frostfs-contract/netmap/netmap_contract.go
Anton Nikiforov 2a889c316c [#7] Remove notary disabled code
Signed-off-by: Anton Nikiforov <an.nikiforov@yadro.com>
2023-03-14 12:04:01 +03:00

569 lines
17 KiB
Go

package netmap
import (
"git.frostfs.info/TrueCloudLab/frostfs-contract/common"
"github.com/nspcc-dev/neo-go/pkg/interop"
"github.com/nspcc-dev/neo-go/pkg/interop/contract"
"github.com/nspcc-dev/neo-go/pkg/interop/iterator"
"github.com/nspcc-dev/neo-go/pkg/interop/native/ledger"
"github.com/nspcc-dev/neo-go/pkg/interop/native/management"
"github.com/nspcc-dev/neo-go/pkg/interop/native/std"
"github.com/nspcc-dev/neo-go/pkg/interop/runtime"
"github.com/nspcc-dev/neo-go/pkg/interop/storage"
)
// NodeState is an enumeration for node states.
type NodeState int
// Various Node states
const (
_ NodeState = iota
// NodeStateOnline stands for nodes that are in full network and
// operational availability.
NodeStateOnline
// NodeStateOffline stands for nodes that are in network unavailability.
NodeStateOffline
// NodeStateMaintenance stands for nodes under maintenance with partial
// network availability.
NodeStateMaintenance
)
// Node groups data related to FrostFS storage nodes registered in the FrostFS
// network. The information is stored in the current contract.
type Node struct {
// Information about the node encoded according to the FrostFS binary
// protocol.
BLOB []byte
// Current node state.
State NodeState
}
const (
notaryDisabledKey = "notary"
innerRingKey = "innerring"
// DefaultSnapshotCount contains the number of previous snapshots stored by this contract.
// Must be less than 255.
DefaultSnapshotCount = 10
snapshotCountKey = "snapshotCount"
snapshotKeyPrefix = "snapshot_"
snapshotCurrentIDKey = "snapshotCurrent"
snapshotEpoch = "snapshotEpoch"
snapshotBlockKey = "snapshotBlock"
containerContractKey = "containerScriptHash"
balanceContractKey = "balanceScriptHash"
cleanupEpochMethod = "newEpoch"
)
var (
configPrefix = []byte("config")
candidatePrefix = []byte("candidate")
)
// _deploy function sets up initial list of inner ring public keys.
func _deploy(data interface{}, isUpdate bool) {
ctx := storage.GetContext()
common.RmAndCheckNotaryDisabledKey(data, notaryDisabledKey)
var args = data.(struct {
//TODO(@acid-ant): #9 remove notaryDisabled in future version
notaryDisabled bool
addrBalance interop.Hash160
addrContainer interop.Hash160
keys []interop.PublicKey
config [][]byte
version int
})
ln := len(args.config)
if ln%2 != 0 {
panic("bad configuration")
}
for i := 0; i < ln/2; i++ {
key := args.config[i*2]
val := args.config[i*2+1]
setConfig(ctx, key, val)
}
if isUpdate {
common.CheckVersion(args.version)
return
}
if len(args.addrBalance) != interop.Hash160Len || len(args.addrContainer) != interop.Hash160Len {
panic("incorrect length of contract script hash")
}
// epoch number is a little endian int, it doesn't need to be serialized
storage.Put(ctx, snapshotCountKey, DefaultSnapshotCount)
storage.Put(ctx, snapshotEpoch, 0)
storage.Put(ctx, snapshotBlockKey, 0)
prefix := []byte(snapshotKeyPrefix)
for i := 0; i < DefaultSnapshotCount; i++ {
common.SetSerialized(ctx, append(prefix, byte(i)), []Node{})
}
storage.Put(ctx, snapshotCurrentIDKey, 0)
storage.Put(ctx, balanceContractKey, args.addrBalance)
storage.Put(ctx, containerContractKey, args.addrContainer)
runtime.Log("netmap contract initialized")
}
// Update method updates contract source code and manifest. It can be invoked
// only by committee.
func Update(script []byte, manifest []byte, data interface{}) {
if !common.HasUpdateAccess() {
panic("only committee can update contract")
}
contract.Call(interop.Hash160(management.Hash), "update",
contract.All, script, manifest, common.AppendVersion(data))
runtime.Log("netmap contract updated")
}
// AddPeerIR accepts Alphabet calls in the notary-enabled contract setting and
// behaves similar to AddPeer in the notary-disabled one.
//
// AddPeerIR MUST NOT be called in notary-disabled contract setting.
// AddPeerIR MUST be called by the Alphabet member only.
func AddPeerIR(nodeInfo []byte) {
ctx := storage.GetContext()
common.CheckAlphabetWitness(common.AlphabetAddress())
publicKey := nodeInfo[2:35] // V2 format: offset:2, len:33
addToNetmap(ctx, publicKey, Node{
BLOB: nodeInfo,
State: NodeStateOnline,
})
}
// AddPeer accepts information about the network map candidate in the FrostFS
// binary protocol format and does nothing. Keep method because storage node
// creates a notary transaction with this method, which produces a notary
// notification (implicit here).
func AddPeer(nodeInfo []byte) {
// V2 format - offset:2, len:33
common.CheckWitness(nodeInfo[2:35])
return
}
// updates state of the network map candidate by its public key in the contract
// storage, and throws UpdateStateSuccess notification after this.
//
// State MUST be from the NodeState enum.
func updateCandidateState(ctx storage.Context, publicKey interop.PublicKey, state NodeState) {
switch state {
case NodeStateOffline:
removeFromNetmap(ctx, publicKey)
runtime.Log("remove storage node from the network map")
case NodeStateOnline, NodeStateMaintenance:
updateNetmapState(ctx, publicKey, state)
runtime.Log("update state of the network map candidate")
default:
panic("unsupported state")
}
runtime.Notify("UpdateStateSuccess", publicKey, state)
}
// UpdateState accepts new state to be assigned to network map candidate
// identified by the given public key, identifies the signer.
// Applicable only for notary-enabled environment.
//
// Signers:
//
// (a) candidate himself only, if provided public key corresponds to the signer
// (b) Alphabet member only
// (ab) both candidate and Alphabet member
// (c) others
//
// UpdateState case-by-case behavior:
//
// (a) panics
// (b) panics
// (ab) updates candidate's state in the contract storage (*), and throws
// UpdateStateSuccess with the provided key and new state
// (c) panics
//
// (*) Candidate is removed from the candidate set if state is NodeStateOffline.
// Any other state is written into candidate's descriptor in the contract storage.
// If requested candidate is missing, panic occurs. Throws UpdateStateSuccess
// notification on success.
//
// State MUST be from the NodeState enum. Public key MUST be
// interop.PublicKeyCompressedLen bytes.
func UpdateState(state NodeState, publicKey interop.PublicKey) {
if len(publicKey) != interop.PublicKeyCompressedLen {
panic("incorrect public key")
}
ctx := storage.GetContext()
common.CheckWitness(publicKey)
common.CheckAlphabetWitness(common.AlphabetAddress())
updateCandidateState(ctx, publicKey, state)
}
// UpdateStateIR accepts Alphabet calls in the notary-enabled contract setting
// and behaves similar to UpdateState, but does not require candidate's
// signature presence.
//
// UpdateStateIR MUST NOT be called in notary-disabled contract setting.
// UpdateStateIR MUST be called by the Alphabet member only.
func UpdateStateIR(state NodeState, publicKey interop.PublicKey) {
ctx := storage.GetContext()
common.CheckAlphabetWitness(common.AlphabetAddress())
updateCandidateState(ctx, publicKey, state)
}
// NewEpoch method changes the epoch number up to the provided epochNum argument. It can
// be invoked only by Alphabet nodes. If provided epoch number is less than the
// current epoch number or equals it, the method throws panic.
//
// When epoch number is updated, the contract sets storage node candidates as the current
// network map. The contract also invokes NewEpoch method on Balance and Container
// contracts.
//
// It produces NewEpoch notification.
func NewEpoch(epochNum int) {
ctx := storage.GetContext()
multiaddr := common.AlphabetAddress()
common.CheckAlphabetWitness(multiaddr)
currentEpoch := storage.Get(ctx, snapshotEpoch).(int)
if epochNum <= currentEpoch {
panic("invalid epoch") // ignore invocations with invalid epoch
}
dataOnlineState := filterNetmap(ctx)
runtime.Log("process new epoch")
// todo: check if provided epoch number is bigger than current
storage.Put(ctx, snapshotEpoch, epochNum)
storage.Put(ctx, snapshotBlockKey, ledger.CurrentIndex())
id := storage.Get(ctx, snapshotCurrentIDKey).(int)
id = (id + 1) % getSnapshotCount(ctx)
storage.Put(ctx, snapshotCurrentIDKey, id)
// put netmap into actual snapshot
common.SetSerialized(ctx, snapshotKeyPrefix+string([]byte{byte(id)}), dataOnlineState)
// make clean up routines in other contracts
cleanup(ctx, epochNum)
runtime.Notify("NewEpoch", epochNum)
}
// Epoch method returns the current epoch number.
func Epoch() int {
ctx := storage.GetReadOnlyContext()
return storage.Get(ctx, snapshotEpoch).(int)
}
// LastEpochBlock method returns the block number when the current epoch was applied.
func LastEpochBlock() int {
ctx := storage.GetReadOnlyContext()
return storage.Get(ctx, snapshotBlockKey).(int)
}
// Netmap returns set of information about the storage nodes representing a network
// map in the current epoch.
//
// Current state of each node is represented in the State field. It MAY differ
// with the state encoded into BLOB field, in this case binary encoded state
// MUST NOT be processed.
func Netmap() []Node {
ctx := storage.GetReadOnlyContext()
id := storage.Get(ctx, snapshotCurrentIDKey).(int)
return getSnapshot(ctx, snapshotKeyPrefix+string([]byte{byte(id)}))
}
// NetmapCandidates returns set of information about the storage nodes
// representing candidates for the network map in the coming epoch.
//
// Current state of each node is represented in the State field. It MAY differ
// with the state encoded into BLOB field, in this case binary encoded state
// MUST NOT be processed.
func NetmapCandidates() []Node {
ctx := storage.GetReadOnlyContext()
return getNetmapNodes(ctx)
}
// Snapshot returns set of information about the storage nodes representing a network
// map in (current-diff)-th epoch.
//
// Diff MUST NOT be negative. Diff MUST be less than maximum number of network
// map snapshots stored in the contract. The limit is a contract setting,
// DefaultSnapshotCount by default. See UpdateSnapshotCount for details.
//
// Current state of each node is represented in the State field. It MAY differ
// with the state encoded into BLOB field, in this case binary encoded state
// MUST NOT be processed.
func Snapshot(diff int) []Node {
ctx := storage.GetReadOnlyContext()
count := getSnapshotCount(ctx)
if diff < 0 || count <= diff {
panic("incorrect diff")
}
id := storage.Get(ctx, snapshotCurrentIDKey).(int)
needID := (id - diff + count) % count
key := snapshotKeyPrefix + string([]byte{byte(needID)})
return getSnapshot(ctx, key)
}
func getSnapshotCount(ctx storage.Context) int {
return storage.Get(ctx, snapshotCountKey).(int)
}
// UpdateSnapshotCount updates the number of the stored snapshots.
// If a new number is less than the old one, old snapshots are removed.
// Otherwise, history is extended with empty snapshots, so
// `Snapshot` method can return invalid results for `diff = new-old` epochs
// until `diff` epochs have passed.
//
// Count MUST NOT be negative.
func UpdateSnapshotCount(count int) {
common.CheckAlphabetWitness(common.AlphabetAddress())
if count < 0 {
panic("count must be positive")
}
ctx := storage.GetContext()
curr := getSnapshotCount(ctx)
if curr == count {
panic("count has not changed")
}
storage.Put(ctx, snapshotCountKey, count)
id := storage.Get(ctx, snapshotCurrentIDKey).(int)
var delStart, delFinish int
if curr < count {
// Increase history size.
//
// Old state (N = count, K = curr, E = current index, C = current epoch)
// KEY INDEX: 0 | 1 | ... | E | E+1 | ... | K-1 | ... | N-1
// EPOCH : C-E | C-E+1 | ... | C | C-K+1 | ... | C-E-1 |
//
// New state:
// KEY INDEX: 0 | 1 | ... | E | E+1 | ... | K-1 | ... | N-1
// EPOCH : C-E | C-E+1 | ... | C | nil | ... | . | ... | C-E-1
//
// So we need to move tail snapshots N-K keys forward,
// i.e. from E+1 .. K to N-K+E+1 .. N
diff := count - curr
lower := diff + id + 1
for k := count - 1; k >= lower; k-- {
moveSnapshot(ctx, k-diff, k)
}
delStart, delFinish = id+1, id+1+diff
if curr < delFinish {
delFinish = curr
}
} else {
// Decrease history size.
//
// Old state (N = curr, K = count)
// KEY INDEX: 0 | 1 | ... K1 ... | E | E+1 | ... K2-1 ... | N-1
// EPOCH : C-E | C-E+1 | ... .. ... | C | C-N+1 | ... ... ... | C-E-1
var step, start int
if id < count {
// K2 case, move snapshots from E+1+N-K .. N-1 range to E+1 .. K-1
// New state:
// KEY INDEX: 0 | 1 | ... | E | E+1 | ... | K-1
// EPOCH : C-E | C-E+1 | ... | C | C-K+1 | ... | C-E-1
step = curr - count
start = id + 1
} else {
// New state:
// KEY INDEX: 0 | 1 | ... | K-1
// EPOCH : C-K+1 | C-K+2 | ... | C
// K1 case, move snapshots from E-K+1 .. E range to 0 .. K-1
// AND replace current id with K-1
step = id - count + 1
storage.Put(ctx, snapshotCurrentIDKey, count-1)
}
for k := start; k < count; k++ {
moveSnapshot(ctx, k+step, k)
}
delStart, delFinish = count, curr
}
for k := delStart; k < delFinish; k++ {
key := snapshotKeyPrefix + string([]byte{byte(k)})
storage.Delete(ctx, key)
}
}
func moveSnapshot(ctx storage.Context, from, to int) {
keyFrom := snapshotKeyPrefix + string([]byte{byte(from)})
keyTo := snapshotKeyPrefix + string([]byte{byte(to)})
data := storage.Get(ctx, keyFrom)
storage.Put(ctx, keyTo, data)
}
// SnapshotByEpoch returns set of information about the storage nodes representing
// a network map in the given epoch.
//
// Behaves like Snapshot: it is called after difference with the current epoch is
// calculated.
func SnapshotByEpoch(epoch int) []Node {
ctx := storage.GetReadOnlyContext()
currentEpoch := storage.Get(ctx, snapshotEpoch).(int)
return Snapshot(currentEpoch - epoch)
}
// Config returns configuration value of FrostFS configuration. If key does
// not exists, returns nil.
func Config(key []byte) interface{} {
ctx := storage.GetReadOnlyContext()
return getConfig(ctx, key)
}
// SetConfig key-value pair as a FrostFS runtime configuration value. It can be invoked
// only by Alphabet nodes.
func SetConfig(id, key, val []byte) {
ctx := storage.GetContext()
multiaddr := common.AlphabetAddress()
common.CheckAlphabetWitness(multiaddr)
setConfig(ctx, key, val)
runtime.Log("configuration has been updated")
}
type record struct {
key []byte
val []byte
}
// ListConfig returns an array of structures that contain key and value of all
// FrostFS configuration records. Key and value are both byte arrays.
func ListConfig() []record {
ctx := storage.GetReadOnlyContext()
var config []record
it := storage.Find(ctx, configPrefix, storage.None)
for iterator.Next(it) {
pair := iterator.Value(it).(struct {
key []byte
val []byte
})
r := record{key: pair.key[len(configPrefix):], val: pair.val}
config = append(config, r)
}
return config
}
// Version returns the version of the contract.
func Version() int {
return common.Version
}
// serializes and stores the given Node by its public key in the contract storage,
// and throws AddPeerSuccess notification after this.
//
// Public key MUST match the one encoded in BLOB field.
func addToNetmap(ctx storage.Context, publicKey []byte, node Node) {
storageKey := append(candidatePrefix, publicKey...)
storage.Put(ctx, storageKey, std.Serialize(node))
runtime.Notify("AddPeerSuccess", interop.PublicKey(publicKey))
}
func removeFromNetmap(ctx storage.Context, key interop.PublicKey) {
storageKey := append(candidatePrefix, key...)
storage.Delete(ctx, storageKey)
}
func updateNetmapState(ctx storage.Context, key interop.PublicKey, state NodeState) {
storageKey := append(candidatePrefix, key...)
raw := storage.Get(ctx, storageKey).([]byte)
if raw == nil {
panic("peer is missing")
}
node := std.Deserialize(raw).(Node)
node.State = state
storage.Put(ctx, storageKey, std.Serialize(node))
}
func filterNetmap(ctx storage.Context) []Node {
var (
netmap = getNetmapNodes(ctx)
result = []Node{}
)
for i := 0; i < len(netmap); i++ {
item := netmap[i]
if item.State != NodeStateOffline {
result = append(result, item)
}
}
return result
}
func getNetmapNodes(ctx storage.Context) []Node {
result := []Node{}
it := storage.Find(ctx, candidatePrefix, storage.ValuesOnly|storage.DeserializeValues)
for iterator.Next(it) {
node := iterator.Value(it).(Node)
result = append(result, node)
}
return result
}
func getSnapshot(ctx storage.Context, key string) []Node {
data := storage.Get(ctx, key)
if data != nil {
return std.Deserialize(data.([]byte)).([]Node)
}
return []Node{}
}
func getConfig(ctx storage.Context, key interface{}) interface{} {
postfix := key.([]byte)
storageKey := append(configPrefix, postfix...)
return storage.Get(ctx, storageKey)
}
func setConfig(ctx storage.Context, key, val interface{}) {
postfix := key.([]byte)
storageKey := append(configPrefix, postfix...)
storage.Put(ctx, storageKey, val)
}
func cleanup(ctx storage.Context, epoch int) {
balanceContractAddr := storage.Get(ctx, balanceContractKey).(interop.Hash160)
contract.Call(balanceContractAddr, cleanupEpochMethod, contract.All, epoch)
containerContractAddr := storage.Get(ctx, containerContractKey).(interop.Hash160)
contract.Call(containerContractAddr, cleanupEpochMethod, contract.All, epoch)
}