xref: /aosp_15_r20/external/leveldb/db/repair.cc (revision 9507f98c5f32dee4b5f9e4a38cd499f3ff5c4490)
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 // We recover the contents of the descriptor from the other files we find.
6 // (1) Any log files are first converted to tables
7 // (2) We scan every table to compute
8 //     (a) smallest/largest for the table
9 //     (b) largest sequence number in the table
10 // (3) We generate descriptor contents:
11 //      - log number is set to zero
12 //      - next-file-number is set to 1 + largest file number we found
13 //      - last-sequence-number is set to largest sequence# found across
14 //        all tables (see 2c)
15 //      - compaction pointers are cleared
16 //      - every table file is added at level 0
17 //
18 // Possible optimization 1:
19 //   (a) Compute total size and use to pick appropriate max-level M
20 //   (b) Sort tables by largest sequence# in the table
21 //   (c) For each table: if it overlaps earlier table, place in level-0,
22 //       else place in level-M.
23 // Possible optimization 2:
24 //   Store per-table metadata (smallest, largest, largest-seq#, ...)
25 //   in the table's meta section to speed up ScanTable.
26 
27 #include "db/builder.h"
28 #include "db/db_impl.h"
29 #include "db/dbformat.h"
30 #include "db/filename.h"
31 #include "db/log_reader.h"
32 #include "db/log_writer.h"
33 #include "db/memtable.h"
34 #include "db/table_cache.h"
35 #include "db/version_edit.h"
36 #include "db/write_batch_internal.h"
37 #include "leveldb/comparator.h"
38 #include "leveldb/db.h"
39 #include "leveldb/env.h"
40 
41 namespace leveldb {
42 
43 namespace {
44 
45 class Repairer {
46  public:
Repairer(const std::string & dbname,const Options & options)47   Repairer(const std::string& dbname, const Options& options)
48       : dbname_(dbname),
49         env_(options.env),
50         icmp_(options.comparator),
51         ipolicy_(options.filter_policy),
52         options_(SanitizeOptions(dbname, &icmp_, &ipolicy_, options)),
53         owns_info_log_(options_.info_log != options.info_log),
54         owns_cache_(options_.block_cache != options.block_cache),
55         next_file_number_(1) {
56     // TableCache can be small since we expect each table to be opened once.
57     table_cache_ = new TableCache(dbname_, options_, 10);
58   }
59 
~Repairer()60   ~Repairer() {
61     delete table_cache_;
62     if (owns_info_log_) {
63       delete options_.info_log;
64     }
65     if (owns_cache_) {
66       delete options_.block_cache;
67     }
68   }
69 
Run()70   Status Run() {
71     Status status = FindFiles();
72     if (status.ok()) {
73       ConvertLogFilesToTables();
74       ExtractMetaData();
75       status = WriteDescriptor();
76     }
77     if (status.ok()) {
78       unsigned long long bytes = 0;
79       for (size_t i = 0; i < tables_.size(); i++) {
80         bytes += tables_[i].meta.file_size;
81       }
82       Log(options_.info_log,
83           "**** Repaired leveldb %s; "
84           "recovered %d files; %llu bytes. "
85           "Some data may have been lost. "
86           "****",
87           dbname_.c_str(), static_cast<int>(tables_.size()), bytes);
88     }
89     return status;
90   }
91 
92  private:
93   struct TableInfo {
94     FileMetaData meta;
95     SequenceNumber max_sequence;
96   };
97 
FindFiles()98   Status FindFiles() {
99     std::vector<std::string> filenames;
100     Status status = env_->GetChildren(dbname_, &filenames);
101     if (!status.ok()) {
102       return status;
103     }
104     if (filenames.empty()) {
105       return Status::IOError(dbname_, "repair found no files");
106     }
107 
108     uint64_t number;
109     FileType type;
110     for (size_t i = 0; i < filenames.size(); i++) {
111       if (ParseFileName(filenames[i], &number, &type)) {
112         if (type == kDescriptorFile) {
113           manifests_.push_back(filenames[i]);
114         } else {
115           if (number + 1 > next_file_number_) {
116             next_file_number_ = number + 1;
117           }
118           if (type == kLogFile) {
119             logs_.push_back(number);
120           } else if (type == kTableFile) {
121             table_numbers_.push_back(number);
122           } else {
123             // Ignore other files
124           }
125         }
126       }
127     }
128     return status;
129   }
130 
ConvertLogFilesToTables()131   void ConvertLogFilesToTables() {
132     for (size_t i = 0; i < logs_.size(); i++) {
133       std::string logname = LogFileName(dbname_, logs_[i]);
134       Status status = ConvertLogToTable(logs_[i]);
135       if (!status.ok()) {
136         Log(options_.info_log, "Log #%llu: ignoring conversion error: %s",
137             (unsigned long long)logs_[i], status.ToString().c_str());
138       }
139       ArchiveFile(logname);
140     }
141   }
142 
ConvertLogToTable(uint64_t log)143   Status ConvertLogToTable(uint64_t log) {
144     struct LogReporter : public log::Reader::Reporter {
145       Env* env;
146       Logger* info_log;
147       uint64_t lognum;
148       void Corruption(size_t bytes, const Status& s) override {
149         // We print error messages for corruption, but continue repairing.
150         Log(info_log, "Log #%llu: dropping %d bytes; %s",
151             (unsigned long long)lognum, static_cast<int>(bytes),
152             s.ToString().c_str());
153       }
154     };
155 
156     // Open the log file
157     std::string logname = LogFileName(dbname_, log);
158     SequentialFile* lfile;
159     Status status = env_->NewSequentialFile(logname, &lfile);
160     if (!status.ok()) {
161       return status;
162     }
163 
164     // Create the log reader.
165     LogReporter reporter;
166     reporter.env = env_;
167     reporter.info_log = options_.info_log;
168     reporter.lognum = log;
169     // We intentionally make log::Reader do checksumming so that
170     // corruptions cause entire commits to be skipped instead of
171     // propagating bad information (like overly large sequence
172     // numbers).
173     log::Reader reader(lfile, &reporter, false /*do not checksum*/,
174                        0 /*initial_offset*/);
175 
176     // Read all the records and add to a memtable
177     std::string scratch;
178     Slice record;
179     WriteBatch batch;
180     MemTable* mem = new MemTable(icmp_);
181     mem->Ref();
182     int counter = 0;
183     while (reader.ReadRecord(&record, &scratch)) {
184       if (record.size() < 12) {
185         reporter.Corruption(record.size(),
186                             Status::Corruption("log record too small"));
187         continue;
188       }
189       WriteBatchInternal::SetContents(&batch, record);
190       status = WriteBatchInternal::InsertInto(&batch, mem);
191       if (status.ok()) {
192         counter += WriteBatchInternal::Count(&batch);
193       } else {
194         Log(options_.info_log, "Log #%llu: ignoring %s",
195             (unsigned long long)log, status.ToString().c_str());
196         status = Status::OK();  // Keep going with rest of file
197       }
198     }
199     delete lfile;
200 
201     // Do not record a version edit for this conversion to a Table
202     // since ExtractMetaData() will also generate edits.
203     FileMetaData meta;
204     meta.number = next_file_number_++;
205     Iterator* iter = mem->NewIterator();
206     status = BuildTable(dbname_, env_, options_, table_cache_, iter, &meta);
207     delete iter;
208     mem->Unref();
209     mem = nullptr;
210     if (status.ok()) {
211       if (meta.file_size > 0) {
212         table_numbers_.push_back(meta.number);
213       }
214     }
215     Log(options_.info_log, "Log #%llu: %d ops saved to Table #%llu %s",
216         (unsigned long long)log, counter, (unsigned long long)meta.number,
217         status.ToString().c_str());
218     return status;
219   }
220 
ExtractMetaData()221   void ExtractMetaData() {
222     for (size_t i = 0; i < table_numbers_.size(); i++) {
223       ScanTable(table_numbers_[i]);
224     }
225   }
226 
NewTableIterator(const FileMetaData & meta)227   Iterator* NewTableIterator(const FileMetaData& meta) {
228     // Same as compaction iterators: if paranoid_checks are on, turn
229     // on checksum verification.
230     ReadOptions r;
231     r.verify_checksums = options_.paranoid_checks;
232     return table_cache_->NewIterator(r, meta.number, meta.file_size);
233   }
234 
ScanTable(uint64_t number)235   void ScanTable(uint64_t number) {
236     TableInfo t;
237     t.meta.number = number;
238     std::string fname = TableFileName(dbname_, number);
239     Status status = env_->GetFileSize(fname, &t.meta.file_size);
240     if (!status.ok()) {
241       // Try alternate file name.
242       fname = SSTTableFileName(dbname_, number);
243       Status s2 = env_->GetFileSize(fname, &t.meta.file_size);
244       if (s2.ok()) {
245         status = Status::OK();
246       }
247     }
248     if (!status.ok()) {
249       ArchiveFile(TableFileName(dbname_, number));
250       ArchiveFile(SSTTableFileName(dbname_, number));
251       Log(options_.info_log, "Table #%llu: dropped: %s",
252           (unsigned long long)t.meta.number, status.ToString().c_str());
253       return;
254     }
255 
256     // Extract metadata by scanning through table.
257     int counter = 0;
258     Iterator* iter = NewTableIterator(t.meta);
259     bool empty = true;
260     ParsedInternalKey parsed;
261     t.max_sequence = 0;
262     for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
263       Slice key = iter->key();
264       if (!ParseInternalKey(key, &parsed)) {
265         Log(options_.info_log, "Table #%llu: unparsable key %s",
266             (unsigned long long)t.meta.number, EscapeString(key).c_str());
267         continue;
268       }
269 
270       counter++;
271       if (empty) {
272         empty = false;
273         t.meta.smallest.DecodeFrom(key);
274       }
275       t.meta.largest.DecodeFrom(key);
276       if (parsed.sequence > t.max_sequence) {
277         t.max_sequence = parsed.sequence;
278       }
279     }
280     if (!iter->status().ok()) {
281       status = iter->status();
282     }
283     delete iter;
284     Log(options_.info_log, "Table #%llu: %d entries %s",
285         (unsigned long long)t.meta.number, counter, status.ToString().c_str());
286 
287     if (status.ok()) {
288       tables_.push_back(t);
289     } else {
290       RepairTable(fname, t);  // RepairTable archives input file.
291     }
292   }
293 
RepairTable(const std::string & src,TableInfo t)294   void RepairTable(const std::string& src, TableInfo t) {
295     // We will copy src contents to a new table and then rename the
296     // new table over the source.
297 
298     // Create builder.
299     std::string copy = TableFileName(dbname_, next_file_number_++);
300     WritableFile* file;
301     Status s = env_->NewWritableFile(copy, &file);
302     if (!s.ok()) {
303       return;
304     }
305     TableBuilder* builder = new TableBuilder(options_, file);
306 
307     // Copy data.
308     Iterator* iter = NewTableIterator(t.meta);
309     int counter = 0;
310     for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
311       builder->Add(iter->key(), iter->value());
312       counter++;
313     }
314     delete iter;
315 
316     ArchiveFile(src);
317     if (counter == 0) {
318       builder->Abandon();  // Nothing to save
319     } else {
320       s = builder->Finish();
321       if (s.ok()) {
322         t.meta.file_size = builder->FileSize();
323       }
324     }
325     delete builder;
326     builder = nullptr;
327 
328     if (s.ok()) {
329       s = file->Close();
330     }
331     delete file;
332     file = nullptr;
333 
334     if (counter > 0 && s.ok()) {
335       std::string orig = TableFileName(dbname_, t.meta.number);
336       s = env_->RenameFile(copy, orig);
337       if (s.ok()) {
338         Log(options_.info_log, "Table #%llu: %d entries repaired",
339             (unsigned long long)t.meta.number, counter);
340         tables_.push_back(t);
341       }
342     }
343     if (!s.ok()) {
344       env_->RemoveFile(copy);
345     }
346   }
347 
WriteDescriptor()348   Status WriteDescriptor() {
349     std::string tmp = TempFileName(dbname_, 1);
350     WritableFile* file;
351     Status status = env_->NewWritableFile(tmp, &file);
352     if (!status.ok()) {
353       return status;
354     }
355 
356     SequenceNumber max_sequence = 0;
357     for (size_t i = 0; i < tables_.size(); i++) {
358       if (max_sequence < tables_[i].max_sequence) {
359         max_sequence = tables_[i].max_sequence;
360       }
361     }
362 
363     edit_.SetComparatorName(icmp_.user_comparator()->Name());
364     edit_.SetLogNumber(0);
365     edit_.SetNextFile(next_file_number_);
366     edit_.SetLastSequence(max_sequence);
367 
368     for (size_t i = 0; i < tables_.size(); i++) {
369       // TODO(opt): separate out into multiple levels
370       const TableInfo& t = tables_[i];
371       edit_.AddFile(0, t.meta.number, t.meta.file_size, t.meta.smallest,
372                     t.meta.largest);
373     }
374 
375     // std::fprintf(stderr,
376     //              "NewDescriptor:\n%s\n", edit_.DebugString().c_str());
377     {
378       log::Writer log(file);
379       std::string record;
380       edit_.EncodeTo(&record);
381       status = log.AddRecord(record);
382     }
383     if (status.ok()) {
384       status = file->Close();
385     }
386     delete file;
387     file = nullptr;
388 
389     if (!status.ok()) {
390       env_->RemoveFile(tmp);
391     } else {
392       // Discard older manifests
393       for (size_t i = 0; i < manifests_.size(); i++) {
394         ArchiveFile(dbname_ + "/" + manifests_[i]);
395       }
396 
397       // Install new manifest
398       status = env_->RenameFile(tmp, DescriptorFileName(dbname_, 1));
399       if (status.ok()) {
400         status = SetCurrentFile(env_, dbname_, 1);
401       } else {
402         env_->RemoveFile(tmp);
403       }
404     }
405     return status;
406   }
407 
ArchiveFile(const std::string & fname)408   void ArchiveFile(const std::string& fname) {
409     // Move into another directory.  E.g., for
410     //    dir/foo
411     // rename to
412     //    dir/lost/foo
413     const char* slash = strrchr(fname.c_str(), '/');
414     std::string new_dir;
415     if (slash != nullptr) {
416       new_dir.assign(fname.data(), slash - fname.data());
417     }
418     new_dir.append("/lost");
419     env_->CreateDir(new_dir);  // Ignore error
420     std::string new_file = new_dir;
421     new_file.append("/");
422     new_file.append((slash == nullptr) ? fname.c_str() : slash + 1);
423     Status s = env_->RenameFile(fname, new_file);
424     Log(options_.info_log, "Archiving %s: %s\n", fname.c_str(),
425         s.ToString().c_str());
426   }
427 
428   const std::string dbname_;
429   Env* const env_;
430   InternalKeyComparator const icmp_;
431   InternalFilterPolicy const ipolicy_;
432   const Options options_;
433   bool owns_info_log_;
434   bool owns_cache_;
435   TableCache* table_cache_;
436   VersionEdit edit_;
437 
438   std::vector<std::string> manifests_;
439   std::vector<uint64_t> table_numbers_;
440   std::vector<uint64_t> logs_;
441   std::vector<TableInfo> tables_;
442   uint64_t next_file_number_;
443 };
444 }  // namespace
445 
RepairDB(const std::string & dbname,const Options & options)446 Status RepairDB(const std::string& dbname, const Options& options) {
447   Repairer repairer(dbname, options);
448   return repairer.Run();
449 }
450 
451 }  // namespace leveldb
452