[opentitantool] Correct the SHA256 hash ordering #25969
Conversation
|
I don't have any comments on the code, it would be good to see a test case when a signature is generated by an external entity and validated by hsmtool for all signature types or better yet by rom_ext when verifying owner's image signature. |
cfrantz
force-pushed
the
fix-tool-endian
branch
from
633c047 to
29f77b4
Compare
|
Addresses #25967 |
cfrantz
force-pushed
the
fix-tool-endian
branch
from
29f77b4 to
712642b
Compare
cfrantz
force-pushed
the
fix-tool-endian
branch
from
712642b to
77b715c
Compare
I've added another commit that adds tests for opentitantool and hsmtool. Opentitantool uses files on disk and hsmtool uses The tests create a signature with one tool and verify it with the other. Interoperability between the tool-generated signatures and the firmware is tested by the |
cfrantz
force-pushed
the
fix-tool-endian
branch
6 times, most recently
from
8b11569 to
b3658a9
Compare
Opentitanlib was computing the SHA256 hash and storing it in a fixed-sized bigint. This led to various problems dealing with the endianness of the resulting digest, including emitting digest in "little-endian" order. A SHA256 digest _is not_ an integer; it is a sequence of bytes. 1. Eliminate the bigint representation of the SHA256 digest. Give the digest struct serialization, display and parsing traits to eliminate the need for dealing with conversions for those ues cases. Add `to_vec` and `to_vec_rev` to produce the raw bytes when needed. 2. Add a `--spx-hash-reversal-bug` switch to opentitantool spx commands for cases where firmware erroneously uses a reversed digest. 3. Eliminate the copious digest reversals present in hsmtool. Add a `Sha256HashReversed` input format for cases where firmware erroneously uses a reversed digest. 4. Fix the signing rules to use the reversed format only for keys tied to a firmware implementation with the reversal bug. Signed-off-by: Chris Frantz <cfrantz@google.com>
1. Add tests that check that signatures generated with opentitantool can be verified by hsmtool. 2. Add tests that check that signatures generated with hsmtool can be verified by opentitantool. 3. Verify the correctness of the manifest digest calculcated by opentitantool. Signed-off-by: Chris Frantz <cfrantz@google.com>
cfrantz
force-pushed
the
fix-tool-endian
branch
from
b3658a9 to
dd7c719
Compare
|
I also manually tested the offline signing flow and the CloudKMS signing flow. |
Opentitanlib was computing the SHA256 hash and storing it in a fixed-sized bigint. This led to various problems dealing with the endianness of the resulting digest, including emitting digest in "little-endian" order. A SHA256 digest is not an integer; it is a sequence of bytes.
to_vecandto_vec_revto produce the raw bytes when needed.--spx-hash-reversal-bugswitch to opentitantool spx commands for cases where firmware erroneously uses a reversed digest.Sha256HashReversedinput format for cases where firmware erroneously uses a reversed digest.Fixes #25870