downloader.go 78 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108
  1. // Copyright 2015 The go-ethereum Authors
  2. // This file is part of the go-ethereum library.
  3. //
  4. // The go-ethereum library is free software: you can redistribute it and/or modify
  5. // it under the terms of the GNU Lesser General Public License as published by
  6. // the Free Software Foundation, either version 3 of the License, or
  7. // (at your option) any later version.
  8. //
  9. // The go-ethereum library is distributed in the hope that it will be useful,
  10. // but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. // GNU Lesser General Public License for more details.
  13. //
  14. // You should have received a copy of the GNU Lesser General Public License
  15. // along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
  16. // Package downloader contains the manual full chain synchronisation.
  17. package downloader
  18. import (
  19. "errors"
  20. "fmt"
  21. "math/big"
  22. "sync"
  23. "sync/atomic"
  24. "time"
  25. "github.com/ethereum/go-ethereum"
  26. "github.com/ethereum/go-ethereum/common"
  27. "github.com/ethereum/go-ethereum/core/rawdb"
  28. "github.com/ethereum/go-ethereum/core/state/snapshot"
  29. "github.com/ethereum/go-ethereum/core/types"
  30. "github.com/ethereum/go-ethereum/eth/protocols/eth"
  31. "github.com/ethereum/go-ethereum/eth/protocols/snap"
  32. "github.com/ethereum/go-ethereum/ethdb"
  33. "github.com/ethereum/go-ethereum/event"
  34. "github.com/ethereum/go-ethereum/log"
  35. "github.com/ethereum/go-ethereum/metrics"
  36. "github.com/ethereum/go-ethereum/params"
  37. "github.com/ethereum/go-ethereum/trie"
  38. )
  39. var (
  40. MaxBlockFetch = 128 // Amount of blocks to be fetched per retrieval request
  41. MaxHeaderFetch = 192 // Amount of block headers to be fetched per retrieval request
  42. MaxSkeletonSize = 128 // Number of header fetches to need for a skeleton assembly
  43. MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request
  44. MaxStateFetch = 384 // Amount of node state values to allow fetching per request
  45. rttMinEstimate = 2 * time.Second // Minimum round-trip time to target for download requests
  46. rttMaxEstimate = 20 * time.Second // Maximum round-trip time to target for download requests
  47. rttMinConfidence = 0.1 // Worse confidence factor in our estimated RTT value
  48. ttlScaling = 3 // Constant scaling factor for RTT -> TTL conversion
  49. ttlLimit = time.Minute // Maximum TTL allowance to prevent reaching crazy timeouts
  50. qosTuningPeers = 5 // Number of peers to tune based on (best peers)
  51. qosConfidenceCap = 10 // Number of peers above which not to modify RTT confidence
  52. qosTuningImpact = 0.25 // Impact that a new tuning target has on the previous value
  53. maxQueuedHeaders = 32 * 1024 // [eth/62] Maximum number of headers to queue for import (DOS protection)
  54. maxHeadersProcess = 2048 // Number of header download results to import at once into the chain
  55. maxResultsProcess = 2048 // Number of content download results to import at once into the chain
  56. fullMaxForkAncestry uint64 = params.FullImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it)
  57. lightMaxForkAncestry uint64 = params.LightImmutabilityThreshold // Maximum chain reorganisation (locally redeclared so tests can reduce it)
  58. reorgProtThreshold = 48 // Threshold number of recent blocks to disable mini reorg protection
  59. reorgProtHeaderDelay = 2 // Number of headers to delay delivering to cover mini reorgs
  60. fsHeaderCheckFrequency = 100 // Verification frequency of the downloaded headers during fast sync
  61. fsHeaderSafetyNet = 2048 // Number of headers to discard in case a chain violation is detected
  62. fsHeaderForceVerify = 24 // Number of headers to verify before and after the pivot to accept it
  63. fsHeaderContCheck = 3 * time.Second // Time interval to check for header continuations during state download
  64. fsMinFullBlocks = 64 // Number of blocks to retrieve fully even in fast sync
  65. )
  66. var (
  67. errBusy = errors.New("busy")
  68. errUnknownPeer = errors.New("peer is unknown or unhealthy")
  69. errBadPeer = errors.New("action from bad peer ignored")
  70. errStallingPeer = errors.New("peer is stalling")
  71. errUnsyncedPeer = errors.New("unsynced peer")
  72. errNoPeers = errors.New("no peers to keep download active")
  73. errTimeout = errors.New("timeout")
  74. errEmptyHeaderSet = errors.New("empty header set by peer")
  75. errPeersUnavailable = errors.New("no peers available or all tried for download")
  76. errInvalidAncestor = errors.New("retrieved ancestor is invalid")
  77. errInvalidChain = errors.New("retrieved hash chain is invalid")
  78. errInvalidBody = errors.New("retrieved block body is invalid")
  79. errInvalidReceipt = errors.New("retrieved receipt is invalid")
  80. errCancelStateFetch = errors.New("state data download canceled (requested)")
  81. errCancelContentProcessing = errors.New("content processing canceled (requested)")
  82. errCanceled = errors.New("syncing canceled (requested)")
  83. errNoSyncActive = errors.New("no sync active")
  84. errTooOld = errors.New("peer's protocol version too old")
  85. errNoAncestorFound = errors.New("no common ancestor found")
  86. )
  87. type Downloader struct {
  88. // WARNING: The `rttEstimate` and `rttConfidence` fields are accessed atomically.
  89. // On 32 bit platforms, only 64-bit aligned fields can be atomic. The struct is
  90. // guaranteed to be so aligned, so take advantage of that. For more information,
  91. // see https://golang.org/pkg/sync/atomic/#pkg-note-BUG.
  92. rttEstimate uint64 // Round trip time to target for download requests
  93. rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops)
  94. mode uint32 // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode
  95. mux *event.TypeMux // Event multiplexer to announce sync operation events
  96. checkpoint uint64 // Checkpoint block number to enforce head against (e.g. fast sync)
  97. genesis uint64 // Genesis block number to limit sync to (e.g. light client CHT)
  98. queue *queue // Scheduler for selecting the hashes to download
  99. peers *peerSet // Set of active peers from which download can proceed
  100. stateDB ethdb.Database // Database to state sync into (and deduplicate via)
  101. stateBloom *trie.SyncBloom // Bloom filter for fast trie node and contract code existence checks
  102. // Statistics
  103. syncStatsChainOrigin uint64 // Origin block number where syncing started at
  104. syncStatsChainHeight uint64 // Highest block number known when syncing started
  105. syncStatsState stateSyncStats
  106. syncStatsLock sync.RWMutex // Lock protecting the sync stats fields
  107. lightchain LightChain
  108. blockchain BlockChain
  109. // Callbacks
  110. dropPeer peerDropFn // Drops a peer for misbehaving
  111. // Status
  112. synchroniseMock func(id string, hash common.Hash) error // Replacement for synchronise during testing
  113. synchronising int32
  114. notified int32
  115. committed int32
  116. ancientLimit uint64 // The maximum block number which can be regarded as ancient data.
  117. // Channels
  118. headerCh chan dataPack // Channel receiving inbound block headers
  119. bodyCh chan dataPack // Channel receiving inbound block bodies
  120. receiptCh chan dataPack // Channel receiving inbound receipts
  121. bodyWakeCh chan bool // Channel to signal the block body fetcher of new tasks
  122. receiptWakeCh chan bool // Channel to signal the receipt fetcher of new tasks
  123. headerProcCh chan []*types.Header // Channel to feed the header processor new tasks
  124. // State sync
  125. pivotHeader *types.Header // Pivot block header to dynamically push the syncing state root
  126. pivotLock sync.RWMutex // Lock protecting pivot header reads from updates
  127. snapSync bool // Whether to run state sync over the snap protocol
  128. SnapSyncer *snap.Syncer // TODO(karalabe): make private! hack for now
  129. stateSyncStart chan *stateSync
  130. trackStateReq chan *stateReq
  131. stateCh chan dataPack // Channel receiving inbound node state data
  132. // Cancellation and termination
  133. cancelPeer string // Identifier of the peer currently being used as the master (cancel on drop)
  134. cancelCh chan struct{} // Channel to cancel mid-flight syncs
  135. cancelLock sync.RWMutex // Lock to protect the cancel channel and peer in delivers
  136. cancelWg sync.WaitGroup // Make sure all fetcher goroutines have exited.
  137. quitCh chan struct{} // Quit channel to signal termination
  138. quitLock sync.Mutex // Lock to prevent double closes
  139. // Testing hooks
  140. syncInitHook func(uint64, uint64) // Method to call upon initiating a new sync run
  141. bodyFetchHook func([]*types.Header) // Method to call upon starting a block body fetch
  142. receiptFetchHook func([]*types.Header) // Method to call upon starting a receipt fetch
  143. chainInsertHook func([]*fetchResult) // Method to call upon inserting a chain of blocks (possibly in multiple invocations)
  144. }
  145. // LightChain encapsulates functions required to synchronise a light chain.
  146. type LightChain interface {
  147. // HasHeader verifies a header's presence in the local chain.
  148. HasHeader(common.Hash, uint64) bool
  149. // GetHeaderByHash retrieves a header from the local chain.
  150. GetHeaderByHash(common.Hash) *types.Header
  151. // CurrentHeader retrieves the head header from the local chain.
  152. CurrentHeader() *types.Header
  153. // GetTd returns the total difficulty of a local block.
  154. GetTd(common.Hash, uint64) *big.Int
  155. // InsertHeaderChain inserts a batch of headers into the local chain.
  156. InsertHeaderChain([]*types.Header, int) (int, error)
  157. // SetHead rewinds the local chain to a new head.
  158. SetHead(uint64) error
  159. }
  160. // BlockChain encapsulates functions required to sync a (full or fast) blockchain.
  161. type BlockChain interface {
  162. LightChain
  163. // HasBlock verifies a block's presence in the local chain.
  164. HasBlock(common.Hash, uint64) bool
  165. // HasFastBlock verifies a fast block's presence in the local chain.
  166. HasFastBlock(common.Hash, uint64) bool
  167. // GetBlockByHash retrieves a block from the local chain.
  168. GetBlockByHash(common.Hash) *types.Block
  169. // CurrentBlock retrieves the head block from the local chain.
  170. CurrentBlock() *types.Block
  171. // CurrentFastBlock retrieves the head fast block from the local chain.
  172. CurrentFastBlock() *types.Block
  173. // FastSyncCommitHead directly commits the head block to a certain entity.
  174. FastSyncCommitHead(common.Hash) error
  175. // InsertChain inserts a batch of blocks into the local chain.
  176. InsertChain(types.Blocks) (int, error)
  177. // InsertReceiptChain inserts a batch of receipts into the local chain.
  178. InsertReceiptChain(types.Blocks, []types.Receipts, uint64) (int, error)
  179. // Snapshots returns the blockchain snapshot tree to paused it during sync.
  180. Snapshots() *snapshot.Tree
  181. }
  182. // New creates a new downloader to fetch hashes and blocks from remote peers.
  183. func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom, mux *event.TypeMux, chain BlockChain, lightchain LightChain, dropPeer peerDropFn) *Downloader {
  184. if lightchain == nil {
  185. lightchain = chain
  186. }
  187. dl := &Downloader{
  188. stateDB: stateDb,
  189. stateBloom: stateBloom,
  190. mux: mux,
  191. checkpoint: checkpoint,
  192. queue: newQueue(blockCacheMaxItems, blockCacheInitialItems),
  193. peers: newPeerSet(),
  194. rttEstimate: uint64(rttMaxEstimate),
  195. rttConfidence: uint64(1000000),
  196. blockchain: chain,
  197. lightchain: lightchain,
  198. dropPeer: dropPeer,
  199. headerCh: make(chan dataPack, 1),
  200. bodyCh: make(chan dataPack, 1),
  201. receiptCh: make(chan dataPack, 1),
  202. bodyWakeCh: make(chan bool, 1),
  203. receiptWakeCh: make(chan bool, 1),
  204. headerProcCh: make(chan []*types.Header, 1),
  205. quitCh: make(chan struct{}),
  206. stateCh: make(chan dataPack),
  207. SnapSyncer: snap.NewSyncer(stateDb),
  208. stateSyncStart: make(chan *stateSync),
  209. syncStatsState: stateSyncStats{
  210. processed: rawdb.ReadFastTrieProgress(stateDb),
  211. },
  212. trackStateReq: make(chan *stateReq),
  213. }
  214. go dl.qosTuner()
  215. go dl.stateFetcher()
  216. return dl
  217. }
  218. // Progress retrieves the synchronisation boundaries, specifically the origin
  219. // block where synchronisation started at (may have failed/suspended); the block
  220. // or header sync is currently at; and the latest known block which the sync targets.
  221. //
  222. // In addition, during the state download phase of fast synchronisation the number
  223. // of processed and the total number of known states are also returned. Otherwise
  224. // these are zero.
  225. func (d *Downloader) Progress() ethereum.SyncProgress {
  226. // Lock the current stats and return the progress
  227. d.syncStatsLock.RLock()
  228. defer d.syncStatsLock.RUnlock()
  229. current := uint64(0)
  230. mode := d.getMode()
  231. switch {
  232. case d.blockchain != nil && mode == FullSync:
  233. current = d.blockchain.CurrentBlock().NumberU64()
  234. case d.blockchain != nil && mode == FastSync:
  235. current = d.blockchain.CurrentFastBlock().NumberU64()
  236. case d.lightchain != nil:
  237. current = d.lightchain.CurrentHeader().Number.Uint64()
  238. default:
  239. log.Error("Unknown downloader chain/mode combo", "light", d.lightchain != nil, "full", d.blockchain != nil, "mode", mode)
  240. }
  241. return ethereum.SyncProgress{
  242. StartingBlock: d.syncStatsChainOrigin,
  243. CurrentBlock: current,
  244. HighestBlock: d.syncStatsChainHeight,
  245. PulledStates: d.syncStatsState.processed,
  246. KnownStates: d.syncStatsState.processed + d.syncStatsState.pending,
  247. }
  248. }
  249. // Synchronising returns whether the downloader is currently retrieving blocks.
  250. func (d *Downloader) Synchronising() bool {
  251. return atomic.LoadInt32(&d.synchronising) > 0
  252. }
  253. // RegisterPeer injects a new download peer into the set of block source to be
  254. // used for fetching hashes and blocks from.
  255. func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error {
  256. var logger log.Logger
  257. if len(id) < 16 {
  258. // Tests use short IDs, don't choke on them
  259. logger = log.New("peer", id)
  260. } else {
  261. logger = log.New("peer", id[:8])
  262. }
  263. logger.Trace("Registering sync peer")
  264. if err := d.peers.Register(newPeerConnection(id, version, peer, logger)); err != nil {
  265. logger.Error("Failed to register sync peer", "err", err)
  266. return err
  267. }
  268. d.qosReduceConfidence()
  269. return nil
  270. }
  271. // RegisterLightPeer injects a light client peer, wrapping it so it appears as a regular peer.
  272. func (d *Downloader) RegisterLightPeer(id string, version uint, peer LightPeer) error {
  273. return d.RegisterPeer(id, version, &lightPeerWrapper{peer})
  274. }
  275. // UnregisterPeer remove a peer from the known list, preventing any action from
  276. // the specified peer. An effort is also made to return any pending fetches into
  277. // the queue.
  278. func (d *Downloader) UnregisterPeer(id string) error {
  279. // Unregister the peer from the active peer set and revoke any fetch tasks
  280. var logger log.Logger
  281. if len(id) < 16 {
  282. // Tests use short IDs, don't choke on them
  283. logger = log.New("peer", id)
  284. } else {
  285. logger = log.New("peer", id[:8])
  286. }
  287. logger.Trace("Unregistering sync peer")
  288. if err := d.peers.Unregister(id); err != nil {
  289. logger.Error("Failed to unregister sync peer", "err", err)
  290. return err
  291. }
  292. d.queue.Revoke(id)
  293. return nil
  294. }
  295. // Synchronise tries to sync up our local block chain with a remote peer, both
  296. // adding various sanity checks as well as wrapping it with various log entries.
  297. func (d *Downloader) Synchronise(id string, head common.Hash, td *big.Int, mode SyncMode) error {
  298. err := d.synchronise(id, head, td, mode)
  299. switch err {
  300. case nil, errBusy, errCanceled:
  301. return err
  302. }
  303. if errors.Is(err, errInvalidChain) || errors.Is(err, errBadPeer) || errors.Is(err, errTimeout) ||
  304. errors.Is(err, errStallingPeer) || errors.Is(err, errUnsyncedPeer) || errors.Is(err, errEmptyHeaderSet) ||
  305. errors.Is(err, errPeersUnavailable) || errors.Is(err, errTooOld) || errors.Is(err, errInvalidAncestor) {
  306. log.Warn("Synchronisation failed, dropping peer", "peer", id, "err", err)
  307. if d.dropPeer == nil {
  308. // The dropPeer method is nil when `--copydb` is used for a local copy.
  309. // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored
  310. log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", id)
  311. } else {
  312. d.dropPeer(id)
  313. }
  314. return err
  315. }
  316. log.Warn("Synchronisation failed, retrying", "err", err)
  317. return err
  318. }
  319. // synchronise will select the peer and use it for synchronising. If an empty string is given
  320. // it will use the best peer possible and synchronize if its TD is higher than our own. If any of the
  321. // checks fail an error will be returned. This method is synchronous
  322. func (d *Downloader) synchronise(id string, hash common.Hash, td *big.Int, mode SyncMode) error {
  323. // Mock out the synchronisation if testing
  324. if d.synchroniseMock != nil {
  325. return d.synchroniseMock(id, hash)
  326. }
  327. // Make sure only one goroutine is ever allowed past this point at once
  328. if !atomic.CompareAndSwapInt32(&d.synchronising, 0, 1) {
  329. return errBusy
  330. }
  331. defer atomic.StoreInt32(&d.synchronising, 0)
  332. // Post a user notification of the sync (only once per session)
  333. if atomic.CompareAndSwapInt32(&d.notified, 0, 1) {
  334. log.Info("Block synchronisation started")
  335. }
  336. // If we are already full syncing, but have a fast-sync bloom filter laying
  337. // around, make sure it doesn't use memory any more. This is a special case
  338. // when the user attempts to fast sync a new empty network.
  339. if mode == FullSync && d.stateBloom != nil {
  340. d.stateBloom.Close()
  341. }
  342. // If snap sync was requested, create the snap scheduler and switch to fast
  343. // sync mode. Long term we could drop fast sync or merge the two together,
  344. // but until snap becomes prevalent, we should support both. TODO(karalabe).
  345. if mode == SnapSync {
  346. if !d.snapSync {
  347. // Snap sync uses the snapshot namespace to store potentially flakey data until
  348. // sync completely heals and finishes. Pause snapshot maintenance in the mean
  349. // time to prevent access.
  350. if snapshots := d.blockchain.Snapshots(); snapshots != nil { // Only nil in tests
  351. snapshots.Disable()
  352. }
  353. log.Warn("Enabling snapshot sync prototype")
  354. d.snapSync = true
  355. }
  356. mode = FastSync
  357. }
  358. // Reset the queue, peer set and wake channels to clean any internal leftover state
  359. d.queue.Reset(blockCacheMaxItems, blockCacheInitialItems)
  360. d.peers.Reset()
  361. for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} {
  362. select {
  363. case <-ch:
  364. default:
  365. }
  366. }
  367. for _, ch := range []chan dataPack{d.headerCh, d.bodyCh, d.receiptCh} {
  368. for empty := false; !empty; {
  369. select {
  370. case <-ch:
  371. default:
  372. empty = true
  373. }
  374. }
  375. }
  376. for empty := false; !empty; {
  377. select {
  378. case <-d.headerProcCh:
  379. default:
  380. empty = true
  381. }
  382. }
  383. // Create cancel channel for aborting mid-flight and mark the master peer
  384. d.cancelLock.Lock()
  385. d.cancelCh = make(chan struct{})
  386. d.cancelPeer = id
  387. d.cancelLock.Unlock()
  388. defer d.Cancel() // No matter what, we can't leave the cancel channel open
  389. // Atomically set the requested sync mode
  390. atomic.StoreUint32(&d.mode, uint32(mode))
  391. // Retrieve the origin peer and initiate the downloading process
  392. p := d.peers.Peer(id)
  393. if p == nil {
  394. return errUnknownPeer
  395. }
  396. return d.syncWithPeer(p, hash, td)
  397. }
  398. func (d *Downloader) getMode() SyncMode {
  399. return SyncMode(atomic.LoadUint32(&d.mode))
  400. }
  401. // syncWithPeer starts a block synchronization based on the hash chain from the
  402. // specified peer and head hash.
  403. func (d *Downloader) syncWithPeer(p *peerConnection, hash common.Hash, td *big.Int) (err error) {
  404. d.mux.Post(StartEvent{})
  405. defer func() {
  406. // reset on error
  407. if err != nil {
  408. d.mux.Post(FailedEvent{err})
  409. } else {
  410. latest := d.lightchain.CurrentHeader()
  411. d.mux.Post(DoneEvent{latest})
  412. }
  413. }()
  414. if p.version < eth.ETH65 {
  415. return fmt.Errorf("%w: advertized %d < required %d", errTooOld, p.version, eth.ETH65)
  416. }
  417. mode := d.getMode()
  418. log.Debug("Synchronising with the network", "peer", p.id, "eth", p.version, "head", hash, "td", td, "mode", mode)
  419. defer func(start time.Time) {
  420. log.Debug("Synchronisation terminated", "elapsed", common.PrettyDuration(time.Since(start)))
  421. }(time.Now())
  422. // Look up the sync boundaries: the common ancestor and the target block
  423. latest, pivot, err := d.fetchHead(p)
  424. if err != nil {
  425. return err
  426. }
  427. if mode == FastSync && pivot == nil {
  428. // If no pivot block was returned, the head is below the min full block
  429. // threshold (i.e. new chian). In that case we won't really fast sync
  430. // anyway, but still need a valid pivot block to avoid some code hitting
  431. // nil panics on an access.
  432. pivot = d.blockchain.CurrentBlock().Header()
  433. }
  434. height := latest.Number.Uint64()
  435. origin, err := d.findAncestor(p, latest)
  436. if err != nil {
  437. return err
  438. }
  439. d.syncStatsLock.Lock()
  440. if d.syncStatsChainHeight <= origin || d.syncStatsChainOrigin > origin {
  441. d.syncStatsChainOrigin = origin
  442. }
  443. d.syncStatsChainHeight = height
  444. d.syncStatsLock.Unlock()
  445. // Ensure our origin point is below any fast sync pivot point
  446. if mode == FastSync {
  447. if height <= uint64(fsMinFullBlocks) {
  448. origin = 0
  449. } else {
  450. pivotNumber := pivot.Number.Uint64()
  451. if pivotNumber <= origin {
  452. origin = pivotNumber - 1
  453. }
  454. // Write out the pivot into the database so a rollback beyond it will
  455. // reenable fast sync
  456. rawdb.WriteLastPivotNumber(d.stateDB, pivotNumber)
  457. }
  458. }
  459. d.committed = 1
  460. if mode == FastSync && pivot.Number.Uint64() != 0 {
  461. d.committed = 0
  462. }
  463. if mode == FastSync {
  464. // Set the ancient data limitation.
  465. // If we are running fast sync, all block data older than ancientLimit will be
  466. // written to the ancient store. More recent data will be written to the active
  467. // database and will wait for the freezer to migrate.
  468. //
  469. // If there is a checkpoint available, then calculate the ancientLimit through
  470. // that. Otherwise calculate the ancient limit through the advertised height
  471. // of the remote peer.
  472. //
  473. // The reason for picking checkpoint first is that a malicious peer can give us
  474. // a fake (very high) height, forcing the ancient limit to also be very high.
  475. // The peer would start to feed us valid blocks until head, resulting in all of
  476. // the blocks might be written into the ancient store. A following mini-reorg
  477. // could cause issues.
  478. if d.checkpoint != 0 && d.checkpoint > fullMaxForkAncestry+1 {
  479. d.ancientLimit = d.checkpoint
  480. } else if height > fullMaxForkAncestry+1 {
  481. d.ancientLimit = height - fullMaxForkAncestry - 1
  482. } else {
  483. d.ancientLimit = 0
  484. }
  485. frozen, _ := d.stateDB.Ancients() // Ignore the error here since light client can also hit here.
  486. // If a part of blockchain data has already been written into active store,
  487. // disable the ancient style insertion explicitly.
  488. if origin >= frozen && frozen != 0 {
  489. d.ancientLimit = 0
  490. log.Info("Disabling direct-ancient mode", "origin", origin, "ancient", frozen-1)
  491. } else if d.ancientLimit > 0 {
  492. log.Debug("Enabling direct-ancient mode", "ancient", d.ancientLimit)
  493. }
  494. // Rewind the ancient store and blockchain if reorg happens.
  495. if origin+1 < frozen {
  496. if err := d.lightchain.SetHead(origin + 1); err != nil {
  497. return err
  498. }
  499. }
  500. }
  501. // Initiate the sync using a concurrent header and content retrieval algorithm
  502. d.queue.Prepare(origin+1, mode)
  503. if d.syncInitHook != nil {
  504. d.syncInitHook(origin, height)
  505. }
  506. fetchers := []func() error{
  507. func() error { return d.fetchHeaders(p, origin+1) }, // Headers are always retrieved
  508. func() error { return d.fetchBodies(origin + 1) }, // Bodies are retrieved during normal and fast sync
  509. func() error { return d.fetchReceipts(origin + 1) }, // Receipts are retrieved during fast sync
  510. func() error { return d.processHeaders(origin+1, td) },
  511. }
  512. if mode == FastSync {
  513. d.pivotLock.Lock()
  514. d.pivotHeader = pivot
  515. d.pivotLock.Unlock()
  516. fetchers = append(fetchers, func() error { return d.processFastSyncContent() })
  517. } else if mode == FullSync {
  518. fetchers = append(fetchers, d.processFullSyncContent)
  519. }
  520. return d.spawnSync(fetchers)
  521. }
  522. // spawnSync runs d.process and all given fetcher functions to completion in
  523. // separate goroutines, returning the first error that appears.
  524. func (d *Downloader) spawnSync(fetchers []func() error) error {
  525. errc := make(chan error, len(fetchers))
  526. d.cancelWg.Add(len(fetchers))
  527. for _, fn := range fetchers {
  528. fn := fn
  529. go func() { defer d.cancelWg.Done(); errc <- fn() }()
  530. }
  531. // Wait for the first error, then terminate the others.
  532. var err error
  533. for i := 0; i < len(fetchers); i++ {
  534. if i == len(fetchers)-1 {
  535. // Close the queue when all fetchers have exited.
  536. // This will cause the block processor to end when
  537. // it has processed the queue.
  538. d.queue.Close()
  539. }
  540. if err = <-errc; err != nil && err != errCanceled {
  541. break
  542. }
  543. }
  544. d.queue.Close()
  545. d.Cancel()
  546. return err
  547. }
  548. // cancel aborts all of the operations and resets the queue. However, cancel does
  549. // not wait for the running download goroutines to finish. This method should be
  550. // used when cancelling the downloads from inside the downloader.
  551. func (d *Downloader) cancel() {
  552. // Close the current cancel channel
  553. d.cancelLock.Lock()
  554. defer d.cancelLock.Unlock()
  555. if d.cancelCh != nil {
  556. select {
  557. case <-d.cancelCh:
  558. // Channel was already closed
  559. default:
  560. close(d.cancelCh)
  561. }
  562. }
  563. }
  564. // Cancel aborts all of the operations and waits for all download goroutines to
  565. // finish before returning.
  566. func (d *Downloader) Cancel() {
  567. d.cancel()
  568. d.cancelWg.Wait()
  569. }
  570. // Terminate interrupts the downloader, canceling all pending operations.
  571. // The downloader cannot be reused after calling Terminate.
  572. func (d *Downloader) Terminate() {
  573. // Close the termination channel (make sure double close is allowed)
  574. d.quitLock.Lock()
  575. select {
  576. case <-d.quitCh:
  577. default:
  578. close(d.quitCh)
  579. }
  580. if d.stateBloom != nil {
  581. d.stateBloom.Close()
  582. }
  583. d.quitLock.Unlock()
  584. // Cancel any pending download requests
  585. d.Cancel()
  586. }
  587. // fetchHead retrieves the head header and prior pivot block (if available) from
  588. // a remote peer.
  589. func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *types.Header, err error) {
  590. p.log.Debug("Retrieving remote chain head")
  591. mode := d.getMode()
  592. // Request the advertised remote head block and wait for the response
  593. latest, _ := p.peer.Head()
  594. fetch := 1
  595. if mode == FastSync {
  596. fetch = 2 // head + pivot headers
  597. }
  598. go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true)
  599. ttl := d.requestTTL()
  600. timeout := time.After(ttl)
  601. for {
  602. select {
  603. case <-d.cancelCh:
  604. return nil, nil, errCanceled
  605. case packet := <-d.headerCh:
  606. // Discard anything not from the origin peer
  607. if packet.PeerId() != p.id {
  608. log.Debug("Received headers from incorrect peer", "peer", packet.PeerId())
  609. break
  610. }
  611. // Make sure the peer gave us at least one and at most the requested headers
  612. headers := packet.(*headerPack).headers
  613. if len(headers) == 0 || len(headers) > fetch {
  614. return nil, nil, fmt.Errorf("%w: returned headers %d != requested %d", errBadPeer, len(headers), fetch)
  615. }
  616. // The first header needs to be the head, validate against the checkpoint
  617. // and request. If only 1 header was returned, make sure there's no pivot
  618. // or there was not one requested.
  619. head := headers[0]
  620. if (mode == FastSync || mode == LightSync) && head.Number.Uint64() < d.checkpoint {
  621. return nil, nil, fmt.Errorf("%w: remote head %d below checkpoint %d", errUnsyncedPeer, head.Number, d.checkpoint)
  622. }
  623. if len(headers) == 1 {
  624. if mode == FastSync && head.Number.Uint64() > uint64(fsMinFullBlocks) {
  625. return nil, nil, fmt.Errorf("%w: no pivot included along head header", errBadPeer)
  626. }
  627. p.log.Debug("Remote head identified, no pivot", "number", head.Number, "hash", head.Hash())
  628. return head, nil, nil
  629. }
  630. // At this point we have 2 headers in total and the first is the
  631. // validated head of the chian. Check the pivot number and return,
  632. pivot := headers[1]
  633. if pivot.Number.Uint64() != head.Number.Uint64()-uint64(fsMinFullBlocks) {
  634. return nil, nil, fmt.Errorf("%w: remote pivot %d != requested %d", errInvalidChain, pivot.Number, head.Number.Uint64()-uint64(fsMinFullBlocks))
  635. }
  636. return head, pivot, nil
  637. case <-timeout:
  638. p.log.Debug("Waiting for head header timed out", "elapsed", ttl)
  639. return nil, nil, errTimeout
  640. case <-d.bodyCh:
  641. case <-d.receiptCh:
  642. // Out of bounds delivery, ignore
  643. }
  644. }
  645. }
  646. // calculateRequestSpan calculates what headers to request from a peer when trying to determine the
  647. // common ancestor.
  648. // It returns parameters to be used for peer.RequestHeadersByNumber:
  649. // from - starting block number
  650. // count - number of headers to request
  651. // skip - number of headers to skip
  652. // and also returns 'max', the last block which is expected to be returned by the remote peers,
  653. // given the (from,count,skip)
  654. func calculateRequestSpan(remoteHeight, localHeight uint64) (int64, int, int, uint64) {
  655. var (
  656. from int
  657. count int
  658. MaxCount = MaxHeaderFetch / 16
  659. )
  660. // requestHead is the highest block that we will ask for. If requestHead is not offset,
  661. // the highest block that we will get is 16 blocks back from head, which means we
  662. // will fetch 14 or 15 blocks unnecessarily in the case the height difference
  663. // between us and the peer is 1-2 blocks, which is most common
  664. requestHead := int(remoteHeight) - 1
  665. if requestHead < 0 {
  666. requestHead = 0
  667. }
  668. // requestBottom is the lowest block we want included in the query
  669. // Ideally, we want to include the one just below our own head
  670. requestBottom := int(localHeight - 1)
  671. if requestBottom < 0 {
  672. requestBottom = 0
  673. }
  674. totalSpan := requestHead - requestBottom
  675. span := 1 + totalSpan/MaxCount
  676. if span < 2 {
  677. span = 2
  678. }
  679. if span > 16 {
  680. span = 16
  681. }
  682. count = 1 + totalSpan/span
  683. if count > MaxCount {
  684. count = MaxCount
  685. }
  686. if count < 2 {
  687. count = 2
  688. }
  689. from = requestHead - (count-1)*span
  690. if from < 0 {
  691. from = 0
  692. }
  693. max := from + (count-1)*span
  694. return int64(from), count, span - 1, uint64(max)
  695. }
  696. // findAncestor tries to locate the common ancestor link of the local chain and
  697. // a remote peers blockchain. In the general case when our node was in sync and
  698. // on the correct chain, checking the top N links should already get us a match.
  699. // In the rare scenario when we ended up on a long reorganisation (i.e. none of
  700. // the head links match), we do a binary search to find the common ancestor.
  701. func (d *Downloader) findAncestor(p *peerConnection, remoteHeader *types.Header) (uint64, error) {
  702. // Figure out the valid ancestor range to prevent rewrite attacks
  703. var (
  704. floor = int64(-1)
  705. localHeight uint64
  706. remoteHeight = remoteHeader.Number.Uint64()
  707. )
  708. mode := d.getMode()
  709. switch mode {
  710. case FullSync:
  711. localHeight = d.blockchain.CurrentBlock().NumberU64()
  712. case FastSync:
  713. localHeight = d.blockchain.CurrentFastBlock().NumberU64()
  714. default:
  715. localHeight = d.lightchain.CurrentHeader().Number.Uint64()
  716. }
  717. p.log.Debug("Looking for common ancestor", "local", localHeight, "remote", remoteHeight)
  718. // Recap floor value for binary search
  719. maxForkAncestry := fullMaxForkAncestry
  720. if d.getMode() == LightSync {
  721. maxForkAncestry = lightMaxForkAncestry
  722. }
  723. if localHeight >= maxForkAncestry {
  724. // We're above the max reorg threshold, find the earliest fork point
  725. floor = int64(localHeight - maxForkAncestry)
  726. }
  727. // If we're doing a light sync, ensure the floor doesn't go below the CHT, as
  728. // all headers before that point will be missing.
  729. if mode == LightSync {
  730. // If we don't know the current CHT position, find it
  731. if d.genesis == 0 {
  732. header := d.lightchain.CurrentHeader()
  733. for header != nil {
  734. d.genesis = header.Number.Uint64()
  735. if floor >= int64(d.genesis)-1 {
  736. break
  737. }
  738. header = d.lightchain.GetHeaderByHash(header.ParentHash)
  739. }
  740. }
  741. // We already know the "genesis" block number, cap floor to that
  742. if floor < int64(d.genesis)-1 {
  743. floor = int64(d.genesis) - 1
  744. }
  745. }
  746. ancestor, err := d.findAncestorSpanSearch(p, mode, remoteHeight, localHeight, floor)
  747. if err == nil {
  748. return ancestor, nil
  749. }
  750. // The returned error was not nil.
  751. // If the error returned does not reflect that a common ancestor was not found, return it.
  752. // If the error reflects that a common ancestor was not found, continue to binary search,
  753. // where the error value will be reassigned.
  754. if !errors.Is(err, errNoAncestorFound) {
  755. return 0, err
  756. }
  757. ancestor, err = d.findAncestorBinarySearch(p, mode, remoteHeight, floor)
  758. if err != nil {
  759. return 0, err
  760. }
  761. return ancestor, nil
  762. }
  763. func (d *Downloader) findAncestorSpanSearch(p *peerConnection, mode SyncMode, remoteHeight, localHeight uint64, floor int64) (commonAncestor uint64, err error) {
  764. from, count, skip, max := calculateRequestSpan(remoteHeight, localHeight)
  765. p.log.Trace("Span searching for common ancestor", "count", count, "from", from, "skip", skip)
  766. go p.peer.RequestHeadersByNumber(uint64(from), count, skip, false)
  767. // Wait for the remote response to the head fetch
  768. number, hash := uint64(0), common.Hash{}
  769. ttl := d.requestTTL()
  770. timeout := time.After(ttl)
  771. for finished := false; !finished; {
  772. select {
  773. case <-d.cancelCh:
  774. return 0, errCanceled
  775. case packet := <-d.headerCh:
  776. // Discard anything not from the origin peer
  777. if packet.PeerId() != p.id {
  778. log.Debug("Received headers from incorrect peer", "peer", packet.PeerId())
  779. break
  780. }
  781. // Make sure the peer actually gave something valid
  782. headers := packet.(*headerPack).headers
  783. if len(headers) == 0 {
  784. p.log.Warn("Empty head header set")
  785. return 0, errEmptyHeaderSet
  786. }
  787. // Make sure the peer's reply conforms to the request
  788. for i, header := range headers {
  789. expectNumber := from + int64(i)*int64(skip+1)
  790. if number := header.Number.Int64(); number != expectNumber {
  791. p.log.Warn("Head headers broke chain ordering", "index", i, "requested", expectNumber, "received", number)
  792. return 0, fmt.Errorf("%w: %v", errInvalidChain, errors.New("head headers broke chain ordering"))
  793. }
  794. }
  795. // Check if a common ancestor was found
  796. finished = true
  797. for i := len(headers) - 1; i >= 0; i-- {
  798. // Skip any headers that underflow/overflow our requested set
  799. if headers[i].Number.Int64() < from || headers[i].Number.Uint64() > max {
  800. continue
  801. }
  802. // Otherwise check if we already know the header or not
  803. h := headers[i].Hash()
  804. n := headers[i].Number.Uint64()
  805. var known bool
  806. switch mode {
  807. case FullSync:
  808. known = d.blockchain.HasBlock(h, n)
  809. case FastSync:
  810. known = d.blockchain.HasFastBlock(h, n)
  811. default:
  812. known = d.lightchain.HasHeader(h, n)
  813. }
  814. if known {
  815. number, hash = n, h
  816. break
  817. }
  818. }
  819. case <-timeout:
  820. p.log.Debug("Waiting for head header timed out", "elapsed", ttl)
  821. return 0, errTimeout
  822. case <-d.bodyCh:
  823. case <-d.receiptCh:
  824. // Out of bounds delivery, ignore
  825. }
  826. }
  827. // If the head fetch already found an ancestor, return
  828. if hash != (common.Hash{}) {
  829. if int64(number) <= floor {
  830. p.log.Warn("Ancestor below allowance", "number", number, "hash", hash, "allowance", floor)
  831. return 0, errInvalidAncestor
  832. }
  833. p.log.Debug("Found common ancestor", "number", number, "hash", hash)
  834. return number, nil
  835. }
  836. return 0, errNoAncestorFound
  837. }
  838. func (d *Downloader) findAncestorBinarySearch(p *peerConnection, mode SyncMode, remoteHeight uint64, floor int64) (commonAncestor uint64, err error) {
  839. hash := common.Hash{}
  840. // Ancestor not found, we need to binary search over our chain
  841. start, end := uint64(0), remoteHeight
  842. if floor > 0 {
  843. start = uint64(floor)
  844. }
  845. p.log.Trace("Binary searching for common ancestor", "start", start, "end", end)
  846. for start+1 < end {
  847. // Split our chain interval in two, and request the hash to cross check
  848. check := (start + end) / 2
  849. ttl := d.requestTTL()
  850. timeout := time.After(ttl)
  851. go p.peer.RequestHeadersByNumber(check, 1, 0, false)
  852. // Wait until a reply arrives to this request
  853. for arrived := false; !arrived; {
  854. select {
  855. case <-d.cancelCh:
  856. return 0, errCanceled
  857. case packet := <-d.headerCh:
  858. // Discard anything not from the origin peer
  859. if packet.PeerId() != p.id {
  860. log.Debug("Received headers from incorrect peer", "peer", packet.PeerId())
  861. break
  862. }
  863. // Make sure the peer actually gave something valid
  864. headers := packet.(*headerPack).headers
  865. if len(headers) != 1 {
  866. p.log.Warn("Multiple headers for single request", "headers", len(headers))
  867. return 0, fmt.Errorf("%w: multiple headers (%d) for single request", errBadPeer, len(headers))
  868. }
  869. arrived = true
  870. // Modify the search interval based on the response
  871. h := headers[0].Hash()
  872. n := headers[0].Number.Uint64()
  873. var known bool
  874. switch mode {
  875. case FullSync:
  876. known = d.blockchain.HasBlock(h, n)
  877. case FastSync:
  878. known = d.blockchain.HasFastBlock(h, n)
  879. default:
  880. known = d.lightchain.HasHeader(h, n)
  881. }
  882. if !known {
  883. end = check
  884. break
  885. }
  886. header := d.lightchain.GetHeaderByHash(h) // Independent of sync mode, header surely exists
  887. if header.Number.Uint64() != check {
  888. p.log.Warn("Received non requested header", "number", header.Number, "hash", header.Hash(), "request", check)
  889. return 0, fmt.Errorf("%w: non-requested header (%d)", errBadPeer, header.Number)
  890. }
  891. start = check
  892. hash = h
  893. case <-timeout:
  894. p.log.Debug("Waiting for search header timed out", "elapsed", ttl)
  895. return 0, errTimeout
  896. case <-d.bodyCh:
  897. case <-d.receiptCh:
  898. // Out of bounds delivery, ignore
  899. }
  900. }
  901. }
  902. // Ensure valid ancestry and return
  903. if int64(start) <= floor {
  904. p.log.Warn("Ancestor below allowance", "number", start, "hash", hash, "allowance", floor)
  905. return 0, errInvalidAncestor
  906. }
  907. p.log.Debug("Found common ancestor", "number", start, "hash", hash)
  908. return start, nil
  909. }
  910. // fetchHeaders keeps retrieving headers concurrently from the number
  911. // requested, until no more are returned, potentially throttling on the way. To
  912. // facilitate concurrency but still protect against malicious nodes sending bad
  913. // headers, we construct a header chain skeleton using the "origin" peer we are
  914. // syncing with, and fill in the missing headers using anyone else. Headers from
  915. // other peers are only accepted if they map cleanly to the skeleton. If no one
  916. // can fill in the skeleton - not even the origin peer - it's assumed invalid and
  917. // the origin is dropped.
  918. func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error {
  919. p.log.Debug("Directing header downloads", "origin", from)
  920. defer p.log.Debug("Header download terminated")
  921. // Create a timeout timer, and the associated header fetcher
  922. skeleton := true // Skeleton assembly phase or finishing up
  923. pivoting := false // Whether the next request is pivot verification
  924. request := time.Now() // time of the last skeleton fetch request
  925. timeout := time.NewTimer(0) // timer to dump a non-responsive active peer
  926. <-timeout.C // timeout channel should be initially empty
  927. defer timeout.Stop()
  928. var ttl time.Duration
  929. getHeaders := func(from uint64) {
  930. request = time.Now()
  931. ttl = d.requestTTL()
  932. timeout.Reset(ttl)
  933. if skeleton {
  934. p.log.Trace("Fetching skeleton headers", "count", MaxHeaderFetch, "from", from)
  935. go p.peer.RequestHeadersByNumber(from+uint64(MaxHeaderFetch)-1, MaxSkeletonSize, MaxHeaderFetch-1, false)
  936. } else {
  937. p.log.Trace("Fetching full headers", "count", MaxHeaderFetch, "from", from)
  938. go p.peer.RequestHeadersByNumber(from, MaxHeaderFetch, 0, false)
  939. }
  940. }
  941. getNextPivot := func() {
  942. pivoting = true
  943. request = time.Now()
  944. ttl = d.requestTTL()
  945. timeout.Reset(ttl)
  946. d.pivotLock.RLock()
  947. pivot := d.pivotHeader.Number.Uint64()
  948. d.pivotLock.RUnlock()
  949. p.log.Trace("Fetching next pivot header", "number", pivot+uint64(fsMinFullBlocks))
  950. go p.peer.RequestHeadersByNumber(pivot+uint64(fsMinFullBlocks), 2, fsMinFullBlocks-9, false) // move +64 when it's 2x64-8 deep
  951. }
  952. // Start pulling the header chain skeleton until all is done
  953. ancestor := from
  954. getHeaders(from)
  955. mode := d.getMode()
  956. for {
  957. select {
  958. case <-d.cancelCh:
  959. return errCanceled
  960. case packet := <-d.headerCh:
  961. // Make sure the active peer is giving us the skeleton headers
  962. if packet.PeerId() != p.id {
  963. log.Debug("Received skeleton from incorrect peer", "peer", packet.PeerId())
  964. break
  965. }
  966. headerReqTimer.UpdateSince(request)
  967. timeout.Stop()
  968. // If the pivot is being checked, move if it became stale and run the real retrieval
  969. var pivot uint64
  970. d.pivotLock.RLock()
  971. if d.pivotHeader != nil {
  972. pivot = d.pivotHeader.Number.Uint64()
  973. }
  974. d.pivotLock.RUnlock()
  975. if pivoting {
  976. if packet.Items() == 2 {
  977. // Retrieve the headers and do some sanity checks, just in case
  978. headers := packet.(*headerPack).headers
  979. if have, want := headers[0].Number.Uint64(), pivot+uint64(fsMinFullBlocks); have != want {
  980. log.Warn("Peer sent invalid next pivot", "have", have, "want", want)
  981. return fmt.Errorf("%w: next pivot number %d != requested %d", errInvalidChain, have, want)
  982. }
  983. if have, want := headers[1].Number.Uint64(), pivot+2*uint64(fsMinFullBlocks)-8; have != want {
  984. log.Warn("Peer sent invalid pivot confirmer", "have", have, "want", want)
  985. return fmt.Errorf("%w: next pivot confirmer number %d != requested %d", errInvalidChain, have, want)
  986. }
  987. log.Warn("Pivot seemingly stale, moving", "old", pivot, "new", headers[0].Number)
  988. pivot = headers[0].Number.Uint64()
  989. d.pivotLock.Lock()
  990. d.pivotHeader = headers[0]
  991. d.pivotLock.Unlock()
  992. // Write out the pivot into the database so a rollback beyond
  993. // it will reenable fast sync and update the state root that
  994. // the state syncer will be downloading.
  995. rawdb.WriteLastPivotNumber(d.stateDB, pivot)
  996. }
  997. pivoting = false
  998. getHeaders(from)
  999. continue
  1000. }
  1001. // If the skeleton's finished, pull any remaining head headers directly from the origin
  1002. if skeleton && packet.Items() == 0 {
  1003. skeleton = false
  1004. getHeaders(from)
  1005. continue
  1006. }
  1007. // If no more headers are inbound, notify the content fetchers and return
  1008. if packet.Items() == 0 {
  1009. // Don't abort header fetches while the pivot is downloading
  1010. if atomic.LoadInt32(&d.committed) == 0 && pivot <= from {
  1011. p.log.Debug("No headers, waiting for pivot commit")
  1012. select {
  1013. case <-time.After(fsHeaderContCheck):
  1014. getHeaders(from)
  1015. continue
  1016. case <-d.cancelCh:
  1017. return errCanceled
  1018. }
  1019. }
  1020. // Pivot done (or not in fast sync) and no more headers, terminate the process
  1021. p.log.Debug("No more headers available")
  1022. select {
  1023. case d.headerProcCh <- nil:
  1024. return nil
  1025. case <-d.cancelCh:
  1026. return errCanceled
  1027. }
  1028. }
  1029. headers := packet.(*headerPack).headers
  1030. // If we received a skeleton batch, resolve internals concurrently
  1031. if skeleton {
  1032. filled, proced, err := d.fillHeaderSkeleton(from, headers)
  1033. if err != nil {
  1034. p.log.Debug("Skeleton chain invalid", "err", err)
  1035. return fmt.Errorf("%w: %v", errInvalidChain, err)
  1036. }
  1037. headers = filled[proced:]
  1038. from += uint64(proced)
  1039. } else {
  1040. // If we're closing in on the chain head, but haven't yet reached it, delay
  1041. // the last few headers so mini reorgs on the head don't cause invalid hash
  1042. // chain errors.
  1043. if n := len(headers); n > 0 {
  1044. // Retrieve the current head we're at
  1045. var head uint64
  1046. if mode == LightSync {
  1047. head = d.lightchain.CurrentHeader().Number.Uint64()
  1048. } else {
  1049. head = d.blockchain.CurrentFastBlock().NumberU64()
  1050. if full := d.blockchain.CurrentBlock().NumberU64(); head < full {
  1051. head = full
  1052. }
  1053. }
  1054. // If the head is below the common ancestor, we're actually deduplicating
  1055. // already existing chain segments, so use the ancestor as the fake head.
  1056. // Otherwise we might end up delaying header deliveries pointlessly.
  1057. if head < ancestor {
  1058. head = ancestor
  1059. }
  1060. // If the head is way older than this batch, delay the last few headers
  1061. if head+uint64(reorgProtThreshold) < headers[n-1].Number.Uint64() {
  1062. delay := reorgProtHeaderDelay
  1063. if delay > n {
  1064. delay = n
  1065. }
  1066. headers = headers[:n-delay]
  1067. }
  1068. }
  1069. }
  1070. // Insert all the new headers and fetch the next batch
  1071. if len(headers) > 0 {
  1072. p.log.Trace("Scheduling new headers", "count", len(headers), "from", from)
  1073. select {
  1074. case d.headerProcCh <- headers:
  1075. case <-d.cancelCh:
  1076. return errCanceled
  1077. }
  1078. from += uint64(len(headers))
  1079. // If we're still skeleton filling fast sync, check pivot staleness
  1080. // before continuing to the next skeleton filling
  1081. if skeleton && pivot > 0 {
  1082. getNextPivot()
  1083. } else {
  1084. getHeaders(from)
  1085. }
  1086. } else {
  1087. // No headers delivered, or all of them being delayed, sleep a bit and retry
  1088. p.log.Trace("All headers delayed, waiting")
  1089. select {
  1090. case <-time.After(fsHeaderContCheck):
  1091. getHeaders(from)
  1092. continue
  1093. case <-d.cancelCh:
  1094. return errCanceled
  1095. }
  1096. }
  1097. case <-timeout.C:
  1098. if d.dropPeer == nil {
  1099. // The dropPeer method is nil when `--copydb` is used for a local copy.
  1100. // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored
  1101. p.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", p.id)
  1102. break
  1103. }
  1104. // Header retrieval timed out, consider the peer bad and drop
  1105. p.log.Debug("Header request timed out", "elapsed", ttl)
  1106. headerTimeoutMeter.Mark(1)
  1107. d.dropPeer(p.id)
  1108. // Finish the sync gracefully instead of dumping the gathered data though
  1109. for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} {
  1110. select {
  1111. case ch <- false:
  1112. case <-d.cancelCh:
  1113. }
  1114. }
  1115. select {
  1116. case d.headerProcCh <- nil:
  1117. case <-d.cancelCh:
  1118. }
  1119. return fmt.Errorf("%w: header request timed out", errBadPeer)
  1120. }
  1121. }
  1122. }
  1123. // fillHeaderSkeleton concurrently retrieves headers from all our available peers
  1124. // and maps them to the provided skeleton header chain.
  1125. //
  1126. // Any partial results from the beginning of the skeleton is (if possible) forwarded
  1127. // immediately to the header processor to keep the rest of the pipeline full even
  1128. // in the case of header stalls.
  1129. //
  1130. // The method returns the entire filled skeleton and also the number of headers
  1131. // already forwarded for processing.
  1132. func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) ([]*types.Header, int, error) {
  1133. log.Debug("Filling up skeleton", "from", from)
  1134. d.queue.ScheduleSkeleton(from, skeleton)
  1135. var (
  1136. deliver = func(packet dataPack) (int, error) {
  1137. pack := packet.(*headerPack)
  1138. return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh)
  1139. }
  1140. expire = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) }
  1141. reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) {
  1142. return d.queue.ReserveHeaders(p, count), false, false
  1143. }
  1144. fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) }
  1145. capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) }
  1146. setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) {
  1147. p.SetHeadersIdle(accepted, deliveryTime)
  1148. }
  1149. )
  1150. err := d.fetchParts(d.headerCh, deliver, d.queue.headerContCh, expire,
  1151. d.queue.PendingHeaders, d.queue.InFlightHeaders, reserve,
  1152. nil, fetch, d.queue.CancelHeaders, capacity, d.peers.HeaderIdlePeers, setIdle, "headers")
  1153. log.Debug("Skeleton fill terminated", "err", err)
  1154. filled, proced := d.queue.RetrieveHeaders()
  1155. return filled, proced, err
  1156. }
  1157. // fetchBodies iteratively downloads the scheduled block bodies, taking any
  1158. // available peers, reserving a chunk of blocks for each, waiting for delivery
  1159. // and also periodically checking for timeouts.
  1160. func (d *Downloader) fetchBodies(from uint64) error {
  1161. log.Debug("Downloading block bodies", "origin", from)
  1162. var (
  1163. deliver = func(packet dataPack) (int, error) {
  1164. pack := packet.(*bodyPack)
  1165. return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles)
  1166. }
  1167. expire = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) }
  1168. fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) }
  1169. capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) }
  1170. setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) }
  1171. )
  1172. err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire,
  1173. d.queue.PendingBlocks, d.queue.InFlightBlocks, d.queue.ReserveBodies,
  1174. d.bodyFetchHook, fetch, d.queue.CancelBodies, capacity, d.peers.BodyIdlePeers, setIdle, "bodies")
  1175. log.Debug("Block body download terminated", "err", err)
  1176. return err
  1177. }
  1178. // fetchReceipts iteratively downloads the scheduled block receipts, taking any
  1179. // available peers, reserving a chunk of receipts for each, waiting for delivery
  1180. // and also periodically checking for timeouts.
  1181. func (d *Downloader) fetchReceipts(from uint64) error {
  1182. log.Debug("Downloading transaction receipts", "origin", from)
  1183. var (
  1184. deliver = func(packet dataPack) (int, error) {
  1185. pack := packet.(*receiptPack)
  1186. return d.queue.DeliverReceipts(pack.peerID, pack.receipts)
  1187. }
  1188. expire = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) }
  1189. fetch = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) }
  1190. capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) }
  1191. setIdle = func(p *peerConnection, accepted int, deliveryTime time.Time) {
  1192. p.SetReceiptsIdle(accepted, deliveryTime)
  1193. }
  1194. )
  1195. err := d.fetchParts(d.receiptCh, deliver, d.receiptWakeCh, expire,
  1196. d.queue.PendingReceipts, d.queue.InFlightReceipts, d.queue.ReserveReceipts,
  1197. d.receiptFetchHook, fetch, d.queue.CancelReceipts, capacity, d.peers.ReceiptIdlePeers, setIdle, "receipts")
  1198. log.Debug("Transaction receipt download terminated", "err", err)
  1199. return err
  1200. }
  1201. // fetchParts iteratively downloads scheduled block parts, taking any available
  1202. // peers, reserving a chunk of fetch requests for each, waiting for delivery and
  1203. // also periodically checking for timeouts.
  1204. //
  1205. // As the scheduling/timeout logic mostly is the same for all downloaded data
  1206. // types, this method is used by each for data gathering and is instrumented with
  1207. // various callbacks to handle the slight differences between processing them.
  1208. //
  1209. // The instrumentation parameters:
  1210. // - errCancel: error type to return if the fetch operation is cancelled (mostly makes logging nicer)
  1211. // - deliveryCh: channel from which to retrieve downloaded data packets (merged from all concurrent peers)
  1212. // - deliver: processing callback to deliver data packets into type specific download queues (usually within `queue`)
  1213. // - wakeCh: notification channel for waking the fetcher when new tasks are available (or sync completed)
  1214. // - expire: task callback method to abort requests that took too long and return the faulty peers (traffic shaping)
  1215. // - pending: task callback for the number of requests still needing download (detect completion/non-completability)
  1216. // - inFlight: task callback for the number of in-progress requests (wait for all active downloads to finish)
  1217. // - throttle: task callback to check if the processing queue is full and activate throttling (bound memory use)
  1218. // - reserve: task callback to reserve new download tasks to a particular peer (also signals partial completions)
  1219. // - fetchHook: tester callback to notify of new tasks being initiated (allows testing the scheduling logic)
  1220. // - fetch: network callback to actually send a particular download request to a physical remote peer
  1221. // - cancel: task callback to abort an in-flight download request and allow rescheduling it (in case of lost peer)
  1222. // - capacity: network callback to retrieve the estimated type-specific bandwidth capacity of a peer (traffic shaping)
  1223. // - idle: network callback to retrieve the currently (type specific) idle peers that can be assigned tasks
  1224. // - setIdle: network callback to set a peer back to idle and update its estimated capacity (traffic shaping)
  1225. // - kind: textual label of the type being downloaded to display in log messages
  1226. func (d *Downloader) fetchParts(deliveryCh chan dataPack, deliver func(dataPack) (int, error), wakeCh chan bool,
  1227. expire func() map[string]int, pending func() int, inFlight func() bool, reserve func(*peerConnection, int) (*fetchRequest, bool, bool),
  1228. fetchHook func([]*types.Header), fetch func(*peerConnection, *fetchRequest) error, cancel func(*fetchRequest), capacity func(*peerConnection) int,
  1229. idle func() ([]*peerConnection, int), setIdle func(*peerConnection, int, time.Time), kind string) error {
  1230. // Create a ticker to detect expired retrieval tasks
  1231. ticker := time.NewTicker(100 * time.Millisecond)
  1232. defer ticker.Stop()
  1233. update := make(chan struct{}, 1)
  1234. // Prepare the queue and fetch block parts until the block header fetcher's done
  1235. finished := false
  1236. for {
  1237. select {
  1238. case <-d.cancelCh:
  1239. return errCanceled
  1240. case packet := <-deliveryCh:
  1241. deliveryTime := time.Now()
  1242. // If the peer was previously banned and failed to deliver its pack
  1243. // in a reasonable time frame, ignore its message.
  1244. if peer := d.peers.Peer(packet.PeerId()); peer != nil {
  1245. // Deliver the received chunk of data and check chain validity
  1246. accepted, err := deliver(packet)
  1247. if errors.Is(err, errInvalidChain) {
  1248. return err
  1249. }
  1250. // Unless a peer delivered something completely else than requested (usually
  1251. // caused by a timed out request which came through in the end), set it to
  1252. // idle. If the delivery's stale, the peer should have already been idled.
  1253. if !errors.Is(err, errStaleDelivery) {
  1254. setIdle(peer, accepted, deliveryTime)
  1255. }
  1256. // Issue a log to the user to see what's going on
  1257. switch {
  1258. case err == nil && packet.Items() == 0:
  1259. peer.log.Trace("Requested data not delivered", "type", kind)
  1260. case err == nil:
  1261. peer.log.Trace("Delivered new batch of data", "type", kind, "count", packet.Stats())
  1262. default:
  1263. peer.log.Debug("Failed to deliver retrieved data", "type", kind, "err", err)
  1264. }
  1265. }
  1266. // Blocks assembled, try to update the progress
  1267. select {
  1268. case update <- struct{}{}:
  1269. default:
  1270. }
  1271. case cont := <-wakeCh:
  1272. // The header fetcher sent a continuation flag, check if it's done
  1273. if !cont {
  1274. finished = true
  1275. }
  1276. // Headers arrive, try to update the progress
  1277. select {
  1278. case update <- struct{}{}:
  1279. default:
  1280. }
  1281. case <-ticker.C:
  1282. // Sanity check update the progress
  1283. select {
  1284. case update <- struct{}{}:
  1285. default:
  1286. }
  1287. case <-update:
  1288. // Short circuit if we lost all our peers
  1289. if d.peers.Len() == 0 {
  1290. return errNoPeers
  1291. }
  1292. // Check for fetch request timeouts and demote the responsible peers
  1293. for pid, fails := range expire() {
  1294. if peer := d.peers.Peer(pid); peer != nil {
  1295. // If a lot of retrieval elements expired, we might have overestimated the remote peer or perhaps
  1296. // ourselves. Only reset to minimal throughput but don't drop just yet. If even the minimal times
  1297. // out that sync wise we need to get rid of the peer.
  1298. //
  1299. // The reason the minimum threshold is 2 is because the downloader tries to estimate the bandwidth
  1300. // and latency of a peer separately, which requires pushing the measures capacity a bit and seeing
  1301. // how response times reacts, to it always requests one more than the minimum (i.e. min 2).
  1302. if fails > 2 {
  1303. peer.log.Trace("Data delivery timed out", "type", kind)
  1304. setIdle(peer, 0, time.Now())
  1305. } else {
  1306. peer.log.Debug("Stalling delivery, dropping", "type", kind)
  1307. if d.dropPeer == nil {
  1308. // The dropPeer method is nil when `--copydb` is used for a local copy.
  1309. // Timeouts can occur if e.g. compaction hits at the wrong time, and can be ignored
  1310. peer.log.Warn("Downloader wants to drop peer, but peerdrop-function is not set", "peer", pid)
  1311. } else {
  1312. d.dropPeer(pid)
  1313. // If this peer was the master peer, abort sync immediately
  1314. d.cancelLock.RLock()
  1315. master := pid == d.cancelPeer
  1316. d.cancelLock.RUnlock()
  1317. if master {
  1318. d.cancel()
  1319. return errTimeout
  1320. }
  1321. }
  1322. }
  1323. }
  1324. }
  1325. // If there's nothing more to fetch, wait or terminate
  1326. if pending() == 0 {
  1327. if !inFlight() && finished {
  1328. log.Debug("Data fetching completed", "type", kind)
  1329. return nil
  1330. }
  1331. break
  1332. }
  1333. // Send a download request to all idle peers, until throttled
  1334. progressed, throttled, running := false, false, inFlight()
  1335. idles, total := idle()
  1336. pendCount := pending()
  1337. for _, peer := range idles {
  1338. // Short circuit if throttling activated
  1339. if throttled {
  1340. break
  1341. }
  1342. // Short circuit if there is no more available task.
  1343. if pendCount = pending(); pendCount == 0 {
  1344. break
  1345. }
  1346. // Reserve a chunk of fetches for a peer. A nil can mean either that
  1347. // no more headers are available, or that the peer is known not to
  1348. // have them.
  1349. request, progress, throttle := reserve(peer, capacity(peer))
  1350. if progress {
  1351. progressed = true
  1352. }
  1353. if throttle {
  1354. throttled = true
  1355. throttleCounter.Inc(1)
  1356. }
  1357. if request == nil {
  1358. continue
  1359. }
  1360. if request.From > 0 {
  1361. peer.log.Trace("Requesting new batch of data", "type", kind, "from", request.From)
  1362. } else {
  1363. peer.log.Trace("Requesting new batch of data", "type", kind, "count", len(request.Headers), "from", request.Headers[0].Number)
  1364. }
  1365. // Fetch the chunk and make sure any errors return the hashes to the queue
  1366. if fetchHook != nil {
  1367. fetchHook(request.Headers)
  1368. }
  1369. if err := fetch(peer, request); err != nil {
  1370. // Although we could try and make an attempt to fix this, this error really
  1371. // means that we've double allocated a fetch task to a peer. If that is the
  1372. // case, the internal state of the downloader and the queue is very wrong so
  1373. // better hard crash and note the error instead of silently accumulating into
  1374. // a much bigger issue.
  1375. panic(fmt.Sprintf("%v: %s fetch assignment failed", peer, kind))
  1376. }
  1377. running = true
  1378. }
  1379. // Make sure that we have peers available for fetching. If all peers have been tried
  1380. // and all failed throw an error
  1381. if !progressed && !throttled && !running && len(idles) == total && pendCount > 0 {
  1382. return errPeersUnavailable
  1383. }
  1384. }
  1385. }
  1386. }
  1387. // processHeaders takes batches of retrieved headers from an input channel and
  1388. // keeps processing and scheduling them into the header chain and downloader's
  1389. // queue until the stream ends or a failure occurs.
  1390. func (d *Downloader) processHeaders(origin uint64, td *big.Int) error {
  1391. // Keep a count of uncertain headers to roll back
  1392. var (
  1393. rollback uint64 // Zero means no rollback (fine as you can't unroll the genesis)
  1394. rollbackErr error
  1395. mode = d.getMode()
  1396. )
  1397. defer func() {
  1398. if rollback > 0 {
  1399. lastHeader, lastFastBlock, lastBlock := d.lightchain.CurrentHeader().Number, common.Big0, common.Big0
  1400. if mode != LightSync {
  1401. lastFastBlock = d.blockchain.CurrentFastBlock().Number()
  1402. lastBlock = d.blockchain.CurrentBlock().Number()
  1403. }
  1404. if err := d.lightchain.SetHead(rollback - 1); err != nil { // -1 to target the parent of the first uncertain block
  1405. // We're already unwinding the stack, only print the error to make it more visible
  1406. log.Error("Failed to roll back chain segment", "head", rollback-1, "err", err)
  1407. }
  1408. curFastBlock, curBlock := common.Big0, common.Big0
  1409. if mode != LightSync {
  1410. curFastBlock = d.blockchain.CurrentFastBlock().Number()
  1411. curBlock = d.blockchain.CurrentBlock().Number()
  1412. }
  1413. log.Warn("Rolled back chain segment",
  1414. "header", fmt.Sprintf("%d->%d", lastHeader, d.lightchain.CurrentHeader().Number),
  1415. "fast", fmt.Sprintf("%d->%d", lastFastBlock, curFastBlock),
  1416. "block", fmt.Sprintf("%d->%d", lastBlock, curBlock), "reason", rollbackErr)
  1417. }
  1418. }()
  1419. // Wait for batches of headers to process
  1420. gotHeaders := false
  1421. for {
  1422. select {
  1423. case <-d.cancelCh:
  1424. rollbackErr = errCanceled
  1425. return errCanceled
  1426. case headers := <-d.headerProcCh:
  1427. // Terminate header processing if we synced up
  1428. if len(headers) == 0 {
  1429. // Notify everyone that headers are fully processed
  1430. for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} {
  1431. select {
  1432. case ch <- false:
  1433. case <-d.cancelCh:
  1434. }
  1435. }
  1436. // If no headers were retrieved at all, the peer violated its TD promise that it had a
  1437. // better chain compared to ours. The only exception is if its promised blocks were
  1438. // already imported by other means (e.g. fetcher):
  1439. //
  1440. // R <remote peer>, L <local node>: Both at block 10
  1441. // R: Mine block 11, and propagate it to L
  1442. // L: Queue block 11 for import
  1443. // L: Notice that R's head and TD increased compared to ours, start sync
  1444. // L: Import of block 11 finishes
  1445. // L: Sync begins, and finds common ancestor at 11
  1446. // L: Request new headers up from 11 (R's TD was higher, it must have something)
  1447. // R: Nothing to give
  1448. if mode != LightSync {
  1449. head := d.blockchain.CurrentBlock()
  1450. if !gotHeaders && td.Cmp(d.blockchain.GetTd(head.Hash(), head.NumberU64())) > 0 {
  1451. return errStallingPeer
  1452. }
  1453. }
  1454. // If fast or light syncing, ensure promised headers are indeed delivered. This is
  1455. // needed to detect scenarios where an attacker feeds a bad pivot and then bails out
  1456. // of delivering the post-pivot blocks that would flag the invalid content.
  1457. //
  1458. // This check cannot be executed "as is" for full imports, since blocks may still be
  1459. // queued for processing when the header download completes. However, as long as the
  1460. // peer gave us something useful, we're already happy/progressed (above check).
  1461. if mode == FastSync || mode == LightSync {
  1462. head := d.lightchain.CurrentHeader()
  1463. if td.Cmp(d.lightchain.GetTd(head.Hash(), head.Number.Uint64())) > 0 {
  1464. return errStallingPeer
  1465. }
  1466. }
  1467. // Disable any rollback and return
  1468. rollback = 0
  1469. return nil
  1470. }
  1471. // Otherwise split the chunk of headers into batches and process them
  1472. gotHeaders = true
  1473. for len(headers) > 0 {
  1474. // Terminate if something failed in between processing chunks
  1475. select {
  1476. case <-d.cancelCh:
  1477. rollbackErr = errCanceled
  1478. return errCanceled
  1479. default:
  1480. }
  1481. // Select the next chunk of headers to import
  1482. limit := maxHeadersProcess
  1483. if limit > len(headers) {
  1484. limit = len(headers)
  1485. }
  1486. chunk := headers[:limit]
  1487. // In case of header only syncing, validate the chunk immediately
  1488. if mode == FastSync || mode == LightSync {
  1489. // If we're importing pure headers, verify based on their recentness
  1490. var pivot uint64
  1491. d.pivotLock.RLock()
  1492. if d.pivotHeader != nil {
  1493. pivot = d.pivotHeader.Number.Uint64()
  1494. }
  1495. d.pivotLock.RUnlock()
  1496. frequency := fsHeaderCheckFrequency
  1497. if chunk[len(chunk)-1].Number.Uint64()+uint64(fsHeaderForceVerify) > pivot {
  1498. frequency = 1
  1499. }
  1500. if n, err := d.lightchain.InsertHeaderChain(chunk, frequency); err != nil {
  1501. rollbackErr = err
  1502. // If some headers were inserted, track them as uncertain
  1503. if (mode == FastSync || frequency > 1) && n > 0 && rollback == 0 {
  1504. rollback = chunk[0].Number.Uint64()
  1505. }
  1506. log.Warn("Invalid header encountered", "number", chunk[n].Number, "hash", chunk[n].Hash(), "parent", chunk[n].ParentHash, "err", err)
  1507. return fmt.Errorf("%w: %v", errInvalidChain, err)
  1508. }
  1509. // All verifications passed, track all headers within the alloted limits
  1510. if mode == FastSync {
  1511. head := chunk[len(chunk)-1].Number.Uint64()
  1512. if head-rollback > uint64(fsHeaderSafetyNet) {
  1513. rollback = head - uint64(fsHeaderSafetyNet)
  1514. } else {
  1515. rollback = 1
  1516. }
  1517. }
  1518. }
  1519. // Unless we're doing light chains, schedule the headers for associated content retrieval
  1520. if mode == FullSync || mode == FastSync {
  1521. // If we've reached the allowed number of pending headers, stall a bit
  1522. for d.queue.PendingBlocks() >= maxQueuedHeaders || d.queue.PendingReceipts() >= maxQueuedHeaders {
  1523. select {
  1524. case <-d.cancelCh:
  1525. rollbackErr = errCanceled
  1526. return errCanceled
  1527. case <-time.After(time.Second):
  1528. }
  1529. }
  1530. // Otherwise insert the headers for content retrieval
  1531. inserts := d.queue.Schedule(chunk, origin)
  1532. if len(inserts) != len(chunk) {
  1533. rollbackErr = fmt.Errorf("stale headers: len inserts %v len(chunk) %v", len(inserts), len(chunk))
  1534. return fmt.Errorf("%w: stale headers", errBadPeer)
  1535. }
  1536. }
  1537. headers = headers[limit:]
  1538. origin += uint64(limit)
  1539. }
  1540. // Update the highest block number we know if a higher one is found.
  1541. d.syncStatsLock.Lock()
  1542. if d.syncStatsChainHeight < origin {
  1543. d.syncStatsChainHeight = origin - 1
  1544. }
  1545. d.syncStatsLock.Unlock()
  1546. // Signal the content downloaders of the availablility of new tasks
  1547. for _, ch := range []chan bool{d.bodyWakeCh, d.receiptWakeCh} {
  1548. select {
  1549. case ch <- true:
  1550. default:
  1551. }
  1552. }
  1553. }
  1554. }
  1555. }
  1556. // processFullSyncContent takes fetch results from the queue and imports them into the chain.
  1557. func (d *Downloader) processFullSyncContent() error {
  1558. for {
  1559. results := d.queue.Results(true)
  1560. if len(results) == 0 {
  1561. return nil
  1562. }
  1563. if d.chainInsertHook != nil {
  1564. d.chainInsertHook(results)
  1565. }
  1566. if err := d.importBlockResults(results); err != nil {
  1567. return err
  1568. }
  1569. }
  1570. }
  1571. func (d *Downloader) importBlockResults(results []*fetchResult) error {
  1572. // Check for any early termination requests
  1573. if len(results) == 0 {
  1574. return nil
  1575. }
  1576. select {
  1577. case <-d.quitCh:
  1578. return errCancelContentProcessing
  1579. default:
  1580. }
  1581. // Retrieve the a batch of results to import
  1582. first, last := results[0].Header, results[len(results)-1].Header
  1583. log.Debug("Inserting downloaded chain", "items", len(results),
  1584. "firstnum", first.Number, "firsthash", first.Hash(),
  1585. "lastnum", last.Number, "lasthash", last.Hash(),
  1586. )
  1587. blocks := make([]*types.Block, len(results))
  1588. for i, result := range results {
  1589. blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles)
  1590. }
  1591. if index, err := d.blockchain.InsertChain(blocks); err != nil {
  1592. if index < len(results) {
  1593. log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err)
  1594. } else {
  1595. // The InsertChain method in blockchain.go will sometimes return an out-of-bounds index,
  1596. // when it needs to preprocess blocks to import a sidechain.
  1597. // The importer will put together a new list of blocks to import, which is a superset
  1598. // of the blocks delivered from the downloader, and the indexing will be off.
  1599. log.Debug("Downloaded item processing failed on sidechain import", "index", index, "err", err)
  1600. }
  1601. return fmt.Errorf("%w: %v", errInvalidChain, err)
  1602. }
  1603. return nil
  1604. }
  1605. // processFastSyncContent takes fetch results from the queue and writes them to the
  1606. // database. It also controls the synchronisation of state nodes of the pivot block.
  1607. func (d *Downloader) processFastSyncContent() error {
  1608. // Start syncing state of the reported head block. This should get us most of
  1609. // the state of the pivot block.
  1610. d.pivotLock.RLock()
  1611. sync := d.syncState(d.pivotHeader.Root)
  1612. d.pivotLock.RUnlock()
  1613. defer func() {
  1614. // The `sync` object is replaced every time the pivot moves. We need to
  1615. // defer close the very last active one, hence the lazy evaluation vs.
  1616. // calling defer sync.Cancel() !!!
  1617. sync.Cancel()
  1618. }()
  1619. closeOnErr := func(s *stateSync) {
  1620. if err := s.Wait(); err != nil && err != errCancelStateFetch && err != errCanceled && err != snap.ErrCancelled {
  1621. d.queue.Close() // wake up Results
  1622. }
  1623. }
  1624. go closeOnErr(sync)
  1625. // To cater for moving pivot points, track the pivot block and subsequently
  1626. // accumulated download results separately.
  1627. var (
  1628. oldPivot *fetchResult // Locked in pivot block, might change eventually
  1629. oldTail []*fetchResult // Downloaded content after the pivot
  1630. )
  1631. for {
  1632. // Wait for the next batch of downloaded data to be available, and if the pivot
  1633. // block became stale, move the goalpost
  1634. results := d.queue.Results(oldPivot == nil) // Block if we're not monitoring pivot staleness
  1635. if len(results) == 0 {
  1636. // If pivot sync is done, stop
  1637. if oldPivot == nil {
  1638. return sync.Cancel()
  1639. }
  1640. // If sync failed, stop
  1641. select {
  1642. case <-d.cancelCh:
  1643. sync.Cancel()
  1644. return errCanceled
  1645. default:
  1646. }
  1647. }
  1648. if d.chainInsertHook != nil {
  1649. d.chainInsertHook(results)
  1650. }
  1651. // If we haven't downloaded the pivot block yet, check pivot staleness
  1652. // notifications from the header downloader
  1653. d.pivotLock.RLock()
  1654. pivot := d.pivotHeader
  1655. d.pivotLock.RUnlock()
  1656. if oldPivot == nil {
  1657. if pivot.Root != sync.root {
  1658. sync.Cancel()
  1659. sync = d.syncState(pivot.Root)
  1660. go closeOnErr(sync)
  1661. }
  1662. } else {
  1663. results = append(append([]*fetchResult{oldPivot}, oldTail...), results...)
  1664. }
  1665. // Split around the pivot block and process the two sides via fast/full sync
  1666. if atomic.LoadInt32(&d.committed) == 0 {
  1667. latest := results[len(results)-1].Header
  1668. // If the height is above the pivot block by 2 sets, it means the pivot
  1669. // become stale in the network and it was garbage collected, move to a
  1670. // new pivot.
  1671. //
  1672. // Note, we have `reorgProtHeaderDelay` number of blocks withheld, Those
  1673. // need to be taken into account, otherwise we're detecting the pivot move
  1674. // late and will drop peers due to unavailable state!!!
  1675. if height := latest.Number.Uint64(); height >= pivot.Number.Uint64()+2*uint64(fsMinFullBlocks)-uint64(reorgProtHeaderDelay) {
  1676. log.Warn("Pivot became stale, moving", "old", pivot.Number.Uint64(), "new", height-uint64(fsMinFullBlocks)+uint64(reorgProtHeaderDelay))
  1677. pivot = results[len(results)-1-fsMinFullBlocks+reorgProtHeaderDelay].Header // must exist as lower old pivot is uncommitted
  1678. d.pivotLock.Lock()
  1679. d.pivotHeader = pivot
  1680. d.pivotLock.Unlock()
  1681. // Write out the pivot into the database so a rollback beyond it will
  1682. // reenable fast sync
  1683. rawdb.WriteLastPivotNumber(d.stateDB, pivot.Number.Uint64())
  1684. }
  1685. }
  1686. P, beforeP, afterP := splitAroundPivot(pivot.Number.Uint64(), results)
  1687. if err := d.commitFastSyncData(beforeP, sync); err != nil {
  1688. return err
  1689. }
  1690. if P != nil {
  1691. // If new pivot block found, cancel old state retrieval and restart
  1692. if oldPivot != P {
  1693. sync.Cancel()
  1694. sync = d.syncState(P.Header.Root)
  1695. go closeOnErr(sync)
  1696. oldPivot = P
  1697. }
  1698. // Wait for completion, occasionally checking for pivot staleness
  1699. select {
  1700. case <-sync.done:
  1701. if sync.err != nil {
  1702. return sync.err
  1703. }
  1704. if err := d.commitPivotBlock(P); err != nil {
  1705. return err
  1706. }
  1707. oldPivot = nil
  1708. case <-time.After(time.Second):
  1709. oldTail = afterP
  1710. continue
  1711. }
  1712. }
  1713. // Fast sync done, pivot commit done, full import
  1714. if err := d.importBlockResults(afterP); err != nil {
  1715. return err
  1716. }
  1717. }
  1718. }
  1719. func splitAroundPivot(pivot uint64, results []*fetchResult) (p *fetchResult, before, after []*fetchResult) {
  1720. if len(results) == 0 {
  1721. return nil, nil, nil
  1722. }
  1723. if lastNum := results[len(results)-1].Header.Number.Uint64(); lastNum < pivot {
  1724. // the pivot is somewhere in the future
  1725. return nil, results, nil
  1726. }
  1727. // This can also be optimized, but only happens very seldom
  1728. for _, result := range results {
  1729. num := result.Header.Number.Uint64()
  1730. switch {
  1731. case num < pivot:
  1732. before = append(before, result)
  1733. case num == pivot:
  1734. p = result
  1735. default:
  1736. after = append(after, result)
  1737. }
  1738. }
  1739. return p, before, after
  1740. }
  1741. func (d *Downloader) commitFastSyncData(results []*fetchResult, stateSync *stateSync) error {
  1742. // Check for any early termination requests
  1743. if len(results) == 0 {
  1744. return nil
  1745. }
  1746. select {
  1747. case <-d.quitCh:
  1748. return errCancelContentProcessing
  1749. case <-stateSync.done:
  1750. if err := stateSync.Wait(); err != nil {
  1751. return err
  1752. }
  1753. default:
  1754. }
  1755. // Retrieve the a batch of results to import
  1756. first, last := results[0].Header, results[len(results)-1].Header
  1757. log.Debug("Inserting fast-sync blocks", "items", len(results),
  1758. "firstnum", first.Number, "firsthash", first.Hash(),
  1759. "lastnumn", last.Number, "lasthash", last.Hash(),
  1760. )
  1761. blocks := make([]*types.Block, len(results))
  1762. receipts := make([]types.Receipts, len(results))
  1763. for i, result := range results {
  1764. blocks[i] = types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles)
  1765. receipts[i] = result.Receipts
  1766. }
  1767. if index, err := d.blockchain.InsertReceiptChain(blocks, receipts, d.ancientLimit); err != nil {
  1768. log.Debug("Downloaded item processing failed", "number", results[index].Header.Number, "hash", results[index].Header.Hash(), "err", err)
  1769. return fmt.Errorf("%w: %v", errInvalidChain, err)
  1770. }
  1771. return nil
  1772. }
  1773. func (d *Downloader) commitPivotBlock(result *fetchResult) error {
  1774. block := types.NewBlockWithHeader(result.Header).WithBody(result.Transactions, result.Uncles)
  1775. log.Debug("Committing fast sync pivot as new head", "number", block.Number(), "hash", block.Hash())
  1776. // Commit the pivot block as the new head, will require full sync from here on
  1777. if _, err := d.blockchain.InsertReceiptChain([]*types.Block{block}, []types.Receipts{result.Receipts}, d.ancientLimit); err != nil {
  1778. return err
  1779. }
  1780. if err := d.blockchain.FastSyncCommitHead(block.Hash()); err != nil {
  1781. return err
  1782. }
  1783. atomic.StoreInt32(&d.committed, 1)
  1784. // If we had a bloom filter for the state sync, deallocate it now. Note, we only
  1785. // deallocate internally, but keep the empty wrapper. This ensures that if we do
  1786. // a rollback after committing the pivot and restarting fast sync, we don't end
  1787. // up using a nil bloom. Empty bloom is fine, it just returns that it does not
  1788. // have the info we need, so reach down to the database instead.
  1789. if d.stateBloom != nil {
  1790. d.stateBloom.Close()
  1791. }
  1792. return nil
  1793. }
  1794. // DeliverHeaders injects a new batch of block headers received from a remote
  1795. // node into the download schedule.
  1796. func (d *Downloader) DeliverHeaders(id string, headers []*types.Header) error {
  1797. return d.deliver(d.headerCh, &headerPack{id, headers}, headerInMeter, headerDropMeter)
  1798. }
  1799. // DeliverBodies injects a new batch of block bodies received from a remote node.
  1800. func (d *Downloader) DeliverBodies(id string, transactions [][]*types.Transaction, uncles [][]*types.Header) error {
  1801. return d.deliver(d.bodyCh, &bodyPack{id, transactions, uncles}, bodyInMeter, bodyDropMeter)
  1802. }
  1803. // DeliverReceipts injects a new batch of receipts received from a remote node.
  1804. func (d *Downloader) DeliverReceipts(id string, receipts [][]*types.Receipt) error {
  1805. return d.deliver(d.receiptCh, &receiptPack{id, receipts}, receiptInMeter, receiptDropMeter)
  1806. }
  1807. // DeliverNodeData injects a new batch of node state data received from a remote node.
  1808. func (d *Downloader) DeliverNodeData(id string, data [][]byte) error {
  1809. return d.deliver(d.stateCh, &statePack{id, data}, stateInMeter, stateDropMeter)
  1810. }
  1811. // DeliverSnapPacket is invoked from a peer's message handler when it transmits a
  1812. // data packet for the local node to consume.
  1813. func (d *Downloader) DeliverSnapPacket(peer *snap.Peer, packet snap.Packet) error {
  1814. switch packet := packet.(type) {
  1815. case *snap.AccountRangePacket:
  1816. hashes, accounts, err := packet.Unpack()
  1817. if err != nil {
  1818. return err
  1819. }
  1820. return d.SnapSyncer.OnAccounts(peer, packet.ID, hashes, accounts, packet.Proof)
  1821. case *snap.StorageRangesPacket:
  1822. hashset, slotset := packet.Unpack()
  1823. return d.SnapSyncer.OnStorage(peer, packet.ID, hashset, slotset, packet.Proof)
  1824. case *snap.ByteCodesPacket:
  1825. return d.SnapSyncer.OnByteCodes(peer, packet.ID, packet.Codes)
  1826. case *snap.TrieNodesPacket:
  1827. return d.SnapSyncer.OnTrieNodes(peer, packet.ID, packet.Nodes)
  1828. default:
  1829. return fmt.Errorf("unexpected snap packet type: %T", packet)
  1830. }
  1831. }
  1832. // deliver injects a new batch of data received from a remote node.
  1833. func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dropMeter metrics.Meter) (err error) {
  1834. // Update the delivery metrics for both good and failed deliveries
  1835. inMeter.Mark(int64(packet.Items()))
  1836. defer func() {
  1837. if err != nil {
  1838. dropMeter.Mark(int64(packet.Items()))
  1839. }
  1840. }()
  1841. // Deliver or abort if the sync is canceled while queuing
  1842. d.cancelLock.RLock()
  1843. cancel := d.cancelCh
  1844. d.cancelLock.RUnlock()
  1845. if cancel == nil {
  1846. return errNoSyncActive
  1847. }
  1848. select {
  1849. case destCh <- packet:
  1850. return nil
  1851. case <-cancel:
  1852. return errNoSyncActive
  1853. }
  1854. }
  1855. // qosTuner is the quality of service tuning loop that occasionally gathers the
  1856. // peer latency statistics and updates the estimated request round trip time.
  1857. func (d *Downloader) qosTuner() {
  1858. for {
  1859. // Retrieve the current median RTT and integrate into the previoust target RTT
  1860. rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT()))
  1861. atomic.StoreUint64(&d.rttEstimate, uint64(rtt))
  1862. // A new RTT cycle passed, increase our confidence in the estimated RTT
  1863. conf := atomic.LoadUint64(&d.rttConfidence)
  1864. conf = conf + (1000000-conf)/2
  1865. atomic.StoreUint64(&d.rttConfidence, conf)
  1866. // Log the new QoS values and sleep until the next RTT
  1867. log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL())
  1868. select {
  1869. case <-d.quitCh:
  1870. return
  1871. case <-time.After(rtt):
  1872. }
  1873. }
  1874. }
  1875. // qosReduceConfidence is meant to be called when a new peer joins the downloader's
  1876. // peer set, needing to reduce the confidence we have in out QoS estimates.
  1877. func (d *Downloader) qosReduceConfidence() {
  1878. // If we have a single peer, confidence is always 1
  1879. peers := uint64(d.peers.Len())
  1880. if peers == 0 {
  1881. // Ensure peer connectivity races don't catch us off guard
  1882. return
  1883. }
  1884. if peers == 1 {
  1885. atomic.StoreUint64(&d.rttConfidence, 1000000)
  1886. return
  1887. }
  1888. // If we have a ton of peers, don't drop confidence)
  1889. if peers >= uint64(qosConfidenceCap) {
  1890. return
  1891. }
  1892. // Otherwise drop the confidence factor
  1893. conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers
  1894. if float64(conf)/1000000 < rttMinConfidence {
  1895. conf = uint64(rttMinConfidence * 1000000)
  1896. }
  1897. atomic.StoreUint64(&d.rttConfidence, conf)
  1898. rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate))
  1899. log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL())
  1900. }
  1901. // requestRTT returns the current target round trip time for a download request
  1902. // to complete in.
  1903. //
  1904. // Note, the returned RTT is .9 of the actually estimated RTT. The reason is that
  1905. // the downloader tries to adapt queries to the RTT, so multiple RTT values can
  1906. // be adapted to, but smaller ones are preferred (stabler download stream).
  1907. func (d *Downloader) requestRTT() time.Duration {
  1908. return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10
  1909. }
  1910. // requestTTL returns the current timeout allowance for a single download request
  1911. // to finish under.
  1912. func (d *Downloader) requestTTL() time.Duration {
  1913. var (
  1914. rtt = time.Duration(atomic.LoadUint64(&d.rttEstimate))
  1915. conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0
  1916. )
  1917. ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf)
  1918. if ttl > ttlLimit {
  1919. ttl = ttlLimit
  1920. }
  1921. return ttl
  1922. }