marks the image as signed with Pure signature and logic in MCUboot that rejects images without the TLV, when expected by MCUboot. The change also slightly improves security of devices, specifically if the verification can be done by dedicated coprocessor.

Both solutions slightly improve performance as they cut on additional processing.

The PureEdDSA commit requires imgtool change: #2063

Notes on stats:

In case when the sha512 is enabled with ed25519, the size of ROM, as build for the nrf52840dk, drops from 40486 to 39414, but there is visible slowdown in boot; for the sha256 the command line was:

west build -d builds/sha_on_board_256 -b nrf52840dk/nrf52840 bootloader/mcuboot/boot/zephyr/ -DCONFIG_BOOT_SIGNATURE_TYPE_ED25519=y -DCONFIG_BOOT_ED25519_TINYCRYPT=y -DCONFIG_BOOT_IMG_HASH_ALG_SHA256=y

for the 512, the change was in the last config:

west build -d builds/sha_on_board_512 -b nrf52840dk/nrf52840 bootloader/mcuboot/boot/zephyr/ -DCONFIG_BOOT_SIGNATURE_TYPE_ED25519=y -DCONFIG_BOOT_ED25519_TINYCRYPT=y -DCONFIG_BOOT_IMG_HASH_ALG_SHA512=y

Enabling the option CONFIG_MCUBOOT_HASH_STORAGE_DIRECTLY, that comes with second commit, reduces code by another ~50 bytes.

Enabling the CONFIG_BOOT_SIGNATURE_TYPE_PURE, provided by the third commit, shaves off additional ~150 bytes.

While SHA directly on device and in RAM works with the same image signed for specified sha, the PureEdDSA requires special signature processing via imgtool and lacks HASH TLV.

de-nordic · 2024-10-02T12:33:03Z

@taltenbach FYI

nordicjm · 2024-10-04T07:42:24Z

rebase needed

de-nordic · 2024-10-04T08:31:54Z

rebase needed

Done

nordicjm · 2024-10-09T06:59:44Z

boot/zephyr/Kconfig

+	  to address space or RAM area, enabling this option allows hash
+	  calculation functions to directly access the storage through that address
+	  space or using its own DMA. This reduces flash read overhead done
+	  by the MCUboot.


Suggested change

by the MCUboot.

by MCUboot.

nordicjm · 2024-10-09T07:00:02Z

boot/zephyr/Kconfig

+	  by the MCUboot.
+	  Notes:
+	    - not supported when encrypted images are in use, because calculating
+	      SHA requires image to be decrypted first, which is done to RAM.


Suggested change

SHA requires image to be decrypted first, which is done to RAM.

SHA requires image to be decrypted first, which is done in RAM.

nordicjm · 2024-10-09T07:00:15Z

boot/zephyr/Kconfig

+	      SHA requires image to be decrypted first, which is done to RAM.
+	    - currently only supported on internal storage of devices; this
+	      option will not work with devices that use external storage for
+	      either of image slots.


Suggested change

either of image slots.

either of the image slots.

nordicjm · 2024-10-09T07:00:51Z

boot/zephyr/include/mcuboot_config/mcuboot_config.h

@@ -141,6 +141,13 @@
 #define MCUBOOT_DECOMPRESS_IMAGES
 #endif

+/* Invoke hashing functions directly on storage. This requires for device


Suggested change

/* Invoke hashing functions directly on storage. This requires for device

/* Invoke hashing functions directly on storage device. This requires the device

nordicjm · 2024-10-09T07:01:08Z

boot/zephyr/include/mcuboot_config/mcuboot_config.h

@@ -141,6 +141,13 @@
 #define MCUBOOT_DECOMPRESS_IMAGES
 #endif

+/* Invoke hashing functions directly on storage. This requires for device
+ * to be able to map storage to address space or RAM.


Suggested change

* to be able to map storage to address space or RAM.

* be able to map storage to address space or RAM.

nordicjm · 2024-10-09T07:02:43Z

boot/bootutil/src/image_validate.c

+
+    /* Search for the TLV */
+    rc = bootutil_tlv_iter_next(&it, &off, &len, NULL);
+


this should check the value too and it should only have one valid value

You mean that the TLV should have a value?

Value and length, even if length is 1 and value is 1

Agree. I also see that layout is not documented.

nordicjm · 2024-10-09T07:03:01Z

boot/bootutil/src/image_validate.c

@@ -425,6 +461,67 @@ bootutil_img_validate(struct enc_key_data *enc_state, int image_index,
    FIH_DECLARE(security_counter_valid, FIH_FAILURE);
 #endif

+#ifdef MCUBOOT_DECOMPRESS_IMAGES


this should not be here

nvlsianpu · 2024-10-09T12:40:35Z

boot/bootutil/src/image_validate.c

+
+    /* Search for the TLV */
+    rc = bootutil_tlv_iter_next(&it, &off, &len, NULL);
+


Agree. I also see that layout is not documented.

nvlsianpu · 2024-10-09T15:51:19Z

boot/bootutil/src/image_validate.c

    rc = !image_hash_valid;
    if (rc) {
        goto out;
    }
+#elif defined(MCUBOOT_SIGN_PURE)
+    /* This returns true on EQ, rc is err on non-0 */
+    rc = !FIH_EQ(valid_signature, FIH_SUCCESS);


use FIH_NOT_EQ()

nvlsianpu · 2024-10-09T15:52:17Z

boot/bootutil/src/bootutil_priv.h

@@ -268,6 +268,9 @@ struct boot_loader_state {
 fih_ret bootutil_verify_sig(uint8_t *hash, uint32_t hlen, uint8_t *sig,
                            size_t slen, uint8_t key_id);

+fih_ret bootutil_verify_img(const uint8_t *img, uint32_t size,
+                            uint8_t *sig, size_t slen, uint8_t key_id);
+


nit: I would appreciate code comment on difference to above function.

Added comment.

d3zd3z · 2024-10-10T17:28:35Z

boot/zephyr/Kconfig

+	bool "Use Pure signature of image"
+	depends on BOOT_SIGNATURE_TYPE_PURE_ALLOW
+	help
+	  The Pure signature is calculated directly over image rather than


I still have some problems with the wording here. I would argue "This is a more secure signature" is a little misleading. It is unclear to me how this affects the security when doing hardware encryption, because we can still do the signature of the hash by the hardware. The only real thing that PureEdDSA does is make the algorithm more resilient to weaknesses found in the hash function used. It doesn't affect the security of the signature itself.

There is actually a bit of a performance hit to Pure mode as well, as it requires hashing the image twice, whereas Hash EdDSA hashes the image once, and then hashes that has twice.

I still have some problems with the wording here. I would argue "This is a more secure signature" is a little misleading. It is unclear to me how this affects the security when doing hardware encryption, because we can still do the signature of the hash by the hardware. The only real thing that PureEdDSA does is make the algorithm more resilient to weaknesses found in the hash function used. It doesn't affect the security of the signature itself.

I can remove the "more secure" info.

There is actually a bit of a performance hit to Pure mode as well, as it requires hashing the image twice, whereas Hash EdDSA hashes the image once, and then hashes that has twice.

A hardware may access an image directly via address space which improves performance , and I do not think that hash is calculated twice, I suspect it to be cached during operation.

Removed the "more secure" or confusing, I think, comments.

The commit enables SHA512 support, for image hashing, with TinyCrypt. Although on 32bit machines the SHA256 will be faster than SHA512, benefit of enabling the SHA512 is that you have only one algorithm compiled in which reduces size of code. Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>

The commit add support for passing storage device address space to hash calculation functions, which allows to use hardware accelerated hash calculation on storage. This feature only works when image encryption is not enabled and all slots are defined within internal storage of device. The feature is enabled with MCUboot configuration option MCUBOOT_HASH_STORAGE_DIRECTLY. Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>

The commit adds support for PureEdDSA, which validates signature of image rather than hash. This is most secure, available, ED25519 usage in MCUboot, but due to requirement of PureEdDSA to be able to calculate signature at whole message at once, here image, it only works on setups where entire image can be mapped to device address space, so that PSA functions calculating the signature can see the whole image at once. The feature is enabled with MCUBOOT_SIGN_PURE MCUboot configuration option. Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>

Select BOOT_IMG_HASH_ALG_SHA512_ALLOW via BOOT_ED25519_TINYCRYPT. Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>

Adds CONFIG_BOOT_IMG_HASH_DIRECTLY_ON_STORAGE, which enables MCUBOOT_HASH_STORAGE_DIRECTLY for Zephyr. Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>

Commit adds CONFIG_BOOT_SIGNATURE_TYPE_PURE Kconfig option, which enables MCUBOOT_SIGN_PURE in MCUboot configuration. Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>

de-nordic · 2024-10-22T16:55:46Z

@d3zd3z Got time to revisit?

de-nordic mentioned this pull request Oct 2, 2024

[nrf fromlist] imgtool: Add pure signature support nrfconnect/sdk-mcuboot#342

Merged

de-nordic force-pushed the sha-direct-and-ed-pure branch from 1d63554 to 46a8117 Compare October 2, 2024 11:18

de-nordic requested a review from d3zd3z October 2, 2024 12:31

de-nordic marked this pull request as ready for review October 2, 2024 12:31

de-nordic requested review from nvlsianpu and nordicjm October 2, 2024 12:31

de-nordic added the crypto Encryption support label Oct 2, 2024

de-nordic force-pushed the sha-direct-and-ed-pure branch 2 times, most recently from 1cef60b to a66656b Compare October 2, 2024 15:37

de-nordic force-pushed the sha-direct-and-ed-pure branch from a66656b to 6fcbb82 Compare October 4, 2024 08:31

de-nordic force-pushed the sha-direct-and-ed-pure branch from 6fcbb82 to 33d2fd5 Compare October 8, 2024 14:11

nordicjm requested changes Oct 9, 2024

View reviewed changes

de-nordic force-pushed the sha-direct-and-ed-pure branch 3 times, most recently from 5054696 to 5cc083f Compare October 9, 2024 14:11

nvlsianpu reviewed Oct 9, 2024

View reviewed changes

de-nordic force-pushed the sha-direct-and-ed-pure branch 3 times, most recently from 5fa4c21 to c9d0e3c Compare October 10, 2024 16:16

d3zd3z requested changes Oct 10, 2024

View reviewed changes

nvlsianpu approved these changes Oct 11, 2024

View reviewed changes

de-nordic force-pushed the sha-direct-and-ed-pure branch 2 times, most recently from 29bb94b to 496f44f Compare October 11, 2024 10:01

de-nordic requested review from nordicjm and d3zd3z October 11, 2024 10:02

de-nordic added 2 commits October 11, 2024 13:45

de-nordic added 4 commits October 11, 2024 13:57

zephyr: Allow SHA512 with TinyCrypt

4eb04b1

Select BOOT_IMG_HASH_ALG_SHA512_ALLOW via BOOT_ED25519_TINYCRYPT. Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>

zephyr: Add Kconfig and configuration for SHA on storage

20c3dab

Adds CONFIG_BOOT_IMG_HASH_DIRECTLY_ON_STORAGE, which enables MCUBOOT_HASH_STORAGE_DIRECTLY for Zephyr. Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>

zephyr: Add Kconfig option to select PureEdDS

75de637

Commit adds CONFIG_BOOT_SIGNATURE_TYPE_PURE Kconfig option, which enables MCUBOOT_SIGN_PURE in MCUboot configuration. Signed-off-by: Dominik Ermel <dominik.ermel@nordicsemi.no>

de-nordic force-pushed the sha-direct-and-ed-pure branch from 496f44f to 75de637 Compare October 11, 2024 13:58

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

PureEdDSA and support for calculating SHA directly on device #2080

PureEdDSA and support for calculating SHA directly on device #2080

de-nordic commented Oct 2, 2024 •

edited

Loading

de-nordic commented Oct 2, 2024

nordicjm commented Oct 4, 2024

de-nordic commented Oct 4, 2024

nordicjm Oct 9, 2024

nordicjm Oct 9, 2024

nordicjm Oct 9, 2024

nordicjm Oct 9, 2024

nordicjm Oct 9, 2024

nordicjm Oct 9, 2024

de-nordic Oct 9, 2024

nordicjm Oct 9, 2024

nvlsianpu Oct 9, 2024

nordicjm Oct 9, 2024

nvlsianpu Oct 9, 2024

nvlsianpu Oct 9, 2024

de-nordic Oct 11, 2024

nvlsianpu Oct 9, 2024

de-nordic Oct 11, 2024

d3zd3z Oct 10, 2024

d3zd3z Oct 10, 2024

de-nordic Oct 10, 2024

de-nordic Oct 10, 2024

de-nordic Oct 11, 2024

de-nordic commented Oct 22, 2024

	SHA requires image to be decrypted first, which is done to RAM.
	SHA requires image to be decrypted first, which is done in RAM.

	/* Invoke hashing functions directly on storage. This requires for device
	/* Invoke hashing functions directly on storage device. This requires the device

	* to be able to map storage to address space or RAM.
	* be able to map storage to address space or RAM.


		/* Search for the TLV */
		rc = bootutil_tlv_iter_next(&it, &off, &len, NULL);

PureEdDSA and support for calculating SHA directly on device #2080

Are you sure you want to change the base?

PureEdDSA and support for calculating SHA directly on device #2080

Conversation

de-nordic commented Oct 2, 2024 • edited Loading

de-nordic commented Oct 2, 2024

nordicjm commented Oct 4, 2024

de-nordic commented Oct 4, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

de-nordic commented Oct 22, 2024

de-nordic commented Oct 2, 2024 •

edited

Loading