restic/vendor/github.com/kurin/blazer/b2/reader.go
Alexander Neumann b9f0f031b6 Update dependencies
Closes #2129
2019-02-10 13:24:37 +01:00

348 lines
7.6 KiB
Go

// Copyright 2016, the Blazer authors
//
// 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.
package b2
import (
"bytes"
"context"
"crypto/sha1"
"errors"
"fmt"
"hash"
"io"
"sync"
"time"
"github.com/kurin/blazer/internal/blog"
)
var errNoMoreContent = errors.New("416: out of content")
// Reader reads files from B2.
type Reader struct {
// ConcurrentDownloads is the number of simultaneous downloads to pull from
// B2. Values greater than one will cause B2 to make multiple HTTP requests
// for a given file, increasing available bandwidth at the cost of buffering
// the downloads in memory.
ConcurrentDownloads int
// ChunkSize is the size to fetch per ConcurrentDownload. The default is
// 10MB.
ChunkSize int
ctx context.Context
cancel context.CancelFunc // cancels ctx
o *Object
name string
offset int64 // the start of the file
length int64 // the length to read, or -1
csize int // chunk size
read int // amount read
chwid int // chunks written
chrid int // chunks read
chbuf chan *rchunk
init sync.Once
chunks map[int]*rchunk
vrfy hash.Hash
readOffEnd bool
sha1 string
rmux sync.Mutex // guards rcond
rcond *sync.Cond
emux sync.RWMutex // guards err, believe it or not
err error
smux sync.Mutex
smap map[int]*meteredReader
}
type rchunk struct {
bytes.Buffer
final bool
}
// Close frees resources associated with the download.
func (r *Reader) Close() error {
r.cancel()
r.o.b.c.removeReader(r)
return nil
}
func (r *Reader) setErr(err error) {
r.emux.Lock()
defer r.emux.Unlock()
if r.err == nil {
r.err = err
r.cancel()
}
}
func (r *Reader) setErrNoCancel(err error) {
r.emux.Lock()
defer r.emux.Unlock()
if r.err == nil {
r.err = err
}
}
func (r *Reader) getErr() error {
r.emux.RLock()
defer r.emux.RUnlock()
return r.err
}
func (r *Reader) thread() {
go func() {
for {
var buf *rchunk
select {
case b, ok := <-r.chbuf:
if !ok {
return
}
buf = b
case <-r.ctx.Done():
return
}
r.rmux.Lock()
chunkID := r.chwid
r.chwid++
r.rmux.Unlock()
offset := int64(chunkID*r.csize) + r.offset
size := int64(r.csize)
if r.length > 0 {
if size > r.length {
buf.final = true
size = r.length
}
r.length -= size
}
var b backoff
redo:
fr, err := r.o.b.b.downloadFileByName(r.ctx, r.name, offset, size)
if err == errNoMoreContent {
// this read generated a 416 so we are entirely past the end of the object
r.readOffEnd = true
buf.final = true
r.rmux.Lock()
r.chunks[chunkID] = buf
r.rmux.Unlock()
r.rcond.Broadcast()
return
}
if err != nil {
r.setErr(err)
r.rcond.Broadcast()
return
}
rsize, _, sha1, _ := fr.stats()
if len(sha1) == 40 && r.sha1 != sha1 {
r.sha1 = sha1
}
mr := &meteredReader{r: noopResetter{fr}, size: int(rsize)}
r.smux.Lock()
r.smap[chunkID] = mr
r.smux.Unlock()
i, err := copyContext(r.ctx, buf, mr)
fr.Close()
r.smux.Lock()
r.smap[chunkID] = nil
r.smux.Unlock()
if i < int64(rsize) || err == io.ErrUnexpectedEOF {
// Probably the network connection was closed early. Retry.
blog.V(1).Infof("b2 reader %d: got %dB of %dB; retrying after %v", chunkID, i, rsize, b)
if err := b.wait(r.ctx); err != nil {
r.setErr(err)
r.rcond.Broadcast()
return
}
buf.Reset()
goto redo
}
if err != nil {
r.setErr(err)
r.rcond.Broadcast()
return
}
r.rmux.Lock()
r.chunks[chunkID] = buf
r.rmux.Unlock()
r.rcond.Broadcast()
}
}()
}
func (r *Reader) curChunk() (*rchunk, error) {
ch := make(chan *rchunk)
go func() {
r.rmux.Lock()
defer r.rmux.Unlock()
for r.chunks[r.chrid] == nil && r.getErr() == nil && r.ctx.Err() == nil {
r.rcond.Wait()
}
select {
case ch <- r.chunks[r.chrid]:
case <-r.ctx.Done():
return
}
}()
select {
case buf := <-ch:
return buf, r.getErr()
case <-r.ctx.Done():
if r.getErr() != nil {
return nil, r.getErr()
}
return nil, r.ctx.Err()
}
}
func (r *Reader) initFunc() {
r.smux.Lock()
r.smap = make(map[int]*meteredReader)
r.smux.Unlock()
r.o.b.c.addReader(r)
r.rcond = sync.NewCond(&r.rmux)
cr := r.ConcurrentDownloads
if cr < 1 {
cr = 1
}
if r.ChunkSize < 1 {
r.ChunkSize = 1e7
}
r.csize = r.ChunkSize
r.chbuf = make(chan *rchunk, cr)
for i := 0; i < cr; i++ {
r.thread()
r.chbuf <- &rchunk{}
}
r.vrfy = sha1.New()
}
func (r *Reader) Read(p []byte) (int, error) {
if err := r.getErr(); err != nil {
return 0, err
}
r.init.Do(r.initFunc)
chunk, err := r.curChunk()
if err != nil {
r.setErrNoCancel(err)
return 0, err
}
n, err := chunk.Read(p)
r.vrfy.Write(p[:n]) // Hash.Write never returns an error.
r.read += n
if err == io.EOF {
if chunk.final {
close(r.chbuf)
r.setErrNoCancel(err)
return n, err
}
r.chrid++
chunk.Reset()
r.chbuf <- chunk
err = nil
}
r.setErrNoCancel(err)
return n, err
}
func (r *Reader) status() *ReaderStatus {
r.smux.Lock()
defer r.smux.Unlock()
rs := &ReaderStatus{
Progress: make([]float64, len(r.smap)),
}
for i := 1; i <= len(r.smap); i++ {
rs.Progress[i-1] = r.smap[i].done()
}
return rs
}
// Verify checks the SHA1 hash on download and compares it to the SHA1 hash
// submitted on upload. If the two differ, this returns an error. If the
// correct hash could not be calculated (if, for example, the entire object was
// not read, or if the object was uploaded as a "large file" and thus the SHA1
// hash was not sent), this returns (nil, false).
func (r *Reader) Verify() (error, bool) {
got := fmt.Sprintf("%x", r.vrfy.Sum(nil))
if r.sha1 == got {
return nil, true
}
// TODO: if the exact length of the file is requested AND the checksum is
// bad, this will return (nil, false) instead of (an error, true). This is
// because there's no good way that I can tell to determine that we've hit
// the end of the file without reading off the end. Consider reading N+1
// bytes at the very end to close this hole.
if r.offset > 0 || !r.readOffEnd || len(r.sha1) != 40 {
return nil, false
}
return fmt.Errorf("bad hash: got %v, want %v", got, r.sha1), true
}
// strip a writer of any non-Write methods
type onlyWriter struct{ w io.Writer }
func (ow onlyWriter) Write(p []byte) (int, error) { return ow.w.Write(p) }
func copyContext(ctx context.Context, w io.Writer, r io.Reader) (int64, error) {
var n int64
var err error
done := make(chan struct{})
go func() {
if _, ok := w.(*Writer); ok {
w = onlyWriter{w}
}
n, err = io.Copy(w, r)
close(done)
}()
select {
case <-done:
return n, err
case <-ctx.Done():
return 0, ctx.Err()
}
}
type noopResetter struct {
io.Reader
}
func (noopResetter) Reset() error { return nil }
type backoff time.Duration
func (b *backoff) wait(ctx context.Context) error {
if *b == 0 {
*b = backoff(time.Millisecond)
}
select {
case <-time.After(time.Duration(*b)):
if time.Duration(*b) < time.Second*10 {
*b <<= 1
}
return nil
case <-ctx.Done():
return ctx.Err()
}
}
func (b backoff) String() string {
return time.Duration(b).String()
}