restic/internal/archiver/archiver.go
Alexander Neumann f99c95c766 archiver: Fix intermediate index upload
A user discovered[1] that when the backup finishes during the upload of
an intermediate index, the upload is cancelled and the index never fully
saved, but the snapshot is saved and the backup finalizes without an
error. This lead to a situation where a snapshot references data that is
contained in the repo, but not referenced in any index, leading to
strange error messages.

This commit uses a dedicated context to signal the intermediate index
uploading routine to terminate after the last index has been uploaded.
This way, an upload running when the backup finishes is completed before
the routine terminates and the snapshot is saved.

[1] https://forum.restic.net/t/error-loading-tree-check-prune-and-forget-gives-error-b2-backend/406
2018-01-26 22:01:07 +01:00

870 lines
20 KiB
Go

package archiver
import (
"context"
"encoding/json"
"fmt"
"io"
"os"
"path/filepath"
"sort"
"sync"
"time"
"github.com/restic/restic/internal/errors"
"github.com/restic/restic/internal/restic"
"github.com/restic/restic/internal/walk"
"github.com/restic/restic/internal/debug"
"github.com/restic/restic/internal/fs"
"github.com/restic/restic/internal/pipe"
"github.com/restic/chunker"
)
const (
maxConcurrentBlobs = 32
maxConcurrency = 10
)
var archiverPrintWarnings = func(path string, fi os.FileInfo, err error) {
fmt.Fprintf(os.Stderr, "warning for %v: %v", path, err)
}
var archiverAllowAllFiles = func(string, os.FileInfo) bool { return true }
// Archiver is used to backup a set of directories.
type Archiver struct {
repo restic.Repository
knownBlobs struct {
restic.IDSet
sync.Mutex
}
blobToken chan struct{}
Warn func(dir string, fi os.FileInfo, err error)
SelectFilter pipe.SelectFunc
Excludes []string
WithAccessTime bool
}
// New returns a new archiver.
func New(repo restic.Repository) *Archiver {
arch := &Archiver{
repo: repo,
blobToken: make(chan struct{}, maxConcurrentBlobs),
knownBlobs: struct {
restic.IDSet
sync.Mutex
}{
IDSet: restic.NewIDSet(),
},
}
for i := 0; i < maxConcurrentBlobs; i++ {
arch.blobToken <- struct{}{}
}
arch.Warn = archiverPrintWarnings
arch.SelectFilter = archiverAllowAllFiles
return arch
}
// isKnownBlob returns true iff the blob is not yet in the list of known blobs.
// When the blob is not known, false is returned and the blob is added to the
// list. This means that the caller false is returned to is responsible to save
// the blob to the backend.
func (arch *Archiver) isKnownBlob(id restic.ID, t restic.BlobType) bool {
arch.knownBlobs.Lock()
defer arch.knownBlobs.Unlock()
if arch.knownBlobs.Has(id) {
return true
}
arch.knownBlobs.Insert(id)
if arch.repo.Index().Has(id, t) {
return true
}
return false
}
// Save stores a blob read from rd in the repository.
func (arch *Archiver) Save(ctx context.Context, t restic.BlobType, data []byte, id restic.ID) error {
debug.Log("Save(%v, %v)\n", t, id)
if arch.isKnownBlob(id, restic.DataBlob) {
debug.Log("blob %v is known\n", id)
return nil
}
_, err := arch.repo.SaveBlob(ctx, t, data, id)
if err != nil {
debug.Log("Save(%v, %v): error %v\n", t, id, err)
return err
}
debug.Log("Save(%v, %v): new blob\n", t, id)
return nil
}
// SaveTreeJSON stores a tree in the repository.
func (arch *Archiver) SaveTreeJSON(ctx context.Context, tree *restic.Tree) (restic.ID, error) {
data, err := json.Marshal(tree)
if err != nil {
return restic.ID{}, errors.Wrap(err, "Marshal")
}
data = append(data, '\n')
// check if tree has been saved before
id := restic.Hash(data)
if arch.isKnownBlob(id, restic.TreeBlob) {
return id, nil
}
return arch.repo.SaveBlob(ctx, restic.TreeBlob, data, id)
}
func (arch *Archiver) reloadFileIfChanged(node *restic.Node, file fs.File) (*restic.Node, error) {
if !arch.WithAccessTime {
node.AccessTime = node.ModTime
}
fi, err := file.Stat()
if err != nil {
return nil, errors.Wrap(err, "restic.Stat")
}
if fi.ModTime().Equal(node.ModTime) {
return node, nil
}
arch.Warn(node.Path, fi, errors.New("file has changed"))
node, err = restic.NodeFromFileInfo(node.Path, fi)
if err != nil {
debug.Log("restic.NodeFromFileInfo returned error for %v: %v", node.Path, err)
arch.Warn(node.Path, fi, err)
}
if !arch.WithAccessTime {
node.AccessTime = node.ModTime
}
return node, nil
}
type saveResult struct {
id restic.ID
bytes uint64
}
func (arch *Archiver) saveChunk(ctx context.Context, chunk chunker.Chunk, p *restic.Progress, token struct{}, file fs.File, resultChannel chan<- saveResult) {
defer freeBuf(chunk.Data)
id := restic.Hash(chunk.Data)
err := arch.Save(ctx, restic.DataBlob, chunk.Data, id)
// TODO handle error
if err != nil {
debug.Log("Save(%v) failed: %v", id, err)
fmt.Printf("\nerror while saving data to the repo: %+v\n", err)
panic(err)
}
p.Report(restic.Stat{Bytes: uint64(chunk.Length)})
arch.blobToken <- token
resultChannel <- saveResult{id: id, bytes: uint64(chunk.Length)}
}
func waitForResults(resultChannels [](<-chan saveResult)) ([]saveResult, error) {
results := []saveResult{}
for _, ch := range resultChannels {
results = append(results, <-ch)
}
if len(results) != len(resultChannels) {
return nil, errors.Errorf("chunker returned %v chunks, but only %v blobs saved", len(resultChannels), len(results))
}
return results, nil
}
func updateNodeContent(node *restic.Node, results []saveResult) error {
debug.Log("checking size for file %s", node.Path)
var bytes uint64
node.Content = make([]restic.ID, len(results))
for i, b := range results {
node.Content[i] = b.id
bytes += b.bytes
debug.Log(" adding blob %s, %d bytes", b.id, b.bytes)
}
if bytes != node.Size {
fmt.Fprintf(os.Stderr, "warning for %v: expected %d bytes, saved %d bytes\n", node.Path, node.Size, bytes)
}
debug.Log("SaveFile(%q): %v blobs\n", node.Path, len(results))
return nil
}
// SaveFile stores the content of the file on the backend as a Blob by calling
// Save for each chunk.
func (arch *Archiver) SaveFile(ctx context.Context, p *restic.Progress, node *restic.Node) (*restic.Node, error) {
file, err := fs.Open(node.Path)
if err != nil {
return node, errors.Wrap(err, "Open")
}
defer file.Close()
debug.RunHook("archiver.SaveFile", node.Path)
node, err = arch.reloadFileIfChanged(node, file)
if err != nil {
return node, err
}
chnker := chunker.New(file, arch.repo.Config().ChunkerPolynomial)
resultChannels := [](<-chan saveResult){}
for {
chunk, err := chnker.Next(getBuf())
if errors.Cause(err) == io.EOF {
break
}
if err != nil {
return node, errors.Wrap(err, "chunker.Next")
}
resCh := make(chan saveResult, 1)
go arch.saveChunk(ctx, chunk, p, <-arch.blobToken, file, resCh)
resultChannels = append(resultChannels, resCh)
}
results, err := waitForResults(resultChannels)
if err != nil {
return node, err
}
err = updateNodeContent(node, results)
return node, err
}
func (arch *Archiver) fileWorker(ctx context.Context, wg *sync.WaitGroup, p *restic.Progress, entCh <-chan pipe.Entry) {
defer func() {
debug.Log("done")
wg.Done()
}()
for {
select {
case e, ok := <-entCh:
if !ok {
// channel is closed
return
}
debug.Log("got job %v", e)
// check for errors
if e.Error() != nil {
debug.Log("job %v has errors: %v", e.Path(), e.Error())
// TODO: integrate error reporting
fmt.Fprintf(os.Stderr, "error for %v: %v\n", e.Path(), e.Error())
// ignore this file
e.Result() <- nil
p.Report(restic.Stat{Errors: 1})
continue
}
node, err := restic.NodeFromFileInfo(e.Fullpath(), e.Info())
if err != nil {
debug.Log("restic.NodeFromFileInfo returned error for %v: %v", node.Path, err)
arch.Warn(e.Fullpath(), e.Info(), err)
}
if !arch.WithAccessTime {
node.AccessTime = node.ModTime
}
// try to use old node, if present
if e.Node != nil {
debug.Log(" %v use old data", e.Path())
oldNode := e.Node.(*restic.Node)
// check if all content is still available in the repository
contentMissing := false
for _, blob := range oldNode.Content {
if !arch.repo.Index().Has(blob, restic.DataBlob) {
debug.Log(" %v not using old data, %v is missing", e.Path(), blob)
contentMissing = true
break
}
}
if !contentMissing {
node.Content = oldNode.Content
debug.Log(" %v content is complete", e.Path())
}
} else {
debug.Log(" %v no old data", e.Path())
}
// otherwise read file normally
if node.Type == "file" && len(node.Content) == 0 {
debug.Log(" read and save %v", e.Path())
node, err = arch.SaveFile(ctx, p, node)
if err != nil {
fmt.Fprintf(os.Stderr, "error for %v: %v\n", node.Path, err)
arch.Warn(e.Path(), nil, err)
// ignore this file
e.Result() <- nil
p.Report(restic.Stat{Errors: 1})
continue
}
} else {
// report old data size
p.Report(restic.Stat{Bytes: node.Size})
}
debug.Log(" processed %v, %d blobs", e.Path(), len(node.Content))
e.Result() <- node
p.Report(restic.Stat{Files: 1})
case <-ctx.Done():
// pipeline was cancelled
return
}
}
}
func (arch *Archiver) dirWorker(ctx context.Context, wg *sync.WaitGroup, p *restic.Progress, dirCh <-chan pipe.Dir) {
debug.Log("start")
defer func() {
debug.Log("done")
wg.Done()
}()
for {
select {
case dir, ok := <-dirCh:
if !ok {
// channel is closed
return
}
debug.Log("save dir %v (%d entries), error %v\n", dir.Path(), len(dir.Entries), dir.Error())
// ignore dir nodes with errors
if dir.Error() != nil {
fmt.Fprintf(os.Stderr, "error walking dir %v: %v\n", dir.Path(), dir.Error())
dir.Result() <- nil
p.Report(restic.Stat{Errors: 1})
continue
}
tree := restic.NewTree()
// wait for all content
for _, ch := range dir.Entries {
debug.Log("receiving result from %v", ch)
res := <-ch
// if we get a nil pointer here, an error has happened while
// processing this entry. Ignore it for now.
if res == nil {
debug.Log("got nil result?")
continue
}
// else insert node
node := res.(*restic.Node)
if node.Type == "dir" {
debug.Log("got tree node for %s: %v", node.Path, node.Subtree)
if node.Subtree == nil {
debug.Log("subtree is nil for node %v", node.Path)
continue
}
if node.Subtree.IsNull() {
panic("invalid null subtree restic.ID")
}
}
// insert node into tree, resolve name collisions
name := node.Name
i := 0
for {
i++
err := tree.Insert(node)
if err == nil {
break
}
newName := fmt.Sprintf("%v-%d", name, i)
fmt.Fprintf(os.Stderr, "%v: name collision for %q, renaming to %q\n", filepath.Dir(node.Path), node.Name, newName)
node.Name = newName
}
}
node := &restic.Node{}
if dir.Path() != "" && dir.Info() != nil {
n, err := restic.NodeFromFileInfo(dir.Fullpath(), dir.Info())
if err != nil {
arch.Warn(dir.Path(), dir.Info(), err)
}
node = n
if !arch.WithAccessTime {
node.AccessTime = node.ModTime
}
}
if err := dir.Error(); err != nil {
node.Error = err.Error()
}
id, err := arch.SaveTreeJSON(ctx, tree)
if err != nil {
panic(err)
}
debug.Log("save tree for %s: %v", dir.Path(), id)
if id.IsNull() {
panic("invalid null subtree restic.ID return from SaveTreeJSON()")
}
node.Subtree = &id
debug.Log("sending result to %v", dir.Result())
dir.Result() <- node
if dir.Path() != "" {
p.Report(restic.Stat{Dirs: 1})
}
case <-ctx.Done():
// pipeline was cancelled
return
}
}
}
type archivePipe struct {
Old <-chan walk.TreeJob
New <-chan pipe.Job
}
func copyJobs(ctx context.Context, in <-chan pipe.Job, out chan<- pipe.Job) {
var (
// disable sending on the outCh until we received a job
outCh chan<- pipe.Job
// enable receiving from in
inCh = in
job pipe.Job
ok bool
)
for {
select {
case <-ctx.Done():
return
case job, ok = <-inCh:
if !ok {
// input channel closed, we're done
debug.Log("input channel closed, we're done")
return
}
inCh = nil
outCh = out
case outCh <- job:
outCh = nil
inCh = in
}
}
}
type archiveJob struct {
hasOld bool
old walk.TreeJob
new pipe.Job
}
func (a *archivePipe) compare(ctx context.Context, out chan<- pipe.Job) {
defer func() {
close(out)
debug.Log("done")
}()
debug.Log("start")
var (
loadOld, loadNew bool = true, true
ok bool
oldJob walk.TreeJob
newJob pipe.Job
)
for {
if loadOld {
oldJob, ok = <-a.Old
// if the old channel is closed, just pass through the new jobs
if !ok {
debug.Log("old channel is closed, copy from new channel")
// handle remaining newJob
if !loadNew {
out <- archiveJob{new: newJob}.Copy()
}
copyJobs(ctx, a.New, out)
return
}
loadOld = false
}
if loadNew {
newJob, ok = <-a.New
// if the new channel is closed, there are no more files in the current snapshot, return
if !ok {
debug.Log("new channel is closed, we're done")
return
}
loadNew = false
}
debug.Log("old job: %v", oldJob.Path)
debug.Log("new job: %v", newJob.Path())
// at this point we have received an old job as well as a new job, compare paths
file1 := oldJob.Path
file2 := newJob.Path()
dir1 := filepath.Dir(file1)
dir2 := filepath.Dir(file2)
if file1 == file2 {
debug.Log(" same filename %q", file1)
// send job
out <- archiveJob{hasOld: true, old: oldJob, new: newJob}.Copy()
loadOld = true
loadNew = true
continue
} else if dir1 < dir2 {
debug.Log(" %q < %q, file %q added", dir1, dir2, file2)
// file is new, send new job and load new
loadNew = true
out <- archiveJob{new: newJob}.Copy()
continue
} else if dir1 == dir2 {
if file1 < file2 {
debug.Log(" %q < %q, file %q removed", file1, file2, file1)
// file has been removed, load new old
loadOld = true
continue
} else {
debug.Log(" %q > %q, file %q added", file1, file2, file2)
// file is new, send new job and load new
loadNew = true
out <- archiveJob{new: newJob}.Copy()
continue
}
}
debug.Log(" %q > %q, file %q removed", file1, file2, file1)
// file has been removed, throw away old job and load new
loadOld = true
}
}
func (j archiveJob) Copy() pipe.Job {
if !j.hasOld {
return j.new
}
// handle files
if isRegularFile(j.new.Info()) {
debug.Log(" job %v is file", j.new.Path())
// if type has changed, return new job directly
if j.old.Node == nil {
return j.new
}
// if file is newer, return the new job
if j.old.Node.IsNewer(j.new.Fullpath(), j.new.Info()) {
debug.Log(" job %v is newer", j.new.Path())
return j.new
}
debug.Log(" job %v add old data", j.new.Path())
// otherwise annotate job with old data
e := j.new.(pipe.Entry)
e.Node = j.old.Node
return e
}
// dirs and other types are just returned
return j.new
}
const saveIndexTime = 30 * time.Second
// saveIndexes regularly queries the master index for full indexes and saves them.
func (arch *Archiver) saveIndexes(saveCtx, shutdownCtx context.Context, wg *sync.WaitGroup) {
defer wg.Done()
ticker := time.NewTicker(saveIndexTime)
defer ticker.Stop()
for {
select {
case <-saveCtx.Done():
return
case <-shutdownCtx.Done():
return
case <-ticker.C:
debug.Log("saving full indexes")
err := arch.repo.SaveFullIndex(saveCtx)
if err != nil {
debug.Log("save indexes returned an error: %v", err)
fmt.Fprintf(os.Stderr, "error saving preliminary index: %v\n", err)
}
}
}
}
// unique returns a slice that only contains unique strings.
func unique(items []string) []string {
seen := make(map[string]struct{})
for _, item := range items {
seen[item] = struct{}{}
}
items = items[:0]
for item := range seen {
items = append(items, item)
}
return items
}
// baseNameSlice allows sorting paths by basename.
//
// Snapshots have contents sorted by basename, but we receive full paths.
// For the archivePipe to advance them in pairs, we traverse the given
// paths in the same order as the snapshot.
type baseNameSlice []string
func (p baseNameSlice) Len() int { return len(p) }
func (p baseNameSlice) Less(i, j int) bool { return filepath.Base(p[i]) < filepath.Base(p[j]) }
func (p baseNameSlice) Swap(i, j int) { p[i], p[j] = p[j], p[i] }
// Snapshot creates a snapshot of the given paths. If parentrestic.ID is set, this is
// used to compare the files to the ones archived at the time this snapshot was
// taken.
func (arch *Archiver) Snapshot(ctx context.Context, p *restic.Progress, paths, tags []string, hostname string, parentID *restic.ID, time time.Time) (*restic.Snapshot, restic.ID, error) {
paths = unique(paths)
sort.Sort(baseNameSlice(paths))
debug.Log("start for %v", paths)
debug.RunHook("Archiver.Snapshot", nil)
// signal the whole pipeline to stop
var err error
p.Start()
defer p.Done()
// create new snapshot
sn, err := restic.NewSnapshot(paths, tags, hostname, time)
if err != nil {
return nil, restic.ID{}, err
}
sn.Excludes = arch.Excludes
jobs := archivePipe{}
// use parent snapshot (if some was given)
if parentID != nil {
sn.Parent = parentID
// load parent snapshot
parent, err := restic.LoadSnapshot(ctx, arch.repo, *parentID)
if err != nil {
return nil, restic.ID{}, err
}
// start walker on old tree
ch := make(chan walk.TreeJob)
go walk.Tree(ctx, arch.repo, *parent.Tree, ch)
jobs.Old = ch
} else {
// use closed channel
ch := make(chan walk.TreeJob)
close(ch)
jobs.Old = ch
}
// start walker
pipeCh := make(chan pipe.Job)
resCh := make(chan pipe.Result, 1)
go func() {
pipe.Walk(ctx, paths, arch.SelectFilter, pipeCh, resCh)
debug.Log("pipe.Walk done")
}()
jobs.New = pipeCh
ch := make(chan pipe.Job)
go jobs.compare(ctx, ch)
var wg sync.WaitGroup
entCh := make(chan pipe.Entry)
dirCh := make(chan pipe.Dir)
// split
wg.Add(1)
go func() {
pipe.Split(ch, dirCh, entCh)
debug.Log("split done")
close(dirCh)
close(entCh)
wg.Done()
}()
// run workers
for i := 0; i < maxConcurrency; i++ {
wg.Add(2)
go arch.fileWorker(ctx, &wg, p, entCh)
go arch.dirWorker(ctx, &wg, p, dirCh)
}
// run index saver
var wgIndexSaver sync.WaitGroup
shutdownCtx, indexShutdown := context.WithCancel(ctx)
wgIndexSaver.Add(1)
go arch.saveIndexes(ctx, shutdownCtx, &wgIndexSaver)
// wait for all workers to terminate
debug.Log("wait for workers")
wg.Wait()
// stop index saver
indexShutdown()
wgIndexSaver.Wait()
debug.Log("workers terminated")
// flush repository
err = arch.repo.Flush(ctx)
if err != nil {
return nil, restic.ID{}, err
}
// receive the top-level tree
root := (<-resCh).(*restic.Node)
debug.Log("root node received: %v", root.Subtree)
sn.Tree = root.Subtree
// load top-level tree again to see if it is empty
toptree, err := arch.repo.LoadTree(ctx, *root.Subtree)
if err != nil {
return nil, restic.ID{}, err
}
if len(toptree.Nodes) == 0 {
return nil, restic.ID{}, errors.Fatal("no files/dirs saved, refusing to create empty snapshot")
}
// save index
err = arch.repo.SaveIndex(ctx)
if err != nil {
debug.Log("error saving index: %v", err)
return nil, restic.ID{}, err
}
debug.Log("saved indexes")
// save snapshot
id, err := arch.repo.SaveJSONUnpacked(ctx, restic.SnapshotFile, sn)
if err != nil {
return nil, restic.ID{}, err
}
debug.Log("saved snapshot %v", id)
return sn, id, nil
}
func isRegularFile(fi os.FileInfo) bool {
if fi == nil {
return false
}
return fi.Mode()&(os.ModeType|os.ModeCharDevice) == 0
}
// Scan traverses the dirs to collect restic.Stat information while emitting progress
// information with p.
func Scan(dirs []string, filter pipe.SelectFunc, p *restic.Progress) (restic.Stat, error) {
p.Start()
defer p.Done()
var stat restic.Stat
for _, dir := range dirs {
debug.Log("Start for %v", dir)
err := fs.Walk(dir, func(str string, fi os.FileInfo, err error) error {
// TODO: integrate error reporting
if err != nil {
fmt.Fprintf(os.Stderr, "error for %v: %v\n", str, err)
return nil
}
if fi == nil {
fmt.Fprintf(os.Stderr, "error for %v: FileInfo is nil\n", str)
return nil
}
if !filter(str, fi) {
debug.Log("path %v excluded", str)
if fi.IsDir() {
return filepath.SkipDir
}
return nil
}
s := restic.Stat{}
if fi.IsDir() {
s.Dirs++
} else {
s.Files++
if isRegularFile(fi) {
s.Bytes += uint64(fi.Size())
}
}
p.Report(s)
stat.Add(s)
// TODO: handle error?
return nil
})
debug.Log("Done for %v, err: %v", dir, err)
if err != nil {
return restic.Stat{}, errors.Wrap(err, "fs.Walk")
}
}
return stat, nil
}