-
Notifications
You must be signed in to change notification settings - Fork 707
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Staking ledger bonding fixes #3639
Staking ledger bonding fixes #3639
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Pretty important fix 🙏 .
The whole stash/controller logic is not so great. Intuitively I would have not expected Ledger
to be keyed by controller which seems the core issue behind this bug. Another observation is, it is not very clear why some extrinsics in staking pallet can be called via stash and others via controller.
Co-authored-by: Kian Paimani <[email protected]>
…es to leger are made through the staking ledger API
set_controller
and more try-runtime checks
@gpestana just for info: |
This pull request has been mentioned on Polkadot Forum. There might be relevant details there: https://forum.polkadot.network/t/parity-tech-update-for-march/7226/1 |
Currently, the staking logic does not prevent a controller from becoming a stash of *another* ledger (introduced by [removing this check](https://github.com/paritytech/polkadot-sdk/pull/1484/files#diff-3aa6ceab5aa4e0ab2ed73a7245e0f5b42e0832d8ca5b1ed85d7b2a52fb196524L850)). Given that the remaining of the code expects that never happens, bonding a ledger with a stash that is a controller of another ledger may lead to data inconsistencies and data losses in bonded ledgers. For more detailed explanation of this issue: https://hackmd.io/@gpestana/HJoBm2tqo/%2FTPdi28H7Qc2mNUqLSMn15w In a nutshell, when fetching a ledger with a given controller, we may be end up getting the wrong ledger which can lead to unexpected ledger states. This PR also ensures that `set_controller` does not lead to data inconsistencies in the staking ledger and bonded storage in the case when a controller of a stash is a stash of *another* ledger. and improves the staking `try-runtime` checks to catch potential issues with the storage preemptively. In summary, there are two important cases here: 1. **"Sane" double bonded ledger** When a controller of a ledger is a stash of *another* ledger. In this case, we have: ``` > Bonded(stash, controller) (A, B) // stash A with controller B (B, C) // B is also a stash of another ledger (C, D) > Ledger(controller) Ledger(B) = L_a (stash = A) Ledger(C) = L_b (stash = B) Ledger(D) = L_c (stash = C) ``` In this case, the ledgers can be mutated and all operations are OK. However, we should not allow `set_controller` to be called if it means it results in a "corrupt" double bonded ledger (see below). 3. **"Corrupt" double bonded ledger** ``` > Bonded(stash, controller) (A, B) // stash A with controller B (B, B) (C, D) ``` In this case, B is a stash and controller AND is corrupted, since B is responsible for 2 ledgers which is not correct and will lead to inconsistent states. Thus, in this case, in this PR we are preventing these ledgers from mutating (i.e. operations like bonding extra etc) until the ledger is brought back to a consistent state. --- **Changes**: - Checks if stash is already a controller when calling `Call::bond` (fixes the regression introduced by [removing this check](https://github.com/paritytech/polkadot-sdk/pull/1484/files#diff-3aa6ceab5aa4e0ab2ed73a7245e0f5b42e0832d8ca5b1ed85d7b2a52fb196524L850)); - Ensures that all fetching ledgers from storage are done through the `StakingLedger` API; - Ensures that -- when fetching a ledger from storage using the `StakingLedger` API --, a `Error::BadState` is returned if the ledger bonding is in a bad state. This prevents bad ledgers from mutating (e.g. `bond_extra`, `set_controller`, etc) its state and avoid further data inconsistencies. - Prevents stashes which are controllers or another ledger from calling `set_controller`, since that may lead to a bad state. - Adds further try-state runtime checks that check if there are ledgers in a bad state based on their bonded metadata. Related to #3245 --------- Co-authored-by: Kian Paimani <[email protected]> Co-authored-by: kianenigma <[email protected]>
Currently, the staking logic does not prevent a controller from becoming a stash of *another* ledger (introduced by [removing this check](https://github.com/paritytech/polkadot-sdk/pull/1484/files#diff-3aa6ceab5aa4e0ab2ed73a7245e0f5b42e0832d8ca5b1ed85d7b2a52fb196524L850)). Given that the remaining of the code expects that never happens, bonding a ledger with a stash that is a controller of another ledger may lead to data inconsistencies and data losses in bonded ledgers. For more detailed explanation of this issue: https://hackmd.io/@gpestana/HJoBm2tqo/%2FTPdi28H7Qc2mNUqLSMn15w In a nutshell, when fetching a ledger with a given controller, we may be end up getting the wrong ledger which can lead to unexpected ledger states. This PR also ensures that `set_controller` does not lead to data inconsistencies in the staking ledger and bonded storage in the case when a controller of a stash is a stash of *another* ledger. and improves the staking `try-runtime` checks to catch potential issues with the storage preemptively. In summary, there are two important cases here: 1. **"Sane" double bonded ledger** When a controller of a ledger is a stash of *another* ledger. In this case, we have: ``` > Bonded(stash, controller) (A, B) // stash A with controller B (B, C) // B is also a stash of another ledger (C, D) > Ledger(controller) Ledger(B) = L_a (stash = A) Ledger(C) = L_b (stash = B) Ledger(D) = L_c (stash = C) ``` In this case, the ledgers can be mutated and all operations are OK. However, we should not allow `set_controller` to be called if it means it results in a "corrupt" double bonded ledger (see below). 3. **"Corrupt" double bonded ledger** ``` > Bonded(stash, controller) (A, B) // stash A with controller B (B, B) (C, D) ``` In this case, B is a stash and controller AND is corrupted, since B is responsible for 2 ledgers which is not correct and will lead to inconsistent states. Thus, in this case, in this PR we are preventing these ledgers from mutating (i.e. operations like bonding extra etc) until the ledger is brought back to a consistent state. --- **Changes**: - Checks if stash is already a controller when calling `Call::bond` (fixes the regression introduced by [removing this check](https://github.com/paritytech/polkadot-sdk/pull/1484/files#diff-3aa6ceab5aa4e0ab2ed73a7245e0f5b42e0832d8ca5b1ed85d7b2a52fb196524L850)); - Ensures that all fetching ledgers from storage are done through the `StakingLedger` API; - Ensures that -- when fetching a ledger from storage using the `StakingLedger` API --, a `Error::BadState` is returned if the ledger bonding is in a bad state. This prevents bad ledgers from mutating (e.g. `bond_extra`, `set_controller`, etc) its state and avoid further data inconsistencies. - Prevents stashes which are controllers or another ledger from calling `set_controller`, since that may lead to a bad state. - Adds further try-state runtime checks that check if there are ledgers in a bad state based on their bonded metadata. Related to #3245 --------- Co-authored-by: Kian Paimani <[email protected]> Co-authored-by: kianenigma <[email protected]>
Currently, the staking logic does not prevent a controller from becoming a stash of *another* ledger (introduced by [removing this check](https://github.com/paritytech/polkadot-sdk/pull/1484/files#diff-3aa6ceab5aa4e0ab2ed73a7245e0f5b42e0832d8ca5b1ed85d7b2a52fb196524L850)). Given that the remaining of the code expects that never happens, bonding a ledger with a stash that is a controller of another ledger may lead to data inconsistencies and data losses in bonded ledgers. For more detailed explanation of this issue: https://hackmd.io/@gpestana/HJoBm2tqo/%2FTPdi28H7Qc2mNUqLSMn15w In a nutshell, when fetching a ledger with a given controller, we may be end up getting the wrong ledger which can lead to unexpected ledger states. This PR also ensures that `set_controller` does not lead to data inconsistencies in the staking ledger and bonded storage in the case when a controller of a stash is a stash of *another* ledger. and improves the staking `try-runtime` checks to catch potential issues with the storage preemptively. In summary, there are two important cases here: 1. **"Sane" double bonded ledger** When a controller of a ledger is a stash of *another* ledger. In this case, we have: ``` > Bonded(stash, controller) (A, B) // stash A with controller B (B, C) // B is also a stash of another ledger (C, D) > Ledger(controller) Ledger(B) = L_a (stash = A) Ledger(C) = L_b (stash = B) Ledger(D) = L_c (stash = C) ``` In this case, the ledgers can be mutated and all operations are OK. However, we should not allow `set_controller` to be called if it means it results in a "corrupt" double bonded ledger (see below). 3. **"Corrupt" double bonded ledger** ``` > Bonded(stash, controller) (A, B) // stash A with controller B (B, B) (C, D) ``` In this case, B is a stash and controller AND is corrupted, since B is responsible for 2 ledgers which is not correct and will lead to inconsistent states. Thus, in this case, in this PR we are preventing these ledgers from mutating (i.e. operations like bonding extra etc) until the ledger is brought back to a consistent state. --- **Changes**: - Checks if stash is already a controller when calling `Call::bond` (fixes the regression introduced by [removing this check](https://github.com/paritytech/polkadot-sdk/pull/1484/files#diff-3aa6ceab5aa4e0ab2ed73a7245e0f5b42e0832d8ca5b1ed85d7b2a52fb196524L850)); - Ensures that all fetching ledgers from storage are done through the `StakingLedger` API; - Ensures that -- when fetching a ledger from storage using the `StakingLedger` API --, a `Error::BadState` is returned if the ledger bonding is in a bad state. This prevents bad ledgers from mutating (e.g. `bond_extra`, `set_controller`, etc) its state and avoid further data inconsistencies. - Prevents stashes which are controllers or another ledger from calling `set_controller`, since that may lead to a bad state. - Adds further try-state runtime checks that check if there are ledgers in a bad state based on their bonded metadata. Related to #3245 --------- Co-authored-by: Kian Paimani <[email protected]> Co-authored-by: kianenigma <[email protected]>
Currently, the staking logic does not prevent a controller from becoming a stash of *another* ledger (introduced by [removing this check](https://github.com/paritytech/polkadot-sdk/pull/1484/files#diff-3aa6ceab5aa4e0ab2ed73a7245e0f5b42e0832d8ca5b1ed85d7b2a52fb196524L850)). Given that the remaining of the code expects that never happens, bonding a ledger with a stash that is a controller of another ledger may lead to data inconsistencies and data losses in bonded ledgers. For more detailed explanation of this issue: https://hackmd.io/@gpestana/HJoBm2tqo/%2FTPdi28H7Qc2mNUqLSMn15w In a nutshell, when fetching a ledger with a given controller, we may be end up getting the wrong ledger which can lead to unexpected ledger states. This PR also ensures that `set_controller` does not lead to data inconsistencies in the staking ledger and bonded storage in the case when a controller of a stash is a stash of *another* ledger. and improves the staking `try-runtime` checks to catch potential issues with the storage preemptively. In summary, there are two important cases here: 1. **"Sane" double bonded ledger** When a controller of a ledger is a stash of *another* ledger. In this case, we have: ``` > Bonded(stash, controller) (A, B) // stash A with controller B (B, C) // B is also a stash of another ledger (C, D) > Ledger(controller) Ledger(B) = L_a (stash = A) Ledger(C) = L_b (stash = B) Ledger(D) = L_c (stash = C) ``` In this case, the ledgers can be mutated and all operations are OK. However, we should not allow `set_controller` to be called if it means it results in a "corrupt" double bonded ledger (see below). 3. **"Corrupt" double bonded ledger** ``` > Bonded(stash, controller) (A, B) // stash A with controller B (B, B) (C, D) ``` In this case, B is a stash and controller AND is corrupted, since B is responsible for 2 ledgers which is not correct and will lead to inconsistent states. Thus, in this case, in this PR we are preventing these ledgers from mutating (i.e. operations like bonding extra etc) until the ledger is brought back to a consistent state. --- **Changes**: - Checks if stash is already a controller when calling `Call::bond` (fixes the regression introduced by [removing this check](https://github.com/paritytech/polkadot-sdk/pull/1484/files#diff-3aa6ceab5aa4e0ab2ed73a7245e0f5b42e0832d8ca5b1ed85d7b2a52fb196524L850)); - Ensures that all fetching ledgers from storage are done through the `StakingLedger` API; - Ensures that -- when fetching a ledger from storage using the `StakingLedger` API --, a `Error::BadState` is returned if the ledger bonding is in a bad state. This prevents bad ledgers from mutating (e.g. `bond_extra`, `set_controller`, etc) its state and avoid further data inconsistencies. - Prevents stashes which are controllers or another ledger from calling `set_controller`, since that may lead to a bad state. - Adds further try-state runtime checks that check if there are ledgers in a bad state based on their bonded metadata. Related to #3245 --------- Co-authored-by: Kian Paimani <[email protected]> Co-authored-by: kianenigma <[email protected]>
This backport PR should bump the `pallet-staking` from 30.0.1 to 30.0.2. Backports for 1.8: - #3639 Relevant Issues: - #3245 --------- Co-authored-by: Kian Paimani <[email protected]> Co-authored-by: kianenigma <[email protected]>
This backport PR should bump the `pallet-staking` from 28.0.0 to 28.0.1. Backports for 1.6: - #3639 Relevant Issues: - #3245 --------- Co-authored-by: Kian Paimani <[email protected]> Co-authored-by: kianenigma <[email protected]>
This backport PR should bump the `pallet-staking` from 27.0.0 to 27.0.1 Backports for 1.5: - #3639 Relevant Issues: - #3245 Co-authored-by: Kian Paimani <[email protected]> Co-authored-by: kianenigma <[email protected]>
This backport PR should bump the `pallet-staking` from 26.0.1 to 26.0.2. Backports for 1.4: - #3639 Relevant Issues: - #3245 Co-authored-by: Kian Paimani <[email protected]> Co-authored-by: kianenigma <[email protected]>
This pull request has been mentioned on Polkadot Forum. There might be relevant details there: https://forum.polkadot.network/t/recover-corrupted-staking-ledgers-in-polkadot-and-kusama/9796/1 |
Currently, the staking logic does not prevent a controller from becoming a stash of another ledger (introduced by removing this check). Given that the remaining of the code expects that never happens, bonding a ledger with a stash that is a controller of another ledger may lead to data inconsistencies and data losses in bonded ledgers. For more detailed explanation of this issue: https://hackmd.io/@gpestana/HJoBm2tqo/%2FTPdi28H7Qc2mNUqLSMn15w
In a nutshell, when fetching a ledger with a given controller, we may be end up getting the wrong ledger which can lead to unexpected ledger states.
This PR also ensures that
set_controller
does not lead to data inconsistencies in the staking ledger and bonded storage in the case when a controller of a stash is a stash of another ledger. and improves the stakingtry-runtime
checks to catch potential issues with the storage preemptively.In summary, there are two important cases here:
When a controller of a ledger is a stash of another ledger. In this case, we have:
In this case, the ledgers can be mutated and all operations are OK. However, we should not allow
set_controller
to be called if it means it results in a "corrupt" double bonded ledger (see below).In this case, B is a stash and controller AND is corrupted, since B is responsible for 2 ledgers which is not correct and will lead to inconsistent states. Thus, in this case, in this PR we are preventing these ledgers from mutating (i.e. operations like bonding extra etc) until the ledger is brought back to a consistent state.
Changes:
Call::bond
(fixes the regression introduced by removing this check);StakingLedger
API;StakingLedger
API --, aError::BadState
is returned if the ledger bonding is in a bad state. This prevents bad ledgers from mutating (e.g.bond_extra
,set_controller
, etc) its state and avoid further data inconsistencies.set_controller
, since that may lead to a bad state.Related to #3245