block.cc 7.9 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267
  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. // Decodes the blocks generated by block_builder.cc.
  6. #include "table/block.h"
  7. #include <algorithm>
  8. #include <cstdint>
  9. #include <vector>
  10. #include "leveldb/comparator.h"
  11. #include "table/format.h"
  12. #include "util/coding.h"
  13. #include "util/logging.h"
  14. namespace leveldb {
  15. inline uint32_t Block::NumRestarts() const {
  16. assert(size_ >= sizeof(uint32_t));
  17. return DecodeFixed32(data_ + size_ - sizeof(uint32_t));
  18. }
  19. Block::Block(const BlockContents& contents)
  20. : data_(contents.data.data()),
  21. size_(contents.data.size()),
  22. owned_(contents.heap_allocated) {
  23. if (size_ < sizeof(uint32_t)) {
  24. size_ = 0; // Error marker
  25. } else {
  26. size_t max_restarts_allowed = (size_ - sizeof(uint32_t)) / sizeof(uint32_t);
  27. if (NumRestarts() > max_restarts_allowed) {
  28. // The size is too small for NumRestarts()
  29. size_ = 0;
  30. } else {
  31. restart_offset_ = (uint32_t)size_ - (1 + NumRestarts()) * sizeof(uint32_t);
  32. }
  33. }
  34. }
  35. Block::~Block() {
  36. if (owned_) {
  37. delete[] data_;
  38. }
  39. }
  40. // Helper routine: decode the next block entry starting at "p",
  41. // storing the number of shared key bytes, non_shared key bytes,
  42. // and the length of the value in "*shared", "*non_shared", and
  43. // "*value_length", respectively. Will not dereference past "limit".
  44. //
  45. // If any errors are detected, returns nullptr. Otherwise, returns a
  46. // pointer to the key delta (just past the three decoded values).
  47. static inline const char* DecodeEntry(const char* p, const char* limit,
  48. uint32_t* shared, uint32_t* non_shared,
  49. uint32_t* value_length) {
  50. if (limit - p < 3) return nullptr;
  51. *shared = reinterpret_cast<const uint8_t*>(p)[0];
  52. *non_shared = reinterpret_cast<const uint8_t*>(p)[1];
  53. *value_length = reinterpret_cast<const uint8_t*>(p)[2];
  54. if ((*shared | *non_shared | *value_length) < 128) {
  55. // Fast path: all three values are encoded in one byte each
  56. p += 3;
  57. } else {
  58. if ((p = GetVarint32Ptr(p, limit, shared)) == nullptr) return nullptr;
  59. if ((p = GetVarint32Ptr(p, limit, non_shared)) == nullptr) return nullptr;
  60. if ((p = GetVarint32Ptr(p, limit, value_length)) == nullptr) return nullptr;
  61. }
  62. if (static_cast<uint32_t>(limit - p) < (*non_shared + *value_length)) {
  63. return nullptr;
  64. }
  65. return p;
  66. }
  67. class Block::Iter : public Iterator {
  68. private:
  69. const Comparator* const comparator_;
  70. const char* const data_; // underlying block contents
  71. uint32_t const restarts_; // Offset of restart array (list of fixed32)
  72. uint32_t const num_restarts_; // Number of uint32_t entries in restart array
  73. // current_ is offset in data_ of current entry. >= restarts_ if !Valid
  74. uint32_t current_;
  75. uint32_t restart_index_; // Index of restart block in which current_ falls
  76. std::string key_;
  77. Slice value_;
  78. Status status_;
  79. inline int Compare(const Slice& a, const Slice& b) const {
  80. return comparator_->Compare(a, b);
  81. }
  82. // Return the offset in data_ just past the end of the current entry.
  83. inline uint32_t NextEntryOffset() const {
  84. return (uint32_t)((value_.data() + value_.size()) - data_);
  85. }
  86. uint32_t GetRestartPoint(uint32_t index) {
  87. assert(index < num_restarts_);
  88. return DecodeFixed32(data_ + restarts_ + index * sizeof(uint32_t));
  89. }
  90. void SeekToRestartPoint(uint32_t index) {
  91. key_.clear();
  92. restart_index_ = index;
  93. // current_ will be fixed by ParseNextKey();
  94. // ParseNextKey() starts at the end of value_, so set value_ accordingly
  95. uint32_t offset = GetRestartPoint(index);
  96. value_ = Slice(data_ + offset, 0);
  97. }
  98. public:
  99. Iter(const Comparator* comparator, const char* data, uint32_t restarts,
  100. uint32_t num_restarts)
  101. : comparator_(comparator),
  102. data_(data),
  103. restarts_(restarts),
  104. num_restarts_(num_restarts),
  105. current_(restarts_),
  106. restart_index_(num_restarts_) {
  107. assert(num_restarts_ > 0);
  108. }
  109. bool Valid() const override { return current_ < restarts_; }
  110. Status status() const override { return status_; }
  111. Slice key() const override {
  112. assert(Valid());
  113. return key_;
  114. }
  115. Slice value() const override {
  116. assert(Valid());
  117. return value_;
  118. }
  119. void Next() override {
  120. assert(Valid());
  121. ParseNextKey();
  122. }
  123. void Prev() override {
  124. assert(Valid());
  125. // Scan backwards to a restart point before current_
  126. const uint32_t original = current_;
  127. while (GetRestartPoint(restart_index_) >= original) {
  128. if (restart_index_ == 0) {
  129. // No more entries
  130. current_ = restarts_;
  131. restart_index_ = num_restarts_;
  132. return;
  133. }
  134. restart_index_--;
  135. }
  136. SeekToRestartPoint(restart_index_);
  137. do {
  138. // Loop until end of current entry hits the start of original entry
  139. } while (ParseNextKey() && NextEntryOffset() < original);
  140. }
  141. void Seek(const Slice& target) override {
  142. // Binary search in restart array to find the last restart point
  143. // with a key < target
  144. uint32_t left = 0;
  145. uint32_t right = num_restarts_ - 1;
  146. while (left < right) {
  147. uint32_t mid = (left + right + 1) / 2;
  148. uint32_t region_offset = GetRestartPoint(mid);
  149. uint32_t shared, non_shared, value_length;
  150. const char* key_ptr =
  151. DecodeEntry(data_ + region_offset, data_ + restarts_, &shared,
  152. &non_shared, &value_length);
  153. if (key_ptr == nullptr || (shared != 0)) {
  154. CorruptionError();
  155. return;
  156. }
  157. Slice mid_key(key_ptr, non_shared);
  158. if (Compare(mid_key, target) < 0) {
  159. // Key at "mid" is smaller than "target". Therefore all
  160. // blocks before "mid" are uninteresting.
  161. left = mid;
  162. } else {
  163. // Key at "mid" is >= "target". Therefore all blocks at or
  164. // after "mid" are uninteresting.
  165. right = mid - 1;
  166. }
  167. }
  168. // Linear search (within restart block) for first key >= target
  169. SeekToRestartPoint(left);
  170. while (true) {
  171. if (!ParseNextKey()) {
  172. return;
  173. }
  174. if (Compare(key_, target) >= 0) {
  175. return;
  176. }
  177. }
  178. }
  179. void SeekToFirst() override {
  180. SeekToRestartPoint(0);
  181. ParseNextKey();
  182. }
  183. void SeekToLast() override {
  184. SeekToRestartPoint(num_restarts_ - 1);
  185. while (ParseNextKey() && NextEntryOffset() < restarts_) {
  186. // Keep skipping
  187. }
  188. }
  189. private:
  190. void CorruptionError() {
  191. current_ = restarts_;
  192. restart_index_ = num_restarts_;
  193. status_ = Status::Corruption("bad entry in block");
  194. key_.clear();
  195. value_.clear();
  196. }
  197. bool ParseNextKey() {
  198. current_ = NextEntryOffset();
  199. const char* p = data_ + current_;
  200. const char* limit = data_ + restarts_; // Restarts come right after data
  201. if (p >= limit) {
  202. // No more entries to return. Mark as invalid.
  203. current_ = restarts_;
  204. restart_index_ = num_restarts_;
  205. return false;
  206. }
  207. // Decode next entry
  208. uint32_t shared, non_shared, value_length;
  209. p = DecodeEntry(p, limit, &shared, &non_shared, &value_length);
  210. if (p == nullptr || key_.size() < shared) {
  211. CorruptionError();
  212. return false;
  213. } else {
  214. key_.resize(shared);
  215. key_.append(p, non_shared);
  216. value_ = Slice(p + non_shared, value_length);
  217. while (restart_index_ + 1 < num_restarts_ &&
  218. GetRestartPoint(restart_index_ + 1) < current_) {
  219. ++restart_index_;
  220. }
  221. return true;
  222. }
  223. }
  224. };
  225. Iterator* Block::NewIterator(const Comparator* comparator) {
  226. if (size_ < sizeof(uint32_t)) {
  227. return NewErrorIterator(Status::Corruption("bad block contents"));
  228. }
  229. const uint32_t num_restarts = NumRestarts();
  230. if (num_restarts == 0) {
  231. return NewEmptyIterator();
  232. } else {
  233. return new Iter(comparator, data_, restart_offset_, num_restarts);
  234. }
  235. }
  236. } // namespace leveldb