Validate finality mroot of Savanna blocks as a part of header validation

[linh2931]

For Savanna blocks, add finality mroot validation to verify_basic_proper_block_invariants() as a part of header validation.

#730 called for validating block ID as as part of block header validation. While investigating the issue, it is discovered that new_qc_claims block_num and finality mroot could be validated during the header validation. new_qc_claim's block_num validation was done by #995. Block ID validation is not possible at header validation stage as new transaction_mroot is not available before transactions are applied.

This PR adds finality mroot validation and a unit test for the validation.

Resolves #730

[heifner]

I may have missed something here, but this doesn't look thread safe. This can be filled in when validating the block which could be happening at the same time this is running on the net thread. Can you instead check validated and then assert(prev.valid)? Is it correct that prev.valid is set when validated==true?

[linh2931]

valid

spring/libraries/chain/include/eosio/chain/block_state.hpp

Line 77 in e43fd1e

std::optional<valid_t> valid;

is for Finality Merkle Tree while validated

spring/libraries/chain/include/eosio/chain/block_state.hpp

Line 81 in e43fd1e

copyable_atomic<bool> validated{false}; // We have executed the block's trxs and verified that action merkle root (block id) matches.

is mostly for forkbd. They are independent. Since I am using previous block's valid structure, it should be already constructed. I don't see a thread safe problem.

[heifner]

During the transition to Savanna it is modified here:

spring/libraries/chain/controller.cpp

Line 1385 in 53ad484

// irreversible apply was just done, calculate new_valid here instead of in transition_to_savanna()

spring/libraries/chain/controller.cpp

Line 1406 in 53ad484

new_bsp->valid = prev->new_valid(*new_bsp, *legacy->action_mroot_savanna, new_bsp->strong_digest);

Even outside Savanna transition it is modified here:

spring/libraries/chain/controller.cpp

Line 804 in 53ad484

validating_bsp->valid = bb.parent.new_valid(bhs, action_mroot, validating_bsp->strong_digest);

This can happen while the net threads are accessing prev.valid.

I believe it is correct that valid will be filled in when validated is true. Since validated is atomic you could check that value here instead of valid.

[greg7mdp]

We should probably use either finality_mroot or action_mroot, but not both.

[linh2931]

Will change to use finality_mroot as it is a Savanna thing.

[ericpassmore]

Note:start
category: Other
component: Internal
summary: Validate finality mroot of Savanna blocks as a part of header validation.
Note:end