balloc.c 45 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410141114121413141414151416141714181419142014211422142314241425142614271428142914301431143214331434143514361437143814391440144114421443144414451446144714481449145014511452145314541455145614571458145914601461146214631464146514661467146814691470147114721473147414751476147714781479148014811482148314841485148614871488148914901491149214931494149514961497149814991500150115021503150415051506150715081509151015111512151315141515151615171518151915201521152215231524152515261527152815291530153115321533153415351536153715381539154015411542154315441545154615471548154915501551155215531554155515561557
  1. /*
  2. * linux/fs/ext2/balloc.c
  3. *
  4. * Copyright (C) 1992, 1993, 1994, 1995
  5. * Remy Card (card@masi.ibp.fr)
  6. * Laboratoire MASI - Institut Blaise Pascal
  7. * Universite Pierre et Marie Curie (Paris VI)
  8. *
  9. * Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
  10. * Big-endian to little-endian byte-swapping/bitmaps by
  11. * David S. Miller (davem@caip.rutgers.edu), 1995
  12. */
  13. #include "ext2.h"
  14. #include <linux/quotaops.h>
  15. #include <linux/slab.h>
  16. #include <linux/sched.h>
  17. #include <linux/buffer_head.h>
  18. #include <linux/capability.h>
  19. /*
  20. * balloc.c contains the blocks allocation and deallocation routines
  21. */
  22. /*
  23. * The free blocks are managed by bitmaps. A file system contains several
  24. * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
  25. * block for inodes, N blocks for the inode table and data blocks.
  26. *
  27. * The file system contains group descriptors which are located after the
  28. * super block. Each descriptor contains the number of the bitmap block and
  29. * the free blocks count in the block. The descriptors are loaded in memory
  30. * when a file system is mounted (see ext2_fill_super).
  31. */
  32. #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
  33. struct ext2_group_desc * ext2_get_group_desc(struct super_block * sb,
  34. unsigned int block_group,
  35. struct buffer_head ** bh)
  36. {
  37. unsigned long group_desc;
  38. unsigned long offset;
  39. struct ext2_group_desc * desc;
  40. struct ext2_sb_info *sbi = EXT2_SB(sb);
  41. if (block_group >= sbi->s_groups_count) {
  42. ext2_error (sb, "ext2_get_group_desc",
  43. "block_group >= groups_count - "
  44. "block_group = %d, groups_count = %lu",
  45. block_group, sbi->s_groups_count);
  46. return NULL;
  47. }
  48. group_desc = block_group >> EXT2_DESC_PER_BLOCK_BITS(sb);
  49. offset = block_group & (EXT2_DESC_PER_BLOCK(sb) - 1);
  50. if (!sbi->s_group_desc[group_desc]) {
  51. ext2_error (sb, "ext2_get_group_desc",
  52. "Group descriptor not loaded - "
  53. "block_group = %d, group_desc = %lu, desc = %lu",
  54. block_group, group_desc, offset);
  55. return NULL;
  56. }
  57. desc = (struct ext2_group_desc *) sbi->s_group_desc[group_desc]->b_data;
  58. if (bh)
  59. *bh = sbi->s_group_desc[group_desc];
  60. return desc + offset;
  61. }
  62. static int ext2_valid_block_bitmap(struct super_block *sb,
  63. struct ext2_group_desc *desc,
  64. unsigned int block_group,
  65. struct buffer_head *bh)
  66. {
  67. ext2_grpblk_t offset;
  68. ext2_grpblk_t next_zero_bit;
  69. ext2_fsblk_t bitmap_blk;
  70. ext2_fsblk_t group_first_block;
  71. group_first_block = ext2_group_first_block_no(sb, block_group);
  72. /* check whether block bitmap block number is set */
  73. bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
  74. offset = bitmap_blk - group_first_block;
  75. if (!ext2_test_bit(offset, bh->b_data))
  76. /* bad block bitmap */
  77. goto err_out;
  78. /* check whether the inode bitmap block number is set */
  79. bitmap_blk = le32_to_cpu(desc->bg_inode_bitmap);
  80. offset = bitmap_blk - group_first_block;
  81. if (!ext2_test_bit(offset, bh->b_data))
  82. /* bad block bitmap */
  83. goto err_out;
  84. /* check whether the inode table block number is set */
  85. bitmap_blk = le32_to_cpu(desc->bg_inode_table);
  86. offset = bitmap_blk - group_first_block;
  87. next_zero_bit = ext2_find_next_zero_bit(bh->b_data,
  88. offset + EXT2_SB(sb)->s_itb_per_group,
  89. offset);
  90. if (next_zero_bit >= offset + EXT2_SB(sb)->s_itb_per_group)
  91. /* good bitmap for inode tables */
  92. return 1;
  93. err_out:
  94. ext2_error(sb, __func__,
  95. "Invalid block bitmap - "
  96. "block_group = %d, block = %lu",
  97. block_group, bitmap_blk);
  98. return 0;
  99. }
  100. /*
  101. * Read the bitmap for a given block_group,and validate the
  102. * bits for block/inode/inode tables are set in the bitmaps
  103. *
  104. * Return buffer_head on success or NULL in case of failure.
  105. */
  106. static struct buffer_head *
  107. read_block_bitmap(struct super_block *sb, unsigned int block_group)
  108. {
  109. struct ext2_group_desc * desc;
  110. struct buffer_head * bh = NULL;
  111. ext2_fsblk_t bitmap_blk;
  112. desc = ext2_get_group_desc(sb, block_group, NULL);
  113. if (!desc)
  114. return NULL;
  115. bitmap_blk = le32_to_cpu(desc->bg_block_bitmap);
  116. bh = sb_getblk(sb, bitmap_blk);
  117. if (unlikely(!bh)) {
  118. ext2_error(sb, __func__,
  119. "Cannot read block bitmap - "
  120. "block_group = %d, block_bitmap = %u",
  121. block_group, le32_to_cpu(desc->bg_block_bitmap));
  122. return NULL;
  123. }
  124. if (likely(bh_uptodate_or_lock(bh)))
  125. return bh;
  126. if (bh_submit_read(bh) < 0) {
  127. brelse(bh);
  128. ext2_error(sb, __func__,
  129. "Cannot read block bitmap - "
  130. "block_group = %d, block_bitmap = %u",
  131. block_group, le32_to_cpu(desc->bg_block_bitmap));
  132. return NULL;
  133. }
  134. ext2_valid_block_bitmap(sb, desc, block_group, bh);
  135. /*
  136. * file system mounted not to panic on error, continue with corrupt
  137. * bitmap
  138. */
  139. return bh;
  140. }
  141. static void release_blocks(struct super_block *sb, int count)
  142. {
  143. if (count) {
  144. struct ext2_sb_info *sbi = EXT2_SB(sb);
  145. percpu_counter_add(&sbi->s_freeblocks_counter, count);
  146. sb->s_dirt = 1;
  147. }
  148. }
  149. static void group_adjust_blocks(struct super_block *sb, int group_no,
  150. struct ext2_group_desc *desc, struct buffer_head *bh, int count)
  151. {
  152. if (count) {
  153. struct ext2_sb_info *sbi = EXT2_SB(sb);
  154. unsigned free_blocks;
  155. spin_lock(sb_bgl_lock(sbi, group_no));
  156. free_blocks = le16_to_cpu(desc->bg_free_blocks_count);
  157. desc->bg_free_blocks_count = cpu_to_le16(free_blocks + count);
  158. spin_unlock(sb_bgl_lock(sbi, group_no));
  159. sb->s_dirt = 1;
  160. mark_buffer_dirty(bh);
  161. }
  162. }
  163. /*
  164. * The reservation window structure operations
  165. * --------------------------------------------
  166. * Operations include:
  167. * dump, find, add, remove, is_empty, find_next_reservable_window, etc.
  168. *
  169. * We use a red-black tree to represent per-filesystem reservation
  170. * windows.
  171. *
  172. */
  173. /**
  174. * __rsv_window_dump() -- Dump the filesystem block allocation reservation map
  175. * @rb_root: root of per-filesystem reservation rb tree
  176. * @verbose: verbose mode
  177. * @fn: function which wishes to dump the reservation map
  178. *
  179. * If verbose is turned on, it will print the whole block reservation
  180. * windows(start, end). Otherwise, it will only print out the "bad" windows,
  181. * those windows that overlap with their immediate neighbors.
  182. */
  183. #if 1
  184. static void __rsv_window_dump(struct rb_root *root, int verbose,
  185. const char *fn)
  186. {
  187. struct rb_node *n;
  188. struct ext2_reserve_window_node *rsv, *prev;
  189. int bad;
  190. restart:
  191. n = rb_first(root);
  192. bad = 0;
  193. prev = NULL;
  194. printk("Block Allocation Reservation Windows Map (%s):\n", fn);
  195. while (n) {
  196. rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
  197. if (verbose)
  198. printk("reservation window 0x%p "
  199. "start: %lu, end: %lu\n",
  200. rsv, rsv->rsv_start, rsv->rsv_end);
  201. if (rsv->rsv_start && rsv->rsv_start >= rsv->rsv_end) {
  202. printk("Bad reservation %p (start >= end)\n",
  203. rsv);
  204. bad = 1;
  205. }
  206. if (prev && prev->rsv_end >= rsv->rsv_start) {
  207. printk("Bad reservation %p (prev->end >= start)\n",
  208. rsv);
  209. bad = 1;
  210. }
  211. if (bad) {
  212. if (!verbose) {
  213. printk("Restarting reservation walk in verbose mode\n");
  214. verbose = 1;
  215. goto restart;
  216. }
  217. }
  218. n = rb_next(n);
  219. prev = rsv;
  220. }
  221. printk("Window map complete.\n");
  222. BUG_ON(bad);
  223. }
  224. #define rsv_window_dump(root, verbose) \
  225. __rsv_window_dump((root), (verbose), __func__)
  226. #else
  227. #define rsv_window_dump(root, verbose) do {} while (0)
  228. #endif
  229. /**
  230. * goal_in_my_reservation()
  231. * @rsv: inode's reservation window
  232. * @grp_goal: given goal block relative to the allocation block group
  233. * @group: the current allocation block group
  234. * @sb: filesystem super block
  235. *
  236. * Test if the given goal block (group relative) is within the file's
  237. * own block reservation window range.
  238. *
  239. * If the reservation window is outside the goal allocation group, return 0;
  240. * grp_goal (given goal block) could be -1, which means no specific
  241. * goal block. In this case, always return 1.
  242. * If the goal block is within the reservation window, return 1;
  243. * otherwise, return 0;
  244. */
  245. static int
  246. goal_in_my_reservation(struct ext2_reserve_window *rsv, ext2_grpblk_t grp_goal,
  247. unsigned int group, struct super_block * sb)
  248. {
  249. ext2_fsblk_t group_first_block, group_last_block;
  250. group_first_block = ext2_group_first_block_no(sb, group);
  251. group_last_block = group_first_block + EXT2_BLOCKS_PER_GROUP(sb) - 1;
  252. if ((rsv->_rsv_start > group_last_block) ||
  253. (rsv->_rsv_end < group_first_block))
  254. return 0;
  255. if ((grp_goal >= 0) && ((grp_goal + group_first_block < rsv->_rsv_start)
  256. || (grp_goal + group_first_block > rsv->_rsv_end)))
  257. return 0;
  258. return 1;
  259. }
  260. /**
  261. * search_reserve_window()
  262. * @rb_root: root of reservation tree
  263. * @goal: target allocation block
  264. *
  265. * Find the reserved window which includes the goal, or the previous one
  266. * if the goal is not in any window.
  267. * Returns NULL if there are no windows or if all windows start after the goal.
  268. */
  269. static struct ext2_reserve_window_node *
  270. search_reserve_window(struct rb_root *root, ext2_fsblk_t goal)
  271. {
  272. struct rb_node *n = root->rb_node;
  273. struct ext2_reserve_window_node *rsv;
  274. if (!n)
  275. return NULL;
  276. do {
  277. rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
  278. if (goal < rsv->rsv_start)
  279. n = n->rb_left;
  280. else if (goal > rsv->rsv_end)
  281. n = n->rb_right;
  282. else
  283. return rsv;
  284. } while (n);
  285. /*
  286. * We've fallen off the end of the tree: the goal wasn't inside
  287. * any particular node. OK, the previous node must be to one
  288. * side of the interval containing the goal. If it's the RHS,
  289. * we need to back up one.
  290. */
  291. if (rsv->rsv_start > goal) {
  292. n = rb_prev(&rsv->rsv_node);
  293. rsv = rb_entry(n, struct ext2_reserve_window_node, rsv_node);
  294. }
  295. return rsv;
  296. }
  297. /*
  298. * ext2_rsv_window_add() -- Insert a window to the block reservation rb tree.
  299. * @sb: super block
  300. * @rsv: reservation window to add
  301. *
  302. * Must be called with rsv_lock held.
  303. */
  304. void ext2_rsv_window_add(struct super_block *sb,
  305. struct ext2_reserve_window_node *rsv)
  306. {
  307. struct rb_root *root = &EXT2_SB(sb)->s_rsv_window_root;
  308. struct rb_node *node = &rsv->rsv_node;
  309. ext2_fsblk_t start = rsv->rsv_start;
  310. struct rb_node ** p = &root->rb_node;
  311. struct rb_node * parent = NULL;
  312. struct ext2_reserve_window_node *this;
  313. while (*p)
  314. {
  315. parent = *p;
  316. this = rb_entry(parent, struct ext2_reserve_window_node, rsv_node);
  317. if (start < this->rsv_start)
  318. p = &(*p)->rb_left;
  319. else if (start > this->rsv_end)
  320. p = &(*p)->rb_right;
  321. else {
  322. rsv_window_dump(root, 1);
  323. BUG();
  324. }
  325. }
  326. rb_link_node(node, parent, p);
  327. rb_insert_color(node, root);
  328. }
  329. /**
  330. * rsv_window_remove() -- unlink a window from the reservation rb tree
  331. * @sb: super block
  332. * @rsv: reservation window to remove
  333. *
  334. * Mark the block reservation window as not allocated, and unlink it
  335. * from the filesystem reservation window rb tree. Must be called with
  336. * rsv_lock held.
  337. */
  338. static void rsv_window_remove(struct super_block *sb,
  339. struct ext2_reserve_window_node *rsv)
  340. {
  341. rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
  342. rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
  343. rsv->rsv_alloc_hit = 0;
  344. rb_erase(&rsv->rsv_node, &EXT2_SB(sb)->s_rsv_window_root);
  345. }
  346. /*
  347. * rsv_is_empty() -- Check if the reservation window is allocated.
  348. * @rsv: given reservation window to check
  349. *
  350. * returns 1 if the end block is EXT2_RESERVE_WINDOW_NOT_ALLOCATED.
  351. */
  352. static inline int rsv_is_empty(struct ext2_reserve_window *rsv)
  353. {
  354. /* a valid reservation end block could not be 0 */
  355. return (rsv->_rsv_end == EXT2_RESERVE_WINDOW_NOT_ALLOCATED);
  356. }
  357. /**
  358. * ext2_init_block_alloc_info()
  359. * @inode: file inode structure
  360. *
  361. * Allocate and initialize the reservation window structure, and
  362. * link the window to the ext2 inode structure at last
  363. *
  364. * The reservation window structure is only dynamically allocated
  365. * and linked to ext2 inode the first time the open file
  366. * needs a new block. So, before every ext2_new_block(s) call, for
  367. * regular files, we should check whether the reservation window
  368. * structure exists or not. In the latter case, this function is called.
  369. * Fail to do so will result in block reservation being turned off for that
  370. * open file.
  371. *
  372. * This function is called from ext2_get_blocks_handle(), also called
  373. * when setting the reservation window size through ioctl before the file
  374. * is open for write (needs block allocation).
  375. *
  376. * Needs truncate_mutex protection prior to calling this function.
  377. */
  378. void ext2_init_block_alloc_info(struct inode *inode)
  379. {
  380. struct ext2_inode_info *ei = EXT2_I(inode);
  381. struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
  382. struct super_block *sb = inode->i_sb;
  383. block_i = kmalloc(sizeof(*block_i), GFP_NOFS);
  384. if (block_i) {
  385. struct ext2_reserve_window_node *rsv = &block_i->rsv_window_node;
  386. rsv->rsv_start = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
  387. rsv->rsv_end = EXT2_RESERVE_WINDOW_NOT_ALLOCATED;
  388. /*
  389. * if filesystem is mounted with NORESERVATION, the goal
  390. * reservation window size is set to zero to indicate
  391. * block reservation is off
  392. */
  393. if (!test_opt(sb, RESERVATION))
  394. rsv->rsv_goal_size = 0;
  395. else
  396. rsv->rsv_goal_size = EXT2_DEFAULT_RESERVE_BLOCKS;
  397. rsv->rsv_alloc_hit = 0;
  398. block_i->last_alloc_logical_block = 0;
  399. block_i->last_alloc_physical_block = 0;
  400. }
  401. ei->i_block_alloc_info = block_i;
  402. }
  403. /**
  404. * ext2_discard_reservation()
  405. * @inode: inode
  406. *
  407. * Discard(free) block reservation window on last file close, or truncate
  408. * or at last iput().
  409. *
  410. * It is being called in three cases:
  411. * ext2_release_file(): last writer closes the file
  412. * ext2_clear_inode(): last iput(), when nobody links to this file.
  413. * ext2_truncate(): when the block indirect map is about to change.
  414. */
  415. void ext2_discard_reservation(struct inode *inode)
  416. {
  417. struct ext2_inode_info *ei = EXT2_I(inode);
  418. struct ext2_block_alloc_info *block_i = ei->i_block_alloc_info;
  419. struct ext2_reserve_window_node *rsv;
  420. spinlock_t *rsv_lock = &EXT2_SB(inode->i_sb)->s_rsv_window_lock;
  421. if (!block_i)
  422. return;
  423. rsv = &block_i->rsv_window_node;
  424. if (!rsv_is_empty(&rsv->rsv_window)) {
  425. spin_lock(rsv_lock);
  426. if (!rsv_is_empty(&rsv->rsv_window))
  427. rsv_window_remove(inode->i_sb, rsv);
  428. spin_unlock(rsv_lock);
  429. }
  430. }
  431. /**
  432. * ext2_free_blocks_sb() -- Free given blocks and update quota and i_blocks
  433. * @inode: inode
  434. * @block: start physcial block to free
  435. * @count: number of blocks to free
  436. */
  437. void ext2_free_blocks (struct inode * inode, unsigned long block,
  438. unsigned long count)
  439. {
  440. struct buffer_head *bitmap_bh = NULL;
  441. struct buffer_head * bh2;
  442. unsigned long block_group;
  443. unsigned long bit;
  444. unsigned long i;
  445. unsigned long overflow;
  446. struct super_block * sb = inode->i_sb;
  447. struct ext2_sb_info * sbi = EXT2_SB(sb);
  448. struct ext2_group_desc * desc;
  449. struct ext2_super_block * es = sbi->s_es;
  450. unsigned freed = 0, group_freed;
  451. if (block < le32_to_cpu(es->s_first_data_block) ||
  452. block + count < block ||
  453. block + count > le32_to_cpu(es->s_blocks_count)) {
  454. ext2_error (sb, "ext2_free_blocks",
  455. "Freeing blocks not in datazone - "
  456. "block = %lu, count = %lu", block, count);
  457. goto error_return;
  458. }
  459. ext2_debug ("freeing block(s) %lu-%lu\n", block, block + count - 1);
  460. do_more:
  461. overflow = 0;
  462. block_group = (block - le32_to_cpu(es->s_first_data_block)) /
  463. EXT2_BLOCKS_PER_GROUP(sb);
  464. bit = (block - le32_to_cpu(es->s_first_data_block)) %
  465. EXT2_BLOCKS_PER_GROUP(sb);
  466. /*
  467. * Check to see if we are freeing blocks across a group
  468. * boundary.
  469. */
  470. if (bit + count > EXT2_BLOCKS_PER_GROUP(sb)) {
  471. overflow = bit + count - EXT2_BLOCKS_PER_GROUP(sb);
  472. count -= overflow;
  473. }
  474. brelse(bitmap_bh);
  475. bitmap_bh = read_block_bitmap(sb, block_group);
  476. if (!bitmap_bh)
  477. goto error_return;
  478. desc = ext2_get_group_desc (sb, block_group, &bh2);
  479. if (!desc)
  480. goto error_return;
  481. if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) ||
  482. in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) ||
  483. in_range (block, le32_to_cpu(desc->bg_inode_table),
  484. sbi->s_itb_per_group) ||
  485. in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table),
  486. sbi->s_itb_per_group)) {
  487. ext2_error (sb, "ext2_free_blocks",
  488. "Freeing blocks in system zones - "
  489. "Block = %lu, count = %lu",
  490. block, count);
  491. goto error_return;
  492. }
  493. for (i = 0, group_freed = 0; i < count; i++) {
  494. if (!ext2_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
  495. bit + i, bitmap_bh->b_data)) {
  496. ext2_error(sb, __func__,
  497. "bit already cleared for block %lu", block + i);
  498. } else {
  499. group_freed++;
  500. }
  501. }
  502. mark_buffer_dirty(bitmap_bh);
  503. if (sb->s_flags & MS_SYNCHRONOUS)
  504. sync_dirty_buffer(bitmap_bh);
  505. group_adjust_blocks(sb, block_group, desc, bh2, group_freed);
  506. freed += group_freed;
  507. if (overflow) {
  508. block += count;
  509. count = overflow;
  510. goto do_more;
  511. }
  512. error_return:
  513. brelse(bitmap_bh);
  514. release_blocks(sb, freed);
  515. dquot_free_block_nodirty(inode, freed);
  516. mark_inode_dirty(inode);
  517. }
  518. /**
  519. * bitmap_search_next_usable_block()
  520. * @start: the starting block (group relative) of the search
  521. * @bh: bufferhead contains the block group bitmap
  522. * @maxblocks: the ending block (group relative) of the reservation
  523. *
  524. * The bitmap search --- search forward through the actual bitmap on disk until
  525. * we find a bit free.
  526. */
  527. static ext2_grpblk_t
  528. bitmap_search_next_usable_block(ext2_grpblk_t start, struct buffer_head *bh,
  529. ext2_grpblk_t maxblocks)
  530. {
  531. ext2_grpblk_t next;
  532. next = ext2_find_next_zero_bit(bh->b_data, maxblocks, start);
  533. if (next >= maxblocks)
  534. return -1;
  535. return next;
  536. }
  537. /**
  538. * find_next_usable_block()
  539. * @start: the starting block (group relative) to find next
  540. * allocatable block in bitmap.
  541. * @bh: bufferhead contains the block group bitmap
  542. * @maxblocks: the ending block (group relative) for the search
  543. *
  544. * Find an allocatable block in a bitmap. We perform the "most
  545. * appropriate allocation" algorithm of looking for a free block near
  546. * the initial goal; then for a free byte somewhere in the bitmap;
  547. * then for any free bit in the bitmap.
  548. */
  549. static ext2_grpblk_t
  550. find_next_usable_block(int start, struct buffer_head *bh, int maxblocks)
  551. {
  552. ext2_grpblk_t here, next;
  553. char *p, *r;
  554. if (start > 0) {
  555. /*
  556. * The goal was occupied; search forward for a free
  557. * block within the next XX blocks.
  558. *
  559. * end_goal is more or less random, but it has to be
  560. * less than EXT2_BLOCKS_PER_GROUP. Aligning up to the
  561. * next 64-bit boundary is simple..
  562. */
  563. ext2_grpblk_t end_goal = (start + 63) & ~63;
  564. if (end_goal > maxblocks)
  565. end_goal = maxblocks;
  566. here = ext2_find_next_zero_bit(bh->b_data, end_goal, start);
  567. if (here < end_goal)
  568. return here;
  569. ext2_debug("Bit not found near goal\n");
  570. }
  571. here = start;
  572. if (here < 0)
  573. here = 0;
  574. p = ((char *)bh->b_data) + (here >> 3);
  575. r = memscan(p, 0, ((maxblocks + 7) >> 3) - (here >> 3));
  576. next = (r - ((char *)bh->b_data)) << 3;
  577. if (next < maxblocks && next >= here)
  578. return next;
  579. here = bitmap_search_next_usable_block(here, bh, maxblocks);
  580. return here;
  581. }
  582. /*
  583. * ext2_try_to_allocate()
  584. * @sb: superblock
  585. * @handle: handle to this transaction
  586. * @group: given allocation block group
  587. * @bitmap_bh: bufferhead holds the block bitmap
  588. * @grp_goal: given target block within the group
  589. * @count: target number of blocks to allocate
  590. * @my_rsv: reservation window
  591. *
  592. * Attempt to allocate blocks within a give range. Set the range of allocation
  593. * first, then find the first free bit(s) from the bitmap (within the range),
  594. * and at last, allocate the blocks by claiming the found free bit as allocated.
  595. *
  596. * To set the range of this allocation:
  597. * if there is a reservation window, only try to allocate block(s)
  598. * from the file's own reservation window;
  599. * Otherwise, the allocation range starts from the give goal block,
  600. * ends at the block group's last block.
  601. *
  602. * If we failed to allocate the desired block then we may end up crossing to a
  603. * new bitmap.
  604. */
  605. static int
  606. ext2_try_to_allocate(struct super_block *sb, int group,
  607. struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
  608. unsigned long *count,
  609. struct ext2_reserve_window *my_rsv)
  610. {
  611. ext2_fsblk_t group_first_block;
  612. ext2_grpblk_t start, end;
  613. unsigned long num = 0;
  614. /* we do allocation within the reservation window if we have a window */
  615. if (my_rsv) {
  616. group_first_block = ext2_group_first_block_no(sb, group);
  617. if (my_rsv->_rsv_start >= group_first_block)
  618. start = my_rsv->_rsv_start - group_first_block;
  619. else
  620. /* reservation window cross group boundary */
  621. start = 0;
  622. end = my_rsv->_rsv_end - group_first_block + 1;
  623. if (end > EXT2_BLOCKS_PER_GROUP(sb))
  624. /* reservation window crosses group boundary */
  625. end = EXT2_BLOCKS_PER_GROUP(sb);
  626. if ((start <= grp_goal) && (grp_goal < end))
  627. start = grp_goal;
  628. else
  629. grp_goal = -1;
  630. } else {
  631. if (grp_goal > 0)
  632. start = grp_goal;
  633. else
  634. start = 0;
  635. end = EXT2_BLOCKS_PER_GROUP(sb);
  636. }
  637. BUG_ON(start > EXT2_BLOCKS_PER_GROUP(sb));
  638. repeat:
  639. if (grp_goal < 0) {
  640. grp_goal = find_next_usable_block(start, bitmap_bh, end);
  641. if (grp_goal < 0)
  642. goto fail_access;
  643. if (!my_rsv) {
  644. int i;
  645. for (i = 0; i < 7 && grp_goal > start &&
  646. !ext2_test_bit(grp_goal - 1,
  647. bitmap_bh->b_data);
  648. i++, grp_goal--)
  649. ;
  650. }
  651. }
  652. start = grp_goal;
  653. if (ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group), grp_goal,
  654. bitmap_bh->b_data)) {
  655. /*
  656. * The block was allocated by another thread, or it was
  657. * allocated and then freed by another thread
  658. */
  659. start++;
  660. grp_goal++;
  661. if (start >= end)
  662. goto fail_access;
  663. goto repeat;
  664. }
  665. num++;
  666. grp_goal++;
  667. while (num < *count && grp_goal < end
  668. && !ext2_set_bit_atomic(sb_bgl_lock(EXT2_SB(sb), group),
  669. grp_goal, bitmap_bh->b_data)) {
  670. num++;
  671. grp_goal++;
  672. }
  673. *count = num;
  674. return grp_goal - num;
  675. fail_access:
  676. *count = num;
  677. return -1;
  678. }
  679. /**
  680. * find_next_reservable_window():
  681. * find a reservable space within the given range.
  682. * It does not allocate the reservation window for now:
  683. * alloc_new_reservation() will do the work later.
  684. *
  685. * @search_head: the head of the searching list;
  686. * This is not necessarily the list head of the whole filesystem
  687. *
  688. * We have both head and start_block to assist the search
  689. * for the reservable space. The list starts from head,
  690. * but we will shift to the place where start_block is,
  691. * then start from there, when looking for a reservable space.
  692. *
  693. * @size: the target new reservation window size
  694. *
  695. * @group_first_block: the first block we consider to start
  696. * the real search from
  697. *
  698. * @last_block:
  699. * the maximum block number that our goal reservable space
  700. * could start from. This is normally the last block in this
  701. * group. The search will end when we found the start of next
  702. * possible reservable space is out of this boundary.
  703. * This could handle the cross boundary reservation window
  704. * request.
  705. *
  706. * basically we search from the given range, rather than the whole
  707. * reservation double linked list, (start_block, last_block)
  708. * to find a free region that is of my size and has not
  709. * been reserved.
  710. *
  711. */
  712. static int find_next_reservable_window(
  713. struct ext2_reserve_window_node *search_head,
  714. struct ext2_reserve_window_node *my_rsv,
  715. struct super_block * sb,
  716. ext2_fsblk_t start_block,
  717. ext2_fsblk_t last_block)
  718. {
  719. struct rb_node *next;
  720. struct ext2_reserve_window_node *rsv, *prev;
  721. ext2_fsblk_t cur;
  722. int size = my_rsv->rsv_goal_size;
  723. /* TODO: make the start of the reservation window byte-aligned */
  724. /* cur = *start_block & ~7;*/
  725. cur = start_block;
  726. rsv = search_head;
  727. if (!rsv)
  728. return -1;
  729. while (1) {
  730. if (cur <= rsv->rsv_end)
  731. cur = rsv->rsv_end + 1;
  732. /* TODO?
  733. * in the case we could not find a reservable space
  734. * that is what is expected, during the re-search, we could
  735. * remember what's the largest reservable space we could have
  736. * and return that one.
  737. *
  738. * For now it will fail if we could not find the reservable
  739. * space with expected-size (or more)...
  740. */
  741. if (cur > last_block)
  742. return -1; /* fail */
  743. prev = rsv;
  744. next = rb_next(&rsv->rsv_node);
  745. rsv = rb_entry(next,struct ext2_reserve_window_node,rsv_node);
  746. /*
  747. * Reached the last reservation, we can just append to the
  748. * previous one.
  749. */
  750. if (!next)
  751. break;
  752. if (cur + size <= rsv->rsv_start) {
  753. /*
  754. * Found a reserveable space big enough. We could
  755. * have a reservation across the group boundary here
  756. */
  757. break;
  758. }
  759. }
  760. /*
  761. * we come here either :
  762. * when we reach the end of the whole list,
  763. * and there is empty reservable space after last entry in the list.
  764. * append it to the end of the list.
  765. *
  766. * or we found one reservable space in the middle of the list,
  767. * return the reservation window that we could append to.
  768. * succeed.
  769. */
  770. if ((prev != my_rsv) && (!rsv_is_empty(&my_rsv->rsv_window)))
  771. rsv_window_remove(sb, my_rsv);
  772. /*
  773. * Let's book the whole avaliable window for now. We will check the
  774. * disk bitmap later and then, if there are free blocks then we adjust
  775. * the window size if it's larger than requested.
  776. * Otherwise, we will remove this node from the tree next time
  777. * call find_next_reservable_window.
  778. */
  779. my_rsv->rsv_start = cur;
  780. my_rsv->rsv_end = cur + size - 1;
  781. my_rsv->rsv_alloc_hit = 0;
  782. if (prev != my_rsv)
  783. ext2_rsv_window_add(sb, my_rsv);
  784. return 0;
  785. }
  786. /**
  787. * alloc_new_reservation()--allocate a new reservation window
  788. *
  789. * To make a new reservation, we search part of the filesystem
  790. * reservation list (the list that inside the group). We try to
  791. * allocate a new reservation window near the allocation goal,
  792. * or the beginning of the group, if there is no goal.
  793. *
  794. * We first find a reservable space after the goal, then from
  795. * there, we check the bitmap for the first free block after
  796. * it. If there is no free block until the end of group, then the
  797. * whole group is full, we failed. Otherwise, check if the free
  798. * block is inside the expected reservable space, if so, we
  799. * succeed.
  800. * If the first free block is outside the reservable space, then
  801. * start from the first free block, we search for next available
  802. * space, and go on.
  803. *
  804. * on succeed, a new reservation will be found and inserted into the list
  805. * It contains at least one free block, and it does not overlap with other
  806. * reservation windows.
  807. *
  808. * failed: we failed to find a reservation window in this group
  809. *
  810. * @rsv: the reservation
  811. *
  812. * @grp_goal: The goal (group-relative). It is where the search for a
  813. * free reservable space should start from.
  814. * if we have a goal(goal >0 ), then start from there,
  815. * no goal(goal = -1), we start from the first block
  816. * of the group.
  817. *
  818. * @sb: the super block
  819. * @group: the group we are trying to allocate in
  820. * @bitmap_bh: the block group block bitmap
  821. *
  822. */
  823. static int alloc_new_reservation(struct ext2_reserve_window_node *my_rsv,
  824. ext2_grpblk_t grp_goal, struct super_block *sb,
  825. unsigned int group, struct buffer_head *bitmap_bh)
  826. {
  827. struct ext2_reserve_window_node *search_head;
  828. ext2_fsblk_t group_first_block, group_end_block, start_block;
  829. ext2_grpblk_t first_free_block;
  830. struct rb_root *fs_rsv_root = &EXT2_SB(sb)->s_rsv_window_root;
  831. unsigned long size;
  832. int ret;
  833. spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
  834. group_first_block = ext2_group_first_block_no(sb, group);
  835. group_end_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
  836. if (grp_goal < 0)
  837. start_block = group_first_block;
  838. else
  839. start_block = grp_goal + group_first_block;
  840. size = my_rsv->rsv_goal_size;
  841. if (!rsv_is_empty(&my_rsv->rsv_window)) {
  842. /*
  843. * if the old reservation is cross group boundary
  844. * and if the goal is inside the old reservation window,
  845. * we will come here when we just failed to allocate from
  846. * the first part of the window. We still have another part
  847. * that belongs to the next group. In this case, there is no
  848. * point to discard our window and try to allocate a new one
  849. * in this group(which will fail). we should
  850. * keep the reservation window, just simply move on.
  851. *
  852. * Maybe we could shift the start block of the reservation
  853. * window to the first block of next group.
  854. */
  855. if ((my_rsv->rsv_start <= group_end_block) &&
  856. (my_rsv->rsv_end > group_end_block) &&
  857. (start_block >= my_rsv->rsv_start))
  858. return -1;
  859. if ((my_rsv->rsv_alloc_hit >
  860. (my_rsv->rsv_end - my_rsv->rsv_start + 1) / 2)) {
  861. /*
  862. * if the previously allocation hit ratio is
  863. * greater than 1/2, then we double the size of
  864. * the reservation window the next time,
  865. * otherwise we keep the same size window
  866. */
  867. size = size * 2;
  868. if (size > EXT2_MAX_RESERVE_BLOCKS)
  869. size = EXT2_MAX_RESERVE_BLOCKS;
  870. my_rsv->rsv_goal_size= size;
  871. }
  872. }
  873. spin_lock(rsv_lock);
  874. /*
  875. * shift the search start to the window near the goal block
  876. */
  877. search_head = search_reserve_window(fs_rsv_root, start_block);
  878. /*
  879. * find_next_reservable_window() simply finds a reservable window
  880. * inside the given range(start_block, group_end_block).
  881. *
  882. * To make sure the reservation window has a free bit inside it, we
  883. * need to check the bitmap after we found a reservable window.
  884. */
  885. retry:
  886. ret = find_next_reservable_window(search_head, my_rsv, sb,
  887. start_block, group_end_block);
  888. if (ret == -1) {
  889. if (!rsv_is_empty(&my_rsv->rsv_window))
  890. rsv_window_remove(sb, my_rsv);
  891. spin_unlock(rsv_lock);
  892. return -1;
  893. }
  894. /*
  895. * On success, find_next_reservable_window() returns the
  896. * reservation window where there is a reservable space after it.
  897. * Before we reserve this reservable space, we need
  898. * to make sure there is at least a free block inside this region.
  899. *
  900. * Search the first free bit on the block bitmap. Search starts from
  901. * the start block of the reservable space we just found.
  902. */
  903. spin_unlock(rsv_lock);
  904. first_free_block = bitmap_search_next_usable_block(
  905. my_rsv->rsv_start - group_first_block,
  906. bitmap_bh, group_end_block - group_first_block + 1);
  907. if (first_free_block < 0) {
  908. /*
  909. * no free block left on the bitmap, no point
  910. * to reserve the space. return failed.
  911. */
  912. spin_lock(rsv_lock);
  913. if (!rsv_is_empty(&my_rsv->rsv_window))
  914. rsv_window_remove(sb, my_rsv);
  915. spin_unlock(rsv_lock);
  916. return -1; /* failed */
  917. }
  918. start_block = first_free_block + group_first_block;
  919. /*
  920. * check if the first free block is within the
  921. * free space we just reserved
  922. */
  923. if (start_block >= my_rsv->rsv_start && start_block <= my_rsv->rsv_end)
  924. return 0; /* success */
  925. /*
  926. * if the first free bit we found is out of the reservable space
  927. * continue search for next reservable space,
  928. * start from where the free block is,
  929. * we also shift the list head to where we stopped last time
  930. */
  931. search_head = my_rsv;
  932. spin_lock(rsv_lock);
  933. goto retry;
  934. }
  935. /**
  936. * try_to_extend_reservation()
  937. * @my_rsv: given reservation window
  938. * @sb: super block
  939. * @size: the delta to extend
  940. *
  941. * Attempt to expand the reservation window large enough to have
  942. * required number of free blocks
  943. *
  944. * Since ext2_try_to_allocate() will always allocate blocks within
  945. * the reservation window range, if the window size is too small,
  946. * multiple blocks allocation has to stop at the end of the reservation
  947. * window. To make this more efficient, given the total number of
  948. * blocks needed and the current size of the window, we try to
  949. * expand the reservation window size if necessary on a best-effort
  950. * basis before ext2_new_blocks() tries to allocate blocks.
  951. */
  952. static void try_to_extend_reservation(struct ext2_reserve_window_node *my_rsv,
  953. struct super_block *sb, int size)
  954. {
  955. struct ext2_reserve_window_node *next_rsv;
  956. struct rb_node *next;
  957. spinlock_t *rsv_lock = &EXT2_SB(sb)->s_rsv_window_lock;
  958. if (!spin_trylock(rsv_lock))
  959. return;
  960. next = rb_next(&my_rsv->rsv_node);
  961. if (!next)
  962. my_rsv->rsv_end += size;
  963. else {
  964. next_rsv = rb_entry(next, struct ext2_reserve_window_node, rsv_node);
  965. if ((next_rsv->rsv_start - my_rsv->rsv_end - 1) >= size)
  966. my_rsv->rsv_end += size;
  967. else
  968. my_rsv->rsv_end = next_rsv->rsv_start - 1;
  969. }
  970. spin_unlock(rsv_lock);
  971. }
  972. /**
  973. * ext2_try_to_allocate_with_rsv()
  974. * @sb: superblock
  975. * @group: given allocation block group
  976. * @bitmap_bh: bufferhead holds the block bitmap
  977. * @grp_goal: given target block within the group
  978. * @count: target number of blocks to allocate
  979. * @my_rsv: reservation window
  980. *
  981. * This is the main function used to allocate a new block and its reservation
  982. * window.
  983. *
  984. * Each time when a new block allocation is need, first try to allocate from
  985. * its own reservation. If it does not have a reservation window, instead of
  986. * looking for a free bit on bitmap first, then look up the reservation list to
  987. * see if it is inside somebody else's reservation window, we try to allocate a
  988. * reservation window for it starting from the goal first. Then do the block
  989. * allocation within the reservation window.
  990. *
  991. * This will avoid keeping on searching the reservation list again and
  992. * again when somebody is looking for a free block (without
  993. * reservation), and there are lots of free blocks, but they are all
  994. * being reserved.
  995. *
  996. * We use a red-black tree for the per-filesystem reservation list.
  997. */
  998. static ext2_grpblk_t
  999. ext2_try_to_allocate_with_rsv(struct super_block *sb, unsigned int group,
  1000. struct buffer_head *bitmap_bh, ext2_grpblk_t grp_goal,
  1001. struct ext2_reserve_window_node * my_rsv,
  1002. unsigned long *count)
  1003. {
  1004. ext2_fsblk_t group_first_block, group_last_block;
  1005. ext2_grpblk_t ret = 0;
  1006. unsigned long num = *count;
  1007. /*
  1008. * we don't deal with reservation when
  1009. * filesystem is mounted without reservation
  1010. * or the file is not a regular file
  1011. * or last attempt to allocate a block with reservation turned on failed
  1012. */
  1013. if (my_rsv == NULL) {
  1014. return ext2_try_to_allocate(sb, group, bitmap_bh,
  1015. grp_goal, count, NULL);
  1016. }
  1017. /*
  1018. * grp_goal is a group relative block number (if there is a goal)
  1019. * 0 <= grp_goal < EXT2_BLOCKS_PER_GROUP(sb)
  1020. * first block is a filesystem wide block number
  1021. * first block is the block number of the first block in this group
  1022. */
  1023. group_first_block = ext2_group_first_block_no(sb, group);
  1024. group_last_block = group_first_block + (EXT2_BLOCKS_PER_GROUP(sb) - 1);
  1025. /*
  1026. * Basically we will allocate a new block from inode's reservation
  1027. * window.
  1028. *
  1029. * We need to allocate a new reservation window, if:
  1030. * a) inode does not have a reservation window; or
  1031. * b) last attempt to allocate a block from existing reservation
  1032. * failed; or
  1033. * c) we come here with a goal and with a reservation window
  1034. *
  1035. * We do not need to allocate a new reservation window if we come here
  1036. * at the beginning with a goal and the goal is inside the window, or
  1037. * we don't have a goal but already have a reservation window.
  1038. * then we could go to allocate from the reservation window directly.
  1039. */
  1040. while (1) {
  1041. if (rsv_is_empty(&my_rsv->rsv_window) || (ret < 0) ||
  1042. !goal_in_my_reservation(&my_rsv->rsv_window,
  1043. grp_goal, group, sb)) {
  1044. if (my_rsv->rsv_goal_size < *count)
  1045. my_rsv->rsv_goal_size = *count;
  1046. ret = alloc_new_reservation(my_rsv, grp_goal, sb,
  1047. group, bitmap_bh);
  1048. if (ret < 0)
  1049. break; /* failed */
  1050. if (!goal_in_my_reservation(&my_rsv->rsv_window,
  1051. grp_goal, group, sb))
  1052. grp_goal = -1;
  1053. } else if (grp_goal >= 0) {
  1054. int curr = my_rsv->rsv_end -
  1055. (grp_goal + group_first_block) + 1;
  1056. if (curr < *count)
  1057. try_to_extend_reservation(my_rsv, sb,
  1058. *count - curr);
  1059. }
  1060. if ((my_rsv->rsv_start > group_last_block) ||
  1061. (my_rsv->rsv_end < group_first_block)) {
  1062. rsv_window_dump(&EXT2_SB(sb)->s_rsv_window_root, 1);
  1063. BUG();
  1064. }
  1065. ret = ext2_try_to_allocate(sb, group, bitmap_bh, grp_goal,
  1066. &num, &my_rsv->rsv_window);
  1067. if (ret >= 0) {
  1068. my_rsv->rsv_alloc_hit += num;
  1069. *count = num;
  1070. break; /* succeed */
  1071. }
  1072. num = *count;
  1073. }
  1074. return ret;
  1075. }
  1076. /**
  1077. * ext2_has_free_blocks()
  1078. * @sbi: in-core super block structure.
  1079. *
  1080. * Check if filesystem has at least 1 free block available for allocation.
  1081. */
  1082. static int ext2_has_free_blocks(struct ext2_sb_info *sbi)
  1083. {
  1084. ext2_fsblk_t free_blocks, root_blocks;
  1085. free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
  1086. root_blocks = le32_to_cpu(sbi->s_es->s_r_blocks_count);
  1087. if (free_blocks < root_blocks + 1 && !capable(CAP_SYS_RESOURCE) &&
  1088. sbi->s_resuid != current_fsuid() &&
  1089. (sbi->s_resgid == 0 || !in_group_p (sbi->s_resgid))) {
  1090. return 0;
  1091. }
  1092. return 1;
  1093. }
  1094. /*
  1095. * ext2_new_blocks() -- core block(s) allocation function
  1096. * @inode: file inode
  1097. * @goal: given target block(filesystem wide)
  1098. * @count: target number of blocks to allocate
  1099. * @errp: error code
  1100. *
  1101. * ext2_new_blocks uses a goal block to assist allocation. If the goal is
  1102. * free, or there is a free block within 32 blocks of the goal, that block
  1103. * is allocated. Otherwise a forward search is made for a free block; within
  1104. * each block group the search first looks for an entire free byte in the block
  1105. * bitmap, and then for any free bit if that fails.
  1106. * This function also updates quota and i_blocks field.
  1107. */
  1108. ext2_fsblk_t ext2_new_blocks(struct inode *inode, ext2_fsblk_t goal,
  1109. unsigned long *count, int *errp)
  1110. {
  1111. struct buffer_head *bitmap_bh = NULL;
  1112. struct buffer_head *gdp_bh;
  1113. int group_no;
  1114. int goal_group;
  1115. ext2_grpblk_t grp_target_blk; /* blockgroup relative goal block */
  1116. ext2_grpblk_t grp_alloc_blk; /* blockgroup-relative allocated block*/
  1117. ext2_fsblk_t ret_block; /* filesyetem-wide allocated block */
  1118. int bgi; /* blockgroup iteration index */
  1119. int performed_allocation = 0;
  1120. ext2_grpblk_t free_blocks; /* number of free blocks in a group */
  1121. struct super_block *sb;
  1122. struct ext2_group_desc *gdp;
  1123. struct ext2_super_block *es;
  1124. struct ext2_sb_info *sbi;
  1125. struct ext2_reserve_window_node *my_rsv = NULL;
  1126. struct ext2_block_alloc_info *block_i;
  1127. unsigned short windowsz = 0;
  1128. unsigned long ngroups;
  1129. unsigned long num = *count;
  1130. int ret;
  1131. *errp = -ENOSPC;
  1132. sb = inode->i_sb;
  1133. if (!sb) {
  1134. printk("ext2_new_blocks: nonexistent device");
  1135. return 0;
  1136. }
  1137. /*
  1138. * Check quota for allocation of this block.
  1139. */
  1140. ret = dquot_alloc_block(inode, num);
  1141. if (ret) {
  1142. *errp = ret;
  1143. return 0;
  1144. }
  1145. sbi = EXT2_SB(sb);
  1146. es = EXT2_SB(sb)->s_es;
  1147. ext2_debug("goal=%lu.\n", goal);
  1148. /*
  1149. * Allocate a block from reservation only when
  1150. * filesystem is mounted with reservation(default,-o reservation), and
  1151. * it's a regular file, and
  1152. * the desired window size is greater than 0 (One could use ioctl
  1153. * command EXT2_IOC_SETRSVSZ to set the window size to 0 to turn off
  1154. * reservation on that particular file)
  1155. */
  1156. block_i = EXT2_I(inode)->i_block_alloc_info;
  1157. if (block_i) {
  1158. windowsz = block_i->rsv_window_node.rsv_goal_size;
  1159. if (windowsz > 0)
  1160. my_rsv = &block_i->rsv_window_node;
  1161. }
  1162. if (!ext2_has_free_blocks(sbi)) {
  1163. *errp = -ENOSPC;
  1164. goto out;
  1165. }
  1166. /*
  1167. * First, test whether the goal block is free.
  1168. */
  1169. if (goal < le32_to_cpu(es->s_first_data_block) ||
  1170. goal >= le32_to_cpu(es->s_blocks_count))
  1171. goal = le32_to_cpu(es->s_first_data_block);
  1172. group_no = (goal - le32_to_cpu(es->s_first_data_block)) /
  1173. EXT2_BLOCKS_PER_GROUP(sb);
  1174. goal_group = group_no;
  1175. retry_alloc:
  1176. gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
  1177. if (!gdp)
  1178. goto io_error;
  1179. free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
  1180. /*
  1181. * if there is not enough free blocks to make a new resevation
  1182. * turn off reservation for this allocation
  1183. */
  1184. if (my_rsv && (free_blocks < windowsz)
  1185. && (free_blocks > 0)
  1186. && (rsv_is_empty(&my_rsv->rsv_window)))
  1187. my_rsv = NULL;
  1188. if (free_blocks > 0) {
  1189. grp_target_blk = ((goal - le32_to_cpu(es->s_first_data_block)) %
  1190. EXT2_BLOCKS_PER_GROUP(sb));
  1191. bitmap_bh = read_block_bitmap(sb, group_no);
  1192. if (!bitmap_bh)
  1193. goto io_error;
  1194. grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
  1195. bitmap_bh, grp_target_blk,
  1196. my_rsv, &num);
  1197. if (grp_alloc_blk >= 0)
  1198. goto allocated;
  1199. }
  1200. ngroups = EXT2_SB(sb)->s_groups_count;
  1201. smp_rmb();
  1202. /*
  1203. * Now search the rest of the groups. We assume that
  1204. * group_no and gdp correctly point to the last group visited.
  1205. */
  1206. for (bgi = 0; bgi < ngroups; bgi++) {
  1207. group_no++;
  1208. if (group_no >= ngroups)
  1209. group_no = 0;
  1210. gdp = ext2_get_group_desc(sb, group_no, &gdp_bh);
  1211. if (!gdp)
  1212. goto io_error;
  1213. free_blocks = le16_to_cpu(gdp->bg_free_blocks_count);
  1214. /*
  1215. * skip this group (and avoid loading bitmap) if there
  1216. * are no free blocks
  1217. */
  1218. if (!free_blocks)
  1219. continue;
  1220. /*
  1221. * skip this group if the number of
  1222. * free blocks is less than half of the reservation
  1223. * window size.
  1224. */
  1225. if (my_rsv && (free_blocks <= (windowsz/2)))
  1226. continue;
  1227. brelse(bitmap_bh);
  1228. bitmap_bh = read_block_bitmap(sb, group_no);
  1229. if (!bitmap_bh)
  1230. goto io_error;
  1231. /*
  1232. * try to allocate block(s) from this group, without a goal(-1).
  1233. */
  1234. grp_alloc_blk = ext2_try_to_allocate_with_rsv(sb, group_no,
  1235. bitmap_bh, -1, my_rsv, &num);
  1236. if (grp_alloc_blk >= 0)
  1237. goto allocated;
  1238. }
  1239. /*
  1240. * We may end up a bogus ealier ENOSPC error due to
  1241. * filesystem is "full" of reservations, but
  1242. * there maybe indeed free blocks avaliable on disk
  1243. * In this case, we just forget about the reservations
  1244. * just do block allocation as without reservations.
  1245. */
  1246. if (my_rsv) {
  1247. my_rsv = NULL;
  1248. windowsz = 0;
  1249. group_no = goal_group;
  1250. goto retry_alloc;
  1251. }
  1252. /* No space left on the device */
  1253. *errp = -ENOSPC;
  1254. goto out;
  1255. allocated:
  1256. ext2_debug("using block group %d(%d)\n",
  1257. group_no, gdp->bg_free_blocks_count);
  1258. ret_block = grp_alloc_blk + ext2_group_first_block_no(sb, group_no);
  1259. if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) ||
  1260. in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) ||
  1261. in_range(ret_block, le32_to_cpu(gdp->bg_inode_table),
  1262. EXT2_SB(sb)->s_itb_per_group) ||
  1263. in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table),
  1264. EXT2_SB(sb)->s_itb_per_group)) {
  1265. ext2_error(sb, "ext2_new_blocks",
  1266. "Allocating block in system zone - "
  1267. "blocks from "E2FSBLK", length %lu",
  1268. ret_block, num);
  1269. /*
  1270. * ext2_try_to_allocate marked the blocks we allocated as in
  1271. * use. So we may want to selectively mark some of the blocks
  1272. * as free
  1273. */
  1274. goto retry_alloc;
  1275. }
  1276. performed_allocation = 1;
  1277. if (ret_block + num - 1 >= le32_to_cpu(es->s_blocks_count)) {
  1278. ext2_error(sb, "ext2_new_blocks",
  1279. "block("E2FSBLK") >= blocks count(%d) - "
  1280. "block_group = %d, es == %p ", ret_block,
  1281. le32_to_cpu(es->s_blocks_count), group_no, es);
  1282. goto out;
  1283. }
  1284. group_adjust_blocks(sb, group_no, gdp, gdp_bh, -num);
  1285. percpu_counter_sub(&sbi->s_freeblocks_counter, num);
  1286. mark_buffer_dirty(bitmap_bh);
  1287. if (sb->s_flags & MS_SYNCHRONOUS)
  1288. sync_dirty_buffer(bitmap_bh);
  1289. *errp = 0;
  1290. brelse(bitmap_bh);
  1291. dquot_free_block_nodirty(inode, *count-num);
  1292. mark_inode_dirty(inode);
  1293. *count = num;
  1294. return ret_block;
  1295. io_error:
  1296. *errp = -EIO;
  1297. out:
  1298. /*
  1299. * Undo the block allocation
  1300. */
  1301. if (!performed_allocation) {
  1302. dquot_free_block_nodirty(inode, *count);
  1303. mark_inode_dirty(inode);
  1304. }
  1305. brelse(bitmap_bh);
  1306. return 0;
  1307. }
  1308. ext2_fsblk_t ext2_new_block(struct inode *inode, unsigned long goal, int *errp)
  1309. {
  1310. unsigned long count = 1;
  1311. return ext2_new_blocks(inode, goal, &count, errp);
  1312. }
  1313. #ifdef EXT2FS_DEBUG
  1314. static const int nibblemap[] = {4, 3, 3, 2, 3, 2, 2, 1, 3, 2, 2, 1, 2, 1, 1, 0};
  1315. unsigned long ext2_count_free (struct buffer_head * map, unsigned int numchars)
  1316. {
  1317. unsigned int i;
  1318. unsigned long sum = 0;
  1319. if (!map)
  1320. return (0);
  1321. for (i = 0; i < numchars; i++)
  1322. sum += nibblemap[map->b_data[i] & 0xf] +
  1323. nibblemap[(map->b_data[i] >> 4) & 0xf];
  1324. return (sum);
  1325. }
  1326. #endif /* EXT2FS_DEBUG */
  1327. unsigned long ext2_count_free_blocks (struct super_block * sb)
  1328. {
  1329. struct ext2_group_desc * desc;
  1330. unsigned long desc_count = 0;
  1331. int i;
  1332. #ifdef EXT2FS_DEBUG
  1333. unsigned long bitmap_count, x;
  1334. struct ext2_super_block *es;
  1335. es = EXT2_SB(sb)->s_es;
  1336. desc_count = 0;
  1337. bitmap_count = 0;
  1338. desc = NULL;
  1339. for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
  1340. struct buffer_head *bitmap_bh;
  1341. desc = ext2_get_group_desc (sb, i, NULL);
  1342. if (!desc)
  1343. continue;
  1344. desc_count += le16_to_cpu(desc->bg_free_blocks_count);
  1345. bitmap_bh = read_block_bitmap(sb, i);
  1346. if (!bitmap_bh)
  1347. continue;
  1348. x = ext2_count_free(bitmap_bh, sb->s_blocksize);
  1349. printk ("group %d: stored = %d, counted = %lu\n",
  1350. i, le16_to_cpu(desc->bg_free_blocks_count), x);
  1351. bitmap_count += x;
  1352. brelse(bitmap_bh);
  1353. }
  1354. printk("ext2_count_free_blocks: stored = %lu, computed = %lu, %lu\n",
  1355. (long)le32_to_cpu(es->s_free_blocks_count),
  1356. desc_count, bitmap_count);
  1357. return bitmap_count;
  1358. #else
  1359. for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
  1360. desc = ext2_get_group_desc (sb, i, NULL);
  1361. if (!desc)
  1362. continue;
  1363. desc_count += le16_to_cpu(desc->bg_free_blocks_count);
  1364. }
  1365. return desc_count;
  1366. #endif
  1367. }
  1368. static inline int test_root(int a, int b)
  1369. {
  1370. int num = b;
  1371. while (a > num)
  1372. num *= b;
  1373. return num == a;
  1374. }
  1375. static int ext2_group_sparse(int group)
  1376. {
  1377. if (group <= 1)
  1378. return 1;
  1379. return (test_root(group, 3) || test_root(group, 5) ||
  1380. test_root(group, 7));
  1381. }
  1382. /**
  1383. * ext2_bg_has_super - number of blocks used by the superblock in group
  1384. * @sb: superblock for filesystem
  1385. * @group: group number to check
  1386. *
  1387. * Return the number of blocks used by the superblock (primary or backup)
  1388. * in this group. Currently this will be only 0 or 1.
  1389. */
  1390. int ext2_bg_has_super(struct super_block *sb, int group)
  1391. {
  1392. if (EXT2_HAS_RO_COMPAT_FEATURE(sb,EXT2_FEATURE_RO_COMPAT_SPARSE_SUPER)&&
  1393. !ext2_group_sparse(group))
  1394. return 0;
  1395. return 1;
  1396. }
  1397. /**
  1398. * ext2_bg_num_gdb - number of blocks used by the group table in group
  1399. * @sb: superblock for filesystem
  1400. * @group: group number to check
  1401. *
  1402. * Return the number of blocks used by the group descriptor table
  1403. * (primary or backup) in this group. In the future there may be a
  1404. * different number of descriptor blocks in each group.
  1405. */
  1406. unsigned long ext2_bg_num_gdb(struct super_block *sb, int group)
  1407. {
  1408. return ext2_bg_has_super(sb, group) ? EXT2_SB(sb)->s_gdb_count : 0;
  1409. }