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.
5 #include "leveldb/table.h"
7 #include "leveldb/cache.h"
8 #include "leveldb/comparator.h"
9 #include "leveldb/env.h"
10 #include "leveldb/filter_policy.h"
11 #include "leveldb/options.h"
12 #include "table/block.h"
13 #include "table/filter_block.h"
14 #include "table/format.h"
15 #include "table/two_level_iterator.h"
16 #include "util/coding.h"
23 delete [] filter_data;
29 RandomAccessFile* file;
31 FilterBlockReader* filter;
32 const char* filter_data;
34 BlockHandle metaindex_handle; // Handle to metaindex_block: saved from footer
38 Status Table::Open(const Options& options,
39 RandomAccessFile* file,
43 if (size < Footer::kEncodedLength) {
44 return Status::InvalidArgument("file is too short to be an sstable");
47 char footer_space[Footer::kEncodedLength];
49 Status s = file->Read(size - Footer::kEncodedLength, Footer::kEncodedLength,
50 &footer_input, footer_space);
51 if (!s.ok()) return s;
54 s = footer.DecodeFrom(&footer_input);
55 if (!s.ok()) return s;
57 // Read the index block
58 BlockContents contents;
59 Block* index_block = NULL;
61 s = ReadBlock(file, ReadOptions(), footer.index_handle(), &contents);
63 index_block = new Block(contents);
68 // We've successfully read the footer and the index block: we're
69 // ready to serve requests.
70 Rep* rep = new Table::Rep;
71 rep->options = options;
73 rep->metaindex_handle = footer.metaindex_handle();
74 rep->index_block = index_block;
75 rep->cache_id = (options.block_cache ? options.block_cache->NewId() : 0);
76 rep->filter_data = NULL;
78 *table = new Table(rep);
79 (*table)->ReadMeta(footer);
81 if (index_block) delete index_block;
87 void Table::ReadMeta(const Footer& footer) {
88 if (rep_->options.filter_policy == NULL) {
89 return; // Do not need any metadata
92 // TODO(sanjay): Skip this if footer.metaindex_handle() size indicates
93 // it is an empty block.
95 BlockContents contents;
96 if (!ReadBlock(rep_->file, opt, footer.metaindex_handle(), &contents).ok()) {
97 // Do not propagate errors since meta info is not needed for operation
100 Block* meta = new Block(contents);
102 Iterator* iter = meta->NewIterator(BytewiseComparator());
103 std::string key = "filter.";
104 key.append(rep_->options.filter_policy->Name());
106 if (iter->Valid() && iter->key() == Slice(key)) {
107 ReadFilter(iter->value());
113 void Table::ReadFilter(const Slice& filter_handle_value) {
114 Slice v = filter_handle_value;
115 BlockHandle filter_handle;
116 if (!filter_handle.DecodeFrom(&v).ok()) {
120 // We might want to unify with ReadBlock() if we start
121 // requiring checksum verification in Table::Open.
124 if (!ReadBlock(rep_->file, opt, filter_handle, &block).ok()) {
127 if (block.heap_allocated) {
128 rep_->filter_data = block.data.data(); // Will need to delete later
130 rep_->filter = new FilterBlockReader(rep_->options.filter_policy, block.data);
137 static void DeleteBlock(void* arg, void* ignored) {
138 delete reinterpret_cast<Block*>(arg);
141 static void DeleteCachedBlock(const Slice& key, void* value) {
142 Block* block = reinterpret_cast<Block*>(value);
146 static void ReleaseBlock(void* arg, void* h) {
147 Cache* cache = reinterpret_cast<Cache*>(arg);
148 Cache::Handle* handle = reinterpret_cast<Cache::Handle*>(h);
149 cache->Release(handle);
152 // Convert an index iterator value (i.e., an encoded BlockHandle)
153 // into an iterator over the contents of the corresponding block.
154 Iterator* Table::BlockReader(void* arg,
155 const ReadOptions& options,
156 const Slice& index_value) {
157 Table* table = reinterpret_cast<Table*>(arg);
158 Cache* block_cache = table->rep_->options.block_cache;
160 Cache::Handle* cache_handle = NULL;
163 Slice input = index_value;
164 Status s = handle.DecodeFrom(&input);
165 // We intentionally allow extra stuff in index_value so that we
166 // can add more features in the future.
169 BlockContents contents;
170 if (block_cache != NULL) {
171 char cache_key_buffer[16];
172 EncodeFixed64(cache_key_buffer, table->rep_->cache_id);
173 EncodeFixed64(cache_key_buffer+8, handle.offset());
174 Slice key(cache_key_buffer, sizeof(cache_key_buffer));
175 cache_handle = block_cache->Lookup(key);
176 if (cache_handle != NULL) {
177 block = reinterpret_cast<Block*>(block_cache->Value(cache_handle));
179 s = ReadBlock(table->rep_->file, options, handle, &contents);
181 block = new Block(contents);
182 if (contents.cachable && options.fill_cache) {
183 cache_handle = block_cache->Insert(
184 key, block, block->size(), &DeleteCachedBlock);
189 s = ReadBlock(table->rep_->file, options, handle, &contents);
191 block = new Block(contents);
198 iter = block->NewIterator(table->rep_->options.comparator);
199 if (cache_handle == NULL) {
200 iter->RegisterCleanup(&DeleteBlock, block, NULL);
202 iter->RegisterCleanup(&ReleaseBlock, block_cache, cache_handle);
205 iter = NewErrorIterator(s);
210 Iterator* Table::NewIterator(const ReadOptions& options) const {
211 return NewTwoLevelIterator(
212 rep_->index_block->NewIterator(rep_->options.comparator),
213 &Table::BlockReader, const_cast<Table*>(this), options);
216 Status Table::InternalGet(const ReadOptions& options, const Slice& k,
218 void (*saver)(void*, const Slice&, const Slice&)) {
220 Iterator* iiter = rep_->index_block->NewIterator(rep_->options.comparator);
222 if (iiter->Valid()) {
223 Slice handle_value = iiter->value();
224 FilterBlockReader* filter = rep_->filter;
226 if (filter != NULL &&
227 handle.DecodeFrom(&handle_value).ok() &&
228 !filter->KeyMayMatch(handle.offset(), k)) {
231 Slice handle = iiter->value();
232 Iterator* block_iter = BlockReader(this, options, iiter->value());
234 if (block_iter->Valid()) {
235 (*saver)(arg, block_iter->key(), block_iter->value());
237 s = block_iter->status();
249 uint64_t Table::ApproximateOffsetOf(const Slice& key) const {
250 Iterator* index_iter =
251 rep_->index_block->NewIterator(rep_->options.comparator);
252 index_iter->Seek(key);
254 if (index_iter->Valid()) {
256 Slice input = index_iter->value();
257 Status s = handle.DecodeFrom(&input);
259 result = handle.offset();
261 // Strange: we can't decode the block handle in the index block.
262 // We'll just return the offset of the metaindex block, which is
263 // close to the whole file size for this case.
264 result = rep_->metaindex_handle.offset();
267 // key is past the last key in the file. Approximate the offset
268 // by returning the offset of the metaindex block (which is
269 // right near the end of the file).
270 result = rep_->metaindex_handle.offset();
276 } // namespace leveldb