forked from NVIDIA/aistore
/
tar.go
255 lines (218 loc) · 7.06 KB
/
tar.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
// Package extract provides provides functions for working with compressed files
/*
* Copyright (c) 2018-2023, NVIDIA CORPORATION. All rights reserved.
*/
package extract
import (
"archive/tar"
"io"
"strconv"
"github.com/artashesbalabekyan/aistore/cluster"
"github.com/artashesbalabekyan/aistore/cmn/archive"
"github.com/artashesbalabekyan/aistore/cmn/cos"
"github.com/artashesbalabekyan/aistore/cmn/debug"
"github.com/artashesbalabekyan/aistore/fs"
"github.com/artashesbalabekyan/aistore/memsys"
jsoniter "github.com/json-iterator/go"
)
var (
// Predefined padding buffer (zero-initialized).
padBuf [archive.TarBlockSize]byte
// interface guard
_ Creator = (*tarExtractCreator)(nil)
)
type (
tarExtractCreator struct {
t cluster.Target
}
// tarRecordDataReader is used for writing metadata as well as data to the buffer.
tarRecordDataReader struct {
slab *memsys.Slab
metadataSize int64
size int64
written int64
metadataBuf []byte
tarWriter *tar.Writer
}
)
func newTarRecordDataReader(t cluster.Target) *tarRecordDataReader {
rd := &tarRecordDataReader{}
rd.metadataBuf, rd.slab = t.ByteMM().Alloc()
return rd
}
func (rd *tarRecordDataReader) reinit(tw *tar.Writer, size, metadataSize int64) {
rd.tarWriter = tw
rd.written = 0
rd.size = size
rd.metadataSize = metadataSize
}
func (rd *tarRecordDataReader) free() {
rd.slab.Free(rd.metadataBuf)
}
func (rd *tarRecordDataReader) Write(p []byte) (int, error) {
// Write header
remainingMetadataSize := rd.metadataSize - rd.written
if remainingMetadataSize > 0 {
writeN := int64(len(p))
if writeN < remainingMetadataSize {
debug.Assert(int64(len(rd.metadataBuf))-rd.written >= writeN)
copy(rd.metadataBuf[rd.written:], p)
rd.written += writeN
return len(p), nil
}
debug.Assert(int64(len(rd.metadataBuf))-rd.written >= remainingMetadataSize)
copy(rd.metadataBuf[rd.written:], p[:remainingMetadataSize])
rd.written += remainingMetadataSize
p = p[remainingMetadataSize:]
var header tar.Header
if err := jsoniter.Unmarshal(rd.metadataBuf[:rd.metadataSize], &header); err != nil {
return int(remainingMetadataSize), err
}
if err := rd.tarWriter.WriteHeader(&header); err != nil {
return int(remainingMetadataSize), err
}
} else {
remainingMetadataSize = 0
}
n, err := rd.tarWriter.Write(p)
rd.written += int64(n)
return n + int(remainingMetadataSize), err
}
// ExtractShard reads the tarball f and extracts its metadata.
func (t *tarExtractCreator) ExtractShard(lom *cluster.LOM, r cos.ReadReaderAt, extractor RecordExtractor,
toDisk bool) (extractedSize int64, extractedCount int, err error) {
var (
size int64
header *tar.Header
tr = tar.NewReader(r)
)
buf, slab := t.t.PageMM().AllocSize(lom.SizeBytes())
defer slab.Free(buf)
offset := int64(0)
for {
header, err = tr.Next()
if err == io.EOF {
return extractedSize, extractedCount, nil
} else if err != nil {
return extractedSize, extractedCount, err
}
bmeta := cos.MustMarshal(header)
offset += t.MetadataSize()
if header.Typeflag == tar.TypeDir {
// We can safely ignore this case because we do `MkdirAll` anyway
// when we create files. And since dirs can appear after all the files
// we must have this `MkdirAll` before files.
continue
} else if header.Format == tar.FormatPAX {
// When dealing with `tar.FormatPAX` we also need to take into
// consideration the `tar.TypeXHeader` that comes before the actual header.
// Together it looks like this: [x-header][pax-records][pax-header][pax-file].
// Since `tar.Reader` skips over this header and writes to `header.PAXRecords`
// we need to manually adjust the offset, otherwise when using the
// offset we will point to totally wrong location.
// Add offset for `tar.TypeXHeader`.
offset += t.MetadataSize()
// Add offset for size of PAX records - there is no way of knowing
// the size, so we must estimate it by ourselves...
size := estimateXHeaderSize(header.PAXRecords)
size = cos.CeilAlignInt64(size, archive.TarBlockSize)
offset += size
}
data := cos.NewSizedReader(tr, header.Size)
extractMethod := ExtractToMem
if toDisk {
extractMethod = ExtractToDisk
}
args := extractRecordArgs{
shardName: lom.ObjName,
fileType: fs.ObjectType,
recordName: header.Name,
r: data,
metadata: bmeta,
extractMethod: extractMethod,
offset: offset,
buf: buf,
}
if size, err = extractor.ExtractRecordWithBuffer(args); err != nil {
return extractedSize, extractedCount, err
}
extractedSize += size
extractedCount++
// .tar format pads all block to 512 bytes
offset += cos.CeilAlignInt64(header.Size, archive.TarBlockSize)
}
}
func NewTarExtractCreator(t cluster.Target) Creator {
return &tarExtractCreator{t: t}
}
// CreateShard creates a new shard locally based on the Shard.
// Note that the order of closing must be trw, gzw, then finally tarball.
func (t *tarExtractCreator) CreateShard(s *Shard, tarball io.Writer, loadContent LoadContentFunc) (written int64, err error) {
var (
n int64
needFlush bool
tw = tar.NewWriter(tarball)
rdReader = newTarRecordDataReader(t.t)
)
defer func() {
rdReader.free()
cos.Close(tw)
}()
for _, rec := range s.Records.All() {
for _, obj := range rec.Objects {
switch obj.StoreType {
case OffsetStoreType:
if needFlush {
// We now will write directly to the tarball file so we need
// to flush everything what we have written so far.
if err := tw.Flush(); err != nil {
return written, err
}
needFlush = false
}
if n, err = loadContent(tarball, rec, obj); err != nil {
return written + n, err
}
// pad to 512 bytes
diff := cos.CeilAlignInt64(n, archive.TarBlockSize) - n
if diff > 0 {
if _, err = tarball.Write(padBuf[:diff]); err != nil {
return written + n, err
}
n += diff
}
debug.Assert(diff >= 0 && diff < archive.TarBlockSize)
case SGLStoreType, DiskStoreType:
rdReader.reinit(tw, obj.Size, obj.MetadataSize)
if n, err = loadContent(rdReader, rec, obj); err != nil {
return written + n, err
}
written += n
needFlush = true
default:
cos.AssertMsg(false, obj.StoreType)
}
written += n
}
}
return written, nil
}
func (*tarExtractCreator) UsingCompression() bool { return false }
func (*tarExtractCreator) SupportsOffset() bool { return true }
func (*tarExtractCreator) MetadataSize() int64 { return archive.TarBlockSize } // size of tar header with padding
// NOTE: Mostly taken from `tar.formatPAXRecord`.
func estimateXHeaderSize(paxRecords map[string]string) int64 {
totalSize := 0
for k, v := range paxRecords {
const padding = 3 // Extra padding for ' ', '=', and '\n'
size := len(k) + len(v) + padding
size += len(strconv.Itoa(size))
record := strconv.Itoa(size) + " " + k + "=" + v + "\n"
// Final adjustment if adding size field increased the record size.
if len(record) != size {
record = strconv.Itoa(len(record)) + " " + k + "=" + v + "\n"
}
totalSize += len(record)
}
return int64(totalSize)
}