2016-06-17 20:57:51 +02:00
|
|
|
/*
|
2018-03-16 23:09:31 +01:00
|
|
|
* Minio Cloud Storage, (C) 2016, 2017, 2018 Minio, Inc.
|
2016-06-17 20:57:51 +02:00
|
|
|
*
|
|
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
* you may not use this file except in compliance with the License.
|
|
|
|
* You may obtain a copy of the License at
|
|
|
|
*
|
|
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
*
|
|
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
* See the License for the specific language governing permissions and
|
|
|
|
* limitations under the License.
|
|
|
|
*/
|
|
|
|
|
2016-08-19 01:23:42 +02:00
|
|
|
package cmd
|
2016-05-26 01:42:31 +02:00
|
|
|
|
2016-11-17 01:42:23 +01:00
|
|
|
import (
|
2018-03-14 20:01:47 +01:00
|
|
|
"context"
|
2016-11-17 01:42:23 +01:00
|
|
|
"fmt"
|
2019-01-17 13:58:18 +01:00
|
|
|
"io"
|
2016-11-17 01:42:23 +01:00
|
|
|
"path"
|
|
|
|
"sync"
|
2017-11-25 20:58:29 +01:00
|
|
|
|
2018-04-06 00:04:40 +02:00
|
|
|
"github.com/minio/minio/cmd/logger"
|
2018-01-22 23:54:55 +01:00
|
|
|
"github.com/minio/minio/pkg/madmin"
|
2016-11-17 01:42:23 +01:00
|
|
|
)
|
2016-10-17 11:10:23 +02:00
|
|
|
|
2018-04-09 19:25:41 +02:00
|
|
|
func (xl xlObjects) ReloadFormat(ctx context.Context, dryRun bool) error {
|
2018-04-10 18:36:37 +02:00
|
|
|
logger.LogIf(ctx, NotImplemented{})
|
|
|
|
return NotImplemented{}
|
2018-04-09 19:25:41 +02:00
|
|
|
}
|
|
|
|
|
2018-03-14 20:01:47 +01:00
|
|
|
func (xl xlObjects) HealFormat(ctx context.Context, dryRun bool) (madmin.HealResultItem, error) {
|
2018-04-06 00:04:40 +02:00
|
|
|
logger.LogIf(ctx, NotImplemented{})
|
|
|
|
return madmin.HealResultItem{}, NotImplemented{}
|
2016-11-17 01:42:23 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
// Heals a bucket if it doesn't exist on one of the disks, additionally
|
|
|
|
// also heals the missing entries for bucket metadata files
|
|
|
|
// `policy.json, notification.xml, listeners.json`.
|
2018-03-14 20:01:47 +01:00
|
|
|
func (xl xlObjects) HealBucket(ctx context.Context, bucket string, dryRun bool) (
|
2018-01-22 23:54:55 +01:00
|
|
|
results []madmin.HealResultItem, err error) {
|
|
|
|
|
2018-03-16 23:09:31 +01:00
|
|
|
storageDisks := xl.getDisks()
|
|
|
|
|
2017-12-22 12:28:13 +01:00
|
|
|
// get write quorum for an object
|
2018-03-16 23:09:31 +01:00
|
|
|
writeQuorum := len(storageDisks)/2 + 1
|
2017-12-22 12:28:13 +01:00
|
|
|
|
2016-11-17 01:42:23 +01:00
|
|
|
// Heal bucket.
|
2018-02-16 02:45:57 +01:00
|
|
|
var result madmin.HealResultItem
|
2018-04-06 00:04:40 +02:00
|
|
|
result, err = healBucket(ctx, storageDisks, bucket, writeQuorum, dryRun)
|
2018-01-22 23:54:55 +01:00
|
|
|
if err != nil {
|
2018-02-16 02:45:57 +01:00
|
|
|
return nil, err
|
2016-11-17 01:42:23 +01:00
|
|
|
}
|
2018-01-22 23:54:55 +01:00
|
|
|
results = append(results, result)
|
2016-11-17 01:42:23 +01:00
|
|
|
|
|
|
|
// Proceed to heal bucket metadata.
|
2018-01-22 23:54:55 +01:00
|
|
|
metaResults, err := healBucketMetadata(xl, bucket, dryRun)
|
|
|
|
results = append(results, metaResults...)
|
|
|
|
return results, err
|
2016-11-17 01:42:23 +01:00
|
|
|
}
|
2016-10-17 11:10:23 +02:00
|
|
|
|
2016-11-20 02:37:57 +01:00
|
|
|
// Heal bucket - create buckets on disks where it does not exist.
|
2018-04-06 00:04:40 +02:00
|
|
|
func healBucket(ctx context.Context, storageDisks []StorageAPI, bucket string, writeQuorum int,
|
2018-01-22 23:54:55 +01:00
|
|
|
dryRun bool) (res madmin.HealResultItem, err error) {
|
2016-10-17 11:10:23 +02:00
|
|
|
|
|
|
|
// Initialize sync waitgroup.
|
|
|
|
var wg = &sync.WaitGroup{}
|
|
|
|
|
|
|
|
// Initialize list of errors.
|
2016-11-17 01:42:23 +01:00
|
|
|
var dErrs = make([]error, len(storageDisks))
|
2016-10-17 11:10:23 +02:00
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// Disk states slices
|
|
|
|
beforeState := make([]string, len(storageDisks))
|
|
|
|
afterState := make([]string, len(storageDisks))
|
|
|
|
|
2016-10-17 11:10:23 +02:00
|
|
|
// Make a volume entry on all underlying storage disks.
|
2016-11-17 01:42:23 +01:00
|
|
|
for index, disk := range storageDisks {
|
2016-10-17 11:10:23 +02:00
|
|
|
if disk == nil {
|
2018-04-06 00:04:40 +02:00
|
|
|
dErrs[index] = errDiskNotFound
|
2018-01-22 23:54:55 +01:00
|
|
|
beforeState[index] = madmin.DriveStateOffline
|
|
|
|
afterState[index] = madmin.DriveStateOffline
|
2016-10-17 11:10:23 +02:00
|
|
|
continue
|
2016-05-26 01:42:31 +02:00
|
|
|
}
|
2016-10-17 11:10:23 +02:00
|
|
|
wg.Add(1)
|
2018-01-22 23:54:55 +01:00
|
|
|
|
2016-10-17 11:10:23 +02:00
|
|
|
// Make a volume inside a go-routine.
|
|
|
|
go func(index int, disk StorageAPI) {
|
|
|
|
defer wg.Done()
|
2018-03-16 23:09:31 +01:00
|
|
|
if _, serr := disk.StatVol(bucket); serr != nil {
|
2018-04-10 18:36:37 +02:00
|
|
|
if serr == errDiskNotFound {
|
2018-02-16 02:45:57 +01:00
|
|
|
beforeState[index] = madmin.DriveStateOffline
|
|
|
|
afterState[index] = madmin.DriveStateOffline
|
2018-03-16 23:09:31 +01:00
|
|
|
dErrs[index] = serr
|
2018-02-16 02:45:57 +01:00
|
|
|
return
|
|
|
|
}
|
2018-04-10 18:36:37 +02:00
|
|
|
if serr != errVolumeNotFound {
|
2018-01-22 23:54:55 +01:00
|
|
|
beforeState[index] = madmin.DriveStateCorrupt
|
|
|
|
afterState[index] = madmin.DriveStateCorrupt
|
2018-03-16 23:09:31 +01:00
|
|
|
dErrs[index] = serr
|
2016-10-17 11:10:23 +02:00
|
|
|
return
|
|
|
|
}
|
2018-01-22 23:54:55 +01:00
|
|
|
|
|
|
|
beforeState[index] = madmin.DriveStateMissing
|
|
|
|
afterState[index] = madmin.DriveStateMissing
|
|
|
|
|
|
|
|
// mutate only if not a dry-run
|
|
|
|
if dryRun {
|
|
|
|
return
|
|
|
|
}
|
|
|
|
|
|
|
|
makeErr := disk.MakeVol(bucket)
|
2018-02-16 02:45:57 +01:00
|
|
|
dErrs[index] = makeErr
|
2018-01-22 23:54:55 +01:00
|
|
|
if makeErr == nil {
|
|
|
|
afterState[index] = madmin.DriveStateOk
|
2016-10-17 11:10:23 +02:00
|
|
|
}
|
2018-02-16 02:45:57 +01:00
|
|
|
return
|
2016-10-17 11:10:23 +02:00
|
|
|
}
|
2018-02-16 02:45:57 +01:00
|
|
|
beforeState[index] = madmin.DriveStateOk
|
|
|
|
afterState[index] = madmin.DriveStateOk
|
2016-10-17 11:10:23 +02:00
|
|
|
}(index, disk)
|
2016-05-26 01:42:31 +02:00
|
|
|
}
|
2016-07-13 00:20:31 +02:00
|
|
|
|
2016-10-17 11:10:23 +02:00
|
|
|
// Wait for all make vol to finish.
|
|
|
|
wg.Wait()
|
2016-07-13 00:20:31 +02:00
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// Initialize heal result info
|
|
|
|
res = madmin.HealResultItem{
|
|
|
|
Type: madmin.HealItemBucket,
|
|
|
|
Bucket: bucket,
|
|
|
|
DiskCount: len(storageDisks),
|
|
|
|
}
|
|
|
|
for i, before := range beforeState {
|
2018-02-16 02:45:57 +01:00
|
|
|
if storageDisks[i] == nil {
|
|
|
|
res.Before.Drives = append(res.Before.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: "",
|
|
|
|
State: before,
|
|
|
|
})
|
|
|
|
res.After.Drives = append(res.After.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: "",
|
|
|
|
State: afterState[i],
|
|
|
|
})
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
drive := storageDisks[i].String()
|
|
|
|
res.Before.Drives = append(res.Before.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: drive,
|
|
|
|
State: before,
|
|
|
|
})
|
|
|
|
res.After.Drives = append(res.After.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: drive,
|
|
|
|
State: afterState[i],
|
|
|
|
})
|
2018-01-22 23:54:55 +01:00
|
|
|
}
|
|
|
|
|
2018-04-06 00:04:40 +02:00
|
|
|
reducedErr := reduceWriteQuorumErrs(ctx, dErrs, bucketOpIgnoredErrs, writeQuorum)
|
2018-04-10 18:36:37 +02:00
|
|
|
if reducedErr == errXLWriteQuorum {
|
2016-10-17 11:10:23 +02:00
|
|
|
// Purge successfully created buckets if we don't have writeQuorum.
|
2016-11-17 01:42:23 +01:00
|
|
|
undoMakeBucket(storageDisks, bucket)
|
2016-07-13 00:20:31 +02:00
|
|
|
}
|
2018-01-22 23:54:55 +01:00
|
|
|
return res, reducedErr
|
2016-05-26 01:42:31 +02:00
|
|
|
}
|
|
|
|
|
2016-11-17 01:42:23 +01:00
|
|
|
// Heals all the metadata associated for a given bucket, this function
|
|
|
|
// heals `policy.json`, `notification.xml` and `listeners.json`.
|
2018-01-22 23:54:55 +01:00
|
|
|
func healBucketMetadata(xl xlObjects, bucket string, dryRun bool) (
|
|
|
|
results []madmin.HealResultItem, err error) {
|
|
|
|
|
2016-11-17 01:42:23 +01:00
|
|
|
healBucketMetaFn := func(metaPath string) error {
|
2018-04-06 00:04:40 +02:00
|
|
|
reqInfo := &logger.ReqInfo{BucketName: bucket}
|
|
|
|
ctx := logger.SetReqInfo(context.Background(), reqInfo)
|
|
|
|
result, healErr := xl.HealObject(ctx, minioMetaBucket, metaPath, dryRun)
|
2018-10-01 18:54:26 +02:00
|
|
|
// If object is not found, skip the file.
|
|
|
|
if isErrObjectNotFound(healErr) {
|
|
|
|
return nil
|
|
|
|
}
|
2018-01-22 23:54:55 +01:00
|
|
|
if healErr != nil {
|
|
|
|
return healErr
|
2017-08-31 20:29:22 +02:00
|
|
|
}
|
2018-01-22 23:54:55 +01:00
|
|
|
result.Type = madmin.HealItemBucketMetadata
|
|
|
|
results = append(results, result)
|
2016-11-17 01:42:23 +01:00
|
|
|
return nil
|
2016-10-17 11:10:23 +02:00
|
|
|
}
|
2016-06-17 20:57:51 +02:00
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// Heal `policy.json` for missing entries, ignores if
|
|
|
|
// `policy.json` is not found.
|
2017-01-17 02:05:00 +01:00
|
|
|
policyPath := pathJoin(bucketConfigPrefix, bucket, bucketPolicyConfig)
|
2018-01-22 23:54:55 +01:00
|
|
|
err = healBucketMetaFn(policyPath)
|
|
|
|
if err != nil {
|
|
|
|
return results, err
|
2016-10-17 11:10:23 +02:00
|
|
|
}
|
2016-05-26 01:42:31 +02:00
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// Heal `notification.xml` for missing entries, ignores if
|
|
|
|
// `notification.xml` is not found.
|
|
|
|
nConfigPath := path.Join(bucketConfigPrefix, bucket,
|
|
|
|
bucketNotificationConfig)
|
|
|
|
err = healBucketMetaFn(nConfigPath)
|
|
|
|
if err != nil {
|
|
|
|
return results, err
|
2016-11-17 01:42:23 +01:00
|
|
|
}
|
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// Heal `listeners.json` for missing entries, ignores if
|
|
|
|
// `listeners.json` is not found.
|
2016-11-17 01:42:23 +01:00
|
|
|
lConfigPath := path.Join(bucketConfigPrefix, bucket, bucketListenerConfig)
|
2018-01-22 23:54:55 +01:00
|
|
|
err = healBucketMetaFn(lConfigPath)
|
|
|
|
return results, err
|
2016-11-17 01:42:23 +01:00
|
|
|
}
|
2016-10-17 11:10:23 +02:00
|
|
|
|
2017-01-19 18:34:18 +01:00
|
|
|
// listAllBuckets lists all buckets from all disks. It also
|
|
|
|
// returns the occurrence of each buckets in all disks
|
2018-01-22 23:54:55 +01:00
|
|
|
func listAllBuckets(storageDisks []StorageAPI) (buckets map[string]VolInfo,
|
|
|
|
bucketsOcc map[string]int, err error) {
|
|
|
|
|
2017-01-19 18:34:18 +01:00
|
|
|
buckets = make(map[string]VolInfo)
|
|
|
|
bucketsOcc = make(map[string]int)
|
2016-11-17 01:42:23 +01:00
|
|
|
for _, disk := range storageDisks {
|
|
|
|
if disk == nil {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
var volsInfo []VolInfo
|
|
|
|
volsInfo, err = disk.ListVols()
|
2018-01-22 23:54:55 +01:00
|
|
|
if err != nil {
|
2018-04-10 18:36:37 +02:00
|
|
|
if IsErrIgnored(err, bucketMetadataOpIgnoredErrs...) {
|
2018-01-22 23:54:55 +01:00
|
|
|
continue
|
2016-11-17 01:42:23 +01:00
|
|
|
}
|
2018-02-14 02:03:50 +01:00
|
|
|
return nil, nil, err
|
2017-01-19 18:34:18 +01:00
|
|
|
}
|
2018-01-22 23:54:55 +01:00
|
|
|
for _, volInfo := range volsInfo {
|
|
|
|
// StorageAPI can send volume names which are
|
|
|
|
// incompatible with buckets - these are
|
|
|
|
// skipped, like the meta-bucket.
|
|
|
|
if !IsValidBucketName(volInfo.Name) ||
|
|
|
|
isMinioMetaBucketName(volInfo.Name) {
|
|
|
|
continue
|
2017-01-19 18:34:18 +01:00
|
|
|
}
|
2018-01-22 23:54:55 +01:00
|
|
|
// Increase counter per bucket name
|
|
|
|
bucketsOcc[volInfo.Name]++
|
|
|
|
// Save volume info under bucket name
|
|
|
|
buckets[volInfo.Name] = volInfo
|
2017-01-19 18:34:18 +01:00
|
|
|
}
|
|
|
|
}
|
2018-02-14 02:03:50 +01:00
|
|
|
return buckets, bucketsOcc, nil
|
2017-01-19 18:34:18 +01:00
|
|
|
}
|
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// Heals an object by re-writing corrupt/missing erasure blocks.
|
2018-04-06 00:04:40 +02:00
|
|
|
func healObject(ctx context.Context, storageDisks []StorageAPI, bucket string, object string,
|
2018-01-22 23:54:55 +01:00
|
|
|
quorum int, dryRun bool) (result madmin.HealResultItem, err error) {
|
2018-04-06 00:04:40 +02:00
|
|
|
partsMetadata, errs := readAllXLMetadata(ctx, storageDisks, bucket, object)
|
2018-01-22 23:54:55 +01:00
|
|
|
|
2018-04-25 20:56:39 +02:00
|
|
|
errCount := 0
|
|
|
|
for _, err := range errs {
|
|
|
|
if err != nil {
|
|
|
|
errCount++
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if errCount == len(errs) {
|
|
|
|
// Only if we get errors from all the disks we return error. Else we need to
|
|
|
|
// continue to return filled madmin.HealResultItem struct which includes info
|
|
|
|
// on what disks the file is available etc.
|
|
|
|
if reducedErr := reduceReadQuorumErrs(ctx, errs, nil, quorum); reducedErr != nil {
|
2018-10-03 02:13:51 +02:00
|
|
|
return defaultHealResult(storageDisks, errs, bucket, object), toObjectErr(reducedErr, bucket, object)
|
2018-04-25 20:56:39 +02:00
|
|
|
}
|
2016-10-17 11:10:23 +02:00
|
|
|
}
|
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// List of disks having latest version of the object xl.json
|
|
|
|
// (by modtime).
|
2016-11-17 01:42:23 +01:00
|
|
|
latestDisks, modTime := listOnlineDisks(storageDisks, partsMetadata, errs)
|
2017-03-04 23:53:28 +01:00
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// List of disks having all parts as per latest xl.json.
|
2018-10-03 02:13:51 +02:00
|
|
|
availableDisks, dataErrs := disksWithAllParts(ctx, latestDisks, partsMetadata, errs, bucket, object)
|
2017-03-04 23:53:28 +01:00
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// Initialize heal result object
|
|
|
|
result = madmin.HealResultItem{
|
|
|
|
Type: madmin.HealItemObject,
|
|
|
|
Bucket: bucket,
|
|
|
|
Object: object,
|
|
|
|
DiskCount: len(storageDisks),
|
|
|
|
|
|
|
|
// Initialize object size to -1, so we can detect if we are
|
|
|
|
// unable to reliably find the object size.
|
|
|
|
ObjectSize: -1,
|
2017-04-14 19:28:35 +02:00
|
|
|
}
|
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// Loop to find number of disks with valid data, per-drive
|
|
|
|
// data state and a list of outdated disks on which data needs
|
|
|
|
// to be healed.
|
|
|
|
outDatedDisks := make([]StorageAPI, len(storageDisks))
|
2017-04-14 19:28:35 +02:00
|
|
|
numAvailableDisks := 0
|
2018-01-22 23:54:55 +01:00
|
|
|
disksToHealCount := 0
|
|
|
|
for i, v := range availableDisks {
|
|
|
|
driveState := ""
|
|
|
|
switch {
|
|
|
|
case v != nil:
|
|
|
|
driveState = madmin.DriveStateOk
|
2017-03-04 23:53:28 +01:00
|
|
|
numAvailableDisks++
|
2018-01-22 23:54:55 +01:00
|
|
|
// If data is sane on any one disk, we can
|
|
|
|
// extract the correct object size.
|
|
|
|
result.ObjectSize = partsMetadata[i].Stat.Size
|
|
|
|
result.ParityBlocks = partsMetadata[i].Erasure.ParityBlocks
|
|
|
|
result.DataBlocks = partsMetadata[i].Erasure.DataBlocks
|
2018-10-03 02:13:51 +02:00
|
|
|
case errs[i] == errDiskNotFound, dataErrs[i] == errDiskNotFound:
|
2018-01-22 23:54:55 +01:00
|
|
|
driveState = madmin.DriveStateOffline
|
2018-04-10 18:36:37 +02:00
|
|
|
case errs[i] == errFileNotFound, errs[i] == errVolumeNotFound:
|
2018-01-22 23:54:55 +01:00
|
|
|
fallthrough
|
2018-04-10 18:36:37 +02:00
|
|
|
case dataErrs[i] == errFileNotFound, dataErrs[i] == errVolumeNotFound:
|
2018-01-22 23:54:55 +01:00
|
|
|
driveState = madmin.DriveStateMissing
|
|
|
|
default:
|
|
|
|
// all remaining cases imply corrupt data/metadata
|
|
|
|
driveState = madmin.DriveStateCorrupt
|
|
|
|
}
|
|
|
|
|
|
|
|
// an online disk without valid data/metadata is
|
|
|
|
// outdated and can be healed.
|
|
|
|
if errs[i] != errDiskNotFound && v == nil {
|
|
|
|
outDatedDisks[i] = storageDisks[i]
|
|
|
|
disksToHealCount++
|
2017-03-04 23:53:28 +01:00
|
|
|
}
|
2018-03-28 03:11:39 +02:00
|
|
|
var drive string
|
2018-02-16 02:45:57 +01:00
|
|
|
if v == nil {
|
2018-03-28 03:11:39 +02:00
|
|
|
if errs[i] != errDiskNotFound {
|
|
|
|
drive = outDatedDisks[i].String()
|
|
|
|
}
|
2018-02-16 02:45:57 +01:00
|
|
|
result.Before.Drives = append(result.Before.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
2018-03-28 03:11:39 +02:00
|
|
|
Endpoint: drive,
|
2018-02-16 02:45:57 +01:00
|
|
|
State: driveState,
|
|
|
|
})
|
|
|
|
result.After.Drives = append(result.After.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
2018-03-28 03:11:39 +02:00
|
|
|
Endpoint: drive,
|
2018-02-16 02:45:57 +01:00
|
|
|
State: driveState,
|
|
|
|
})
|
|
|
|
continue
|
|
|
|
}
|
2018-03-28 03:11:39 +02:00
|
|
|
drive = v.String()
|
2018-02-16 02:45:57 +01:00
|
|
|
result.Before.Drives = append(result.Before.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: drive,
|
|
|
|
State: driveState,
|
|
|
|
})
|
|
|
|
result.After.Drives = append(result.After.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: drive,
|
|
|
|
State: driveState,
|
|
|
|
})
|
2017-09-29 00:57:19 +02:00
|
|
|
}
|
|
|
|
|
2017-03-04 23:53:28 +01:00
|
|
|
// If less than read quorum number of disks have all the parts
|
|
|
|
// of the data, we can't reconstruct the erasure-coded data.
|
|
|
|
if numAvailableDisks < quorum {
|
2018-10-03 02:13:51 +02:00
|
|
|
// Default to most common configuration for erasure
|
|
|
|
// blocks upon returning quorum error.
|
|
|
|
result.ParityBlocks = len(storageDisks) / 2
|
|
|
|
result.DataBlocks = len(storageDisks) / 2
|
2018-01-22 23:54:55 +01:00
|
|
|
return result, toObjectErr(errXLReadQuorum, bucket, object)
|
2017-03-04 23:53:28 +01:00
|
|
|
}
|
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
if disksToHealCount == 0 {
|
|
|
|
// Nothing to heal!
|
|
|
|
return result, nil
|
|
|
|
}
|
2017-03-04 23:53:28 +01:00
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// After this point, only have to repair data on disk - so
|
|
|
|
// return if it is a dry-run
|
|
|
|
if dryRun {
|
|
|
|
return result, nil
|
2017-04-01 02:55:15 +02:00
|
|
|
}
|
|
|
|
|
2017-03-04 23:53:28 +01:00
|
|
|
// Latest xlMetaV1 for reference. If a valid metadata is not
|
|
|
|
// present, it is as good as object not found.
|
2018-08-17 23:42:04 +02:00
|
|
|
latestMeta, pErr := pickValidXLMeta(ctx, partsMetadata, modTime, quorum)
|
2016-11-21 05:56:44 +01:00
|
|
|
if pErr != nil {
|
2018-01-22 23:54:55 +01:00
|
|
|
return result, toObjectErr(pErr, bucket, object)
|
2016-11-21 05:56:44 +01:00
|
|
|
}
|
2016-10-17 11:10:23 +02:00
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// Clear data files of the object on outdated disks
|
|
|
|
for _, disk := range outDatedDisks {
|
2017-09-29 00:57:19 +02:00
|
|
|
// Before healing outdated disks, we need to remove
|
|
|
|
// xl.json and part files from "bucket/object/" so
|
|
|
|
// that rename(minioMetaBucket, "tmp/tmpuuid/",
|
|
|
|
// "bucket", "object/") succeeds.
|
2016-10-17 11:10:23 +02:00
|
|
|
if disk == nil {
|
|
|
|
// Not an outdated disk.
|
2016-08-17 20:36:33 +02:00
|
|
|
continue
|
|
|
|
}
|
2017-01-30 09:45:56 +01:00
|
|
|
|
2018-04-10 18:36:37 +02:00
|
|
|
// List and delete the object directory,
|
2018-05-09 04:08:21 +02:00
|
|
|
files, derr := disk.ListDir(bucket, object, -1)
|
2018-01-22 23:54:55 +01:00
|
|
|
if derr == nil {
|
2017-09-29 00:57:19 +02:00
|
|
|
for _, entry := range files {
|
|
|
|
_ = disk.DeleteFile(bucket,
|
|
|
|
pathJoin(object, entry))
|
2016-10-17 11:10:23 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Reorder so that we have data disks first and parity disks next.
|
2017-02-24 18:20:40 +01:00
|
|
|
latestDisks = shuffleDisks(latestDisks, latestMeta.Erasure.Distribution)
|
|
|
|
outDatedDisks = shuffleDisks(outDatedDisks, latestMeta.Erasure.Distribution)
|
|
|
|
partsMetadata = shufflePartsMetadata(partsMetadata, latestMeta.Erasure.Distribution)
|
2019-01-17 13:58:18 +01:00
|
|
|
for i := range outDatedDisks {
|
|
|
|
if outDatedDisks[i] == nil {
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
partsMetadata[i] = newXLMetaFromXLMeta(latestMeta)
|
|
|
|
}
|
2016-10-17 11:10:23 +02:00
|
|
|
|
2017-09-29 00:57:19 +02:00
|
|
|
// We write at temporary location and then rename to final location.
|
2016-11-23 01:52:37 +01:00
|
|
|
tmpID := mustGetUUID()
|
2016-10-17 11:10:23 +02:00
|
|
|
|
2017-09-29 00:57:19 +02:00
|
|
|
// Heal each part. erasureHealFile() will write the healed
|
|
|
|
// part to .minio/tmp/uuid/ which needs to be renamed later to
|
|
|
|
// the final location.
|
2018-08-24 08:35:37 +02:00
|
|
|
erasure, err := NewErasure(ctx, latestMeta.Erasure.DataBlocks,
|
2018-01-22 23:54:55 +01:00
|
|
|
latestMeta.Erasure.ParityBlocks, latestMeta.Erasure.BlockSize)
|
2017-08-15 03:08:42 +02:00
|
|
|
if err != nil {
|
2018-01-22 23:54:55 +01:00
|
|
|
return result, toObjectErr(err, bucket, object)
|
2017-08-15 03:08:42 +02:00
|
|
|
}
|
2018-08-07 00:14:08 +02:00
|
|
|
|
2019-01-17 13:58:18 +01:00
|
|
|
erasureInfo := latestMeta.Erasure
|
2016-10-17 11:10:23 +02:00
|
|
|
for partIndex := 0; partIndex < len(latestMeta.Parts); partIndex++ {
|
|
|
|
partName := latestMeta.Parts[partIndex].Name
|
|
|
|
partSize := latestMeta.Parts[partIndex].Size
|
2019-01-17 13:58:18 +01:00
|
|
|
partActualSize := latestMeta.Parts[partIndex].ActualSize
|
|
|
|
partNumber := latestMeta.Parts[partIndex].Number
|
|
|
|
tillOffset := erasure.ShardFileTillOffset(0, partSize, partSize)
|
|
|
|
readers := make([]io.ReaderAt, len(latestDisks))
|
2019-01-20 03:28:40 +01:00
|
|
|
checksumAlgo := erasureInfo.GetChecksumInfo(partName).Algorithm
|
2018-08-07 00:14:08 +02:00
|
|
|
for i, disk := range latestDisks {
|
|
|
|
if disk == OfflineDisk {
|
|
|
|
continue
|
|
|
|
}
|
2019-01-20 03:28:40 +01:00
|
|
|
checksumInfo := partsMetadata[i].Erasure.GetChecksumInfo(partName)
|
|
|
|
readers[i] = newBitrotReader(disk, bucket, pathJoin(object, partName), tillOffset, checksumAlgo, checksumInfo.Hash, erasure.ShardSize())
|
2018-08-07 00:14:08 +02:00
|
|
|
}
|
2019-01-17 13:58:18 +01:00
|
|
|
writers := make([]io.Writer, len(outDatedDisks))
|
2018-08-07 00:14:08 +02:00
|
|
|
for i, disk := range outDatedDisks {
|
|
|
|
if disk == OfflineDisk {
|
|
|
|
continue
|
2017-08-15 03:08:42 +02:00
|
|
|
}
|
2019-01-20 03:28:40 +01:00
|
|
|
writers[i] = newBitrotWriter(disk, minioMetaTmpBucket, pathJoin(tmpID, partName), tillOffset, checksumAlgo, erasure.ShardSize())
|
2017-08-15 03:08:42 +02:00
|
|
|
}
|
2019-01-17 13:58:18 +01:00
|
|
|
hErr := erasure.Heal(ctx, readers, writers, partSize)
|
|
|
|
closeBitrotReaders(readers)
|
|
|
|
closeBitrotWriters(writers)
|
2017-03-04 23:53:28 +01:00
|
|
|
if hErr != nil {
|
2018-01-22 23:54:55 +01:00
|
|
|
return result, toObjectErr(hErr, bucket, object)
|
2016-10-17 11:10:23 +02:00
|
|
|
}
|
2017-09-29 00:57:19 +02:00
|
|
|
// outDatedDisks that had write errors should not be
|
|
|
|
// written to for remaining parts, so we nil it out.
|
|
|
|
for i, disk := range outDatedDisks {
|
|
|
|
if disk == nil {
|
|
|
|
continue
|
2016-10-17 11:10:23 +02:00
|
|
|
}
|
2017-09-29 00:57:19 +02:00
|
|
|
// A non-nil stale disk which did not receive
|
|
|
|
// a healed part checksum had a write error.
|
2019-01-17 13:58:18 +01:00
|
|
|
if writers[i] == nil {
|
2017-09-29 00:57:19 +02:00
|
|
|
outDatedDisks[i] = nil
|
2018-01-22 23:54:55 +01:00
|
|
|
disksToHealCount--
|
2017-09-29 00:57:19 +02:00
|
|
|
continue
|
|
|
|
}
|
2019-01-17 13:58:18 +01:00
|
|
|
partsMetadata[i].AddObjectPart(partNumber, partName, "", partSize, partActualSize)
|
2019-01-20 03:28:40 +01:00
|
|
|
partsMetadata[i].Erasure.AddChecksumInfo(ChecksumInfo{partName, checksumAlgo, bitrotWriterSum(writers[i])})
|
2017-09-29 00:57:19 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
// If all disks are having errors, we give up.
|
2018-01-22 23:54:55 +01:00
|
|
|
if disksToHealCount == 0 {
|
|
|
|
return result, fmt.Errorf("all disks without up-to-date data had write errors")
|
2016-10-17 11:10:23 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Generate and write `xl.json` generated from other disks.
|
2018-10-03 02:13:51 +02:00
|
|
|
outDatedDisks, aErr := writeUniqueXLMetadata(ctx, outDatedDisks, minioMetaTmpBucket, tmpID,
|
2017-09-29 00:57:19 +02:00
|
|
|
partsMetadata, diskCount(outDatedDisks))
|
2017-03-04 23:53:28 +01:00
|
|
|
if aErr != nil {
|
2018-01-22 23:54:55 +01:00
|
|
|
return result, toObjectErr(aErr, bucket, object)
|
2016-08-17 20:36:33 +02:00
|
|
|
}
|
|
|
|
|
2016-10-17 11:10:23 +02:00
|
|
|
// Rename from tmp location to the actual location.
|
2018-03-28 03:11:39 +02:00
|
|
|
for _, disk := range outDatedDisks {
|
2016-10-17 11:10:23 +02:00
|
|
|
if disk == nil {
|
2016-08-31 20:42:57 +02:00
|
|
|
continue
|
2016-08-17 20:36:33 +02:00
|
|
|
}
|
2017-09-29 00:57:19 +02:00
|
|
|
|
2016-10-17 11:10:23 +02:00
|
|
|
// Attempt a rename now from healed data to final location.
|
2017-09-29 00:57:19 +02:00
|
|
|
aErr = disk.RenameFile(minioMetaTmpBucket, retainSlash(tmpID), bucket,
|
|
|
|
retainSlash(object))
|
2017-03-04 23:53:28 +01:00
|
|
|
if aErr != nil {
|
2018-04-06 00:04:40 +02:00
|
|
|
logger.LogIf(ctx, aErr)
|
|
|
|
return result, toObjectErr(aErr, bucket, object)
|
2016-08-17 20:36:33 +02:00
|
|
|
}
|
2018-01-22 23:54:55 +01:00
|
|
|
|
2018-03-28 03:11:39 +02:00
|
|
|
for i, v := range result.Before.Drives {
|
|
|
|
if v.Endpoint == disk.String() {
|
|
|
|
result.After.Drives[i].State = madmin.DriveStateOk
|
2018-02-16 02:45:57 +01:00
|
|
|
}
|
|
|
|
}
|
2016-08-17 20:36:33 +02:00
|
|
|
}
|
2018-01-22 23:54:55 +01:00
|
|
|
|
|
|
|
// Set the size of the object in the heal result
|
|
|
|
result.ObjectSize = latestMeta.Stat.Size
|
|
|
|
|
|
|
|
return result, nil
|
2016-08-17 20:36:33 +02:00
|
|
|
}
|
2016-11-17 01:42:23 +01:00
|
|
|
|
2018-05-11 01:53:42 +02:00
|
|
|
// healObjectDir - heals object directory specifically, this special call
|
|
|
|
// is needed since we do not have a special backend format for directories.
|
|
|
|
func (xl xlObjects) healObjectDir(ctx context.Context, bucket, object string, dryRun bool) (hr madmin.HealResultItem, err error) {
|
|
|
|
storageDisks := xl.getDisks()
|
|
|
|
|
|
|
|
// Initialize heal result object
|
|
|
|
hr = madmin.HealResultItem{
|
|
|
|
Type: madmin.HealItemObject,
|
|
|
|
Bucket: bucket,
|
|
|
|
Object: object,
|
|
|
|
DiskCount: len(storageDisks),
|
|
|
|
ParityBlocks: len(storageDisks) / 2,
|
|
|
|
DataBlocks: len(storageDisks) / 2,
|
|
|
|
ObjectSize: 0,
|
|
|
|
}
|
|
|
|
|
|
|
|
// Prepare object creation in all disks
|
|
|
|
for _, disk := range storageDisks {
|
|
|
|
if disk == nil {
|
|
|
|
hr.Before.Drives = append(hr.Before.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
State: madmin.DriveStateOffline,
|
|
|
|
})
|
|
|
|
hr.After.Drives = append(hr.After.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
State: madmin.DriveStateMissing,
|
|
|
|
})
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
|
|
|
|
drive := disk.String()
|
|
|
|
hr.Before.Drives = append(hr.Before.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: drive,
|
|
|
|
State: madmin.DriveStateMissing,
|
|
|
|
})
|
|
|
|
hr.After.Drives = append(hr.After.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: drive,
|
|
|
|
State: madmin.DriveStateMissing,
|
|
|
|
})
|
|
|
|
|
|
|
|
if !dryRun {
|
|
|
|
if err := disk.MakeVol(pathJoin(bucket, object)); err != nil && err != errVolumeExists {
|
|
|
|
return hr, toObjectErr(err, bucket, object)
|
|
|
|
}
|
|
|
|
|
|
|
|
for i, v := range hr.Before.Drives {
|
|
|
|
if v.Endpoint == drive {
|
|
|
|
hr.After.Drives[i].State = madmin.DriveStateOk
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return hr, nil
|
|
|
|
}
|
|
|
|
|
2018-10-03 02:13:51 +02:00
|
|
|
// Populates default heal result item entries with possible values when we are returning prematurely.
|
|
|
|
// This is to ensure that in any circumstance we are not returning empty arrays with wrong values.
|
|
|
|
func defaultHealResult(storageDisks []StorageAPI, errs []error, bucket, object string) madmin.HealResultItem {
|
|
|
|
// Initialize heal result object
|
|
|
|
result := madmin.HealResultItem{
|
|
|
|
Type: madmin.HealItemObject,
|
|
|
|
Bucket: bucket,
|
|
|
|
Object: object,
|
|
|
|
DiskCount: len(storageDisks),
|
|
|
|
|
|
|
|
// Initialize object size to -1, so we can detect if we are
|
|
|
|
// unable to reliably find the object size.
|
|
|
|
ObjectSize: -1,
|
|
|
|
}
|
|
|
|
|
|
|
|
for index, disk := range storageDisks {
|
|
|
|
if disk == nil {
|
|
|
|
result.Before.Drives = append(result.Before.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
State: madmin.DriveStateOffline,
|
|
|
|
})
|
|
|
|
result.After.Drives = append(result.After.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
State: madmin.DriveStateOffline,
|
|
|
|
})
|
|
|
|
continue
|
|
|
|
}
|
|
|
|
drive := disk.String()
|
|
|
|
driveState := madmin.DriveStateCorrupt
|
|
|
|
switch errs[index] {
|
|
|
|
case errFileNotFound, errVolumeNotFound:
|
|
|
|
driveState = madmin.DriveStateMissing
|
|
|
|
}
|
|
|
|
result.Before.Drives = append(result.Before.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: drive,
|
|
|
|
State: driveState,
|
|
|
|
})
|
|
|
|
result.After.Drives = append(result.After.Drives, madmin.HealDriveInfo{
|
|
|
|
UUID: "",
|
|
|
|
Endpoint: drive,
|
|
|
|
State: driveState,
|
|
|
|
})
|
|
|
|
}
|
|
|
|
|
|
|
|
// Default to most common configuration for erasure blocks.
|
|
|
|
result.ParityBlocks = len(storageDisks) / 2
|
|
|
|
result.DataBlocks = len(storageDisks) / 2
|
|
|
|
|
|
|
|
return result
|
|
|
|
}
|
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// HealObject - heal the given object.
|
|
|
|
//
|
2016-11-17 01:42:23 +01:00
|
|
|
// FIXME: If an object object was deleted and one disk was down,
|
|
|
|
// and later the disk comes back up again, heal on the object
|
|
|
|
// should delete it.
|
2018-03-14 20:01:47 +01:00
|
|
|
func (xl xlObjects) HealObject(ctx context.Context, bucket, object string, dryRun bool) (hr madmin.HealResultItem, err error) {
|
2018-08-21 01:58:47 +02:00
|
|
|
// Create context that also contains information about the object and bucket.
|
|
|
|
// The top level handler might not have this information.
|
|
|
|
reqInfo := logger.GetReqInfo(ctx)
|
|
|
|
var newReqInfo *logger.ReqInfo
|
|
|
|
if reqInfo != nil {
|
2018-11-19 23:47:03 +01:00
|
|
|
newReqInfo = logger.NewReqInfo(reqInfo.RemoteHost, reqInfo.UserAgent, reqInfo.DeploymentID, reqInfo.RequestID, reqInfo.API, bucket, object)
|
2018-08-21 01:58:47 +02:00
|
|
|
} else {
|
2018-11-19 23:47:03 +01:00
|
|
|
newReqInfo = logger.NewReqInfo("", "", globalDeploymentID, "", "Heal", bucket, object)
|
2018-08-21 01:58:47 +02:00
|
|
|
}
|
|
|
|
healCtx := logger.SetReqInfo(context.Background(), newReqInfo)
|
|
|
|
|
2018-05-11 01:53:42 +02:00
|
|
|
// Healing directories handle it separately.
|
|
|
|
if hasSuffix(object, slashSeparator) {
|
2018-08-21 01:58:47 +02:00
|
|
|
return xl.healObjectDir(healCtx, bucket, object, dryRun)
|
2018-05-11 01:53:42 +02:00
|
|
|
}
|
|
|
|
|
2018-10-03 02:13:51 +02:00
|
|
|
storageDisks := xl.getDisks()
|
|
|
|
|
2018-01-22 23:54:55 +01:00
|
|
|
// FIXME: Metadata is read again in the healObject() call below.
|
2017-12-22 12:28:13 +01:00
|
|
|
// Read metadata files from all the disks
|
2018-10-03 02:13:51 +02:00
|
|
|
partsMetadata, errs := readAllXLMetadata(healCtx, storageDisks, bucket, object)
|
2017-12-22 12:28:13 +01:00
|
|
|
|
2018-08-21 01:58:47 +02:00
|
|
|
latestXLMeta, err := getLatestXLMeta(healCtx, partsMetadata, errs)
|
2018-07-31 09:23:29 +02:00
|
|
|
if err != nil {
|
2018-10-03 02:13:51 +02:00
|
|
|
return defaultHealResult(storageDisks, errs, bucket, object), toObjectErr(err, bucket, object)
|
2018-07-31 09:23:29 +02:00
|
|
|
}
|
2017-12-22 12:28:13 +01:00
|
|
|
|
2016-11-17 01:42:23 +01:00
|
|
|
// Lock the object before healing.
|
2018-01-13 05:34:52 +01:00
|
|
|
objectLock := xl.nsMutex.NewNSLock(bucket, object)
|
2018-01-22 23:54:55 +01:00
|
|
|
if lerr := objectLock.GetRLock(globalHealingTimeout); lerr != nil {
|
2018-10-03 02:13:51 +02:00
|
|
|
return defaultHealResult(storageDisks, errs, bucket, object), lerr
|
2017-08-31 20:29:22 +02:00
|
|
|
}
|
2016-11-17 01:42:23 +01:00
|
|
|
defer objectLock.RUnlock()
|
|
|
|
|
|
|
|
// Heal the object.
|
2018-08-21 01:58:47 +02:00
|
|
|
return healObject(healCtx, xl.getDisks(), bucket, object, latestXLMeta.Erasure.DataBlocks, dryRun)
|
2016-11-17 01:42:23 +01:00
|
|
|
}
|