Hi Alexei,
Thanks for your response.
1) This is an implementation issue that can and probably should be fixed. If i understand correctly, the point of having mbedtls_md_info_from_type() and cousins is so that you don't have to hard code hash algorithms. I assume the reason TF-A is written this way is to minimize size.
Upgrading/changing the hash algorithm will require re-building of TF-A and re-flashing BL1 in ROM.
True, the way it is implemented today. I'm saying it might be worth removing this limitation. I missed the fact that this is a limitation even on TRUSTED_BOARD_BOOT today, let alone measured boot, so this is a general problem that needs to be solved. Adding sha512.c into the BL1 when MEASURED_BOOT is enabled may not be such a huge increase in size after all and may be acceptable. This will remove the limitation of tying down the code in ROM to a particular crytpo algorithm, ie make it crypto agile. Making the crypto algorithms used data driven through configs/eFuses is desirable over having them hard coded. This has a code size trade-off but if a signature scheme, key size, or hash algorithm is broken, you can create config/eFuse update without having to change ROM code, and hence NOT leaving older devices vulnerable/unsupported. What you have now is good given what we have today, but i strongly encourage the community to consider crypto-agility for BL1(which btw is not novel). This is inline with what fconf is aiming to do.
2) Thanks for clarifying. With this design, the Root-of-Trust for measurement(RTM) is the entire secure world software stack, including the config files. If the config files are in the RTM, you can use fields in the config file to indicate the crypto algoirthm, without hard-coding them at compile time. Also, note that having a huge RTM like the entire secure software stack is okay for now but definitely increases the attack surface to defeat measured boot. Ideally, you want nothing outside of BL1 to be the RTM. This can be done by providing platform hooks to "extend" the measurement into a platform specific "PCR" equivalent with IMPDEF protections(or a real TPM for that matter).
3,4,5,6) Thanks for clarifying. Question: Is it accurate to say that, ultimately, in this design, a platform will require a trusted application that talks to a real TPM or implements a fTPM, to provide the RTS(Root-of-Trust for storage) and RTR(Root-of-Trust for reporting)?
Thanks Raghu
On 3/25/20 9:01 AM, Alexei Fedorov wrote:
Hi Raghu,
Thanks for reviewing the proposal. Please find my answers below.
- The idea was to use the same hash algorithm throughout all TF-A
code for consistency and not introduce any new build flags. One of the initial implementations even didn't calculate the hash itself but was reading verified data provided by the Chain of Trust (CoT) for the purpose of optimisation.
Existing definition of TF_MBEDTLS_HASH_ALG_ID in 'drivers\auth\mbedtls\mbedtls_common.mk' at line #76:
ifeq (${HASH_ALG}, sha384) TF_MBEDTLS_HASH_ALG_ID := TF_MBEDTLS_SHA384 else ifeq (${HASH_ALG}, sha512) TF_MBEDTLS_HASH_ALG_ID := TF_MBEDTLS_SHA512 else TF_MBEDTLS_HASH_ALG_ID := TF_MBEDTLS_SHA256 endif
passed to 'include\drivers\auth\mbedtls\mbedtls_config.h', line #72
#define MBEDTLS_SHA256_C #if (TF_MBEDTLS_HASH_ALG_ID != TF_MBEDTLS_SHA256) #define MBEDTLS_SHA512_C #endif
and used in Mbed TLS to define MBEDTLS_MD_MAX_SIZE in 'include\mbedtls\md.h':
#if defined(MBEDTLS_SHA512_C) #define MBEDTLS_MD_MAX_SIZE 64 /* longest known is SHA512 */ #else #define MBEDTLS_MD_MAX_SIZE 32 /* longest known is SHA256 or less */ #endif
makes impossible usage HASH_ALG=sha256 for TF-A and sha512 for Measured Boot calculations, because the following chain of function calls arm_bl1_set_bl2_hash() -> crypto_mod_calc_hash() -> crypto_lib_desc.calc_hash() -> calc_hash() -> mbedtls_md_info_from_type() returns CRYPTO_ERR_HASH error caused by insufficient space in internal Mbed TLS buffers and fixing this issue needs extra modifications in make and header files.
Upgrading/changing the hash algorithm will require re-building of TF-A and re-flashing BL1 in ROM, so please explain what you mean by "potentially break measured boot on old devices in case a hash algorithm is broken"
The functionality for getting the hash algorithm from the platform (e.g. eFuses) can be added later as a platform build option and requires fixing the issue described above.
- Yes, Measured Boot requires TF-A built with TRUSTED_BOARD_BOOT option
enabled, and as BL2 image is a part of CoT it is verified by BL1.
- Yes. Event Log implementation is based on TCG Specifications.
BL2 loads images, calculates their hashes and writes data into Event Log stored in Secure memory.
- It is planned to add fTPM service implementation to TF-A, see
Javier's message: https://lists.trustedfirmware.org/pipermail/tf-a/2020-March/000339.html
Stuart could also comment on the naming convention.
N/A
Event Log is a complex structure with entries of different lengths, and
TFTF test checks the length of each field against the remaining size of the Event Log's data to be processed before accessing and printing the actual data.
Thanks. Alexei.
*From:* TF-A tf-a-bounces@lists.trustedfirmware.org on behalf of Raghu Krishnamurthy via TF-A tf-a@lists.trustedfirmware.org *Sent:* 21 March 2020 05:53 *To:* tf-a@lists.trustedfirmware.org tf-a@lists.trustedfirmware.org *Subject:* Re: [TF-A] Proposal for Measured Boot Implementation Hi Alexei,
Thanks. This looks good at first glance. However, i do have some questions that aren't obvious to me by reading the description below and looking at code. Questions are numbered based on your original email. Perhaps these can be discussed in the TF-A forum if it is inconvenient over email.
- Would be good if the hash alg comes from the config file. This will
make the implementation "crypto agile" from the very beginning. It is common to want to upgrade/change the hash algorithm and since BL1 is in ROM, you potentially break measured boot on old devices in case a hash algorithm is broken. The other option is to get the hash algorithm from the platform, perhaps a platform gets it from eFuses as opposed to config files. 2) It looks like you are using memory allocated in the loaded DTB as the equivalent of a TPM "PCR". How is this protected from direct modification by BL2? Or is it not protected because BL2 forms a part of the Root-of-Trust for Measurement(RTM)?(since it's signature is verified by BL1?) 3) What does "Event Log" refer to? Is it the same event log proposed by TCG in the platform firmware profile ? As a general question, how close is the measured boot in TF-A/PSA going to be to TCG ? Will BL2 extend measurements for other images ? 4) Would be great not to refer to "TPM" in the measured boot implementation. Here we are implementing measured boot without a TPM, but it could be implemented with a TPM. Maybe it should be tcg event log? 5) OK. 6) What does validate event log mean here? More details ?
Thanks -Raghu
On 3/20/20 7:15 AM, Alexei Fedorov via TF-A wrote:
Hello,
I'm preparing the next set of patches for Measured Boot support in TF-A, please find some details on design and implementation below.
- SHA256/384/512 hash algorithm for Measured Boot related hash calculations
is passed as an existing build 'HASH_ALG' build parameter.
- BL1 calculates BL2 image hash and passes these data to BL2 via
FW_CONFIG/TB_FW_CONFIG device tree in new 'bl2_hash_data' byte array added in 'fvp_fw_config.dts'.
These changes are part of the patch under review, please see https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/3555
- Event Log is calculated by BL2 in Secure Memory and copied to
Non-secure memory. Address in Non-secure memory is calculated as:
"nt_fw_config_addr + nt_fw_config_max_size"
with values obtained from 'tb_fw_config':
nt_fw_config_addr = <0x0 0x80000000>; nt_fw_config_max_size = <0x200>;
- Event Log address and size is passed by TOS_FW_CONFIG and NT_FW_CONFIG
device tree in 2 new added properties:
Property name: 'tpm_event_log_addr' Value type is an unsigned 64-bit integer specifying the physical address of the Event Log.
Property name: 'tpm_event_log_size' Value type is an unsigned 32-bit integer specifying the size of the Event Log.
/* TPM Event Log Config */ tpm_event_log { compatible = "arm,nt_fw"; tpm_event_log_addr = <0x0 0x0>; tpm_event_log_size = <0x0>; };
- TF-A provides Event Log to the BL33 (TFTF/UEFI/U-boot) in 'nt_fw_config'
device tree, which address is passed by BL31 as 'arg0' parameter, see TFTF patch:
https://review.trustedfirmware.org/c/TF-A/tf-a-tests/+/3327
- A new test which validates and prints Event Log data passed
in 'nt_fw_config' to BL33 will be added to TFTF.
Please review and provide your comments on the proposed design.
Regards. Alexei.
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
-- TF-A mailing list TF-A@lists.trustedfirmware.org https://lists.trustedfirmware.org/mailman/listinfo/tf-a IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.