go-ethereum/p2p/dnsdisc/sync.go
Felix Lange 191364c350 p2p/dnsdisc: add enode.Iterator API (#20437)
* p2p/dnsdisc: add support for enode.Iterator

This changes the dnsdisc.Client API to support the enode.Iterator
interface.

* p2p/dnsdisc: rate-limit DNS requests

* p2p/dnsdisc: preserve linked trees across root updates

This improves the way links are handled when the link root changes.
Previously, sync would simply remove all links from the current tree and
garbage-collect all unreachable trees before syncing the new list of
links.

This behavior isn't great in certain cases: Consider a structure where
trees A, B, and C reference each other and D links to A. If D's link
root changed, the sync code would first remove trees A, B and C, only to
re-sync them later when the link to A was found again.

The fix for this problem is to track the current set of links in each
clientTree and removing old links only AFTER all links are synced.

* p2p/dnsdisc: deflake iterator test

* cmd/devp2p: adapt dnsClient to new p2p/dnsdisc API

* p2p/dnsdisc: tiny comment fix
2019-12-12 11:15:36 +02:00

267 lines
6.8 KiB
Go

// Copyright 2019 The go-ethereum Authors
// This file is part of the go-ethereum library.
//
// The go-ethereum library is free software: you can redistribute it and/or modify
// it under the terms of the GNU Lesser General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// The go-ethereum library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public License
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
package dnsdisc
import (
"context"
"math/rand"
"time"
"github.com/ethereum/go-ethereum/common/mclock"
"github.com/ethereum/go-ethereum/p2p/enode"
)
// clientTree is a full tree being synced.
type clientTree struct {
c *Client
loc *linkEntry // link to this tree
lastRootCheck mclock.AbsTime // last revalidation of root
root *rootEntry
enrs *subtreeSync
links *subtreeSync
lc *linkCache // tracks all links between all trees
curLinks map[string]struct{} // links contained in this tree
linkGCRoot string // root on which last link GC has run
}
func newClientTree(c *Client, lc *linkCache, loc *linkEntry) *clientTree {
return &clientTree{c: c, lc: lc, loc: loc}
}
// syncAll retrieves all entries of the tree.
func (ct *clientTree) syncAll(dest map[string]entry) error {
if err := ct.updateRoot(); err != nil {
return err
}
if err := ct.links.resolveAll(dest); err != nil {
return err
}
if err := ct.enrs.resolveAll(dest); err != nil {
return err
}
return nil
}
// syncRandom retrieves a single entry of the tree. The Node return value
// is non-nil if the entry was a node.
func (ct *clientTree) syncRandom(ctx context.Context) (*enode.Node, error) {
if ct.rootUpdateDue() {
if err := ct.updateRoot(); err != nil {
return nil, err
}
}
// Link tree sync has priority, run it to completion before syncing ENRs.
if !ct.links.done() {
err := ct.syncNextLink(ctx)
return nil, err
}
ct.gcLinks()
// Sync next random entry in ENR tree. Once every node has been visited, we simply
// start over. This is fine because entries are cached.
if ct.enrs.done() {
ct.enrs = newSubtreeSync(ct.c, ct.loc, ct.root.eroot, false)
}
return ct.syncNextRandomENR(ctx)
}
// gcLinks removes outdated links from the global link cache. GC runs once
// when the link sync finishes.
func (ct *clientTree) gcLinks() {
if !ct.links.done() || ct.root.lroot == ct.linkGCRoot {
return
}
ct.lc.resetLinks(ct.loc.str, ct.curLinks)
ct.linkGCRoot = ct.root.lroot
}
func (ct *clientTree) syncNextLink(ctx context.Context) error {
hash := ct.links.missing[0]
e, err := ct.links.resolveNext(ctx, hash)
if err != nil {
return err
}
ct.links.missing = ct.links.missing[1:]
if dest, ok := e.(*linkEntry); ok {
ct.lc.addLink(ct.loc.str, dest.str)
ct.curLinks[dest.str] = struct{}{}
}
return nil
}
func (ct *clientTree) syncNextRandomENR(ctx context.Context) (*enode.Node, error) {
index := rand.Intn(len(ct.enrs.missing))
hash := ct.enrs.missing[index]
e, err := ct.enrs.resolveNext(ctx, hash)
if err != nil {
return nil, err
}
ct.enrs.missing = removeHash(ct.enrs.missing, index)
if ee, ok := e.(*enrEntry); ok {
return ee.node, nil
}
return nil, nil
}
func (ct *clientTree) String() string {
return ct.loc.String()
}
// removeHash removes the element at index from h.
func removeHash(h []string, index int) []string {
if len(h) == 1 {
return nil
}
last := len(h) - 1
if index < last {
h[index] = h[last]
h[last] = ""
}
return h[:last]
}
// updateRoot ensures that the given tree has an up-to-date root.
func (ct *clientTree) updateRoot() error {
ct.lastRootCheck = ct.c.clock.Now()
ctx, cancel := context.WithTimeout(context.Background(), ct.c.cfg.Timeout)
defer cancel()
root, err := ct.c.resolveRoot(ctx, ct.loc)
if err != nil {
return err
}
ct.root = &root
// Invalidate subtrees if changed.
if ct.links == nil || root.lroot != ct.links.root {
ct.links = newSubtreeSync(ct.c, ct.loc, root.lroot, true)
ct.curLinks = make(map[string]struct{})
}
if ct.enrs == nil || root.eroot != ct.enrs.root {
ct.enrs = newSubtreeSync(ct.c, ct.loc, root.eroot, false)
}
return nil
}
// rootUpdateDue returns true when a root update is needed.
func (ct *clientTree) rootUpdateDue() bool {
return ct.root == nil || time.Duration(ct.c.clock.Now()-ct.lastRootCheck) > ct.c.cfg.RecheckInterval
}
// subtreeSync is the sync of an ENR or link subtree.
type subtreeSync struct {
c *Client
loc *linkEntry
root string
missing []string // missing tree node hashes
link bool // true if this sync is for the link tree
}
func newSubtreeSync(c *Client, loc *linkEntry, root string, link bool) *subtreeSync {
return &subtreeSync{c, loc, root, []string{root}, link}
}
func (ts *subtreeSync) done() bool {
return len(ts.missing) == 0
}
func (ts *subtreeSync) resolveAll(dest map[string]entry) error {
for !ts.done() {
hash := ts.missing[0]
ctx, cancel := context.WithTimeout(context.Background(), ts.c.cfg.Timeout)
e, err := ts.resolveNext(ctx, hash)
cancel()
if err != nil {
return err
}
dest[hash] = e
ts.missing = ts.missing[1:]
}
return nil
}
func (ts *subtreeSync) resolveNext(ctx context.Context, hash string) (entry, error) {
e, err := ts.c.resolveEntry(ctx, ts.loc.domain, hash)
if err != nil {
return nil, err
}
switch e := e.(type) {
case *enrEntry:
if ts.link {
return nil, errENRInLinkTree
}
case *linkEntry:
if !ts.link {
return nil, errLinkInENRTree
}
case *branchEntry:
ts.missing = append(ts.missing, e.children...)
}
return e, nil
}
// linkCache tracks links between trees.
type linkCache struct {
backrefs map[string]map[string]struct{}
changed bool
}
func (lc *linkCache) isReferenced(r string) bool {
return len(lc.backrefs[r]) != 0
}
func (lc *linkCache) addLink(from, to string) {
if _, ok := lc.backrefs[to][from]; ok {
return
}
if lc.backrefs == nil {
lc.backrefs = make(map[string]map[string]struct{})
}
if _, ok := lc.backrefs[to]; !ok {
lc.backrefs[to] = make(map[string]struct{})
}
lc.backrefs[to][from] = struct{}{}
lc.changed = true
}
// resetLinks clears all links of the given tree.
func (lc *linkCache) resetLinks(from string, keep map[string]struct{}) {
stk := []string{from}
for len(stk) > 0 {
item := stk[len(stk)-1]
stk = stk[:len(stk)-1]
for r, refs := range lc.backrefs {
if _, ok := keep[r]; ok {
continue
}
if _, ok := refs[item]; !ok {
continue
}
lc.changed = true
delete(refs, item)
if len(refs) == 0 {
delete(lc.backrefs, r)
stk = append(stk, r)
}
}
}
}