/bitcoin/src/leveldb/db/log_reader.cc
Line | Count | Source |
1 | | // Copyright (c) 2011 The LevelDB Authors. All rights reserved. |
2 | | // Use of this source code is governed by a BSD-style license that can be |
3 | | // found in the LICENSE file. See the AUTHORS file for names of contributors. |
4 | | |
5 | | #include "db/log_reader.h" |
6 | | |
7 | | #include <stdio.h> |
8 | | |
9 | | #include "leveldb/env.h" |
10 | | #include "util/coding.h" |
11 | | #include "util/crc32c.h" |
12 | | |
13 | | namespace leveldb { |
14 | | namespace log { |
15 | | |
16 | 33.2k | Reader::Reporter::~Reporter() = default; |
17 | | |
18 | | Reader::Reader(SequentialFile* file, Reporter* reporter, bool checksum, |
19 | | uint64_t initial_offset) |
20 | 33.2k | : file_(file), |
21 | 33.2k | reporter_(reporter), |
22 | 33.2k | checksum_(checksum), |
23 | 33.2k | backing_store_(new char[kBlockSize]), |
24 | 33.2k | buffer_(), |
25 | 33.2k | eof_(false), |
26 | 33.2k | last_record_offset_(0), |
27 | 33.2k | end_of_buffer_offset_(0), |
28 | 33.2k | initial_offset_(initial_offset), |
29 | 33.2k | resyncing_(initial_offset > 0) {} |
30 | | |
31 | 33.2k | Reader::~Reader() { delete[] backing_store_; } |
32 | | |
33 | 0 | bool Reader::SkipToInitialBlock() { |
34 | 0 | const size_t offset_in_block = initial_offset_ % kBlockSize; |
35 | 0 | uint64_t block_start_location = initial_offset_ - offset_in_block; |
36 | | |
37 | | // Don't search a block if we'd be in the trailer |
38 | 0 | if (offset_in_block > kBlockSize - 6) { Branch (38:7): [True: 0, False: 0]
|
39 | 0 | block_start_location += kBlockSize; |
40 | 0 | } |
41 | |
|
42 | 0 | end_of_buffer_offset_ = block_start_location; |
43 | | |
44 | | // Skip to start of first block that can contain the initial record |
45 | 0 | if (block_start_location > 0) { Branch (45:7): [True: 0, False: 0]
|
46 | 0 | Status skip_status = file_->Skip(block_start_location); |
47 | 0 | if (!skip_status.ok()) { Branch (47:9): [True: 0, False: 0]
|
48 | 0 | ReportDrop(block_start_location, skip_status); |
49 | 0 | return false; |
50 | 0 | } |
51 | 0 | } |
52 | | |
53 | 0 | return true; |
54 | 0 | } |
55 | | |
56 | 66.5k | bool Reader::ReadRecord(Slice* record, std::string* scratch) { |
57 | 66.5k | if (last_record_offset_ < initial_offset_) { Branch (57:7): [True: 0, False: 66.5k]
|
58 | 0 | if (!SkipToInitialBlock()) { Branch (58:9): [True: 0, False: 0]
|
59 | 0 | return false; |
60 | 0 | } |
61 | 0 | } |
62 | | |
63 | 66.5k | scratch->clear(); |
64 | 66.5k | record->clear(); |
65 | 66.5k | bool in_fragmented_record = false; |
66 | | // Record offset of the logical record that we're reading |
67 | | // 0 is a dummy value to make compilers happy |
68 | 66.5k | uint64_t prospective_record_offset = 0; |
69 | | |
70 | 66.5k | Slice fragment; |
71 | 66.5k | while (true) { Branch (71:10): [Folded - Ignored]
|
72 | 66.5k | const unsigned int record_type = ReadPhysicalRecord(&fragment); |
73 | | |
74 | | // ReadPhysicalRecord may have only had an empty trailer remaining in its |
75 | | // internal buffer. Calculate the offset of the next physical record now |
76 | | // that it has returned, properly accounting for its header size. |
77 | 66.5k | uint64_t physical_record_offset = |
78 | 66.5k | end_of_buffer_offset_ - buffer_.size() - kHeaderSize - fragment.size(); |
79 | | |
80 | 66.5k | if (resyncing_) { Branch (80:9): [True: 0, False: 66.5k]
|
81 | 0 | if (record_type == kMiddleType) { Branch (81:11): [True: 0, False: 0]
|
82 | 0 | continue; |
83 | 0 | } else if (record_type == kLastType) { Branch (83:18): [True: 0, False: 0]
|
84 | 0 | resyncing_ = false; |
85 | 0 | continue; |
86 | 0 | } else { |
87 | 0 | resyncing_ = false; |
88 | 0 | } |
89 | 0 | } |
90 | | |
91 | 66.5k | switch (record_type) { |
92 | 33.2k | case kFullType: Branch (92:7): [True: 33.2k, False: 33.2k]
|
93 | 33.2k | if (in_fragmented_record) { Branch (93:13): [True: 0, False: 33.2k]
|
94 | | // Handle bug in earlier versions of log::Writer where |
95 | | // it could emit an empty kFirstType record at the tail end |
96 | | // of a block followed by a kFullType or kFirstType record |
97 | | // at the beginning of the next block. |
98 | 0 | if (!scratch->empty()) { Branch (98:15): [True: 0, False: 0]
|
99 | 0 | ReportCorruption(scratch->size(), "partial record without end(1)"); |
100 | 0 | } |
101 | 0 | } |
102 | 33.2k | prospective_record_offset = physical_record_offset; |
103 | 33.2k | scratch->clear(); |
104 | 33.2k | *record = fragment; |
105 | 33.2k | last_record_offset_ = prospective_record_offset; |
106 | 33.2k | return true; |
107 | | |
108 | 0 | case kFirstType: Branch (108:7): [True: 0, False: 66.5k]
|
109 | 0 | if (in_fragmented_record) { Branch (109:13): [True: 0, False: 0]
|
110 | | // Handle bug in earlier versions of log::Writer where |
111 | | // it could emit an empty kFirstType record at the tail end |
112 | | // of a block followed by a kFullType or kFirstType record |
113 | | // at the beginning of the next block. |
114 | 0 | if (!scratch->empty()) { Branch (114:15): [True: 0, False: 0]
|
115 | 0 | ReportCorruption(scratch->size(), "partial record without end(2)"); |
116 | 0 | } |
117 | 0 | } |
118 | 0 | prospective_record_offset = physical_record_offset; |
119 | 0 | scratch->assign(fragment.data(), fragment.size()); |
120 | 0 | in_fragmented_record = true; |
121 | 0 | break; |
122 | | |
123 | 0 | case kMiddleType: Branch (123:7): [True: 0, False: 66.5k]
|
124 | 0 | if (!in_fragmented_record) { Branch (124:13): [True: 0, False: 0]
|
125 | 0 | ReportCorruption(fragment.size(), |
126 | 0 | "missing start of fragmented record(1)"); |
127 | 0 | } else { |
128 | 0 | scratch->append(fragment.data(), fragment.size()); |
129 | 0 | } |
130 | 0 | break; |
131 | | |
132 | 0 | case kLastType: Branch (132:7): [True: 0, False: 66.5k]
|
133 | 0 | if (!in_fragmented_record) { Branch (133:13): [True: 0, False: 0]
|
134 | 0 | ReportCorruption(fragment.size(), |
135 | 0 | "missing start of fragmented record(2)"); |
136 | 0 | } else { |
137 | 0 | scratch->append(fragment.data(), fragment.size()); |
138 | 0 | *record = Slice(*scratch); |
139 | 0 | last_record_offset_ = prospective_record_offset; |
140 | 0 | return true; |
141 | 0 | } |
142 | 0 | break; |
143 | | |
144 | 33.2k | case kEof: Branch (144:7): [True: 33.2k, False: 33.2k]
|
145 | 33.2k | if (in_fragmented_record) { Branch (145:13): [True: 0, False: 33.2k]
|
146 | | // This can be caused by the writer dying immediately after |
147 | | // writing a physical record but before completing the next; don't |
148 | | // treat it as a corruption, just ignore the entire logical record. |
149 | 0 | scratch->clear(); |
150 | 0 | } |
151 | 33.2k | return false; |
152 | | |
153 | 0 | case kBadRecord: Branch (153:7): [True: 0, False: 66.5k]
|
154 | 0 | if (in_fragmented_record) { Branch (154:13): [True: 0, False: 0]
|
155 | 0 | ReportCorruption(scratch->size(), "error in middle of record"); |
156 | 0 | in_fragmented_record = false; |
157 | 0 | scratch->clear(); |
158 | 0 | } |
159 | 0 | break; |
160 | | |
161 | 0 | default: { Branch (161:7): [True: 0, False: 66.5k]
|
162 | 0 | char buf[40]; |
163 | 0 | snprintf(buf, sizeof(buf), "unknown record type %u", record_type); |
164 | 0 | ReportCorruption( |
165 | 0 | (fragment.size() + (in_fragmented_record ? scratch->size() : 0)), Branch (165:33): [True: 0, False: 0]
|
166 | 0 | buf); |
167 | 0 | in_fragmented_record = false; |
168 | 0 | scratch->clear(); |
169 | 0 | break; |
170 | 0 | } |
171 | 66.5k | } |
172 | 66.5k | } |
173 | 0 | return false; |
174 | 66.5k | } |
175 | | |
176 | 0 | uint64_t Reader::LastRecordOffset() { return last_record_offset_; } |
177 | | |
178 | 0 | void Reader::ReportCorruption(uint64_t bytes, const char* reason) { |
179 | 0 | ReportDrop(bytes, Status::Corruption(reason, file_->GetName())); |
180 | 0 | } |
181 | | |
182 | 0 | void Reader::ReportDrop(uint64_t bytes, const Status& reason) { |
183 | 0 | if (reporter_ != nullptr && Branch (183:7): [True: 0, False: 0]
|
184 | 0 | end_of_buffer_offset_ - buffer_.size() - bytes >= initial_offset_) { Branch (184:7): [True: 0, False: 0]
|
185 | 0 | reporter_->Corruption(static_cast<size_t>(bytes), reason); |
186 | 0 | } |
187 | 0 | } |
188 | | |
189 | 66.5k | unsigned int Reader::ReadPhysicalRecord(Slice* result) { |
190 | 99.8k | while (true) { Branch (190:10): [Folded - Ignored]
|
191 | 99.8k | if (buffer_.size() < kHeaderSize) { Branch (191:9): [True: 66.5k, False: 33.2k]
|
192 | 66.5k | if (!eof_) { Branch (192:11): [True: 33.2k, False: 33.2k]
|
193 | | // Last read was a full read, so this is a trailer to skip |
194 | 33.2k | buffer_.clear(); |
195 | 33.2k | Status status = file_->Read(kBlockSize, &buffer_, backing_store_); |
196 | 33.2k | end_of_buffer_offset_ += buffer_.size(); |
197 | 33.2k | if (!status.ok()) { Branch (197:13): [True: 0, False: 33.2k]
|
198 | 0 | buffer_.clear(); |
199 | 0 | ReportDrop(kBlockSize, status); |
200 | 0 | eof_ = true; |
201 | 0 | return kEof; |
202 | 33.2k | } else if (buffer_.size() < kBlockSize) { Branch (202:20): [True: 33.2k, False: 0]
|
203 | 33.2k | eof_ = true; |
204 | 33.2k | } |
205 | 33.2k | continue; |
206 | 33.2k | } else { |
207 | | // Note that if buffer_ is non-empty, we have a truncated header at the |
208 | | // end of the file, which can be caused by the writer crashing in the |
209 | | // middle of writing the header. Instead of considering this an error, |
210 | | // just report EOF. |
211 | 33.2k | buffer_.clear(); |
212 | 33.2k | return kEof; |
213 | 33.2k | } |
214 | 66.5k | } |
215 | | |
216 | | // Parse the header |
217 | 33.2k | const char* header = buffer_.data(); |
218 | 33.2k | const uint32_t a = static_cast<uint32_t>(header[4]) & 0xff; |
219 | 33.2k | const uint32_t b = static_cast<uint32_t>(header[5]) & 0xff; |
220 | 33.2k | const unsigned int type = header[6]; |
221 | 33.2k | const uint32_t length = a | (b << 8); |
222 | 33.2k | if (kHeaderSize + length > buffer_.size()) { Branch (222:9): [True: 0, False: 33.2k]
|
223 | 0 | size_t drop_size = buffer_.size(); |
224 | 0 | buffer_.clear(); |
225 | 0 | if (!eof_) { Branch (225:11): [True: 0, False: 0]
|
226 | 0 | ReportCorruption(drop_size, "bad record length"); |
227 | 0 | return kBadRecord; |
228 | 0 | } |
229 | | // If the end of the file has been reached without reading |length| bytes |
230 | | // of payload, assume the writer died in the middle of writing the record. |
231 | | // Don't report a corruption. |
232 | 0 | return kEof; |
233 | 0 | } |
234 | | |
235 | 33.2k | if (type == kZeroType && length == 0) { Branch (235:9): [True: 0, False: 33.2k]
Branch (235:30): [True: 0, False: 0]
|
236 | | // Skip zero length record without reporting any drops since |
237 | | // such records are produced by the mmap based writing code in |
238 | | // env_posix.cc that preallocates file regions. |
239 | 0 | buffer_.clear(); |
240 | 0 | return kBadRecord; |
241 | 0 | } |
242 | | |
243 | | // Check crc |
244 | 33.2k | if (checksum_) { Branch (244:9): [True: 33.2k, False: 0]
|
245 | 33.2k | uint32_t expected_crc = crc32c::Unmask(DecodeFixed32(header)); |
246 | 33.2k | uint32_t actual_crc = crc32c::Value(header + 6, 1 + length); |
247 | 33.2k | if (actual_crc != expected_crc) { Branch (247:11): [True: 0, False: 33.2k]
|
248 | | // Drop the rest of the buffer since "length" itself may have |
249 | | // been corrupted and if we trust it, we could find some |
250 | | // fragment of a real log record that just happens to look |
251 | | // like a valid log record. |
252 | 0 | size_t drop_size = buffer_.size(); |
253 | 0 | buffer_.clear(); |
254 | 0 | ReportCorruption(drop_size, "checksum mismatch"); |
255 | 0 | return kBadRecord; |
256 | 0 | } |
257 | 33.2k | } |
258 | | |
259 | 33.2k | buffer_.remove_prefix(kHeaderSize + length); |
260 | | |
261 | | // Skip physical record that started before initial_offset_ |
262 | 33.2k | if (end_of_buffer_offset_ - buffer_.size() - kHeaderSize - length < Branch (262:9): [True: 0, False: 33.2k]
|
263 | 33.2k | initial_offset_) { |
264 | 0 | result->clear(); |
265 | 0 | return kBadRecord; |
266 | 0 | } |
267 | | |
268 | 33.2k | *result = Slice(header + kHeaderSize, length); |
269 | 33.2k | return type; |
270 | 33.2k | } |
271 | 66.5k | } |
272 | | |
273 | | } // namespace log |
274 | | } // namespace leveldb |