Ext2FileSystem.cpp 68 KB

12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721
  1. /*
  2. * Copyright (c) 2018-2020, Andreas Kling <kling@serenityos.org>
  3. * Copyright (c) 2021, sin-ack <sin-ack@protonmail.com>
  4. *
  5. * SPDX-License-Identifier: BSD-2-Clause
  6. */
  7. #include <AK/HashMap.h>
  8. #include <AK/MemoryStream.h>
  9. #include <AK/StdLibExtras.h>
  10. #include <AK/StringView.h>
  11. #include <Kernel/Debug.h>
  12. #include <Kernel/Devices/BlockDevice.h>
  13. #include <Kernel/FileSystem/Ext2FileSystem.h>
  14. #include <Kernel/FileSystem/OpenFileDescription.h>
  15. #include <Kernel/FileSystem/ext2_fs.h>
  16. #include <Kernel/Process.h>
  17. #include <Kernel/UnixTypes.h>
  18. #include <LibC/errno_numbers.h>
  19. namespace Kernel {
  20. static constexpr size_t max_block_size = 4096;
  21. static constexpr size_t max_inline_symlink_length = 60;
  22. struct Ext2FSDirectoryEntry {
  23. String name;
  24. InodeIndex inode_index { 0 };
  25. u8 file_type { 0 };
  26. u16 record_length { 0 };
  27. };
  28. static u8 to_ext2_file_type(mode_t mode)
  29. {
  30. if (is_regular_file(mode))
  31. return EXT2_FT_REG_FILE;
  32. if (is_directory(mode))
  33. return EXT2_FT_DIR;
  34. if (is_character_device(mode))
  35. return EXT2_FT_CHRDEV;
  36. if (is_block_device(mode))
  37. return EXT2_FT_BLKDEV;
  38. if (is_fifo(mode))
  39. return EXT2_FT_FIFO;
  40. if (is_socket(mode))
  41. return EXT2_FT_SOCK;
  42. if (is_symlink(mode))
  43. return EXT2_FT_SYMLINK;
  44. return EXT2_FT_UNKNOWN;
  45. }
  46. static unsigned divide_rounded_up(unsigned a, unsigned b)
  47. {
  48. return (a / b) + (a % b != 0);
  49. }
  50. ErrorOr<NonnullRefPtr<Ext2FS>> Ext2FS::try_create(OpenFileDescription& file_description)
  51. {
  52. return adopt_nonnull_ref_or_enomem(new (nothrow) Ext2FS(file_description));
  53. }
  54. Ext2FS::Ext2FS(OpenFileDescription& file_description)
  55. : BlockBasedFileSystem(file_description)
  56. {
  57. }
  58. Ext2FS::~Ext2FS()
  59. {
  60. }
  61. bool Ext2FS::flush_super_block()
  62. {
  63. MutexLocker locker(m_lock);
  64. VERIFY((sizeof(ext2_super_block) % logical_block_size()) == 0);
  65. auto super_block_buffer = UserOrKernelBuffer::for_kernel_buffer((u8*)&m_super_block);
  66. bool success = raw_write_blocks(2, (sizeof(ext2_super_block) / logical_block_size()), super_block_buffer);
  67. VERIFY(success);
  68. return true;
  69. }
  70. const ext2_group_desc& Ext2FS::group_descriptor(GroupIndex group_index) const
  71. {
  72. // FIXME: Should this fail gracefully somehow?
  73. VERIFY(group_index <= m_block_group_count);
  74. VERIFY(group_index > 0);
  75. return block_group_descriptors()[group_index.value() - 1];
  76. }
  77. ErrorOr<void> Ext2FS::initialize()
  78. {
  79. MutexLocker locker(m_lock);
  80. VERIFY((sizeof(ext2_super_block) % logical_block_size()) == 0);
  81. auto super_block_buffer = UserOrKernelBuffer::for_kernel_buffer((u8*)&m_super_block);
  82. bool success = raw_read_blocks(2, (sizeof(ext2_super_block) / logical_block_size()), super_block_buffer);
  83. VERIFY(success);
  84. auto const& super_block = this->super_block();
  85. if constexpr (EXT2_DEBUG) {
  86. dmesgln("Ext2FS: super block magic: {:04x} (super block size: {})", super_block.s_magic, sizeof(ext2_super_block));
  87. }
  88. if (super_block.s_magic != EXT2_SUPER_MAGIC) {
  89. dmesgln("Ext2FS: Bad super block magic");
  90. return EINVAL;
  91. }
  92. if constexpr (EXT2_DEBUG) {
  93. dmesgln("Ext2FS: {} inodes, {} blocks", super_block.s_inodes_count, super_block.s_blocks_count);
  94. dmesgln("Ext2FS: Block size: {}", EXT2_BLOCK_SIZE(&super_block));
  95. dmesgln("Ext2FS: First data block: {}", super_block.s_first_data_block);
  96. dmesgln("Ext2FS: Inodes per block: {}", inodes_per_block());
  97. dmesgln("Ext2FS: Inodes per group: {}", inodes_per_group());
  98. dmesgln("Ext2FS: Free inodes: {}", super_block.s_free_inodes_count);
  99. dmesgln("Ext2FS: Descriptors per block: {}", EXT2_DESC_PER_BLOCK(&super_block));
  100. dmesgln("Ext2FS: Descriptor size: {}", EXT2_DESC_SIZE(&super_block));
  101. }
  102. set_block_size(EXT2_BLOCK_SIZE(&super_block));
  103. set_fragment_size(EXT2_FRAG_SIZE(&super_block));
  104. // Note: This depends on the block size being available.
  105. TRY(BlockBasedFileSystem::initialize());
  106. VERIFY(block_size() <= (int)max_block_size);
  107. m_block_group_count = ceil_div(super_block.s_blocks_count, super_block.s_blocks_per_group);
  108. if (m_block_group_count == 0) {
  109. dmesgln("Ext2FS: no block groups :(");
  110. return EINVAL;
  111. }
  112. auto blocks_to_read = ceil_div(m_block_group_count * sizeof(ext2_group_desc), block_size());
  113. BlockIndex first_block_of_bgdt = block_size() == 1024 ? 2 : 1;
  114. m_cached_group_descriptor_table = TRY(KBuffer::try_create_with_size(block_size() * blocks_to_read, Memory::Region::Access::ReadWrite, "Ext2FS: Block group descriptors"));
  115. auto buffer = UserOrKernelBuffer::for_kernel_buffer(m_cached_group_descriptor_table->data());
  116. TRY(read_blocks(first_block_of_bgdt, blocks_to_read, buffer));
  117. if constexpr (EXT2_DEBUG) {
  118. for (unsigned i = 1; i <= m_block_group_count; ++i) {
  119. auto const& group = group_descriptor(i);
  120. dbgln("Ext2FS: group[{}] ( block_bitmap: {}, inode_bitmap: {}, inode_table: {} )", i, group.bg_block_bitmap, group.bg_inode_bitmap, group.bg_inode_table);
  121. }
  122. }
  123. m_root_inode = static_ptr_cast<Ext2FSInode>(TRY(get_inode({ fsid(), EXT2_ROOT_INO })));
  124. return {};
  125. }
  126. Ext2FSInode& Ext2FS::root_inode()
  127. {
  128. return *m_root_inode;
  129. }
  130. bool Ext2FS::find_block_containing_inode(InodeIndex inode, BlockIndex& block_index, unsigned& offset) const
  131. {
  132. auto const& super_block = this->super_block();
  133. if (inode != EXT2_ROOT_INO && inode < EXT2_FIRST_INO(&super_block))
  134. return false;
  135. if (inode > super_block.s_inodes_count)
  136. return false;
  137. auto const& bgd = group_descriptor(group_index_from_inode(inode));
  138. u64 full_offset = ((inode.value() - 1) % inodes_per_group()) * inode_size();
  139. block_index = bgd.bg_inode_table + (full_offset >> EXT2_BLOCK_SIZE_BITS(&super_block));
  140. offset = full_offset & (block_size() - 1);
  141. return true;
  142. }
  143. Ext2FS::BlockListShape Ext2FS::compute_block_list_shape(unsigned blocks) const
  144. {
  145. BlockListShape shape;
  146. const unsigned entries_per_block = EXT2_ADDR_PER_BLOCK(&super_block());
  147. unsigned blocks_remaining = blocks;
  148. shape.direct_blocks = min((unsigned)EXT2_NDIR_BLOCKS, blocks_remaining);
  149. blocks_remaining -= shape.direct_blocks;
  150. if (!blocks_remaining)
  151. return shape;
  152. shape.indirect_blocks = min(blocks_remaining, entries_per_block);
  153. shape.meta_blocks += 1;
  154. blocks_remaining -= shape.indirect_blocks;
  155. if (!blocks_remaining)
  156. return shape;
  157. shape.doubly_indirect_blocks = min(blocks_remaining, entries_per_block * entries_per_block);
  158. shape.meta_blocks += 1;
  159. shape.meta_blocks += divide_rounded_up(shape.doubly_indirect_blocks, entries_per_block);
  160. blocks_remaining -= shape.doubly_indirect_blocks;
  161. if (!blocks_remaining)
  162. return shape;
  163. shape.triply_indirect_blocks = min(blocks_remaining, entries_per_block * entries_per_block * entries_per_block);
  164. shape.meta_blocks += 1;
  165. shape.meta_blocks += divide_rounded_up(shape.triply_indirect_blocks, entries_per_block * entries_per_block);
  166. shape.meta_blocks += divide_rounded_up(shape.triply_indirect_blocks, entries_per_block);
  167. blocks_remaining -= shape.triply_indirect_blocks;
  168. VERIFY(blocks_remaining == 0);
  169. return shape;
  170. }
  171. ErrorOr<void> Ext2FSInode::write_indirect_block(BlockBasedFileSystem::BlockIndex block, Span<BlockBasedFileSystem::BlockIndex> blocks_indices)
  172. {
  173. const auto entries_per_block = EXT2_ADDR_PER_BLOCK(&fs().super_block());
  174. VERIFY(blocks_indices.size() <= entries_per_block);
  175. auto block_contents_result = ByteBuffer::create_uninitialized(fs().block_size());
  176. if (!block_contents_result.has_value())
  177. return ENOMEM;
  178. auto block_contents = block_contents_result.release_value();
  179. OutputMemoryStream stream { block_contents };
  180. auto buffer = UserOrKernelBuffer::for_kernel_buffer(stream.data());
  181. VERIFY(blocks_indices.size() <= EXT2_ADDR_PER_BLOCK(&fs().super_block()));
  182. for (unsigned i = 0; i < blocks_indices.size(); ++i)
  183. stream << static_cast<u32>(blocks_indices[i].value());
  184. stream.fill_to_end(0);
  185. return fs().write_block(block, buffer, stream.size());
  186. }
  187. ErrorOr<void> Ext2FSInode::grow_doubly_indirect_block(BlockBasedFileSystem::BlockIndex block, size_t old_blocks_length, Span<BlockBasedFileSystem::BlockIndex> blocks_indices, Vector<Ext2FS::BlockIndex>& new_meta_blocks, unsigned& meta_blocks)
  188. {
  189. const auto entries_per_block = EXT2_ADDR_PER_BLOCK(&fs().super_block());
  190. const auto entries_per_doubly_indirect_block = entries_per_block * entries_per_block;
  191. const auto old_indirect_blocks_length = divide_rounded_up(old_blocks_length, entries_per_block);
  192. const auto new_indirect_blocks_length = divide_rounded_up(blocks_indices.size(), entries_per_block);
  193. VERIFY(blocks_indices.size() > 0);
  194. VERIFY(blocks_indices.size() > old_blocks_length);
  195. VERIFY(blocks_indices.size() <= entries_per_doubly_indirect_block);
  196. auto block_contents_result = ByteBuffer::create_uninitialized(fs().block_size());
  197. if (!block_contents_result.has_value())
  198. return ENOMEM;
  199. auto block_contents = block_contents_result.release_value();
  200. auto* block_as_pointers = (unsigned*)block_contents.data();
  201. OutputMemoryStream stream { block_contents };
  202. auto buffer = UserOrKernelBuffer::for_kernel_buffer(stream.data());
  203. if (old_blocks_length > 0) {
  204. TRY(fs().read_block(block, &buffer, fs().block_size()));
  205. }
  206. // Grow the doubly indirect block.
  207. for (unsigned i = 0; i < old_indirect_blocks_length; i++)
  208. stream << static_cast<u32>(block_as_pointers[i]);
  209. for (unsigned i = old_indirect_blocks_length; i < new_indirect_blocks_length; i++) {
  210. auto new_block = new_meta_blocks.take_last().value();
  211. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::grow_doubly_indirect_block(): Allocating indirect block {} at index {}", identifier(), new_block, i);
  212. stream << static_cast<u32>(new_block);
  213. meta_blocks++;
  214. }
  215. stream.fill_to_end(0);
  216. // Write out the indirect blocks.
  217. for (unsigned i = old_blocks_length / entries_per_block; i < new_indirect_blocks_length; i++) {
  218. const auto offset_block = i * entries_per_block;
  219. TRY(write_indirect_block(block_as_pointers[i], blocks_indices.slice(offset_block, min(blocks_indices.size() - offset_block, entries_per_block))));
  220. }
  221. // Write out the doubly indirect block.
  222. return fs().write_block(block, buffer, stream.size());
  223. }
  224. ErrorOr<void> Ext2FSInode::shrink_doubly_indirect_block(BlockBasedFileSystem::BlockIndex block, size_t old_blocks_length, size_t new_blocks_length, unsigned& meta_blocks)
  225. {
  226. const auto entries_per_block = EXT2_ADDR_PER_BLOCK(&fs().super_block());
  227. const auto entries_per_doubly_indirect_block = entries_per_block * entries_per_block;
  228. const auto old_indirect_blocks_length = divide_rounded_up(old_blocks_length, entries_per_block);
  229. const auto new_indirect_blocks_length = divide_rounded_up(new_blocks_length, entries_per_block);
  230. VERIFY(old_blocks_length > 0);
  231. VERIFY(old_blocks_length >= new_blocks_length);
  232. VERIFY(new_blocks_length <= entries_per_doubly_indirect_block);
  233. auto block_contents_result = ByteBuffer::create_uninitialized(fs().block_size());
  234. if (!block_contents_result.has_value())
  235. return ENOMEM;
  236. auto block_contents = block_contents_result.release_value();
  237. auto* block_as_pointers = (unsigned*)block_contents.data();
  238. auto buffer = UserOrKernelBuffer::for_kernel_buffer(reinterpret_cast<u8*>(block_as_pointers));
  239. TRY(fs().read_block(block, &buffer, fs().block_size()));
  240. // Free the unused indirect blocks.
  241. for (unsigned i = new_indirect_blocks_length; i < old_indirect_blocks_length; i++) {
  242. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::shrink_doubly_indirect_block(): Freeing indirect block {} at index {}", identifier(), block_as_pointers[i], i);
  243. TRY(fs().set_block_allocation_state(block_as_pointers[i], false));
  244. meta_blocks--;
  245. }
  246. // Free the doubly indirect block if no longer needed.
  247. if (new_blocks_length == 0) {
  248. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::shrink_doubly_indirect_block(): Freeing doubly indirect block {}", identifier(), block);
  249. TRY(fs().set_block_allocation_state(block, false));
  250. meta_blocks--;
  251. }
  252. return {};
  253. }
  254. ErrorOr<void> Ext2FSInode::grow_triply_indirect_block(BlockBasedFileSystem::BlockIndex block, size_t old_blocks_length, Span<BlockBasedFileSystem::BlockIndex> blocks_indices, Vector<Ext2FS::BlockIndex>& new_meta_blocks, unsigned& meta_blocks)
  255. {
  256. const auto entries_per_block = EXT2_ADDR_PER_BLOCK(&fs().super_block());
  257. const auto entries_per_doubly_indirect_block = entries_per_block * entries_per_block;
  258. const auto entries_per_triply_indirect_block = entries_per_block * entries_per_block;
  259. const auto old_doubly_indirect_blocks_length = divide_rounded_up(old_blocks_length, entries_per_doubly_indirect_block);
  260. const auto new_doubly_indirect_blocks_length = divide_rounded_up(blocks_indices.size(), entries_per_doubly_indirect_block);
  261. VERIFY(blocks_indices.size() > 0);
  262. VERIFY(blocks_indices.size() > old_blocks_length);
  263. VERIFY(blocks_indices.size() <= entries_per_triply_indirect_block);
  264. auto block_contents_result = ByteBuffer::create_uninitialized(fs().block_size());
  265. if (!block_contents_result.has_value())
  266. return ENOMEM;
  267. auto block_contents = block_contents_result.release_value();
  268. auto* block_as_pointers = (unsigned*)block_contents.data();
  269. OutputMemoryStream stream { block_contents };
  270. auto buffer = UserOrKernelBuffer::for_kernel_buffer(stream.data());
  271. if (old_blocks_length > 0) {
  272. TRY(fs().read_block(block, &buffer, fs().block_size()));
  273. }
  274. // Grow the triply indirect block.
  275. for (unsigned i = 0; i < old_doubly_indirect_blocks_length; i++)
  276. stream << static_cast<u32>(block_as_pointers[i]);
  277. for (unsigned i = old_doubly_indirect_blocks_length; i < new_doubly_indirect_blocks_length; i++) {
  278. auto new_block = new_meta_blocks.take_last().value();
  279. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::grow_triply_indirect_block(): Allocating doubly indirect block {} at index {}", identifier(), new_block, i);
  280. stream << static_cast<u32>(new_block);
  281. meta_blocks++;
  282. }
  283. stream.fill_to_end(0);
  284. // Write out the doubly indirect blocks.
  285. for (unsigned i = old_blocks_length / entries_per_doubly_indirect_block; i < new_doubly_indirect_blocks_length; i++) {
  286. const auto processed_blocks = i * entries_per_doubly_indirect_block;
  287. const auto old_doubly_indirect_blocks_length = min(old_blocks_length > processed_blocks ? old_blocks_length - processed_blocks : 0, entries_per_doubly_indirect_block);
  288. const auto new_doubly_indirect_blocks_length = min(blocks_indices.size() > processed_blocks ? blocks_indices.size() - processed_blocks : 0, entries_per_doubly_indirect_block);
  289. TRY(grow_doubly_indirect_block(block_as_pointers[i], old_doubly_indirect_blocks_length, blocks_indices.slice(processed_blocks, new_doubly_indirect_blocks_length), new_meta_blocks, meta_blocks));
  290. }
  291. // Write out the triply indirect block.
  292. return fs().write_block(block, buffer, stream.size());
  293. }
  294. ErrorOr<void> Ext2FSInode::shrink_triply_indirect_block(BlockBasedFileSystem::BlockIndex block, size_t old_blocks_length, size_t new_blocks_length, unsigned& meta_blocks)
  295. {
  296. const auto entries_per_block = EXT2_ADDR_PER_BLOCK(&fs().super_block());
  297. const auto entries_per_doubly_indirect_block = entries_per_block * entries_per_block;
  298. const auto entries_per_triply_indirect_block = entries_per_doubly_indirect_block * entries_per_block;
  299. const auto old_triply_indirect_blocks_length = divide_rounded_up(old_blocks_length, entries_per_doubly_indirect_block);
  300. const auto new_triply_indirect_blocks_length = new_blocks_length / entries_per_doubly_indirect_block;
  301. VERIFY(old_blocks_length > 0);
  302. VERIFY(old_blocks_length >= new_blocks_length);
  303. VERIFY(new_blocks_length <= entries_per_triply_indirect_block);
  304. auto block_contents_result = ByteBuffer::create_uninitialized(fs().block_size());
  305. if (!block_contents_result.has_value())
  306. return ENOMEM;
  307. auto block_contents = block_contents_result.release_value();
  308. auto* block_as_pointers = (unsigned*)block_contents.data();
  309. auto buffer = UserOrKernelBuffer::for_kernel_buffer(reinterpret_cast<u8*>(block_as_pointers));
  310. TRY(fs().read_block(block, &buffer, fs().block_size()));
  311. // Shrink the doubly indirect blocks.
  312. for (unsigned i = new_triply_indirect_blocks_length; i < old_triply_indirect_blocks_length; i++) {
  313. const auto processed_blocks = i * entries_per_doubly_indirect_block;
  314. const auto old_doubly_indirect_blocks_length = min(old_blocks_length > processed_blocks ? old_blocks_length - processed_blocks : 0, entries_per_doubly_indirect_block);
  315. const auto new_doubly_indirect_blocks_length = min(new_blocks_length > processed_blocks ? new_blocks_length - processed_blocks : 0, entries_per_doubly_indirect_block);
  316. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::shrink_triply_indirect_block(): Shrinking doubly indirect block {} at index {}", identifier(), block_as_pointers[i], i);
  317. TRY(shrink_doubly_indirect_block(block_as_pointers[i], old_doubly_indirect_blocks_length, new_doubly_indirect_blocks_length, meta_blocks));
  318. }
  319. // Free the triply indirect block if no longer needed.
  320. if (new_blocks_length == 0) {
  321. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::shrink_triply_indirect_block(): Freeing triply indirect block {}", identifier(), block);
  322. TRY(fs().set_block_allocation_state(block, false));
  323. meta_blocks--;
  324. }
  325. return {};
  326. }
  327. ErrorOr<void> Ext2FSInode::flush_block_list()
  328. {
  329. MutexLocker locker(m_inode_lock);
  330. if (m_block_list.is_empty()) {
  331. m_raw_inode.i_blocks = 0;
  332. memset(m_raw_inode.i_block, 0, sizeof(m_raw_inode.i_block));
  333. set_metadata_dirty(true);
  334. return {};
  335. }
  336. // NOTE: There is a mismatch between i_blocks and blocks.size() since i_blocks includes meta blocks and blocks.size() does not.
  337. const auto old_block_count = ceil_div(size(), static_cast<u64>(fs().block_size()));
  338. auto old_shape = fs().compute_block_list_shape(old_block_count);
  339. const auto new_shape = fs().compute_block_list_shape(m_block_list.size());
  340. Vector<Ext2FS::BlockIndex> new_meta_blocks;
  341. if (new_shape.meta_blocks > old_shape.meta_blocks) {
  342. new_meta_blocks = TRY(fs().allocate_blocks(fs().group_index_from_inode(index()), new_shape.meta_blocks - old_shape.meta_blocks));
  343. }
  344. m_raw_inode.i_blocks = (m_block_list.size() + new_shape.meta_blocks) * (fs().block_size() / 512);
  345. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::flush_block_list(): Old shape=({};{};{};{}:{}), new shape=({};{};{};{}:{})", identifier(), old_shape.direct_blocks, old_shape.indirect_blocks, old_shape.doubly_indirect_blocks, old_shape.triply_indirect_blocks, old_shape.meta_blocks, new_shape.direct_blocks, new_shape.indirect_blocks, new_shape.doubly_indirect_blocks, new_shape.triply_indirect_blocks, new_shape.meta_blocks);
  346. unsigned output_block_index = 0;
  347. unsigned remaining_blocks = m_block_list.size();
  348. // Deal with direct blocks.
  349. bool inode_dirty = false;
  350. VERIFY(new_shape.direct_blocks <= EXT2_NDIR_BLOCKS);
  351. for (unsigned i = 0; i < new_shape.direct_blocks; ++i) {
  352. if (BlockBasedFileSystem::BlockIndex(m_raw_inode.i_block[i]) != m_block_list[output_block_index])
  353. inode_dirty = true;
  354. m_raw_inode.i_block[i] = m_block_list[output_block_index].value();
  355. ++output_block_index;
  356. --remaining_blocks;
  357. }
  358. // e2fsck considers all blocks reachable through any of the pointers in
  359. // m_raw_inode.i_block as part of this inode regardless of the value in
  360. // m_raw_inode.i_size. When it finds more blocks than the amount that
  361. // is indicated by i_size or i_blocks it offers to repair the filesystem
  362. // by changing those values. That will actually cause further corruption.
  363. // So we must zero all pointers to blocks that are now unused.
  364. for (unsigned i = new_shape.direct_blocks; i < EXT2_NDIR_BLOCKS; ++i) {
  365. m_raw_inode.i_block[i] = 0;
  366. }
  367. if (inode_dirty) {
  368. if constexpr (EXT2_DEBUG) {
  369. dbgln("Ext2FSInode[{}]::flush_block_list(): Writing {} direct block(s) to i_block array of inode {}", identifier(), min((size_t)EXT2_NDIR_BLOCKS, m_block_list.size()), index());
  370. for (size_t i = 0; i < min((size_t)EXT2_NDIR_BLOCKS, m_block_list.size()); ++i)
  371. dbgln(" + {}", m_block_list[i]);
  372. }
  373. set_metadata_dirty(true);
  374. }
  375. // Deal with indirect blocks.
  376. if (old_shape.indirect_blocks != new_shape.indirect_blocks) {
  377. if (new_shape.indirect_blocks > old_shape.indirect_blocks) {
  378. // Write out the indirect block.
  379. if (old_shape.indirect_blocks == 0) {
  380. auto new_block = new_meta_blocks.take_last().value();
  381. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::flush_block_list(): Allocating indirect block: {}", identifier(), new_block);
  382. m_raw_inode.i_block[EXT2_IND_BLOCK] = new_block;
  383. set_metadata_dirty(true);
  384. old_shape.meta_blocks++;
  385. }
  386. TRY(write_indirect_block(m_raw_inode.i_block[EXT2_IND_BLOCK], m_block_list.span().slice(output_block_index, new_shape.indirect_blocks)));
  387. } else if ((new_shape.indirect_blocks == 0) && (old_shape.indirect_blocks != 0)) {
  388. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::flush_block_list(): Freeing indirect block: {}", identifier(), m_raw_inode.i_block[EXT2_IND_BLOCK]);
  389. TRY(fs().set_block_allocation_state(m_raw_inode.i_block[EXT2_IND_BLOCK], false));
  390. old_shape.meta_blocks--;
  391. m_raw_inode.i_block[EXT2_IND_BLOCK] = 0;
  392. }
  393. }
  394. remaining_blocks -= new_shape.indirect_blocks;
  395. output_block_index += new_shape.indirect_blocks;
  396. if (old_shape.doubly_indirect_blocks != new_shape.doubly_indirect_blocks) {
  397. // Write out the doubly indirect block.
  398. if (new_shape.doubly_indirect_blocks > old_shape.doubly_indirect_blocks) {
  399. if (old_shape.doubly_indirect_blocks == 0) {
  400. auto new_block = new_meta_blocks.take_last().value();
  401. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::flush_block_list(): Allocating doubly indirect block: {}", identifier(), new_block);
  402. m_raw_inode.i_block[EXT2_DIND_BLOCK] = new_block;
  403. set_metadata_dirty(true);
  404. old_shape.meta_blocks++;
  405. }
  406. TRY(grow_doubly_indirect_block(m_raw_inode.i_block[EXT2_DIND_BLOCK], old_shape.doubly_indirect_blocks, m_block_list.span().slice(output_block_index, new_shape.doubly_indirect_blocks), new_meta_blocks, old_shape.meta_blocks));
  407. } else {
  408. TRY(shrink_doubly_indirect_block(m_raw_inode.i_block[EXT2_DIND_BLOCK], old_shape.doubly_indirect_blocks, new_shape.doubly_indirect_blocks, old_shape.meta_blocks));
  409. if (new_shape.doubly_indirect_blocks == 0)
  410. m_raw_inode.i_block[EXT2_DIND_BLOCK] = 0;
  411. }
  412. }
  413. remaining_blocks -= new_shape.doubly_indirect_blocks;
  414. output_block_index += new_shape.doubly_indirect_blocks;
  415. if (old_shape.triply_indirect_blocks != new_shape.triply_indirect_blocks) {
  416. // Write out the triply indirect block.
  417. if (new_shape.triply_indirect_blocks > old_shape.triply_indirect_blocks) {
  418. if (old_shape.triply_indirect_blocks == 0) {
  419. auto new_block = new_meta_blocks.take_last().value();
  420. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::flush_block_list(): Allocating triply indirect block: {}", identifier(), new_block);
  421. m_raw_inode.i_block[EXT2_TIND_BLOCK] = new_block;
  422. set_metadata_dirty(true);
  423. old_shape.meta_blocks++;
  424. }
  425. TRY(grow_triply_indirect_block(m_raw_inode.i_block[EXT2_TIND_BLOCK], old_shape.triply_indirect_blocks, m_block_list.span().slice(output_block_index, new_shape.triply_indirect_blocks), new_meta_blocks, old_shape.meta_blocks));
  426. } else {
  427. TRY(shrink_triply_indirect_block(m_raw_inode.i_block[EXT2_TIND_BLOCK], old_shape.triply_indirect_blocks, new_shape.triply_indirect_blocks, old_shape.meta_blocks));
  428. if (new_shape.triply_indirect_blocks == 0)
  429. m_raw_inode.i_block[EXT2_TIND_BLOCK] = 0;
  430. }
  431. }
  432. remaining_blocks -= new_shape.triply_indirect_blocks;
  433. output_block_index += new_shape.triply_indirect_blocks;
  434. dbgln_if(EXT2_BLOCKLIST_DEBUG, "Ext2FSInode[{}]::flush_block_list(): New meta blocks count at {}, expecting {}", identifier(), old_shape.meta_blocks, new_shape.meta_blocks);
  435. VERIFY(new_meta_blocks.size() == 0);
  436. VERIFY(old_shape.meta_blocks == new_shape.meta_blocks);
  437. if (!remaining_blocks)
  438. return {};
  439. dbgln("we don't know how to write qind ext2fs blocks, they don't exist anyway!");
  440. VERIFY_NOT_REACHED();
  441. }
  442. ErrorOr<Vector<Ext2FS::BlockIndex>> Ext2FSInode::compute_block_list() const
  443. {
  444. return compute_block_list_impl(false);
  445. }
  446. ErrorOr<Vector<Ext2FS::BlockIndex>> Ext2FSInode::compute_block_list_with_meta_blocks() const
  447. {
  448. return compute_block_list_impl(true);
  449. }
  450. ErrorOr<Vector<Ext2FS::BlockIndex>> Ext2FSInode::compute_block_list_impl(bool include_block_list_blocks) const
  451. {
  452. // FIXME: This is really awkwardly factored.. foo_impl_internal :|
  453. auto block_list = TRY(compute_block_list_impl_internal(m_raw_inode, include_block_list_blocks));
  454. while (!block_list.is_empty() && block_list.last() == 0)
  455. block_list.take_last();
  456. return block_list;
  457. }
  458. ErrorOr<Vector<Ext2FS::BlockIndex>> Ext2FSInode::compute_block_list_impl_internal(ext2_inode const& e2inode, bool include_block_list_blocks) const
  459. {
  460. unsigned entries_per_block = EXT2_ADDR_PER_BLOCK(&fs().super_block());
  461. unsigned block_count = ceil_div(size(), static_cast<u64>(fs().block_size()));
  462. // If we are handling a symbolic link, the path is stored in the 60 bytes in
  463. // the inode that are used for the 12 direct and 3 indirect block pointers,
  464. // If the path is longer than 60 characters, a block is allocated, and the
  465. // block contains the destination path. The file size corresponds to the
  466. // path length of the destination.
  467. if (::is_symlink(e2inode.i_mode) && e2inode.i_blocks == 0)
  468. block_count = 0;
  469. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]::block_list_for_inode(): i_size={}, i_blocks={}, block_count={}", identifier(), e2inode.i_size, e2inode.i_blocks, block_count);
  470. unsigned blocks_remaining = block_count;
  471. if (include_block_list_blocks) {
  472. auto shape = fs().compute_block_list_shape(block_count);
  473. blocks_remaining += shape.meta_blocks;
  474. }
  475. Vector<Ext2FS::BlockIndex> list;
  476. auto add_block = [&](auto bi) -> ErrorOr<void> {
  477. if (blocks_remaining) {
  478. TRY(list.try_append(bi));
  479. --blocks_remaining;
  480. }
  481. return {};
  482. };
  483. if (include_block_list_blocks) {
  484. // This seems like an excessive over-estimate but w/e.
  485. TRY(list.try_ensure_capacity(blocks_remaining * 2));
  486. } else {
  487. TRY(list.try_ensure_capacity(blocks_remaining));
  488. }
  489. unsigned direct_count = min(block_count, (unsigned)EXT2_NDIR_BLOCKS);
  490. for (unsigned i = 0; i < direct_count; ++i) {
  491. auto block_index = e2inode.i_block[i];
  492. TRY(add_block(block_index));
  493. }
  494. if (!blocks_remaining)
  495. return list;
  496. // Don't need to make copy of add_block, since this capture will only
  497. // be called before compute_block_list_impl_internal finishes.
  498. auto process_block_array = [&](auto array_block_index, auto&& callback) -> ErrorOr<void> {
  499. if (include_block_list_blocks)
  500. TRY(add_block(array_block_index));
  501. auto count = min(blocks_remaining, entries_per_block);
  502. if (!count)
  503. return {};
  504. size_t read_size = count * sizeof(u32);
  505. auto maybe_array_storage = ByteBuffer::create_uninitialized(read_size);
  506. if (!maybe_array_storage.has_value())
  507. return ENOMEM;
  508. auto array_storage = maybe_array_storage.release_value();
  509. auto* array = (u32*)array_storage.data();
  510. auto buffer = UserOrKernelBuffer::for_kernel_buffer((u8*)array);
  511. TRY(fs().read_block(array_block_index, &buffer, read_size, 0));
  512. for (unsigned i = 0; i < count; ++i)
  513. TRY(callback(Ext2FS::BlockIndex(array[i])));
  514. return {};
  515. };
  516. TRY(process_block_array(e2inode.i_block[EXT2_IND_BLOCK], [&](auto block_index) -> ErrorOr<void> {
  517. return add_block(block_index);
  518. }));
  519. if (!blocks_remaining)
  520. return list;
  521. TRY(process_block_array(e2inode.i_block[EXT2_DIND_BLOCK], [&](auto block_index) -> ErrorOr<void> {
  522. return process_block_array(block_index, [&](auto block_index2) -> ErrorOr<void> {
  523. return add_block(block_index2);
  524. });
  525. }));
  526. if (!blocks_remaining)
  527. return list;
  528. TRY(process_block_array(e2inode.i_block[EXT2_TIND_BLOCK], [&](auto block_index) -> ErrorOr<void> {
  529. return process_block_array(block_index, [&](auto block_index2) -> ErrorOr<void> {
  530. return process_block_array(block_index2, [&](auto block_index3) -> ErrorOr<void> {
  531. return add_block(block_index3);
  532. });
  533. });
  534. }));
  535. return list;
  536. }
  537. ErrorOr<void> Ext2FS::free_inode(Ext2FSInode& inode)
  538. {
  539. MutexLocker locker(m_lock);
  540. VERIFY(inode.m_raw_inode.i_links_count == 0);
  541. dbgln_if(EXT2_DEBUG, "Ext2FS[{}]::free_inode(): Inode {} has no more links, time to delete!", fsid(), inode.index());
  542. // Mark all blocks used by this inode as free.
  543. {
  544. auto blocks = TRY(inode.compute_block_list_with_meta_blocks());
  545. for (auto block_index : blocks) {
  546. VERIFY(block_index <= super_block().s_blocks_count);
  547. if (block_index.value())
  548. TRY(set_block_allocation_state(block_index, false));
  549. }
  550. }
  551. // If the inode being freed is a directory, update block group directory counter.
  552. if (inode.is_directory()) {
  553. auto& bgd = const_cast<ext2_group_desc&>(group_descriptor(group_index_from_inode(inode.index())));
  554. --bgd.bg_used_dirs_count;
  555. dbgln_if(EXT2_DEBUG, "Ext2FS[{}]::free_inode(): Decremented bg_used_dirs_count to {} for inode {}", fsid(), bgd.bg_used_dirs_count, inode.index());
  556. m_block_group_descriptors_dirty = true;
  557. }
  558. // NOTE: After this point, the inode metadata is wiped.
  559. memset(&inode.m_raw_inode, 0, sizeof(ext2_inode));
  560. inode.m_raw_inode.i_dtime = kgettimeofday().to_truncated_seconds();
  561. TRY(write_ext2_inode(inode.index(), inode.m_raw_inode));
  562. // Mark the inode as free.
  563. TRY(set_inode_allocation_state(inode.index(), false));
  564. return {};
  565. }
  566. void Ext2FS::flush_block_group_descriptor_table()
  567. {
  568. MutexLocker locker(m_lock);
  569. auto blocks_to_write = ceil_div(m_block_group_count * sizeof(ext2_group_desc), block_size());
  570. auto first_block_of_bgdt = block_size() == 1024 ? 2 : 1;
  571. auto buffer = UserOrKernelBuffer::for_kernel_buffer((u8*)block_group_descriptors());
  572. if (auto result = write_blocks(first_block_of_bgdt, blocks_to_write, buffer); result.is_error())
  573. dbgln("Ext2FS[{}]::flush_block_group_descriptor_table(): Failed to write blocks: {}", fsid(), result.error());
  574. }
  575. void Ext2FS::flush_writes()
  576. {
  577. {
  578. MutexLocker locker(m_lock);
  579. if (m_super_block_dirty) {
  580. flush_super_block();
  581. m_super_block_dirty = false;
  582. }
  583. if (m_block_group_descriptors_dirty) {
  584. flush_block_group_descriptor_table();
  585. m_block_group_descriptors_dirty = false;
  586. }
  587. for (auto& cached_bitmap : m_cached_bitmaps) {
  588. if (cached_bitmap->dirty) {
  589. auto buffer = UserOrKernelBuffer::for_kernel_buffer(cached_bitmap->buffer->data());
  590. if (auto result = write_block(cached_bitmap->bitmap_block_index, buffer, block_size()); result.is_error()) {
  591. dbgln("Ext2FS[{}]::flush_writes(): Failed to write blocks: {}", fsid(), result.error());
  592. }
  593. cached_bitmap->dirty = false;
  594. dbgln_if(EXT2_DEBUG, "Ext2FS[{}]::flush_writes(): Flushed bitmap block {}", fsid(), cached_bitmap->bitmap_block_index);
  595. }
  596. }
  597. // Uncache Inodes that are only kept alive by the index-to-inode lookup cache.
  598. // We don't uncache Inodes that are being watched by at least one InodeWatcher.
  599. // FIXME: It would be better to keep a capped number of Inodes around.
  600. // The problem is that they are quite heavy objects, and use a lot of heap memory
  601. // for their (child name lookup) and (block list) caches.
  602. Vector<InodeIndex> unused_inodes;
  603. for (auto& it : m_inode_cache) {
  604. // NOTE: If we're asked to look up an inode by number (via get_inode) and it turns out
  605. // to not exist, we remember the fact that it doesn't exist by caching a nullptr.
  606. // This seems like a reasonable time to uncache ideas about unknown inodes, so do that.
  607. if (!it.value) {
  608. unused_inodes.append(it.key);
  609. continue;
  610. }
  611. if (it.value->ref_count() != 1)
  612. continue;
  613. if (it.value->has_watchers())
  614. continue;
  615. unused_inodes.append(it.key);
  616. }
  617. for (auto index : unused_inodes)
  618. uncache_inode(index);
  619. }
  620. BlockBasedFileSystem::flush_writes();
  621. }
  622. Ext2FSInode::Ext2FSInode(Ext2FS& fs, InodeIndex index)
  623. : Inode(fs, index)
  624. {
  625. }
  626. Ext2FSInode::~Ext2FSInode()
  627. {
  628. if (m_raw_inode.i_links_count == 0) {
  629. // Alas, we have nowhere to propagate any errors that occur here.
  630. (void)fs().free_inode(*this);
  631. }
  632. }
  633. u64 Ext2FSInode::size() const
  634. {
  635. if (Kernel::is_regular_file(m_raw_inode.i_mode) && ((u32)fs().get_features_readonly() & (u32)Ext2FS::FeaturesReadOnly::FileSize64bits))
  636. return static_cast<u64>(m_raw_inode.i_dir_acl) << 32 | m_raw_inode.i_size;
  637. return m_raw_inode.i_size;
  638. }
  639. InodeMetadata Ext2FSInode::metadata() const
  640. {
  641. MutexLocker locker(m_inode_lock);
  642. InodeMetadata metadata;
  643. metadata.inode = identifier();
  644. metadata.size = size();
  645. metadata.mode = m_raw_inode.i_mode;
  646. metadata.uid = m_raw_inode.i_uid;
  647. metadata.gid = m_raw_inode.i_gid;
  648. metadata.link_count = m_raw_inode.i_links_count;
  649. metadata.atime = m_raw_inode.i_atime;
  650. metadata.ctime = m_raw_inode.i_ctime;
  651. metadata.mtime = m_raw_inode.i_mtime;
  652. metadata.dtime = m_raw_inode.i_dtime;
  653. metadata.block_size = fs().block_size();
  654. metadata.block_count = m_raw_inode.i_blocks;
  655. if (Kernel::is_character_device(m_raw_inode.i_mode) || Kernel::is_block_device(m_raw_inode.i_mode)) {
  656. unsigned dev = m_raw_inode.i_block[0];
  657. if (!dev)
  658. dev = m_raw_inode.i_block[1];
  659. metadata.major_device = (dev & 0xfff00) >> 8;
  660. metadata.minor_device = (dev & 0xff) | ((dev >> 12) & 0xfff00);
  661. }
  662. return metadata;
  663. }
  664. ErrorOr<void> Ext2FSInode::flush_metadata()
  665. {
  666. MutexLocker locker(m_inode_lock);
  667. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]::flush_metadata(): Flushing inode", identifier());
  668. TRY(fs().write_ext2_inode(index(), m_raw_inode));
  669. if (is_directory()) {
  670. // Unless we're about to go away permanently, invalidate the lookup cache.
  671. if (m_raw_inode.i_links_count != 0) {
  672. // FIXME: This invalidation is way too hardcore. It's sad to throw away the whole cache.
  673. m_lookup_cache.clear();
  674. }
  675. }
  676. set_metadata_dirty(false);
  677. return {};
  678. }
  679. ErrorOr<NonnullRefPtr<Inode>> Ext2FS::get_inode(InodeIdentifier inode) const
  680. {
  681. MutexLocker locker(m_lock);
  682. VERIFY(inode.fsid() == fsid());
  683. {
  684. auto it = m_inode_cache.find(inode.index());
  685. if (it != m_inode_cache.end()) {
  686. if (!it->value)
  687. return ENOENT;
  688. return NonnullRefPtr<Inode> { *it->value };
  689. }
  690. }
  691. auto inode_allocation_state = TRY(get_inode_allocation_state(inode.index()));
  692. if (!inode_allocation_state) {
  693. TRY(m_inode_cache.try_set(inode.index(), nullptr));
  694. return ENOENT;
  695. }
  696. BlockIndex block_index;
  697. unsigned offset;
  698. if (!find_block_containing_inode(inode.index(), block_index, offset))
  699. return EINVAL;
  700. auto new_inode = TRY(adopt_nonnull_ref_or_enomem(new (nothrow) Ext2FSInode(const_cast<Ext2FS&>(*this), inode.index())));
  701. auto buffer = UserOrKernelBuffer::for_kernel_buffer(reinterpret_cast<u8*>(&new_inode->m_raw_inode));
  702. TRY(read_block(block_index, &buffer, sizeof(ext2_inode), offset));
  703. TRY(m_inode_cache.try_set(inode.index(), new_inode));
  704. return new_inode;
  705. }
  706. ErrorOr<size_t> Ext2FSInode::read_bytes(off_t offset, size_t count, UserOrKernelBuffer& buffer, OpenFileDescription* description) const
  707. {
  708. MutexLocker inode_locker(m_inode_lock);
  709. VERIFY(offset >= 0);
  710. if (m_raw_inode.i_size == 0)
  711. return 0;
  712. if (static_cast<u64>(offset) >= size())
  713. return 0;
  714. // Symbolic links shorter than 60 characters are store inline inside the i_block array.
  715. // This avoids wasting an entire block on short links. (Most links are short.)
  716. if (is_symlink() && size() < max_inline_symlink_length) {
  717. VERIFY(offset == 0);
  718. size_t nread = min((off_t)size() - offset, static_cast<off_t>(count));
  719. TRY(buffer.write(((const u8*)m_raw_inode.i_block) + offset, nread));
  720. return nread;
  721. }
  722. if (m_block_list.is_empty())
  723. m_block_list = TRY(compute_block_list());
  724. if (m_block_list.is_empty()) {
  725. dmesgln("Ext2FSInode[{}]::read_bytes(): Empty block list", identifier());
  726. return EIO;
  727. }
  728. bool allow_cache = !description || !description->is_direct();
  729. const int block_size = fs().block_size();
  730. BlockBasedFileSystem::BlockIndex first_block_logical_index = offset / block_size;
  731. BlockBasedFileSystem::BlockIndex last_block_logical_index = (offset + count) / block_size;
  732. if (last_block_logical_index >= m_block_list.size())
  733. last_block_logical_index = m_block_list.size() - 1;
  734. int offset_into_first_block = offset % block_size;
  735. size_t nread = 0;
  736. auto remaining_count = min((off_t)count, (off_t)size() - offset);
  737. dbgln_if(EXT2_VERY_DEBUG, "Ext2FSInode[{}]::read_bytes(): Reading up to {} bytes, {} bytes into inode to {}", identifier(), count, offset, buffer.user_or_kernel_ptr());
  738. for (auto bi = first_block_logical_index; remaining_count && bi <= last_block_logical_index; bi = bi.value() + 1) {
  739. auto block_index = m_block_list[bi.value()];
  740. size_t offset_into_block = (bi == first_block_logical_index) ? offset_into_first_block : 0;
  741. size_t num_bytes_to_copy = min((size_t)block_size - offset_into_block, (size_t)remaining_count);
  742. auto buffer_offset = buffer.offset(nread);
  743. if (block_index.value() == 0) {
  744. // This is a hole, act as if it's filled with zeroes.
  745. TRY(buffer_offset.memset(0, num_bytes_to_copy));
  746. } else {
  747. if (auto result = fs().read_block(block_index, &buffer_offset, num_bytes_to_copy, offset_into_block, allow_cache); result.is_error()) {
  748. dmesgln("Ext2FSInode[{}]::read_bytes(): Failed to read block {} (index {})", identifier(), block_index.value(), bi);
  749. return result.release_error();
  750. }
  751. }
  752. remaining_count -= num_bytes_to_copy;
  753. nread += num_bytes_to_copy;
  754. }
  755. return nread;
  756. }
  757. ErrorOr<void> Ext2FSInode::resize(u64 new_size)
  758. {
  759. auto old_size = size();
  760. if (old_size == new_size)
  761. return {};
  762. if (!((u32)fs().get_features_readonly() & (u32)Ext2FS::FeaturesReadOnly::FileSize64bits) && (new_size >= static_cast<u32>(-1)))
  763. return ENOSPC;
  764. u64 block_size = fs().block_size();
  765. auto blocks_needed_before = ceil_div(old_size, block_size);
  766. auto blocks_needed_after = ceil_div(new_size, block_size);
  767. if constexpr (EXT2_DEBUG) {
  768. dbgln("Ext2FSInode[{}]::resize(): Blocks needed before (size was {}): {}", identifier(), old_size, blocks_needed_before);
  769. dbgln("Ext2FSInode[{}]::resize(): Blocks needed after (size is {}): {}", identifier(), new_size, blocks_needed_after);
  770. }
  771. if (blocks_needed_after > blocks_needed_before) {
  772. auto additional_blocks_needed = blocks_needed_after - blocks_needed_before;
  773. if (additional_blocks_needed > fs().super_block().s_free_blocks_count)
  774. return ENOSPC;
  775. }
  776. if (m_block_list.is_empty())
  777. m_block_list = TRY(compute_block_list());
  778. if (blocks_needed_after > blocks_needed_before) {
  779. auto blocks = TRY(fs().allocate_blocks(fs().group_index_from_inode(index()), blocks_needed_after - blocks_needed_before));
  780. TRY(m_block_list.try_extend(move(blocks)));
  781. } else if (blocks_needed_after < blocks_needed_before) {
  782. if constexpr (EXT2_VERY_DEBUG) {
  783. dbgln("Ext2FSInode[{}]::resize(): Shrinking inode, old block list is {} entries:", identifier(), m_block_list.size());
  784. for (auto block_index : m_block_list) {
  785. dbgln(" # {}", block_index);
  786. }
  787. }
  788. while (m_block_list.size() != blocks_needed_after) {
  789. auto block_index = m_block_list.take_last();
  790. if (block_index.value()) {
  791. if (auto result = fs().set_block_allocation_state(block_index, false); result.is_error()) {
  792. dbgln("Ext2FSInode[{}]::resize(): Failed to free block {}: {}", identifier(), block_index, result.error());
  793. return result;
  794. }
  795. }
  796. }
  797. }
  798. TRY(flush_block_list());
  799. m_raw_inode.i_size = new_size;
  800. if (Kernel::is_regular_file(m_raw_inode.i_mode))
  801. m_raw_inode.i_dir_acl = new_size >> 32;
  802. set_metadata_dirty(true);
  803. if (new_size > old_size) {
  804. // If we're growing the inode, make sure we zero out all the new space.
  805. // FIXME: There are definitely more efficient ways to achieve this.
  806. auto bytes_to_clear = new_size - old_size;
  807. auto clear_from = old_size;
  808. u8 zero_buffer[PAGE_SIZE] {};
  809. while (bytes_to_clear) {
  810. auto nwritten = TRY(write_bytes(clear_from, min(static_cast<u64>(sizeof(zero_buffer)), bytes_to_clear), UserOrKernelBuffer::for_kernel_buffer(zero_buffer), nullptr));
  811. VERIFY(nwritten != 0);
  812. bytes_to_clear -= nwritten;
  813. clear_from += nwritten;
  814. }
  815. }
  816. return {};
  817. }
  818. ErrorOr<size_t> Ext2FSInode::write_bytes(off_t offset, size_t count, const UserOrKernelBuffer& data, OpenFileDescription* description)
  819. {
  820. VERIFY(offset >= 0);
  821. if (count == 0)
  822. return 0;
  823. MutexLocker inode_locker(m_inode_lock);
  824. TRY(prepare_to_write_data());
  825. if (is_symlink()) {
  826. VERIFY(offset == 0);
  827. if (max((size_t)(offset + count), (size_t)m_raw_inode.i_size) < max_inline_symlink_length) {
  828. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]::write_bytes(): Poking into i_block array for inline symlink ({} bytes)", identifier(), count);
  829. TRY(data.read(((u8*)m_raw_inode.i_block) + offset, count));
  830. if ((size_t)(offset + count) > (size_t)m_raw_inode.i_size)
  831. m_raw_inode.i_size = offset + count;
  832. set_metadata_dirty(true);
  833. return count;
  834. }
  835. }
  836. bool allow_cache = !description || !description->is_direct();
  837. const auto block_size = fs().block_size();
  838. auto new_size = max(static_cast<u64>(offset) + count, size());
  839. TRY(resize(new_size));
  840. if (m_block_list.is_empty())
  841. m_block_list = TRY(compute_block_list());
  842. if (m_block_list.is_empty()) {
  843. dbgln("Ext2FSInode[{}]::write_bytes(): Empty block list", identifier());
  844. return EIO;
  845. }
  846. BlockBasedFileSystem::BlockIndex first_block_logical_index = offset / block_size;
  847. BlockBasedFileSystem::BlockIndex last_block_logical_index = (offset + count) / block_size;
  848. if (last_block_logical_index >= m_block_list.size())
  849. last_block_logical_index = m_block_list.size() - 1;
  850. size_t offset_into_first_block = offset % block_size;
  851. size_t nwritten = 0;
  852. auto remaining_count = min((off_t)count, (off_t)new_size - offset);
  853. dbgln_if(EXT2_VERY_DEBUG, "Ext2FSInode[{}]::write_bytes(): Writing {} bytes, {} bytes into inode from {}", identifier(), count, offset, data.user_or_kernel_ptr());
  854. for (auto bi = first_block_logical_index; remaining_count && bi <= last_block_logical_index; bi = bi.value() + 1) {
  855. size_t offset_into_block = (bi == first_block_logical_index) ? offset_into_first_block : 0;
  856. size_t num_bytes_to_copy = min((size_t)block_size - offset_into_block, (size_t)remaining_count);
  857. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]::write_bytes(): Writing block {} (offset_into_block: {})", identifier(), m_block_list[bi.value()], offset_into_block);
  858. if (auto result = fs().write_block(m_block_list[bi.value()], data.offset(nwritten), num_bytes_to_copy, offset_into_block, allow_cache); result.is_error()) {
  859. dbgln("Ext2FSInode[{}]::write_bytes(): Failed to write block {} (index {})", identifier(), m_block_list[bi.value()], bi);
  860. return result.release_error();
  861. }
  862. remaining_count -= num_bytes_to_copy;
  863. nwritten += num_bytes_to_copy;
  864. }
  865. did_modify_contents();
  866. dbgln_if(EXT2_VERY_DEBUG, "Ext2FSInode[{}]::write_bytes(): After write, i_size={}, i_blocks={} ({} blocks in list)", identifier(), size(), m_raw_inode.i_blocks, m_block_list.size());
  867. return nwritten;
  868. }
  869. u8 Ext2FS::internal_file_type_to_directory_entry_type(const DirectoryEntryView& entry) const
  870. {
  871. switch (entry.file_type) {
  872. case EXT2_FT_REG_FILE:
  873. return DT_REG;
  874. case EXT2_FT_DIR:
  875. return DT_DIR;
  876. case EXT2_FT_CHRDEV:
  877. return DT_CHR;
  878. case EXT2_FT_BLKDEV:
  879. return DT_BLK;
  880. case EXT2_FT_FIFO:
  881. return DT_FIFO;
  882. case EXT2_FT_SOCK:
  883. return DT_SOCK;
  884. case EXT2_FT_SYMLINK:
  885. return DT_LNK;
  886. default:
  887. return DT_UNKNOWN;
  888. }
  889. }
  890. Ext2FS::FeaturesReadOnly Ext2FS::get_features_readonly() const
  891. {
  892. if (m_super_block.s_rev_level > 0)
  893. return static_cast<Ext2FS::FeaturesReadOnly>(m_super_block.s_feature_ro_compat);
  894. return Ext2FS::FeaturesReadOnly::None;
  895. }
  896. ErrorOr<void> Ext2FSInode::traverse_as_directory(Function<ErrorOr<void>(FileSystem::DirectoryEntryView const&)> callback) const
  897. {
  898. VERIFY(is_directory());
  899. u8 buffer[max_block_size];
  900. auto buf = UserOrKernelBuffer::for_kernel_buffer(buffer);
  901. auto block_size = fs().block_size();
  902. auto file_size = size();
  903. // Directory entries are guaranteed not to span multiple blocks,
  904. // so we can iterate over blocks separately.
  905. for (u64 offset = 0; offset < file_size; offset += block_size) {
  906. TRY(read_bytes(offset, block_size, buf, nullptr));
  907. auto* entry = reinterpret_cast<ext2_dir_entry_2*>(buffer);
  908. auto* entries_end = reinterpret_cast<ext2_dir_entry_2*>(buffer + block_size);
  909. while (entry < entries_end) {
  910. if (entry->inode != 0) {
  911. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]::traverse_as_directory(): inode {}, name_len: {}, rec_len: {}, file_type: {}, name: {}", identifier(), entry->inode, entry->name_len, entry->rec_len, entry->file_type, StringView(entry->name, entry->name_len));
  912. TRY(callback({ { entry->name, entry->name_len }, { fsid(), entry->inode }, entry->file_type }));
  913. }
  914. entry = (ext2_dir_entry_2*)((char*)entry + entry->rec_len);
  915. }
  916. }
  917. return {};
  918. }
  919. ErrorOr<void> Ext2FSInode::write_directory(Vector<Ext2FSDirectoryEntry>& entries)
  920. {
  921. MutexLocker locker(m_inode_lock);
  922. auto block_size = fs().block_size();
  923. // Calculate directory size and record length of entries so that
  924. // the following constraints are met:
  925. // - All used blocks must be entirely filled.
  926. // - Entries are aligned on a 4-byte boundary.
  927. // - No entry may span multiple blocks.
  928. size_t directory_size = 0;
  929. size_t space_in_block = block_size;
  930. for (size_t i = 0; i < entries.size(); ++i) {
  931. auto& entry = entries[i];
  932. entry.record_length = EXT2_DIR_REC_LEN(entry.name.length());
  933. space_in_block -= entry.record_length;
  934. if (i + 1 < entries.size()) {
  935. if (EXT2_DIR_REC_LEN(entries[i + 1].name.length()) > space_in_block) {
  936. entry.record_length += space_in_block;
  937. space_in_block = block_size;
  938. }
  939. } else {
  940. entry.record_length += space_in_block;
  941. }
  942. directory_size += entry.record_length;
  943. }
  944. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]::write_directory(): New directory contents to write (size {}):", identifier(), directory_size);
  945. auto directory_data_result = ByteBuffer::create_uninitialized(directory_size);
  946. if (!directory_data_result.has_value())
  947. return ENOMEM;
  948. auto directory_data = directory_data_result.release_value();
  949. OutputMemoryStream stream { directory_data };
  950. for (auto& entry : entries) {
  951. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]::write_directory(): Writing inode: {}, name_len: {}, rec_len: {}, file_type: {}, name: {}", identifier(), entry.inode_index, u16(entry.name.length()), u16(entry.record_length), u8(entry.file_type), entry.name);
  952. stream << u32(entry.inode_index.value());
  953. stream << u16(entry.record_length);
  954. stream << u8(entry.name.length());
  955. stream << u8(entry.file_type);
  956. stream << entry.name.bytes();
  957. int padding = entry.record_length - entry.name.length() - 8;
  958. for (int j = 0; j < padding; ++j)
  959. stream << u8(0);
  960. }
  961. VERIFY(stream.is_end());
  962. TRY(resize(stream.size()));
  963. auto buffer = UserOrKernelBuffer::for_kernel_buffer(stream.data());
  964. auto nwritten = TRY(write_bytes(0, stream.size(), buffer, nullptr));
  965. set_metadata_dirty(true);
  966. if (nwritten != directory_data.size())
  967. return EIO;
  968. return {};
  969. }
  970. ErrorOr<NonnullRefPtr<Inode>> Ext2FSInode::create_child(StringView name, mode_t mode, dev_t dev, UserID uid, GroupID gid)
  971. {
  972. if (::is_directory(mode))
  973. return fs().create_directory(*this, name, mode, uid, gid);
  974. return fs().create_inode(*this, name, mode, dev, uid, gid);
  975. }
  976. ErrorOr<void> Ext2FSInode::add_child(Inode& child, StringView name, mode_t mode)
  977. {
  978. MutexLocker locker(m_inode_lock);
  979. VERIFY(is_directory());
  980. if (name.length() > EXT2_NAME_LEN)
  981. return ENAMETOOLONG;
  982. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]::add_child(): Adding inode {} with name '{}' and mode {:o} to directory {}", identifier(), child.index(), name, mode, index());
  983. Vector<Ext2FSDirectoryEntry> entries;
  984. TRY(traverse_as_directory([&](auto& entry) -> ErrorOr<void> {
  985. if (name == entry.name)
  986. return EEXIST;
  987. TRY(entries.try_append({ entry.name, entry.inode.index(), entry.file_type }));
  988. return {};
  989. }));
  990. TRY(child.increment_link_count());
  991. entries.empend(name, child.index(), to_ext2_file_type(mode));
  992. TRY(write_directory(entries));
  993. TRY(populate_lookup_cache());
  994. TRY(m_lookup_cache.try_set(name, child.index()));
  995. did_add_child(child.identifier(), name);
  996. return {};
  997. }
  998. ErrorOr<void> Ext2FSInode::remove_child(StringView name)
  999. {
  1000. MutexLocker locker(m_inode_lock);
  1001. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]::remove_child(): Removing '{}'", identifier(), name);
  1002. VERIFY(is_directory());
  1003. TRY(populate_lookup_cache());
  1004. auto it = m_lookup_cache.find(name);
  1005. if (it == m_lookup_cache.end())
  1006. return ENOENT;
  1007. auto child_inode_index = (*it).value;
  1008. InodeIdentifier child_id { fsid(), child_inode_index };
  1009. Vector<Ext2FSDirectoryEntry> entries;
  1010. TRY(traverse_as_directory([&](auto& entry) -> ErrorOr<void> {
  1011. if (name != entry.name)
  1012. TRY(entries.try_append({ entry.name, entry.inode.index(), entry.file_type }));
  1013. return {};
  1014. }));
  1015. TRY(write_directory(entries));
  1016. m_lookup_cache.remove(name);
  1017. auto child_inode = TRY(fs().get_inode(child_id));
  1018. TRY(child_inode->decrement_link_count());
  1019. did_remove_child(child_id, name);
  1020. return {};
  1021. }
  1022. u64 Ext2FS::inodes_per_block() const
  1023. {
  1024. return EXT2_INODES_PER_BLOCK(&super_block());
  1025. }
  1026. u64 Ext2FS::inodes_per_group() const
  1027. {
  1028. return EXT2_INODES_PER_GROUP(&super_block());
  1029. }
  1030. u64 Ext2FS::inode_size() const
  1031. {
  1032. return EXT2_INODE_SIZE(&super_block());
  1033. }
  1034. u64 Ext2FS::blocks_per_group() const
  1035. {
  1036. return EXT2_BLOCKS_PER_GROUP(&super_block());
  1037. }
  1038. ErrorOr<void> Ext2FS::write_ext2_inode(InodeIndex inode, ext2_inode const& e2inode)
  1039. {
  1040. BlockIndex block_index;
  1041. unsigned offset;
  1042. if (!find_block_containing_inode(inode, block_index, offset))
  1043. return EINVAL;
  1044. auto buffer = UserOrKernelBuffer::for_kernel_buffer(const_cast<u8*>((const u8*)&e2inode));
  1045. return write_block(block_index, buffer, inode_size(), offset);
  1046. }
  1047. auto Ext2FS::allocate_blocks(GroupIndex preferred_group_index, size_t count) -> ErrorOr<Vector<BlockIndex>>
  1048. {
  1049. dbgln_if(EXT2_DEBUG, "Ext2FS: allocate_blocks(preferred group: {}, count {})", preferred_group_index, count);
  1050. if (count == 0)
  1051. return Vector<BlockIndex> {};
  1052. Vector<BlockIndex> blocks;
  1053. TRY(blocks.try_ensure_capacity(count));
  1054. MutexLocker locker(m_lock);
  1055. auto group_index = preferred_group_index;
  1056. if (!group_descriptor(preferred_group_index).bg_free_blocks_count) {
  1057. group_index = 1;
  1058. }
  1059. while (blocks.size() < count) {
  1060. bool found_a_group = false;
  1061. if (group_descriptor(group_index).bg_free_blocks_count) {
  1062. found_a_group = true;
  1063. } else {
  1064. if (group_index == preferred_group_index)
  1065. group_index = 1;
  1066. for (; group_index <= m_block_group_count; group_index = GroupIndex { group_index.value() + 1 }) {
  1067. if (group_descriptor(group_index).bg_free_blocks_count) {
  1068. found_a_group = true;
  1069. break;
  1070. }
  1071. }
  1072. }
  1073. VERIFY(found_a_group);
  1074. auto const& bgd = group_descriptor(group_index);
  1075. auto* cached_bitmap = TRY(get_bitmap_block(bgd.bg_block_bitmap));
  1076. int blocks_in_group = min(blocks_per_group(), super_block().s_blocks_count);
  1077. auto block_bitmap = cached_bitmap->bitmap(blocks_in_group);
  1078. BlockIndex first_block_in_group = (group_index.value() - 1) * blocks_per_group() + first_block_index().value();
  1079. size_t free_region_size = 0;
  1080. auto first_unset_bit_index = block_bitmap.find_longest_range_of_unset_bits(count - blocks.size(), free_region_size);
  1081. VERIFY(first_unset_bit_index.has_value());
  1082. dbgln_if(EXT2_DEBUG, "Ext2FS: allocating free region of size: {} [{}]", free_region_size, group_index);
  1083. for (size_t i = 0; i < free_region_size; ++i) {
  1084. BlockIndex block_index = (first_unset_bit_index.value() + i) + first_block_in_group.value();
  1085. TRY(set_block_allocation_state(block_index, true));
  1086. blocks.unchecked_append(block_index);
  1087. dbgln_if(EXT2_DEBUG, " allocated > {}", block_index);
  1088. }
  1089. }
  1090. VERIFY(blocks.size() == count);
  1091. return blocks;
  1092. }
  1093. ErrorOr<InodeIndex> Ext2FS::allocate_inode(GroupIndex preferred_group)
  1094. {
  1095. dbgln_if(EXT2_DEBUG, "Ext2FS: allocate_inode(preferred_group: {})", preferred_group);
  1096. MutexLocker locker(m_lock);
  1097. // FIXME: We shouldn't refuse to allocate an inode if there is no group that can house the whole thing.
  1098. // In those cases we should just spread it across multiple groups.
  1099. auto is_suitable_group = [this](auto group_index) {
  1100. auto& bgd = group_descriptor(group_index);
  1101. return bgd.bg_free_inodes_count && bgd.bg_free_blocks_count >= 1;
  1102. };
  1103. GroupIndex group_index;
  1104. if (preferred_group.value() && is_suitable_group(preferred_group)) {
  1105. group_index = preferred_group;
  1106. } else {
  1107. for (unsigned i = 1; i <= m_block_group_count; ++i) {
  1108. if (is_suitable_group(i)) {
  1109. group_index = i;
  1110. break;
  1111. }
  1112. }
  1113. }
  1114. if (!group_index) {
  1115. dmesgln("Ext2FS: allocate_inode: no suitable group found for new inode");
  1116. return ENOSPC;
  1117. }
  1118. dbgln_if(EXT2_DEBUG, "Ext2FS: allocate_inode: found suitable group [{}] for new inode :^)", group_index);
  1119. auto const& bgd = group_descriptor(group_index);
  1120. unsigned inodes_in_group = min(inodes_per_group(), super_block().s_inodes_count);
  1121. InodeIndex first_inode_in_group = (group_index.value() - 1) * inodes_per_group() + 1;
  1122. auto* cached_bitmap = TRY(get_bitmap_block(bgd.bg_inode_bitmap));
  1123. auto inode_bitmap = cached_bitmap->bitmap(inodes_in_group);
  1124. for (size_t i = 0; i < inode_bitmap.size(); ++i) {
  1125. if (inode_bitmap.get(i))
  1126. continue;
  1127. inode_bitmap.set(i, true);
  1128. auto inode_index = InodeIndex(first_inode_in_group.value() + i);
  1129. cached_bitmap->dirty = true;
  1130. m_super_block.s_free_inodes_count--;
  1131. m_super_block_dirty = true;
  1132. const_cast<ext2_group_desc&>(bgd).bg_free_inodes_count--;
  1133. m_block_group_descriptors_dirty = true;
  1134. // In case the inode cache had this cached as "non-existent", uncache that info.
  1135. m_inode_cache.remove(inode_index.value());
  1136. return inode_index;
  1137. }
  1138. dmesgln("Ext2FS: allocate_inode found no available inode, despite bgd claiming there are inodes :(");
  1139. return EIO;
  1140. }
  1141. Ext2FS::GroupIndex Ext2FS::group_index_from_block_index(BlockIndex block_index) const
  1142. {
  1143. if (!block_index)
  1144. return 0;
  1145. return (block_index.value() - 1) / blocks_per_group() + 1;
  1146. }
  1147. auto Ext2FS::group_index_from_inode(InodeIndex inode) const -> GroupIndex
  1148. {
  1149. if (!inode)
  1150. return 0;
  1151. return (inode.value() - 1) / inodes_per_group() + 1;
  1152. }
  1153. ErrorOr<bool> Ext2FS::get_inode_allocation_state(InodeIndex index) const
  1154. {
  1155. MutexLocker locker(m_lock);
  1156. if (index == 0)
  1157. return EINVAL;
  1158. auto group_index = group_index_from_inode(index);
  1159. auto const& bgd = group_descriptor(group_index);
  1160. unsigned index_in_group = index.value() - ((group_index.value() - 1) * inodes_per_group());
  1161. unsigned bit_index = (index_in_group - 1) % inodes_per_group();
  1162. auto* cached_bitmap = TRY(const_cast<Ext2FS&>(*this).get_bitmap_block(bgd.bg_inode_bitmap));
  1163. return cached_bitmap->bitmap(inodes_per_group()).get(bit_index);
  1164. }
  1165. ErrorOr<void> Ext2FS::update_bitmap_block(BlockIndex bitmap_block, size_t bit_index, bool new_state, u32& super_block_counter, u16& group_descriptor_counter)
  1166. {
  1167. auto* cached_bitmap = TRY(get_bitmap_block(bitmap_block));
  1168. bool current_state = cached_bitmap->bitmap(blocks_per_group()).get(bit_index);
  1169. if (current_state == new_state) {
  1170. dbgln("Ext2FS: Bit {} in bitmap block {} had unexpected state {}", bit_index, bitmap_block, current_state);
  1171. return EIO;
  1172. }
  1173. cached_bitmap->bitmap(blocks_per_group()).set(bit_index, new_state);
  1174. cached_bitmap->dirty = true;
  1175. if (new_state) {
  1176. --super_block_counter;
  1177. --group_descriptor_counter;
  1178. } else {
  1179. ++super_block_counter;
  1180. ++group_descriptor_counter;
  1181. }
  1182. m_super_block_dirty = true;
  1183. m_block_group_descriptors_dirty = true;
  1184. return {};
  1185. }
  1186. ErrorOr<void> Ext2FS::set_inode_allocation_state(InodeIndex inode_index, bool new_state)
  1187. {
  1188. MutexLocker locker(m_lock);
  1189. auto group_index = group_index_from_inode(inode_index);
  1190. unsigned index_in_group = inode_index.value() - ((group_index.value() - 1) * inodes_per_group());
  1191. unsigned bit_index = (index_in_group - 1) % inodes_per_group();
  1192. dbgln_if(EXT2_DEBUG, "Ext2FS: set_inode_allocation_state: Inode {} -> {}", inode_index, new_state);
  1193. auto& bgd = const_cast<ext2_group_desc&>(group_descriptor(group_index));
  1194. return update_bitmap_block(bgd.bg_inode_bitmap, bit_index, new_state, m_super_block.s_free_inodes_count, bgd.bg_free_inodes_count);
  1195. }
  1196. Ext2FS::BlockIndex Ext2FS::first_block_index() const
  1197. {
  1198. return block_size() == 1024 ? 1 : 0;
  1199. }
  1200. ErrorOr<Ext2FS::CachedBitmap*> Ext2FS::get_bitmap_block(BlockIndex bitmap_block_index)
  1201. {
  1202. for (auto& cached_bitmap : m_cached_bitmaps) {
  1203. if (cached_bitmap->bitmap_block_index == bitmap_block_index)
  1204. return cached_bitmap.ptr();
  1205. }
  1206. auto block = TRY(KBuffer::try_create_with_size(block_size(), Memory::Region::Access::ReadWrite, "Ext2FS: Cached bitmap block"));
  1207. auto buffer = UserOrKernelBuffer::for_kernel_buffer(block->data());
  1208. TRY(read_block(bitmap_block_index, &buffer, block_size()));
  1209. auto new_bitmap = TRY(adopt_nonnull_own_or_enomem(new (nothrow) CachedBitmap(bitmap_block_index, move(block))));
  1210. TRY(m_cached_bitmaps.try_append(move(new_bitmap)));
  1211. return m_cached_bitmaps.last().ptr();
  1212. }
  1213. ErrorOr<void> Ext2FS::set_block_allocation_state(BlockIndex block_index, bool new_state)
  1214. {
  1215. VERIFY(block_index != 0);
  1216. MutexLocker locker(m_lock);
  1217. auto group_index = group_index_from_block_index(block_index);
  1218. unsigned index_in_group = (block_index.value() - first_block_index().value()) - ((group_index.value() - 1) * blocks_per_group());
  1219. unsigned bit_index = index_in_group % blocks_per_group();
  1220. auto& bgd = const_cast<ext2_group_desc&>(group_descriptor(group_index));
  1221. dbgln_if(EXT2_DEBUG, "Ext2FS: Block {} state -> {} (in bitmap block {})", block_index, new_state, bgd.bg_block_bitmap);
  1222. return update_bitmap_block(bgd.bg_block_bitmap, bit_index, new_state, m_super_block.s_free_blocks_count, bgd.bg_free_blocks_count);
  1223. }
  1224. ErrorOr<NonnullRefPtr<Inode>> Ext2FS::create_directory(Ext2FSInode& parent_inode, StringView name, mode_t mode, UserID uid, GroupID gid)
  1225. {
  1226. MutexLocker locker(m_lock);
  1227. VERIFY(is_directory(mode));
  1228. auto inode = TRY(create_inode(parent_inode, name, mode, 0, uid, gid));
  1229. dbgln_if(EXT2_DEBUG, "Ext2FS: create_directory: created new directory named '{} with inode {}", name, inode->index());
  1230. Vector<Ext2FSDirectoryEntry> entries;
  1231. entries.empend(".", inode->index(), static_cast<u8>(EXT2_FT_DIR));
  1232. entries.empend("..", parent_inode.index(), static_cast<u8>(EXT2_FT_DIR));
  1233. TRY(static_cast<Ext2FSInode&>(*inode).write_directory(entries));
  1234. TRY(parent_inode.increment_link_count());
  1235. auto& bgd = const_cast<ext2_group_desc&>(group_descriptor(group_index_from_inode(inode->identifier().index())));
  1236. ++bgd.bg_used_dirs_count;
  1237. m_block_group_descriptors_dirty = true;
  1238. return inode;
  1239. }
  1240. ErrorOr<NonnullRefPtr<Inode>> Ext2FS::create_inode(Ext2FSInode& parent_inode, StringView name, mode_t mode, dev_t dev, UserID uid, GroupID gid)
  1241. {
  1242. if (name.length() > EXT2_NAME_LEN)
  1243. return ENAMETOOLONG;
  1244. if (parent_inode.m_raw_inode.i_links_count == 0)
  1245. return ENOENT;
  1246. ext2_inode e2inode {};
  1247. auto now = kgettimeofday().to_truncated_seconds();
  1248. e2inode.i_mode = mode;
  1249. e2inode.i_uid = uid.value();
  1250. e2inode.i_gid = gid.value();
  1251. e2inode.i_size = 0;
  1252. e2inode.i_atime = now;
  1253. e2inode.i_ctime = now;
  1254. e2inode.i_mtime = now;
  1255. e2inode.i_dtime = 0;
  1256. e2inode.i_flags = 0;
  1257. // For directories, add +1 link count for the "." entry in self.
  1258. e2inode.i_links_count = is_directory(mode);
  1259. if (is_character_device(mode))
  1260. e2inode.i_block[0] = dev;
  1261. else if (is_block_device(mode))
  1262. e2inode.i_block[1] = dev;
  1263. auto inode_id = TRY(allocate_inode());
  1264. dbgln_if(EXT2_DEBUG, "Ext2FS: writing initial metadata for inode {}", inode_id.value());
  1265. TRY(write_ext2_inode(inode_id, e2inode));
  1266. auto new_inode = TRY(get_inode({ fsid(), inode_id }));
  1267. dbgln_if(EXT2_DEBUG, "Ext2FS: Adding inode '{}' (mode {:o}) to parent directory {}", name, mode, parent_inode.index());
  1268. TRY(parent_inode.add_child(*new_inode, name, mode));
  1269. return new_inode;
  1270. }
  1271. ErrorOr<void> Ext2FSInode::populate_lookup_cache() const
  1272. {
  1273. MutexLocker locker(m_inode_lock);
  1274. if (!m_lookup_cache.is_empty())
  1275. return {};
  1276. HashMap<String, InodeIndex> children;
  1277. TRY(traverse_as_directory([&children](auto& entry) -> ErrorOr<void> {
  1278. TRY(children.try_set(entry.name, entry.inode.index()));
  1279. return {};
  1280. }));
  1281. VERIFY(m_lookup_cache.is_empty());
  1282. m_lookup_cache = move(children);
  1283. return {};
  1284. }
  1285. ErrorOr<NonnullRefPtr<Inode>> Ext2FSInode::lookup(StringView name)
  1286. {
  1287. VERIFY(is_directory());
  1288. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]:lookup(): Looking up '{}'", identifier(), name);
  1289. TRY(populate_lookup_cache());
  1290. InodeIndex inode_index;
  1291. {
  1292. MutexLocker locker(m_inode_lock);
  1293. auto it = m_lookup_cache.find(name.hash(), [&](auto& entry) { return entry.key == name; });
  1294. if (it == m_lookup_cache.end()) {
  1295. dbgln_if(EXT2_DEBUG, "Ext2FSInode[{}]:lookup(): '{}' not found", identifier(), name);
  1296. return ENOENT;
  1297. }
  1298. inode_index = it->value;
  1299. }
  1300. return fs().get_inode({ fsid(), inode_index });
  1301. }
  1302. void Ext2FSInode::one_ref_left()
  1303. {
  1304. // FIXME: I would like to not live forever, but uncached Ext2FS is fucking painful right now.
  1305. }
  1306. ErrorOr<void> Ext2FSInode::set_atime(time_t t)
  1307. {
  1308. MutexLocker locker(m_inode_lock);
  1309. if (fs().is_readonly())
  1310. return EROFS;
  1311. m_raw_inode.i_atime = t;
  1312. set_metadata_dirty(true);
  1313. return {};
  1314. }
  1315. ErrorOr<void> Ext2FSInode::set_ctime(time_t t)
  1316. {
  1317. MutexLocker locker(m_inode_lock);
  1318. if (fs().is_readonly())
  1319. return EROFS;
  1320. m_raw_inode.i_ctime = t;
  1321. set_metadata_dirty(true);
  1322. return {};
  1323. }
  1324. ErrorOr<void> Ext2FSInode::set_mtime(time_t t)
  1325. {
  1326. MutexLocker locker(m_inode_lock);
  1327. if (fs().is_readonly())
  1328. return EROFS;
  1329. m_raw_inode.i_mtime = t;
  1330. set_metadata_dirty(true);
  1331. return {};
  1332. }
  1333. ErrorOr<void> Ext2FSInode::increment_link_count()
  1334. {
  1335. MutexLocker locker(m_inode_lock);
  1336. if (fs().is_readonly())
  1337. return EROFS;
  1338. constexpr size_t max_link_count = 65535;
  1339. if (m_raw_inode.i_links_count == max_link_count)
  1340. return EMLINK;
  1341. ++m_raw_inode.i_links_count;
  1342. set_metadata_dirty(true);
  1343. return {};
  1344. }
  1345. ErrorOr<void> Ext2FSInode::decrement_link_count()
  1346. {
  1347. MutexLocker locker(m_inode_lock);
  1348. if (fs().is_readonly())
  1349. return EROFS;
  1350. VERIFY(m_raw_inode.i_links_count);
  1351. --m_raw_inode.i_links_count;
  1352. set_metadata_dirty(true);
  1353. if (m_raw_inode.i_links_count == 0)
  1354. did_delete_self();
  1355. if (ref_count() == 1 && m_raw_inode.i_links_count == 0)
  1356. fs().uncache_inode(index());
  1357. return {};
  1358. }
  1359. void Ext2FS::uncache_inode(InodeIndex index)
  1360. {
  1361. MutexLocker locker(m_lock);
  1362. m_inode_cache.remove(index);
  1363. }
  1364. ErrorOr<void> Ext2FSInode::chmod(mode_t mode)
  1365. {
  1366. MutexLocker locker(m_inode_lock);
  1367. if (m_raw_inode.i_mode == mode)
  1368. return {};
  1369. m_raw_inode.i_mode = mode;
  1370. set_metadata_dirty(true);
  1371. return {};
  1372. }
  1373. ErrorOr<void> Ext2FSInode::chown(UserID uid, GroupID gid)
  1374. {
  1375. MutexLocker locker(m_inode_lock);
  1376. if (m_raw_inode.i_uid == uid && m_raw_inode.i_gid == gid)
  1377. return {};
  1378. m_raw_inode.i_uid = uid.value();
  1379. m_raw_inode.i_gid = gid.value();
  1380. set_metadata_dirty(true);
  1381. return {};
  1382. }
  1383. ErrorOr<void> Ext2FSInode::truncate(u64 size)
  1384. {
  1385. MutexLocker locker(m_inode_lock);
  1386. if (static_cast<u64>(m_raw_inode.i_size) == size)
  1387. return {};
  1388. TRY(resize(size));
  1389. set_metadata_dirty(true);
  1390. return {};
  1391. }
  1392. ErrorOr<int> Ext2FSInode::get_block_address(int index)
  1393. {
  1394. MutexLocker locker(m_inode_lock);
  1395. if (m_block_list.is_empty())
  1396. m_block_list = TRY(compute_block_list());
  1397. if (index < 0 || (size_t)index >= m_block_list.size())
  1398. return 0;
  1399. return m_block_list[index].value();
  1400. }
  1401. unsigned Ext2FS::total_block_count() const
  1402. {
  1403. MutexLocker locker(m_lock);
  1404. return super_block().s_blocks_count;
  1405. }
  1406. unsigned Ext2FS::free_block_count() const
  1407. {
  1408. MutexLocker locker(m_lock);
  1409. return super_block().s_free_blocks_count;
  1410. }
  1411. unsigned Ext2FS::total_inode_count() const
  1412. {
  1413. MutexLocker locker(m_lock);
  1414. return super_block().s_inodes_count;
  1415. }
  1416. unsigned Ext2FS::free_inode_count() const
  1417. {
  1418. MutexLocker locker(m_lock);
  1419. return super_block().s_free_inodes_count;
  1420. }
  1421. ErrorOr<void> Ext2FS::prepare_to_unmount()
  1422. {
  1423. MutexLocker locker(m_lock);
  1424. for (auto& it : m_inode_cache) {
  1425. if (it.value->ref_count() > 1)
  1426. return EBUSY;
  1427. }
  1428. m_inode_cache.clear();
  1429. m_root_inode = nullptr;
  1430. return {};
  1431. }
  1432. }