Heap.cpp 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343
  1. /*
  2. * Copyright (c) 2021, Jan de Visser <jan@de-visser.net>
  3. * Copyright (c) 2023, Jelle Raaijmakers <jelle@gmta.nl>
  4. *
  5. * SPDX-License-Identifier: BSD-2-Clause
  6. */
  7. #include <AK/DeprecatedString.h>
  8. #include <AK/Format.h>
  9. #include <AK/QuickSort.h>
  10. #include <LibCore/System.h>
  11. #include <LibSQL/Heap.h>
  12. #include <sys/stat.h>
  13. namespace SQL {
  14. Heap::Heap(DeprecatedString file_name)
  15. {
  16. set_name(move(file_name));
  17. }
  18. Heap::~Heap()
  19. {
  20. if (m_file && !m_write_ahead_log.is_empty()) {
  21. if (auto maybe_error = flush(); maybe_error.is_error())
  22. warnln("~Heap({}): {}", name(), maybe_error.error());
  23. }
  24. }
  25. ErrorOr<void> Heap::open()
  26. {
  27. size_t file_size = 0;
  28. struct stat stat_buffer;
  29. if (stat(name().characters(), &stat_buffer) != 0) {
  30. if (errno != ENOENT) {
  31. warnln("Heap::open({}): could not stat: {}"sv, name(), strerror(errno));
  32. return Error::from_string_literal("Heap::open(): could not stat file");
  33. }
  34. } else if (!S_ISREG(stat_buffer.st_mode)) {
  35. warnln("Heap::open({}): can only use regular files"sv, name());
  36. return Error::from_string_literal("Heap::open(): can only use regular files");
  37. } else {
  38. file_size = stat_buffer.st_size;
  39. }
  40. if (file_size > 0) {
  41. m_next_block = file_size / Block::SIZE;
  42. m_highest_block_written = m_next_block - 1;
  43. }
  44. auto file = TRY(Core::File::open(name(), Core::File::OpenMode::ReadWrite));
  45. m_file = TRY(Core::InputBufferedFile::create(move(file)));
  46. if (file_size > 0) {
  47. if (auto error_maybe = read_zero_block(); error_maybe.is_error()) {
  48. m_file = nullptr;
  49. return error_maybe.release_error();
  50. }
  51. } else {
  52. TRY(initialize_zero_block());
  53. }
  54. // FIXME: We should more gracefully handle version incompatibilities. For now, we drop the database.
  55. if (m_version != VERSION) {
  56. dbgln_if(SQL_DEBUG, "Heap file {} opened has incompatible version {}. Deleting for version {}.", name(), m_version, VERSION);
  57. m_file = nullptr;
  58. TRY(Core::System::unlink(name()));
  59. return open();
  60. }
  61. // Perform a heap scan to find all free blocks
  62. // FIXME: this is very inefficient; store free blocks in a persistent heap structure
  63. for (Block::Index index = 1; index <= m_highest_block_written; ++index) {
  64. auto block_data = TRY(read_raw_block(index));
  65. auto size_in_bytes = *reinterpret_cast<u32*>(block_data.data());
  66. if (size_in_bytes == 0)
  67. TRY(m_free_block_indices.try_append(index));
  68. }
  69. dbgln_if(SQL_DEBUG, "Heap file {} opened; number of blocks = {}; free blocks = {}", name(), m_highest_block_written, m_free_block_indices.size());
  70. return {};
  71. }
  72. ErrorOr<size_t> Heap::file_size_in_bytes() const
  73. {
  74. TRY(m_file->seek(0, SeekMode::FromEndPosition));
  75. return TRY(m_file->tell());
  76. }
  77. bool Heap::has_block(Block::Index index) const
  78. {
  79. return (index <= m_highest_block_written || m_write_ahead_log.contains(index))
  80. && !m_free_block_indices.contains_slow(index);
  81. }
  82. Block::Index Heap::request_new_block_index()
  83. {
  84. if (!m_free_block_indices.is_empty())
  85. return m_free_block_indices.take_last();
  86. return m_next_block++;
  87. }
  88. ErrorOr<ByteBuffer> Heap::read_storage(Block::Index index)
  89. {
  90. dbgln_if(SQL_DEBUG, "{}({})", __FUNCTION__, index);
  91. // Reconstruct the data storage from a potential chain of blocks
  92. ByteBuffer data;
  93. while (index > 0) {
  94. auto block = TRY(read_block(index));
  95. dbgln_if(SQL_DEBUG, " -> {} bytes", block.size_in_bytes());
  96. TRY(data.try_append(block.data().bytes().slice(0, block.size_in_bytes())));
  97. index = block.next_block();
  98. }
  99. return data;
  100. }
  101. ErrorOr<void> Heap::write_storage(Block::Index index, ReadonlyBytes data)
  102. {
  103. dbgln_if(SQL_DEBUG, "{}({}, {} bytes)", __FUNCTION__, index, data.size());
  104. VERIFY(index > 0);
  105. VERIFY(data.size() > 0);
  106. // Split up the storage across multiple blocks if necessary, creating a chain
  107. u32 remaining_size = static_cast<u32>(data.size());
  108. u32 offset_in_data = 0;
  109. Block::Index existing_next_block_index = 0;
  110. while (remaining_size > 0) {
  111. auto block_data_size = AK::min(remaining_size, Block::DATA_SIZE);
  112. remaining_size -= block_data_size;
  113. ByteBuffer block_data;
  114. if (has_block(index)) {
  115. auto existing_block = TRY(read_block(index));
  116. block_data = existing_block.data();
  117. TRY(block_data.try_resize(block_data_size));
  118. existing_next_block_index = existing_block.next_block();
  119. } else {
  120. block_data = TRY(ByteBuffer::create_uninitialized(block_data_size));
  121. existing_next_block_index = 0;
  122. }
  123. Block::Index next_block_index = existing_next_block_index;
  124. if (next_block_index == 0 && remaining_size > 0)
  125. next_block_index = request_new_block_index();
  126. else if (remaining_size == 0)
  127. next_block_index = 0;
  128. block_data.bytes().overwrite(0, data.offset(offset_in_data), block_data_size);
  129. TRY(write_block({ index, block_data_size, next_block_index, move(block_data) }));
  130. index = next_block_index;
  131. offset_in_data += block_data_size;
  132. }
  133. // Free remaining blocks in existing chain, if any
  134. while (existing_next_block_index > 0) {
  135. auto existing_block = TRY(read_block(existing_next_block_index));
  136. existing_next_block_index = existing_block.next_block();
  137. TRY(free_block(existing_block));
  138. }
  139. return {};
  140. }
  141. ErrorOr<ByteBuffer> Heap::read_raw_block(Block::Index index)
  142. {
  143. VERIFY(m_file);
  144. VERIFY(index < m_next_block);
  145. if (auto wal_entry = m_write_ahead_log.get(index); wal_entry.has_value())
  146. return wal_entry.value();
  147. TRY(m_file->seek(index * Block::SIZE, SeekMode::SetPosition));
  148. auto buffer = TRY(ByteBuffer::create_uninitialized(Block::SIZE));
  149. TRY(m_file->read_until_filled(buffer));
  150. return buffer;
  151. }
  152. ErrorOr<Block> Heap::read_block(Block::Index index)
  153. {
  154. dbgln_if(SQL_DEBUG, "Read heap block {}", index);
  155. auto buffer = TRY(read_raw_block(index));
  156. auto size_in_bytes = *reinterpret_cast<u32*>(buffer.offset_pointer(0));
  157. auto next_block = *reinterpret_cast<Block::Index*>(buffer.offset_pointer(sizeof(u32)));
  158. auto data = TRY(buffer.slice(Block::HEADER_SIZE, Block::DATA_SIZE));
  159. return Block { index, size_in_bytes, next_block, move(data) };
  160. }
  161. ErrorOr<void> Heap::write_raw_block(Block::Index index, ReadonlyBytes data)
  162. {
  163. dbgln_if(SQL_DEBUG, "Write raw block {}", index);
  164. VERIFY(m_file);
  165. VERIFY(data.size() == Block::SIZE);
  166. TRY(m_file->seek(index * Block::SIZE, SeekMode::SetPosition));
  167. TRY(m_file->write_until_depleted(data));
  168. if (index > m_highest_block_written)
  169. m_highest_block_written = index;
  170. return {};
  171. }
  172. ErrorOr<void> Heap::write_raw_block_to_wal(Block::Index index, ByteBuffer&& data)
  173. {
  174. dbgln_if(SQL_DEBUG, "{}(): adding raw block {} to WAL", __FUNCTION__, index);
  175. VERIFY(index < m_next_block);
  176. VERIFY(data.size() == Block::SIZE);
  177. TRY(m_write_ahead_log.try_set(index, move(data)));
  178. return {};
  179. }
  180. ErrorOr<void> Heap::write_block(Block const& block)
  181. {
  182. VERIFY(block.index() < m_next_block);
  183. VERIFY(block.next_block() < m_next_block);
  184. VERIFY(block.size_in_bytes() > 0);
  185. VERIFY(block.data().size() <= Block::DATA_SIZE);
  186. auto size_in_bytes = block.size_in_bytes();
  187. auto next_block = block.next_block();
  188. auto heap_data = TRY(ByteBuffer::create_zeroed(Block::SIZE));
  189. heap_data.overwrite(0, &size_in_bytes, sizeof(size_in_bytes));
  190. heap_data.overwrite(sizeof(size_in_bytes), &next_block, sizeof(next_block));
  191. block.data().bytes().copy_to(heap_data.bytes().slice(Block::HEADER_SIZE));
  192. return write_raw_block_to_wal(block.index(), move(heap_data));
  193. }
  194. ErrorOr<void> Heap::free_block(Block const& block)
  195. {
  196. auto index = block.index();
  197. dbgln_if(SQL_DEBUG, "{}({})", __FUNCTION__, index);
  198. VERIFY(index > 0);
  199. VERIFY(has_block(index));
  200. // Zero out freed blocks to facilitate a free block scan upon opening the database later
  201. auto zeroed_data = TRY(ByteBuffer::create_zeroed(Block::SIZE));
  202. TRY(write_raw_block_to_wal(index, move(zeroed_data)));
  203. return m_free_block_indices.try_append(index);
  204. }
  205. ErrorOr<void> Heap::flush()
  206. {
  207. VERIFY(m_file);
  208. auto indices = m_write_ahead_log.keys();
  209. quick_sort(indices);
  210. for (auto index : indices) {
  211. dbgln_if(SQL_DEBUG, "Flushing block {}", index);
  212. auto& data = m_write_ahead_log.get(index).value();
  213. TRY(write_raw_block(index, data));
  214. }
  215. m_write_ahead_log.clear();
  216. dbgln_if(SQL_DEBUG, "WAL flushed; new number of blocks = {}", m_highest_block_written);
  217. return {};
  218. }
  219. constexpr static auto FILE_ID = "SerenitySQL "sv;
  220. constexpr static auto VERSION_OFFSET = FILE_ID.length();
  221. constexpr static auto SCHEMAS_ROOT_OFFSET = VERSION_OFFSET + sizeof(u32);
  222. constexpr static auto TABLES_ROOT_OFFSET = SCHEMAS_ROOT_OFFSET + sizeof(u32);
  223. constexpr static auto TABLE_COLUMNS_ROOT_OFFSET = TABLES_ROOT_OFFSET + sizeof(u32);
  224. constexpr static auto USER_VALUES_OFFSET = TABLE_COLUMNS_ROOT_OFFSET + sizeof(u32);
  225. ErrorOr<void> Heap::read_zero_block()
  226. {
  227. dbgln_if(SQL_DEBUG, "Read zero block from {}", name());
  228. auto block = TRY(read_raw_block(0));
  229. auto file_id_buffer = TRY(block.slice(0, FILE_ID.length()));
  230. auto file_id = StringView(file_id_buffer);
  231. if (file_id != FILE_ID) {
  232. warnln("{}: Zero page corrupt. This is probably not a {} heap file"sv, name(), FILE_ID);
  233. return Error::from_string_literal("Heap()::read_zero_block(): Zero page corrupt. This is probably not a SerenitySQL heap file");
  234. }
  235. memcpy(&m_version, block.offset_pointer(VERSION_OFFSET), sizeof(u32));
  236. dbgln_if(SQL_DEBUG, "Version: {}.{}", (m_version & 0xFFFF0000) >> 16, (m_version & 0x0000FFFF));
  237. memcpy(&m_schemas_root, block.offset_pointer(SCHEMAS_ROOT_OFFSET), sizeof(u32));
  238. dbgln_if(SQL_DEBUG, "Schemas root node: {}", m_schemas_root);
  239. memcpy(&m_tables_root, block.offset_pointer(TABLES_ROOT_OFFSET), sizeof(u32));
  240. dbgln_if(SQL_DEBUG, "Tables root node: {}", m_tables_root);
  241. memcpy(&m_table_columns_root, block.offset_pointer(TABLE_COLUMNS_ROOT_OFFSET), sizeof(u32));
  242. dbgln_if(SQL_DEBUG, "Table columns root node: {}", m_table_columns_root);
  243. memcpy(m_user_values.data(), block.offset_pointer(USER_VALUES_OFFSET), m_user_values.size() * sizeof(u32));
  244. for (auto ix = 0u; ix < m_user_values.size(); ix++) {
  245. if (m_user_values[ix])
  246. dbgln_if(SQL_DEBUG, "User value {}: {}", ix, m_user_values[ix]);
  247. }
  248. return {};
  249. }
  250. ErrorOr<void> Heap::update_zero_block()
  251. {
  252. dbgln_if(SQL_DEBUG, "Write zero block to {}", name());
  253. dbgln_if(SQL_DEBUG, "Version: {}.{}", (m_version & 0xFFFF0000) >> 16, (m_version & 0x0000FFFF));
  254. dbgln_if(SQL_DEBUG, "Schemas root node: {}", m_schemas_root);
  255. dbgln_if(SQL_DEBUG, "Tables root node: {}", m_tables_root);
  256. dbgln_if(SQL_DEBUG, "Table Columns root node: {}", m_table_columns_root);
  257. for (auto ix = 0u; ix < m_user_values.size(); ix++) {
  258. if (m_user_values[ix] > 0)
  259. dbgln_if(SQL_DEBUG, "User value {}: {}", ix, m_user_values[ix]);
  260. }
  261. auto buffer = TRY(ByteBuffer::create_zeroed(Block::SIZE));
  262. auto buffer_bytes = buffer.bytes();
  263. buffer_bytes.overwrite(0, FILE_ID.characters_without_null_termination(), FILE_ID.length());
  264. buffer_bytes.overwrite(VERSION_OFFSET, &m_version, sizeof(u32));
  265. buffer_bytes.overwrite(SCHEMAS_ROOT_OFFSET, &m_schemas_root, sizeof(u32));
  266. buffer_bytes.overwrite(TABLES_ROOT_OFFSET, &m_tables_root, sizeof(u32));
  267. buffer_bytes.overwrite(TABLE_COLUMNS_ROOT_OFFSET, &m_table_columns_root, sizeof(u32));
  268. buffer_bytes.overwrite(USER_VALUES_OFFSET, m_user_values.data(), m_user_values.size() * sizeof(u32));
  269. return write_raw_block_to_wal(0, move(buffer));
  270. }
  271. ErrorOr<void> Heap::initialize_zero_block()
  272. {
  273. m_version = VERSION;
  274. m_schemas_root = 0;
  275. m_tables_root = 0;
  276. m_table_columns_root = 0;
  277. m_next_block = 1;
  278. for (auto& user : m_user_values)
  279. user = 0u;
  280. return update_zero_block();
  281. }
  282. }