1// Copyright 2014 The Go Authors. All rights reserved.
2// Use of this source code is governed by a BSD-style
3// license that can be found in the LICENSE file.
4
5package hpack
6
7import (
8 "bytes"
9 "errors"
10 "io"
11 "sync"
12)
13
14var bufPool = sync.Pool{
15 New: func() interface{} { return new(bytes.Buffer) },
16}
17
18// HuffmanDecode decodes the string in v and writes the expanded
19// result to w, returning the number of bytes written to w and the
20// Write call's return value. At most one Write call is made.
21func HuffmanDecode(w io.Writer, v []byte) (int, error) {
22 buf := bufPool.Get().(*bytes.Buffer)
23 buf.Reset()
24 defer bufPool.Put(buf)
25 if err := huffmanDecode(buf, 0, v); err != nil {
26 return 0, err
27 }
28 return w.Write(buf.Bytes())
29}
30
31// HuffmanDecodeToString decodes the string in v.
32func HuffmanDecodeToString(v []byte) (string, error) {
33 buf := bufPool.Get().(*bytes.Buffer)
34 buf.Reset()
35 defer bufPool.Put(buf)
36 if err := huffmanDecode(buf, 0, v); err != nil {
37 return "", err
38 }
39 return buf.String(), nil
40}
41
42// ErrInvalidHuffman is returned for errors found decoding
43// Huffman-encoded strings.
44var ErrInvalidHuffman = errors.New("hpack: invalid Huffman-encoded data")
45
46// huffmanDecode decodes v to buf.
47// If maxLen is greater than 0, attempts to write more to buf than
48// maxLen bytes will return ErrStringLength.
49func huffmanDecode(buf *bytes.Buffer, maxLen int, v []byte) error {
50 rootHuffmanNode := getRootHuffmanNode()
51 n := rootHuffmanNode
52 // cur is the bit buffer that has not been fed into n.
53 // cbits is the number of low order bits in cur that are valid.
54 // sbits is the number of bits of the symbol prefix being decoded.
55 cur, cbits, sbits := uint(0), uint8(0), uint8(0)
56 for _, b := range v {
57 cur = cur<<8 | uint(b)
58 cbits += 8
59 sbits += 8
60 for cbits >= 8 {
61 idx := byte(cur >> (cbits - 8))
62 n = n.children[idx]
63 if n == nil {
64 return ErrInvalidHuffman
65 }
66 if n.children == nil {
67 if maxLen != 0 && buf.Len() == maxLen {
68 return ErrStringLength
69 }
70 buf.WriteByte(n.sym)
71 cbits -= n.codeLen
72 n = rootHuffmanNode
73 sbits = cbits
74 } else {
75 cbits -= 8
76 }
77 }
78 }
79 for cbits > 0 {
80 n = n.children[byte(cur<<(8-cbits))]
81 if n == nil {
82 return ErrInvalidHuffman
83 }
84 if n.children != nil || n.codeLen > cbits {
85 break
86 }
87 if maxLen != 0 && buf.Len() == maxLen {
88 return ErrStringLength
89 }
90 buf.WriteByte(n.sym)
91 cbits -= n.codeLen
92 n = rootHuffmanNode
93 sbits = cbits
94 }
95 if sbits > 7 {
96 // Either there was an incomplete symbol, or overlong padding.
97 // Both are decoding errors per RFC 7541 section 5.2.
98 return ErrInvalidHuffman
99 }
100 if mask := uint(1<<cbits - 1); cur&mask != mask {
101 // Trailing bits must be a prefix of EOS per RFC 7541 section 5.2.
102 return ErrInvalidHuffman
103 }
104
105 return nil
106}
107
108// incomparable is a zero-width, non-comparable type. Adding it to a struct
109// makes that struct also non-comparable, and generally doesn't add
110// any size (as long as it's first).
111type incomparable [0]func()
112
113type node struct {
114 _ incomparable
115
116 // children is non-nil for internal nodes
117 children *[256]*node
118
119 // The following are only valid if children is nil:
120 codeLen uint8 // number of bits that led to the output of sym
121 sym byte // output symbol
122}
123
124func newInternalNode() *node {
125 return &node{children: new([256]*node)}
126}
127
128var (
129 buildRootOnce sync.Once
130 lazyRootHuffmanNode *node
131)
132
133func getRootHuffmanNode() *node {
134 buildRootOnce.Do(buildRootHuffmanNode)
135 return lazyRootHuffmanNode
136}
137
138func buildRootHuffmanNode() {
139 if len(huffmanCodes) != 256 {
140 panic("unexpected size")
141 }
142 lazyRootHuffmanNode = newInternalNode()
143 // allocate a leaf node for each of the 256 symbols
144 leaves := new([256]node)
145
146 for sym, code := range huffmanCodes {
147 codeLen := huffmanCodeLen[sym]
148
149 cur := lazyRootHuffmanNode
150 for codeLen > 8 {
151 codeLen -= 8
152 i := uint8(code >> codeLen)
153 if cur.children[i] == nil {
154 cur.children[i] = newInternalNode()
155 }
156 cur = cur.children[i]
157 }
158 shift := 8 - codeLen
159 start, end := int(uint8(code<<shift)), int(1<<shift)
160
161 leaves[sym].sym = byte(sym)
162 leaves[sym].codeLen = codeLen
163 for i := start; i < start+end; i++ {
164 cur.children[i] = &leaves[sym]
165 }
166 }
167}
168
169// AppendHuffmanString appends s, as encoded in Huffman codes, to dst
170// and returns the extended buffer.
171func AppendHuffmanString(dst []byte, s string) []byte {
172 // This relies on the maximum huffman code length being 30 (See tables.go huffmanCodeLen array)
173 // So if a uint64 buffer has less than 32 valid bits can always accommodate another huffmanCode.
174 var (
175 x uint64 // buffer
176 n uint // number valid of bits present in x
177 )
178 for i := 0; i < len(s); i++ {
179 c := s[i]
180 n += uint(huffmanCodeLen[c])
181 x <<= huffmanCodeLen[c] % 64
182 x |= uint64(huffmanCodes[c])
183 if n >= 32 {
184 n %= 32 // Normally would be -= 32 but %= 32 informs compiler 0 <= n <= 31 for upcoming shift
185 y := uint32(x >> n) // Compiler doesn't combine memory writes if y isn't uint32
186 dst = append(dst, byte(y>>24), byte(y>>16), byte(y>>8), byte(y))
187 }
188 }
189 // Add padding bits if necessary
190 if over := n % 8; over > 0 {
191 const (
192 eosCode = 0x3fffffff
193 eosNBits = 30
194 eosPadByte = eosCode >> (eosNBits - 8)
195 )
196 pad := 8 - over
197 x = (x << pad) | (eosPadByte >> over)
198 n += pad // 8 now divides into n exactly
199 }
200 // n in (0, 8, 16, 24, 32)
201 switch n / 8 {
202 case 0:
203 return dst
204 case 1:
205 return append(dst, byte(x))
206 case 2:
207 y := uint16(x)
208 return append(dst, byte(y>>8), byte(y))
209 case 3:
210 y := uint16(x >> 8)
211 return append(dst, byte(y>>8), byte(y), byte(x))
212 }
213 // case 4:
214 y := uint32(x)
215 return append(dst, byte(y>>24), byte(y>>16), byte(y>>8), byte(y))
216}
217
218// HuffmanEncodeLength returns the number of bytes required to encode
219// s in Huffman codes. The result is round up to byte boundary.
220func HuffmanEncodeLength(s string) uint64 {
221 n := uint64(0)
222 for i := 0; i < len(s); i++ {
223 n += uint64(huffmanCodeLen[s[i]])
224 }
225 return (n + 7) / 8
226}