storj/satellite/audit/verifier.go

668 lines
22 KiB
Go
Raw Normal View History

2019-01-24 20:15:10 +00:00
// Copyright (C) 2019 Storj Labs, Inc.
2018-10-09 22:10:37 +01:00
// See LICENSE for copying information.
package audit
import (
"bytes"
"context"
"errors"
"fmt"
2018-10-09 22:10:37 +01:00
"io"
"math/rand"
"time"
2018-10-09 22:10:37 +01:00
"github.com/davecgh/go-spew/spew"
"github.com/spacemonkeygo/monkit/v3"
2018-10-09 22:10:37 +01:00
"github.com/vivint/infectious"
2019-01-29 20:42:27 +00:00
"github.com/zeebo/errs"
"go.uber.org/zap"
2018-10-09 22:10:37 +01:00
"storj.io/common/errs2"
"storj.io/common/identity"
"storj.io/common/memory"
"storj.io/common/pb"
"storj.io/common/rpc"
"storj.io/common/rpc/rpcpool"
"storj.io/common/rpc/rpcstatus"
"storj.io/common/storj"
"storj.io/storj/satellite/metabase"
2019-03-28 20:09:23 +00:00
"storj.io/storj/satellite/orders"
"storj.io/storj/satellite/overlay"
"storj.io/uplink/private/piecestore"
2018-10-09 22:10:37 +01:00
)
var (
mon = monkit.Package()
// ErrNotEnoughShares is the errs class for when not enough shares are available to do an audit.
ErrNotEnoughShares = errs.Class("not enough shares for successful audit")
// ErrSegmentDeleted is the errs class when the audited segment was deleted during the audit.
ErrSegmentDeleted = errs.Class("segment deleted during audit")
// ErrSegmentModified is the errs class used when a segment has been changed in any way.
ErrSegmentModified = errs.Class("segment has been modified")
)
2018-10-09 22:10:37 +01:00
// FailurePhase indicates during which phase a GET_SHARE operation failed.
type FailurePhase int
const (
// NoFailure indicates there was no failure during a GET_SHARE operation.
NoFailure FailurePhase = 0
// DialFailure indicates a GET_SHARE operation failed during Dial.
DialFailure FailurePhase = 1
// RequestFailure indicates a GET_SHARE operation failed to make its RPC request, or the request failed.
RequestFailure FailurePhase = 2
)
// Share represents required information about an audited share.
2019-01-23 19:58:44 +00:00
type Share struct {
Error error
FailurePhase FailurePhase
PieceNum int
NodeID storj.NodeID
Data []byte
2018-10-09 22:10:37 +01:00
}
// Verifier helps verify the correctness of a given stripe.
2019-09-10 14:24:16 +01:00
//
// architecture: Worker
type Verifier struct {
log *zap.Logger
metabase *metabase.DB
orders *orders.Service
auditor *identity.PeerIdentity
dialer rpc.Dialer
overlay *overlay.Service
containment Containment
minBytesPerSecond memory.Size
minDownloadTimeout time.Duration
nowFn func() time.Time
OnTestingCheckSegmentAlteredHook func()
2018-10-09 22:10:37 +01:00
}
// NewVerifier creates a Verifier.
func NewVerifier(log *zap.Logger, metabase *metabase.DB, dialer rpc.Dialer, overlay *overlay.Service, containment Containment, orders *orders.Service, id *identity.FullIdentity, minBytesPerSecond memory.Size, minDownloadTimeout time.Duration) *Verifier {
return &Verifier{
log: log,
metabase: metabase,
orders: orders,
auditor: id.PeerIdentity(),
dialer: dialer,
overlay: overlay,
containment: containment,
minBytesPerSecond: minBytesPerSecond,
minDownloadTimeout: minDownloadTimeout,
nowFn: time.Now,
}
2018-10-09 22:10:37 +01:00
}
// Verify downloads shares then verifies the data correctness at a random stripe.
func (verifier *Verifier) Verify(ctx context.Context, segment Segment, skip map[storj.NodeID]bool) (report Report, err error) {
2019-03-20 10:54:37 +00:00
defer mon.Task()(&ctx)(&err)
var segmentInfo metabase.Segment
defer func() {
recordStats(report, len(segmentInfo.Pieces), err)
}()
if segment.Expired(verifier.nowFn()) {
verifier.log.Debug("segment expired before Verify")
return Report{}, nil
}
segmentInfo, err = verifier.metabase.GetSegmentByPosition(ctx, metabase.GetSegmentByPosition{
StreamID: segment.StreamID,
Position: segment.Position,
})
if err != nil {
if metabase.ErrSegmentNotFound.Has(err) {
verifier.log.Debug("segment deleted before Verify")
return Report{}, nil
}
return Report{}, err
}
randomIndex, err := GetRandomStripe(ctx, segmentInfo)
if err != nil {
return Report{}, err
}
2019-03-28 20:09:23 +00:00
var offlineNodes storj.NodeIDList
var failedNodes metabase.Pieces
var unknownNodes storj.NodeIDList
containedNodes := make(map[int]storj.NodeID)
sharesToAudit := make(map[int]Share)
orderLimits, privateKey, cachedNodesInfo, err := verifier.orders.CreateAuditOrderLimits(ctx, segmentInfo, skip)
2019-03-28 20:09:23 +00:00
if err != nil {
if orders.ErrDownloadFailedNotEnoughPieces.Has(err) {
mon.Counter("not_enough_shares_for_audit").Inc(1) //mon:locked
mon.Counter("audit_not_enough_nodes_online").Inc(1) //mon:locked
err = ErrNotEnoughShares.Wrap(err)
}
return Report{}, err
2019-03-28 20:09:23 +00:00
}
cachedNodesReputation := make(map[storj.NodeID]overlay.ReputationStatus, len(cachedNodesInfo))
for id, info := range cachedNodesInfo {
cachedNodesReputation[id] = info.Reputation
}
defer func() { report.NodesReputation = cachedNodesReputation }()
2019-03-20 10:54:37 +00:00
// NOTE offlineNodes will include disqualified nodes because they aren't in
// the skip list
offlineNodes = getOfflineNodes(segmentInfo, orderLimits, skip)
if len(offlineNodes) > 0 {
verifier.log.Debug("Verify: order limits not created for some nodes (offline/disqualified)",
zap.Strings("Node IDs", offlineNodes.Strings()),
zap.String("Segment", segmentInfoString(segment)))
}
shares, err := verifier.DownloadShares(ctx, orderLimits, privateKey, cachedNodesInfo, randomIndex, segmentInfo.Redundancy.ShareSize)
2019-03-20 10:54:37 +00:00
if err != nil {
return Report{
Offlines: offlineNodes,
}, err
}
err = verifier.checkIfSegmentAltered(ctx, segmentInfo)
if err != nil {
if ErrSegmentDeleted.Has(err) {
verifier.log.Debug("segment deleted during Verify")
return Report{}, nil
}
if ErrSegmentModified.Has(err) {
verifier.log.Debug("segment modified during Verify")
return Report{}, nil
}
return Report{
Offlines: offlineNodes,
}, err
2019-03-20 10:54:37 +00:00
}
for pieceNum, share := range shares {
if share.Error == nil {
// no error -- share downloaded successfully
sharesToAudit[pieceNum] = share
continue
}
pieceID := orderLimits[pieceNum].Limit.PieceId
errLogger := verifier.log.With(
zap.Stringer("Node ID", share.NodeID),
zap.String("Segment", segmentInfoString(segment)),
zap.Stringer("Piece ID", pieceID),
zap.Error(share.Error),
)
switch share.FailurePhase {
case DialFailure:
// dial failed -- offline node
offlineNodes = append(offlineNodes, share.NodeID)
errLogger.Debug("Verify: dial failed (offline)")
continue
case RequestFailure:
if errs2.IsRPC(share.Error, rpcstatus.NotFound) {
// missing share
failedNodes = append(failedNodes, metabase.Piece{
Number: uint16(share.PieceNum),
StorageNode: share.NodeID,
})
errLogger.Info("Verify: piece not found (audit failed)")
continue
2019-03-20 10:54:37 +00:00
}
if errs2.IsRPC(share.Error, rpcstatus.DeadlineExceeded) {
// dial successful, but download timed out
containedNodes[pieceNum] = share.NodeID
errLogger.Info("Verify: download timeout (contained)")
continue
}
// unknown error
unknownNodes = append(unknownNodes, share.NodeID)
errLogger.Info("Verify: unknown error (skipped)",
zap.String("ErrorType", spew.Sprintf("%#+v", share.Error)))
2019-03-20 10:54:37 +00:00
}
}
mon.IntVal("verify_shares_downloaded_successfully").Observe(int64(len(sharesToAudit))) //mon:locked
required := segmentInfo.Redundancy.RequiredShares
total := segmentInfo.Redundancy.TotalShares
2019-03-20 10:54:37 +00:00
if len(sharesToAudit) < int(required) {
mon.Counter("not_enough_shares_for_audit").Inc(1) //mon:locked
// if we have reached this point, most likely something went wrong
// like a network problem or a forgotten delete. Don't fail nodes.
// We have an alert on this. Check the logs and see what happened.
if len(offlineNodes)+len(containedNodes) > len(sharesToAudit)+len(failedNodes)+len(unknownNodes) {
mon.Counter("audit_suspected_network_problem").Inc(1) //mon:locked
} else {
mon.Counter("audit_not_enough_shares_acquired").Inc(1) //mon:locked
}
report := Report{
Offlines: offlineNodes,
Unknown: unknownNodes,
}
return report, ErrNotEnoughShares.New("got: %d, required: %d, failed: %d, offline: %d, unknown: %d, contained: %d",
len(sharesToAudit), required, len(failedNodes), len(offlineNodes), len(unknownNodes), len(containedNodes))
2019-03-20 10:54:37 +00:00
}
// ensure we get values, even if only zero values, so that redash can have an alert based on these
mon.Counter("not_enough_shares_for_audit").Inc(0) //mon:locked
mon.Counter("audit_not_enough_nodes_online").Inc(0) //mon:locked
mon.Counter("audit_not_enough_shares_acquired").Inc(0) //mon:locked
mon.Counter("could_not_verify_audit_shares").Inc(0) //mon:locked
mon.Counter("audit_suspected_network_problem").Inc(0) //mon:locked
2019-03-20 10:54:37 +00:00
pieceNums, _, err := auditShares(ctx, required, total, sharesToAudit)
2019-03-20 10:54:37 +00:00
if err != nil {
mon.Counter("could_not_verify_audit_shares").Inc(1) //mon:locked
verifier.log.Error("could not verify shares", zap.String("Segment", segmentInfoString(segment)), zap.Error(err))
return Report{
Segment: &segmentInfo,
Fails: failedNodes,
Offlines: offlineNodes,
Unknown: unknownNodes,
}, err
2019-03-20 10:54:37 +00:00
}
for _, pieceNum := range pieceNums {
verifier.log.Info("Verify: share data altered (audit failed)",
zap.Stringer("Node ID", shares[pieceNum].NodeID),
zap.String("Segment", segmentInfoString(segment)))
failedNodes = append(failedNodes, metabase.Piece{
StorageNode: shares[pieceNum].NodeID,
Number: uint16(pieceNum),
})
2019-03-20 10:54:37 +00:00
}
successNodes := getSuccessNodes(ctx, shares, failedNodes, offlineNodes, unknownNodes, containedNodes)
pendingAudits, err := createPendingAudits(ctx, containedNodes, segment)
if err != nil {
return Report{
Segment: &segmentInfo,
Successes: successNodes,
Fails: failedNodes,
Offlines: offlineNodes,
Unknown: unknownNodes,
}, err
}
2019-03-20 10:54:37 +00:00
return Report{
Segment: &segmentInfo,
Successes: successNodes,
Fails: failedNodes,
Offlines: offlineNodes,
PendingAudits: pendingAudits,
Unknown: unknownNodes,
2019-03-20 10:54:37 +00:00
}, nil
}
func segmentInfoString(segment Segment) string {
return fmt.Sprintf("%s/%d",
segment.StreamID.String(),
segment.Position.Encode(),
)
}
// DownloadShares downloads shares from the nodes where remote pieces are located.
func (verifier *Verifier) DownloadShares(ctx context.Context, limits []*pb.AddressedOrderLimit, piecePrivateKey storj.PiecePrivateKey, cachedNodesInfo map[storj.NodeID]overlay.NodeReputation, stripeIndex int32, shareSize int32) (shares map[int]Share, err error) {
2019-03-20 10:54:37 +00:00
defer mon.Task()(&ctx)(&err)
shares = make(map[int]Share, len(limits))
ch := make(chan *Share, len(limits))
2019-03-20 10:54:37 +00:00
for i, limit := range limits {
if limit == nil {
ch <- nil
2019-03-20 10:54:37 +00:00
continue
}
var ipPort string
node, ok := cachedNodesInfo[limit.Limit.StorageNodeId]
if ok && node.LastIPPort != "" {
ipPort = node.LastIPPort
}
go func(i int, limit *pb.AddressedOrderLimit) {
share := verifier.GetShare(ctx, limit, piecePrivateKey, ipPort, stripeIndex, shareSize, i)
ch <- &share
}(i, limit)
}
2019-03-20 10:54:37 +00:00
for range limits {
share := <-ch
if share != nil {
shares[share.PieceNum] = *share
}
2019-03-20 10:54:37 +00:00
}
return shares, nil
2019-03-20 10:54:37 +00:00
}
// IdentifyContainedNodes returns the set of all contained nodes out of the
// holders of pieces in the given segment.
func (verifier *Verifier) IdentifyContainedNodes(ctx context.Context, segment Segment) (skipList map[storj.NodeID]bool, err error) {
segmentInfo, err := verifier.metabase.GetSegmentByPosition(ctx, metabase.GetSegmentByPosition{
StreamID: segment.StreamID,
Position: segment.Position,
})
if err != nil {
return nil, err
}
skipList = make(map[storj.NodeID]bool)
for _, piece := range segmentInfo.Pieces {
_, err := verifier.containment.Get(ctx, piece.StorageNode)
if err != nil {
if ErrContainedNotFound.Has(err) {
continue
}
verifier.log.Error("can not determine if node is contained", zap.Stringer("node-id", piece.StorageNode), zap.Error(err))
continue
}
skipList[piece.StorageNode] = true
}
return skipList, nil
}
// GetShare use piece store client to download shares from nodes.
func (verifier *Verifier) GetShare(ctx context.Context, limit *pb.AddressedOrderLimit, piecePrivateKey storj.PiecePrivateKey, cachedIPAndPort string, stripeIndex, shareSize int32, pieceNum int) (share Share) {
defer mon.Task()(&ctx)(&share.Error)
share.PieceNum = pieceNum
share.NodeID = limit.GetLimit().StorageNodeId
share.FailurePhase = DialFailure
bandwidthMsgSize := shareSize
// determines number of seconds allotted for receiving data from a storage node
2019-03-22 13:14:17 +00:00
timedCtx := ctx
if verifier.minBytesPerSecond > 0 {
maxTransferTime := time.Duration(int64(time.Second) * int64(bandwidthMsgSize) / verifier.minBytesPerSecond.Int64())
if maxTransferTime < verifier.minDownloadTimeout {
maxTransferTime = verifier.minDownloadTimeout
}
2019-03-22 13:14:17 +00:00
var cancel func()
timedCtx, cancel = context.WithTimeout(ctx, maxTransferTime)
defer cancel()
}
satellite/audit: use LastIPAndPort preferentially This preserves the last_ip_and_port field from node lookups through CreateAuditOrderLimits() and CreateAuditOrderLimit(), so that later calls to (*Verifier).GetShare() can try to use that IP and port. If a connection to the given IP and port cannot be made, or the connection cannot be verified and secured with the target node identity, an attempt is made to connect to the original node address instead. A similar change is not necessary to the other Create*OrderLimits functions, because they already replace node addresses with the cached IP and port as appropriate. We might want to consider making a similar change to CreateGetRepairOrderLimits(), though. The audit situation is unique because the ramifications are especially powerful when we get the address wrong. Failing a single audit can have a heavy cost to a storage node. We need to make extra effort in order to avoid imposing that cost unfairly. Situation 1: If an audit fails because the repair worker failed to make a DNS query (which might well be the fault on the satellite side), and we have last_ip_and_port information available for the target node, it would be unfair not to try connecting to that last_ip_and_port address. Situation 2: If a node has changed addresses recently and the operator correctly changed its DNS entry, but we don't bother querying DNS, it would be unfair to penalize the node for our failure to connect to it. So the audit worker must try both last_ip_and_port _and_ the node address as supplied by the SNO. We elect here to try last_ip_and_port first, on the grounds that (a) it is expected to work in the large majority of cases, and (b) there should not be any security concerns with connecting to an out-or-date address, and (c) avoiding DNS queries on the satellite side helps alleviate satellite operational load. Change-Id: I9bf6c6c79866d879adecac6144a6c346f4f61200
2020-09-30 05:53:43 +01:00
targetNodeID := limit.GetLimit().StorageNodeId
log := verifier.log.Named(targetNodeID.String())
var ps *piecestore.Client
var err error
satellite/audit: use LastIPAndPort preferentially This preserves the last_ip_and_port field from node lookups through CreateAuditOrderLimits() and CreateAuditOrderLimit(), so that later calls to (*Verifier).GetShare() can try to use that IP and port. If a connection to the given IP and port cannot be made, or the connection cannot be verified and secured with the target node identity, an attempt is made to connect to the original node address instead. A similar change is not necessary to the other Create*OrderLimits functions, because they already replace node addresses with the cached IP and port as appropriate. We might want to consider making a similar change to CreateGetRepairOrderLimits(), though. The audit situation is unique because the ramifications are especially powerful when we get the address wrong. Failing a single audit can have a heavy cost to a storage node. We need to make extra effort in order to avoid imposing that cost unfairly. Situation 1: If an audit fails because the repair worker failed to make a DNS query (which might well be the fault on the satellite side), and we have last_ip_and_port information available for the target node, it would be unfair not to try connecting to that last_ip_and_port address. Situation 2: If a node has changed addresses recently and the operator correctly changed its DNS entry, but we don't bother querying DNS, it would be unfair to penalize the node for our failure to connect to it. So the audit worker must try both last_ip_and_port _and_ the node address as supplied by the SNO. We elect here to try last_ip_and_port first, on the grounds that (a) it is expected to work in the large majority of cases, and (b) there should not be any security concerns with connecting to an out-or-date address, and (c) avoiding DNS queries on the satellite side helps alleviate satellite operational load. Change-Id: I9bf6c6c79866d879adecac6144a6c346f4f61200
2020-09-30 05:53:43 +01:00
// if cached IP is given, try connecting there first
if cachedIPAndPort != "" {
nodeAddr := storj.NodeURL{
ID: targetNodeID,
Address: cachedIPAndPort,
}
ps, err = piecestore.Dial(rpcpool.WithForceDial(timedCtx), verifier.dialer, nodeAddr, piecestore.DefaultConfig)
satellite/audit: use LastIPAndPort preferentially This preserves the last_ip_and_port field from node lookups through CreateAuditOrderLimits() and CreateAuditOrderLimit(), so that later calls to (*Verifier).GetShare() can try to use that IP and port. If a connection to the given IP and port cannot be made, or the connection cannot be verified and secured with the target node identity, an attempt is made to connect to the original node address instead. A similar change is not necessary to the other Create*OrderLimits functions, because they already replace node addresses with the cached IP and port as appropriate. We might want to consider making a similar change to CreateGetRepairOrderLimits(), though. The audit situation is unique because the ramifications are especially powerful when we get the address wrong. Failing a single audit can have a heavy cost to a storage node. We need to make extra effort in order to avoid imposing that cost unfairly. Situation 1: If an audit fails because the repair worker failed to make a DNS query (which might well be the fault on the satellite side), and we have last_ip_and_port information available for the target node, it would be unfair not to try connecting to that last_ip_and_port address. Situation 2: If a node has changed addresses recently and the operator correctly changed its DNS entry, but we don't bother querying DNS, it would be unfair to penalize the node for our failure to connect to it. So the audit worker must try both last_ip_and_port _and_ the node address as supplied by the SNO. We elect here to try last_ip_and_port first, on the grounds that (a) it is expected to work in the large majority of cases, and (b) there should not be any security concerns with connecting to an out-or-date address, and (c) avoiding DNS queries on the satellite side helps alleviate satellite operational load. Change-Id: I9bf6c6c79866d879adecac6144a6c346f4f61200
2020-09-30 05:53:43 +01:00
if err != nil {
log.Debug("failed to connect to audit target node at cached IP", zap.String("cached-ip-and-port", cachedIPAndPort), zap.Error(err))
}
}
satellite/audit: use LastIPAndPort preferentially This preserves the last_ip_and_port field from node lookups through CreateAuditOrderLimits() and CreateAuditOrderLimit(), so that later calls to (*Verifier).GetShare() can try to use that IP and port. If a connection to the given IP and port cannot be made, or the connection cannot be verified and secured with the target node identity, an attempt is made to connect to the original node address instead. A similar change is not necessary to the other Create*OrderLimits functions, because they already replace node addresses with the cached IP and port as appropriate. We might want to consider making a similar change to CreateGetRepairOrderLimits(), though. The audit situation is unique because the ramifications are especially powerful when we get the address wrong. Failing a single audit can have a heavy cost to a storage node. We need to make extra effort in order to avoid imposing that cost unfairly. Situation 1: If an audit fails because the repair worker failed to make a DNS query (which might well be the fault on the satellite side), and we have last_ip_and_port information available for the target node, it would be unfair not to try connecting to that last_ip_and_port address. Situation 2: If a node has changed addresses recently and the operator correctly changed its DNS entry, but we don't bother querying DNS, it would be unfair to penalize the node for our failure to connect to it. So the audit worker must try both last_ip_and_port _and_ the node address as supplied by the SNO. We elect here to try last_ip_and_port first, on the grounds that (a) it is expected to work in the large majority of cases, and (b) there should not be any security concerns with connecting to an out-or-date address, and (c) avoiding DNS queries on the satellite side helps alleviate satellite operational load. Change-Id: I9bf6c6c79866d879adecac6144a6c346f4f61200
2020-09-30 05:53:43 +01:00
// if no cached IP was given, or connecting to cached IP failed, use node address
if ps == nil {
nodeAddr := storj.NodeURL{
ID: targetNodeID,
Address: limit.GetStorageNodeAddress().Address,
}
ps, err = piecestore.Dial(rpcpool.WithForceDial(timedCtx), verifier.dialer, nodeAddr, piecestore.DefaultConfig)
satellite/audit: use LastIPAndPort preferentially This preserves the last_ip_and_port field from node lookups through CreateAuditOrderLimits() and CreateAuditOrderLimit(), so that later calls to (*Verifier).GetShare() can try to use that IP and port. If a connection to the given IP and port cannot be made, or the connection cannot be verified and secured with the target node identity, an attempt is made to connect to the original node address instead. A similar change is not necessary to the other Create*OrderLimits functions, because they already replace node addresses with the cached IP and port as appropriate. We might want to consider making a similar change to CreateGetRepairOrderLimits(), though. The audit situation is unique because the ramifications are especially powerful when we get the address wrong. Failing a single audit can have a heavy cost to a storage node. We need to make extra effort in order to avoid imposing that cost unfairly. Situation 1: If an audit fails because the repair worker failed to make a DNS query (which might well be the fault on the satellite side), and we have last_ip_and_port information available for the target node, it would be unfair not to try connecting to that last_ip_and_port address. Situation 2: If a node has changed addresses recently and the operator correctly changed its DNS entry, but we don't bother querying DNS, it would be unfair to penalize the node for our failure to connect to it. So the audit worker must try both last_ip_and_port _and_ the node address as supplied by the SNO. We elect here to try last_ip_and_port first, on the grounds that (a) it is expected to work in the large majority of cases, and (b) there should not be any security concerns with connecting to an out-or-date address, and (c) avoiding DNS queries on the satellite side helps alleviate satellite operational load. Change-Id: I9bf6c6c79866d879adecac6144a6c346f4f61200
2020-09-30 05:53:43 +01:00
if err != nil {
share.Error = Error.Wrap(err)
return share
satellite/audit: use LastIPAndPort preferentially This preserves the last_ip_and_port field from node lookups through CreateAuditOrderLimits() and CreateAuditOrderLimit(), so that later calls to (*Verifier).GetShare() can try to use that IP and port. If a connection to the given IP and port cannot be made, or the connection cannot be verified and secured with the target node identity, an attempt is made to connect to the original node address instead. A similar change is not necessary to the other Create*OrderLimits functions, because they already replace node addresses with the cached IP and port as appropriate. We might want to consider making a similar change to CreateGetRepairOrderLimits(), though. The audit situation is unique because the ramifications are especially powerful when we get the address wrong. Failing a single audit can have a heavy cost to a storage node. We need to make extra effort in order to avoid imposing that cost unfairly. Situation 1: If an audit fails because the repair worker failed to make a DNS query (which might well be the fault on the satellite side), and we have last_ip_and_port information available for the target node, it would be unfair not to try connecting to that last_ip_and_port address. Situation 2: If a node has changed addresses recently and the operator correctly changed its DNS entry, but we don't bother querying DNS, it would be unfair to penalize the node for our failure to connect to it. So the audit worker must try both last_ip_and_port _and_ the node address as supplied by the SNO. We elect here to try last_ip_and_port first, on the grounds that (a) it is expected to work in the large majority of cases, and (b) there should not be any security concerns with connecting to an out-or-date address, and (c) avoiding DNS queries on the satellite side helps alleviate satellite operational load. Change-Id: I9bf6c6c79866d879adecac6144a6c346f4f61200
2020-09-30 05:53:43 +01:00
}
2018-10-09 22:10:37 +01:00
}
satellite/audit: use LastIPAndPort preferentially This preserves the last_ip_and_port field from node lookups through CreateAuditOrderLimits() and CreateAuditOrderLimit(), so that later calls to (*Verifier).GetShare() can try to use that IP and port. If a connection to the given IP and port cannot be made, or the connection cannot be verified and secured with the target node identity, an attempt is made to connect to the original node address instead. A similar change is not necessary to the other Create*OrderLimits functions, because they already replace node addresses with the cached IP and port as appropriate. We might want to consider making a similar change to CreateGetRepairOrderLimits(), though. The audit situation is unique because the ramifications are especially powerful when we get the address wrong. Failing a single audit can have a heavy cost to a storage node. We need to make extra effort in order to avoid imposing that cost unfairly. Situation 1: If an audit fails because the repair worker failed to make a DNS query (which might well be the fault on the satellite side), and we have last_ip_and_port information available for the target node, it would be unfair not to try connecting to that last_ip_and_port address. Situation 2: If a node has changed addresses recently and the operator correctly changed its DNS entry, but we don't bother querying DNS, it would be unfair to penalize the node for our failure to connect to it. So the audit worker must try both last_ip_and_port _and_ the node address as supplied by the SNO. We elect here to try last_ip_and_port first, on the grounds that (a) it is expected to work in the large majority of cases, and (b) there should not be any security concerns with connecting to an out-or-date address, and (c) avoiding DNS queries on the satellite side helps alleviate satellite operational load. Change-Id: I9bf6c6c79866d879adecac6144a6c346f4f61200
2020-09-30 05:53:43 +01:00
share.FailurePhase = RequestFailure
defer func() {
err := ps.Close()
if err != nil {
verifier.log.Error("audit verifier failed to close conn to node: %+v", zap.Error(err))
}
}()
2018-10-09 22:10:37 +01:00
offset := int64(shareSize) * int64(stripeIndex)
2018-10-09 22:10:37 +01:00
downloader, err := ps.Download(timedCtx, limit.GetLimit(), piecePrivateKey, offset, int64(shareSize))
2018-10-09 22:10:37 +01:00
if err != nil {
share.Error = err
return share
2018-10-09 22:10:37 +01:00
}
buf := make([]byte, shareSize)
_, err = io.ReadFull(downloader, buf)
closeErr := downloader.Close()
if err != nil || closeErr != nil {
if errors.Is(err, io.ErrClosedPipe) {
// in some circumstances, this can be returned from the piecestore
// when the peer returned a different error. The downloader gets
// marked as being closed, even though we haven't closed it from
// this side, and ErrClosedPipe gets returned on the next Read
// instead of the actual error. We'll get the real error from
// downloader.Close().
err = nil
}
share.Error = errs.Combine(err, closeErr)
return share
2018-10-09 22:10:37 +01:00
}
share.Data = buf
share.FailurePhase = NoFailure
2018-10-09 22:10:37 +01:00
return share
2018-10-09 22:10:37 +01:00
}
// checkIfSegmentAltered checks if oldSegment has been altered since it was selected for audit.
func (verifier *Verifier) checkIfSegmentAltered(ctx context.Context, oldSegment metabase.Segment) (err error) {
defer mon.Task()(&ctx)(&err)
if verifier.OnTestingCheckSegmentAlteredHook != nil {
verifier.OnTestingCheckSegmentAlteredHook()
}
newSegment, err := verifier.metabase.GetSegmentByPosition(ctx, metabase.GetSegmentByPosition{
StreamID: oldSegment.StreamID,
Position: oldSegment.Position,
})
if err != nil {
if metabase.ErrSegmentNotFound.Has(err) {
return ErrSegmentDeleted.New("StreamID: %q Position: %d", oldSegment.StreamID.String(), oldSegment.Position.Encode())
}
return err
}
if !oldSegment.Pieces.Equal(newSegment.Pieces) {
return ErrSegmentModified.New("StreamID: %q Position: %d", oldSegment.StreamID.String(), oldSegment.Position.Encode())
}
return nil
}
// SetNow allows tests to have the server act as if the current time is whatever they want.
func (verifier *Verifier) SetNow(nowFn func() time.Time) {
verifier.nowFn = nowFn
}
2018-10-09 22:10:37 +01:00
// auditShares takes the downloaded shares and uses infectious's Correct function to check that they
// haven't been altered. auditShares returns a slice containing the piece numbers of altered shares,
// and a slice of the corrected shares.
func auditShares(ctx context.Context, required, total int16, originals map[int]Share) (pieceNums []int, corrected []infectious.Share, err error) {
2018-10-09 22:10:37 +01:00
defer mon.Task()(&ctx)(&err)
f, err := infectious.NewFEC(int(required), int(total))
2018-10-09 22:10:37 +01:00
if err != nil {
return nil, nil, err
2018-10-09 22:10:37 +01:00
}
2018-11-07 01:16:43 +00:00
2018-10-09 22:10:37 +01:00
copies, err := makeCopies(ctx, originals)
if err != nil {
return nil, nil, err
2018-10-09 22:10:37 +01:00
}
err = f.Correct(copies)
if err != nil {
return nil, nil, err
2018-10-09 22:10:37 +01:00
}
2018-11-28 07:33:17 +00:00
for _, share := range copies {
if !bytes.Equal(originals[share.Number].Data, share.Data) {
2018-10-09 22:10:37 +01:00
pieceNums = append(pieceNums, share.Number)
}
}
return pieceNums, copies, nil
2018-10-09 22:10:37 +01:00
}
// makeCopies takes in a map of audit Shares and deep copies their data to a slice of infectious Shares.
2019-03-20 10:54:37 +00:00
func makeCopies(ctx context.Context, originals map[int]Share) (copies []infectious.Share, err error) {
2018-10-09 22:10:37 +01:00
defer mon.Task()(&ctx)(&err)
2019-03-20 10:54:37 +00:00
copies = make([]infectious.Share, 0, len(originals))
for _, original := range originals {
copies = append(copies, infectious.Share{
Data: append([]byte{}, original.Data...),
Number: original.PieceNum})
}
2019-03-20 10:54:37 +00:00
return copies, nil
}
// getOfflineNodes returns those storage nodes from the segment which have no
// order limit nor are skipped.
func getOfflineNodes(segment metabase.Segment, limits []*pb.AddressedOrderLimit, skip map[storj.NodeID]bool) storj.NodeIDList {
var offlines storj.NodeIDList
nodesWithLimit := make(map[storj.NodeID]bool, len(limits))
for _, limit := range limits {
if limit != nil {
nodesWithLimit[limit.GetLimit().StorageNodeId] = true
}
}
for _, piece := range segment.Pieces {
if !nodesWithLimit[piece.StorageNode] && !skip[piece.StorageNode] {
offlines = append(offlines, piece.StorageNode)
}
}
return offlines
}
// getSuccessNodes uses the failed nodes, offline nodes and contained nodes arrays to determine which nodes passed the audit.
func getSuccessNodes(ctx context.Context, shares map[int]Share, failedNodes metabase.Pieces, offlineNodes, unknownNodes storj.NodeIDList, containedNodes map[int]storj.NodeID) (successNodes storj.NodeIDList) {
defer mon.Task()(&ctx)(nil)
fails := make(map[storj.NodeID]bool)
for _, fail := range failedNodes {
fails[fail.StorageNode] = true
2018-10-09 22:10:37 +01:00
}
for _, offline := range offlineNodes {
fails[offline] = true
}
for _, unknown := range unknownNodes {
fails[unknown] = true
}
for _, contained := range containedNodes {
fails[contained] = true
}
for _, share := range shares {
if !fails[share.NodeID] {
successNodes = append(successNodes, share.NodeID)
}
}
2019-02-01 14:48:57 +00:00
return successNodes
}
2019-03-28 20:09:23 +00:00
func createPendingAudits(ctx context.Context, containedNodes map[int]storj.NodeID, segment Segment) (pending []*ReverificationJob, err error) {
defer mon.Task()(&ctx)(&err)
if len(containedNodes) == 0 {
return nil, nil
}
pending = make([]*ReverificationJob, 0, len(containedNodes))
for pieceNum, nodeID := range containedNodes {
pending = append(pending, &ReverificationJob{
Locator: PieceLocator{
NodeID: nodeID,
StreamID: segment.StreamID,
Position: segment.Position,
PieceNum: pieceNum,
},
})
}
return pending, nil
}
// GetRandomStripe takes a segment and returns a random stripe index within that segment.
func GetRandomStripe(ctx context.Context, segment metabase.Segment) (index int32, err error) {
defer mon.Task()(&ctx)(&err)
// the last segment could be smaller than stripe size
if segment.EncryptedSize < segment.Redundancy.StripeSize() {
return 0, nil
}
var src cryptoSource
rnd := rand.New(src)
numStripes := segment.Redundancy.StripeCount(segment.EncryptedSize)
randomStripeIndex := rnd.Int31n(numStripes)
return randomStripeIndex, nil
}
func recordStats(report Report, totalPieces int, verifyErr error) {
// If an audit was able to complete without auditing any nodes, that means
// the segment has been altered.
if verifyErr == nil && len(report.Successes) == 0 {
return
}
numOffline := len(report.Offlines)
numSuccessful := len(report.Successes)
numFailed := len(report.Fails)
numContained := len(report.PendingAudits)
numUnknown := len(report.Unknown)
totalAudited := numSuccessful + numFailed + numOffline + numContained
auditedPercentage := float64(totalAudited) / float64(totalPieces)
offlinePercentage := float64(0)
successfulPercentage := float64(0)
failedPercentage := float64(0)
containedPercentage := float64(0)
unknownPercentage := float64(0)
if totalAudited > 0 {
offlinePercentage = float64(numOffline) / float64(totalAudited)
successfulPercentage = float64(numSuccessful) / float64(totalAudited)
failedPercentage = float64(numFailed) / float64(totalAudited)
containedPercentage = float64(numContained) / float64(totalAudited)
unknownPercentage = float64(numUnknown) / float64(totalAudited)
}
mon.Meter("audit_success_nodes_global").Mark(numSuccessful) //mon:locked
mon.Meter("audit_fail_nodes_global").Mark(numFailed) //mon:locked
mon.Meter("audit_offline_nodes_global").Mark(numOffline) //mon:locked
mon.Meter("audit_contained_nodes_global").Mark(numContained) //mon:locked
mon.Meter("audit_unknown_nodes_global").Mark(numUnknown) //mon:locked
mon.Meter("audit_total_nodes_global").Mark(totalAudited) //mon:locked
mon.Meter("audit_total_pointer_nodes_global").Mark(totalPieces) //mon:locked
mon.IntVal("audit_success_nodes").Observe(int64(numSuccessful)) //mon:locked
mon.IntVal("audit_fail_nodes").Observe(int64(numFailed)) //mon:locked
mon.IntVal("audit_offline_nodes").Observe(int64(numOffline)) //mon:locked
mon.IntVal("audit_contained_nodes").Observe(int64(numContained)) //mon:locked
mon.IntVal("audit_unknown_nodes").Observe(int64(numUnknown)) //mon:locked
mon.IntVal("audit_total_nodes").Observe(int64(totalAudited)) //mon:locked
mon.IntVal("audit_total_pointer_nodes").Observe(int64(totalPieces)) //mon:locked
mon.FloatVal("audited_percentage").Observe(auditedPercentage) //mon:locked
mon.FloatVal("audit_offline_percentage").Observe(offlinePercentage) //mon:locked
mon.FloatVal("audit_successful_percentage").Observe(successfulPercentage) //mon:locked
mon.FloatVal("audit_failed_percentage").Observe(failedPercentage) //mon:locked
mon.FloatVal("audit_contained_percentage").Observe(containedPercentage) //mon:locked
mon.FloatVal("audit_unknown_percentage").Observe(unknownPercentage) //mon:locked
}