Abstract
Iroha Special Instructions processing requires to have a permissions based security model + a way to store some "triggers" on the block chain.
Whitepaper and Iroha v1 documentation were researched. The proposal is to use already existing Iroha Special Instructions + Assets mechanisms for Permissions implementation.
Introduction
Whitepaper gave some information about https://github.com/hyperledger/iroha/blob/iroha2-dev/iroha_2_whitepaper.md#251-event-listeners, and the need to trigger them based on time, block or "condition".
https://github.com/hyperledger/iroha/blob/iroha2-dev/iroha_2_whitepaper.md#211-data-permissions require protection of data from read and write cases.
https://iroha.readthedocs.io/en/master/concepts_architecture/glossary.html#permission gives a Permission definition:
A named rule that gives the privilege to perform a command. Permission cannot be granted to an account directly, instead, account has roles, which are collections of permissions. Although, there is an exception, see Grantable Permission.
and Grant-able Permission:
Only grantable permission is given to anaccountdirectly. An account that holds grantable permission is allowed to perform some particular action on behalf of another account. For example, if accounta@domain1gives the accountb@domain2a permission that it can transfer assets — thenb@domain2can transfer assets ofa@domain1to anyone.
As you can see permissions were a first-level entities in Iroha v1 while they can be easily implemented by Iroha Special Instructions.
pub mod isi {
...
enum Instruction {
Add(...),
Register(...),
...
Check(permissions::isi::CheckInstruction),
}
}
pub mod permissions::isi {
pub struct CheckInstruction<C, O> {
condition: C,
object: O,
}
}
Signature of `Instruction::execute` method provides an ability to return error result which for `Check` variant will prevent execution of Instructions for which account has no permissions. Each OOB Instruction and Query will include `Check` instructions in it:
impl Register {
fn execute(...) -> Result<...> {
Check{|account| account.asset("role") == ADMIN }.execute(...)?;
...
}
}
In this example you can see that we do not need to add permission related attributes to the `Account` and can use assets instead.
The same can be done with custom permissions, storing them in assets components of the account.
This whole concept can be used for event listeners, `Instruction::Listener` can be executed for their configuration, but additionally to storing them in assets components of the account we need to find a way to store global Listeners (Domain level, Peer level) and define cases where we need time based listeners.
"Listeners" store
Current design assumes that we provide instructions with a mutable reference to `WorldStateView` after their validation and store on the block chain. So the only way to trigger Listeners of Conditions (changes in World State View) and Blocks and, possibly, Time is to store them in `WorldStateView`:
6 /// Current state of the blockchain alligned with `Iroha` module.
7 #[derive(Debug, Clone)]
8 pub struct WorldStateView {
9 peer: Peer,
listeners: Vec<ListenerInstruction>,
10 }
11
12 impl WorldStateView {
13 /// Default `WorldStateView` constructor.
14 +--- 3 lines: pub fn new(peer: Peer) -> Self {··························································································································
17
18 /// Put `ValidBlock` of information with changes in form of **Iroha Special Instructions**
19 /// into the world.
20 pub async fn put(&mut self, block: &CommittedBlock) {
21 for transaction in &block.transactions {
22 if let Err(e) = &transaction.proceed(self) {
23 eprintln!("Failed to procced transaction on WSV: {}", e);
24 }
25 }
26 }
As you can see - `put` method may invoke (trigger) listeners execution (because they are just a set of out of the box instructions). The question is - should we execute them here or send a new transaction to the ledger?
ASAP execution
Let's imagine as soon as possible execution and consequences:
12 impl WorldStateView {
13 /// Default `WorldStateView` constructor.
14 +--- 3 lines: pub fn new(peer: Peer) -> Self {··························································································································
17
18 /// Put `ValidBlock` of information with changes in form of **Iroha Special Instructions**
19 /// into the world.
20 pub async fn put(&mut self, block: &CommittedBlock) {
21 for transaction in &block.transactions {
22 if let Err(e) = &transaction.proceed(self) {
23 eprintln!("Failed to procced transaction on WSV: {}", e);
24 }
for listener in self.listeners {
listener.execute(self);
}
25 }
26 }
First question - how should listener understand was it triggered or not? It has only current state, not the delta of changes.
Second one - how should we spread results of it's execution between peers and synchronize I/O actions like notifications, etc.?
Third one - how should we check that listener execution will not break next transactions?
This questions gave us following requirements:
- Listeners should be executed only once in the ledger
- Listeners should be able to track changes (in time, blockchain and world_state_view)
- Listeners should be not executed during validation phase (to prevent unexpected I/O)
- Listeners should be executed after all block's transactions execution
Follow-up execution
What if instead of asap execution we will only form a new transaction with a set of instructions to execute?
18 /// Put `ValidBlock` of information with changes in form of **Iroha Special Instructions**
19 /// into the world.
20 pub async fn put(&mut self, block: &CommittedBlock) {
21 for transaction in &block.transactions {
22 if let Err(e) = &transaction.proceed(self) {
23 eprintln!("Failed to procced transaction on WSV: {}", e);
24 }
25 }
client.submit_all(
listeners.execute(self, block)
);
26 }
But almost all requirements stay open with this solution.
The problematic
The main problem is how Iroha Special Instructions with only World State View as input can be triggered by World State View changes and where to store such Iroha Special Instructions.
World State View by itself held all information needed for the triggering but the current ISI model doesn't provide a way to use them instead of "hardcoded" entities.
If we will be able to make new Out of the Box set of ISI - `ListenerInstruction` with possible variants like `On{trigger: Box<Instruction>, generators: Box<Fn(?)→[Instruction]}` we will be able to store them on the Peer and execute on every `WorldStateView:put` which will result in a new set of generated instructions. This brings us to the next question - what this triggered instructions should know about their triggers - if nothing than we are good with current `ISI::execute` signature. But if we need to pass some information, we should store this info. And to do not change good (in my humble opinion) signature of `ISI::execute` we should store this trigger-related information in generated Instructions. Which means that for cases where we need this information we need special Iroha Special Instructions =)
Let's try to simplify these thoughts:
BLOCK --> WorldStateView WorldStateView --check triggers & generate [ISI]--> WorldStateView::Vec<ListenerInstruction> WorldStateView --execute instructions--> BLOCK WorldStateView --send new transaction--> Iroha
This approach fits almost all our requirements, the problem is to not send new transactions on validation stage and prevent execution of I\O related instructions multiple times.
The second one should be moved into separate RFC because it also implied to regular ISI.
The first one should be addressed here.