[Bug] A malicious validator can broadcast certificate with Transaction::Fee and block the network

[feezybabee]

https://hackerone.com/reports/2310746

Summary:

Currently, Aleo does not speculate on Transaction::Fee:

                    // There are no finalize operations here.
                    // Note: This will abort the entire atomic batch.
                    Transaction::Fee(..) => Err("Cannot speculate on a fee transaction".to_string()),

However, malicious can propose a BatchPropose with Fee transaction. Other validators won't check if a BatchPropose contains a Fee transaction and will sign the proposal. Once the Certificate with Fee transaction is committed, it will block the execution of the block. All the validators will be blocked.

Steps To Reproduce:

1. Clone snarkOS
2. Checkout testnet3: git checkout 052457bddbf736f88227c543646890f8588a0efb
3. Apply the patch unchecked_fee_transaction.patch in the attachment: git apply unchecked_fee_transaction.patch
4. Run ./devnet.sh (To make it easy to reproduce, run all 4 validators with the patch)
5. Wait some round and find the network stuck:
   

Also see the attached sample log sample_log.zip
sample_log.zip

Proof-of-Concept (PoC)

See above

Fix Suggestions:

Check if a BatchPropose contains Transaction::Fee before signing.

Impact

A malicious validator can broadcast certificate with Transaction::Fee and block the whole network

[randomsleep]

@feezybabee It seems that unchecked_fee_transaction.patch is not attached. Let me put it here:
unchecked_fee_transaction.patch

[vicsn]

This looks legit, great work @randomsleep . @ljedrz do you have time coming week to test and propose the fix? See also #2991

[ljedrz]

I can reproduce the issue with the attached patch.

[raychu86]

Please try to reproduce this bug with the changes here - #3083

[ljedrz]

@raychu86 I tried to reproduce it with the mainnet branch, your PR, and the patch attached to this issue, and the network no longer halts; however, I'll retry it with a larger committee now, as a 4-member one had difficulty choosing a leader with the patch in place, once the fee txs were distributed.

[ljedrz]

Hmm, the way the patch works seems to interfere with the current consensus setup; while I'm pretty sure that we no longer hit the critical speculation bug, it would be best if we could come up with an updated way to reproduce the issue. @raychu86 could you suggest some method that would be the least "intrusive"?

[raychu86]

@ljedrz @randomsleep I've created a new proposed fix for this to just do verification for the proposals. Take a look here - #3098. Would be helpful to think about the implications of this fix.

[ozoekwe]

I tried to reproduce it with the mainnet branch, still not working but will find a way surely

[raychu86]

Closed with #3104.