That doesn't answer my question - if theft of this server would not compromise security in any way, then why is this server needed at all and how does this server improve security?
If something improves security, then removing (stealing) that thing, would decrease security. Am I wrong?
It's quite simple, there are only a few rules:
1) you
can create checkpoint, if it has no conflicts with existing checkpoints;
2) you can't replace existing checkpoint with the new one;
3) you can't create checkpoint for the block with height before existing checkpoint, if this block belongs to another chain;
4) you can't create checkpoint for non-existing block (actually you can submit checkpoint with the random block hash value, but this checkpoint will never mature).
This allows attacker to perform some scenarios:
1. Without checkpoints, or with
patched client:
1) 7 blocks found on the main chain;
2) attacker generates 8 blocks in offline, and then publishes his block chain;
3) 7 blocks from the main chain are getting orphaned and replaced by the 8 blocks, which generated by attacker;
4) the miners or a scam victims are crashing their heads against the wall.
2. With checkpoints:
1) 7 blocks found on the main chain;
2) checkpointing node sends checkpoint for the 2nd block;
3) attacker generates 8 blocks in offline, and then publishes his chain;
4) attacker's block chain conflicts with the checkpoint and rejected by network, the main chain is unchanged.
5) attacker is crashing his head against the wall.
3. With compromised checkpointing key:
1) 7 blocks found on the main chain;
2) real checkpointing node sends checkpoint for the 2nd block;
3) attacker generates 8 blocks in offline, and then publishes his chain;
4) attacker's block chain conflicts with the checkpoint and rejected by network, the main chain is unchanged;
5) attacker uses compromised key and trying to submit checkpoint for the first block of his chain, in order to overtake existing checkpoint;
6) the new checkpoint has conflict with already existing checkpoint and as the result, it's rejected by the network;
7) attacker is crashing his head against the wall.
4. With compromised checkpointing key, stolen or DDoS'd server:
1) Users are getting message that existing checkpoint is too old. This message convinces them to use escrow, while this issue isn't resolved.
5. With compromised checkpointing key and stolen or DDoS'd server, which replaced with the new one, belongs to attacker (1st scenario):
1) 7 blocks found on the main chain;
2) attacker node sends checkpoint for the 2nd block;
3) attacker generates 8 blocks in offline, and then publishes his chain;
4) attacker uses compromised key and tries to submit checkpoint for the first block of his chain;
6) the new checkpoint has conflict with already existing checkpoint and as the result, it's rejected by the network;
7) attacker is crashing his head against the wall.
6. With compromised checkpointing key and stolen or DDoS'd server, which replaced with the new one, belongs to attacker (2nd scenario):
1) 7 blocks found on the main chain;
2) attacker ignores them, and generates own block instead;
3) attacker uses compromised key and tries to submit checkpoint for the first block of his own chain;
4) users are syncronizing with attacker's chain instead the original one, because original one conflicts with the checkpoint.
As the result, key holder has very restricted rights in the system. His abilities are limited with existing history protection, he can't alter network history if it was checkpointed before.
Maybe I missed something, haven't slept for 47h. %)
Anyway, I think that client
must include an option, which allows user to bypass checkpoints without applying the custom patches.
