balloc.c 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795
  1. /*
  2. * linux/fs/ext4/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 <linux/time.h>
  14. #include <linux/capability.h>
  15. #include <linux/fs.h>
  16. #include <linux/jbd2.h>
  17. #include <linux/quotaops.h>
  18. #include <linux/buffer_head.h>
  19. #include "ext4.h"
  20. #include "ext4_jbd2.h"
  21. #include "group.h"
  22. #include "mballoc.h"
  23. /*
  24. * balloc.c contains the blocks allocation and deallocation routines
  25. */
  26. /*
  27. * Calculate the block group number and offset, given a block number
  28. */
  29. void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
  30. ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
  31. {
  32. struct ext4_super_block *es = EXT4_SB(sb)->s_es;
  33. ext4_grpblk_t offset;
  34. blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
  35. offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
  36. if (offsetp)
  37. *offsetp = offset;
  38. if (blockgrpp)
  39. *blockgrpp = blocknr;
  40. }
  41. static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
  42. ext4_group_t block_group)
  43. {
  44. ext4_group_t actual_group;
  45. ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
  46. if (actual_group == block_group)
  47. return 1;
  48. return 0;
  49. }
  50. static int ext4_group_used_meta_blocks(struct super_block *sb,
  51. ext4_group_t block_group)
  52. {
  53. ext4_fsblk_t tmp;
  54. struct ext4_sb_info *sbi = EXT4_SB(sb);
  55. /* block bitmap, inode bitmap, and inode table blocks */
  56. int used_blocks = sbi->s_itb_per_group + 2;
  57. if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
  58. struct ext4_group_desc *gdp;
  59. struct buffer_head *bh;
  60. gdp = ext4_get_group_desc(sb, block_group, &bh);
  61. if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
  62. block_group))
  63. used_blocks--;
  64. if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
  65. block_group))
  66. used_blocks--;
  67. tmp = ext4_inode_table(sb, gdp);
  68. for (; tmp < ext4_inode_table(sb, gdp) +
  69. sbi->s_itb_per_group; tmp++) {
  70. if (!ext4_block_in_group(sb, tmp, block_group))
  71. used_blocks -= 1;
  72. }
  73. }
  74. return used_blocks;
  75. }
  76. /* Initializes an uninitialized block bitmap if given, and returns the
  77. * number of blocks free in the group. */
  78. unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
  79. ext4_group_t block_group, struct ext4_group_desc *gdp)
  80. {
  81. int bit, bit_max;
  82. unsigned free_blocks, group_blocks;
  83. struct ext4_sb_info *sbi = EXT4_SB(sb);
  84. if (bh) {
  85. J_ASSERT_BH(bh, buffer_locked(bh));
  86. /* If checksum is bad mark all blocks used to prevent allocation
  87. * essentially implementing a per-group read-only flag. */
  88. if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
  89. ext4_error(sb, __func__,
  90. "Checksum bad for group %u", block_group);
  91. ext4_free_blks_set(sb, gdp, 0);
  92. ext4_free_inodes_set(sb, gdp, 0);
  93. ext4_itable_unused_set(sb, gdp, 0);
  94. memset(bh->b_data, 0xff, sb->s_blocksize);
  95. return 0;
  96. }
  97. memset(bh->b_data, 0, sb->s_blocksize);
  98. }
  99. /* Check for superblock and gdt backups in this group */
  100. bit_max = ext4_bg_has_super(sb, block_group);
  101. if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
  102. block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
  103. sbi->s_desc_per_block) {
  104. if (bit_max) {
  105. bit_max += ext4_bg_num_gdb(sb, block_group);
  106. bit_max +=
  107. le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
  108. }
  109. } else { /* For META_BG_BLOCK_GROUPS */
  110. bit_max += ext4_bg_num_gdb(sb, block_group);
  111. }
  112. if (block_group == sbi->s_groups_count - 1) {
  113. /*
  114. * Even though mke2fs always initialize first and last group
  115. * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
  116. * to make sure we calculate the right free blocks
  117. */
  118. group_blocks = ext4_blocks_count(sbi->s_es) -
  119. le32_to_cpu(sbi->s_es->s_first_data_block) -
  120. (EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count - 1));
  121. } else {
  122. group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
  123. }
  124. free_blocks = group_blocks - bit_max;
  125. if (bh) {
  126. ext4_fsblk_t start, tmp;
  127. int flex_bg = 0;
  128. for (bit = 0; bit < bit_max; bit++)
  129. ext4_set_bit(bit, bh->b_data);
  130. start = ext4_group_first_block_no(sb, block_group);
  131. if (EXT4_HAS_INCOMPAT_FEATURE(sb,
  132. EXT4_FEATURE_INCOMPAT_FLEX_BG))
  133. flex_bg = 1;
  134. /* Set bits for block and inode bitmaps, and inode table */
  135. tmp = ext4_block_bitmap(sb, gdp);
  136. if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
  137. ext4_set_bit(tmp - start, bh->b_data);
  138. tmp = ext4_inode_bitmap(sb, gdp);
  139. if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
  140. ext4_set_bit(tmp - start, bh->b_data);
  141. tmp = ext4_inode_table(sb, gdp);
  142. for (; tmp < ext4_inode_table(sb, gdp) +
  143. sbi->s_itb_per_group; tmp++) {
  144. if (!flex_bg ||
  145. ext4_block_in_group(sb, tmp, block_group))
  146. ext4_set_bit(tmp - start, bh->b_data);
  147. }
  148. /*
  149. * Also if the number of blocks within the group is
  150. * less than the blocksize * 8 ( which is the size
  151. * of bitmap ), set rest of the block bitmap to 1
  152. */
  153. mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
  154. }
  155. return free_blocks - ext4_group_used_meta_blocks(sb, block_group);
  156. }
  157. /*
  158. * The free blocks are managed by bitmaps. A file system contains several
  159. * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
  160. * block for inodes, N blocks for the inode table and data blocks.
  161. *
  162. * The file system contains group descriptors which are located after the
  163. * super block. Each descriptor contains the number of the bitmap block and
  164. * the free blocks count in the block. The descriptors are loaded in memory
  165. * when a file system is mounted (see ext4_fill_super).
  166. */
  167. #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
  168. /**
  169. * ext4_get_group_desc() -- load group descriptor from disk
  170. * @sb: super block
  171. * @block_group: given block group
  172. * @bh: pointer to the buffer head to store the block
  173. * group descriptor
  174. */
  175. struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
  176. ext4_group_t block_group,
  177. struct buffer_head **bh)
  178. {
  179. unsigned int group_desc;
  180. unsigned int offset;
  181. struct ext4_group_desc *desc;
  182. struct ext4_sb_info *sbi = EXT4_SB(sb);
  183. if (block_group >= sbi->s_groups_count) {
  184. ext4_error(sb, "ext4_get_group_desc",
  185. "block_group >= groups_count - "
  186. "block_group = %u, groups_count = %u",
  187. block_group, sbi->s_groups_count);
  188. return NULL;
  189. }
  190. smp_rmb();
  191. group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
  192. offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
  193. if (!sbi->s_group_desc[group_desc]) {
  194. ext4_error(sb, "ext4_get_group_desc",
  195. "Group descriptor not loaded - "
  196. "block_group = %u, group_desc = %u, desc = %u",
  197. block_group, group_desc, offset);
  198. return NULL;
  199. }
  200. desc = (struct ext4_group_desc *)(
  201. (__u8 *)sbi->s_group_desc[group_desc]->b_data +
  202. offset * EXT4_DESC_SIZE(sb));
  203. if (bh)
  204. *bh = sbi->s_group_desc[group_desc];
  205. return desc;
  206. }
  207. static int ext4_valid_block_bitmap(struct super_block *sb,
  208. struct ext4_group_desc *desc,
  209. unsigned int block_group,
  210. struct buffer_head *bh)
  211. {
  212. ext4_grpblk_t offset;
  213. ext4_grpblk_t next_zero_bit;
  214. ext4_fsblk_t bitmap_blk;
  215. ext4_fsblk_t group_first_block;
  216. if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
  217. /* with FLEX_BG, the inode/block bitmaps and itable
  218. * blocks may not be in the group at all
  219. * so the bitmap validation will be skipped for those groups
  220. * or it has to also read the block group where the bitmaps
  221. * are located to verify they are set.
  222. */
  223. return 1;
  224. }
  225. group_first_block = ext4_group_first_block_no(sb, block_group);
  226. /* check whether block bitmap block number is set */
  227. bitmap_blk = ext4_block_bitmap(sb, desc);
  228. offset = bitmap_blk - group_first_block;
  229. if (!ext4_test_bit(offset, bh->b_data))
  230. /* bad block bitmap */
  231. goto err_out;
  232. /* check whether the inode bitmap block number is set */
  233. bitmap_blk = ext4_inode_bitmap(sb, desc);
  234. offset = bitmap_blk - group_first_block;
  235. if (!ext4_test_bit(offset, bh->b_data))
  236. /* bad block bitmap */
  237. goto err_out;
  238. /* check whether the inode table block number is set */
  239. bitmap_blk = ext4_inode_table(sb, desc);
  240. offset = bitmap_blk - group_first_block;
  241. next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
  242. offset + EXT4_SB(sb)->s_itb_per_group,
  243. offset);
  244. if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
  245. /* good bitmap for inode tables */
  246. return 1;
  247. err_out:
  248. ext4_error(sb, __func__,
  249. "Invalid block bitmap - "
  250. "block_group = %d, block = %llu",
  251. block_group, bitmap_blk);
  252. return 0;
  253. }
  254. /**
  255. * ext4_read_block_bitmap()
  256. * @sb: super block
  257. * @block_group: given block group
  258. *
  259. * Read the bitmap for a given block_group,and validate the
  260. * bits for block/inode/inode tables are set in the bitmaps
  261. *
  262. * Return buffer_head on success or NULL in case of failure.
  263. */
  264. struct buffer_head *
  265. ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
  266. {
  267. struct ext4_group_desc *desc;
  268. struct buffer_head *bh = NULL;
  269. ext4_fsblk_t bitmap_blk;
  270. desc = ext4_get_group_desc(sb, block_group, NULL);
  271. if (!desc)
  272. return NULL;
  273. bitmap_blk = ext4_block_bitmap(sb, desc);
  274. bh = sb_getblk(sb, bitmap_blk);
  275. if (unlikely(!bh)) {
  276. ext4_error(sb, __func__,
  277. "Cannot read block bitmap - "
  278. "block_group = %u, block_bitmap = %llu",
  279. block_group, bitmap_blk);
  280. return NULL;
  281. }
  282. if (bitmap_uptodate(bh))
  283. return bh;
  284. lock_buffer(bh);
  285. if (bitmap_uptodate(bh)) {
  286. unlock_buffer(bh);
  287. return bh;
  288. }
  289. spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
  290. if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
  291. ext4_init_block_bitmap(sb, bh, block_group, desc);
  292. set_bitmap_uptodate(bh);
  293. set_buffer_uptodate(bh);
  294. spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
  295. unlock_buffer(bh);
  296. return bh;
  297. }
  298. spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
  299. if (buffer_uptodate(bh)) {
  300. /*
  301. * if not uninit if bh is uptodate,
  302. * bitmap is also uptodate
  303. */
  304. set_bitmap_uptodate(bh);
  305. unlock_buffer(bh);
  306. return bh;
  307. }
  308. /*
  309. * submit the buffer_head for read. We can
  310. * safely mark the bitmap as uptodate now.
  311. * We do it here so the bitmap uptodate bit
  312. * get set with buffer lock held.
  313. */
  314. set_bitmap_uptodate(bh);
  315. if (bh_submit_read(bh) < 0) {
  316. put_bh(bh);
  317. ext4_error(sb, __func__,
  318. "Cannot read block bitmap - "
  319. "block_group = %u, block_bitmap = %llu",
  320. block_group, bitmap_blk);
  321. return NULL;
  322. }
  323. ext4_valid_block_bitmap(sb, desc, block_group, bh);
  324. /*
  325. * file system mounted not to panic on error,
  326. * continue with corrupt bitmap
  327. */
  328. return bh;
  329. }
  330. /**
  331. * ext4_add_groupblocks() -- Add given blocks to an existing group
  332. * @handle: handle to this transaction
  333. * @sb: super block
  334. * @block: start physcial block to add to the block group
  335. * @count: number of blocks to free
  336. *
  337. * This marks the blocks as free in the bitmap. We ask the
  338. * mballoc to reload the buddy after this by setting group
  339. * EXT4_GROUP_INFO_NEED_INIT_BIT flag
  340. */
  341. void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
  342. ext4_fsblk_t block, unsigned long count)
  343. {
  344. struct buffer_head *bitmap_bh = NULL;
  345. struct buffer_head *gd_bh;
  346. ext4_group_t block_group;
  347. ext4_grpblk_t bit;
  348. unsigned int i;
  349. struct ext4_group_desc *desc;
  350. struct ext4_super_block *es;
  351. struct ext4_sb_info *sbi;
  352. int err = 0, ret, blk_free_count;
  353. ext4_grpblk_t blocks_freed;
  354. struct ext4_group_info *grp;
  355. sbi = EXT4_SB(sb);
  356. es = sbi->s_es;
  357. ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
  358. ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
  359. grp = ext4_get_group_info(sb, block_group);
  360. /*
  361. * Check to see if we are freeing blocks across a group
  362. * boundary.
  363. */
  364. if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
  365. goto error_return;
  366. }
  367. bitmap_bh = ext4_read_block_bitmap(sb, block_group);
  368. if (!bitmap_bh)
  369. goto error_return;
  370. desc = ext4_get_group_desc(sb, block_group, &gd_bh);
  371. if (!desc)
  372. goto error_return;
  373. if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
  374. in_range(ext4_inode_bitmap(sb, desc), block, count) ||
  375. in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
  376. in_range(block + count - 1, ext4_inode_table(sb, desc),
  377. sbi->s_itb_per_group)) {
  378. ext4_error(sb, __func__,
  379. "Adding blocks in system zones - "
  380. "Block = %llu, count = %lu",
  381. block, count);
  382. goto error_return;
  383. }
  384. /*
  385. * We are about to add blocks to the bitmap,
  386. * so we need undo access.
  387. */
  388. BUFFER_TRACE(bitmap_bh, "getting undo access");
  389. err = ext4_journal_get_undo_access(handle, bitmap_bh);
  390. if (err)
  391. goto error_return;
  392. /*
  393. * We are about to modify some metadata. Call the journal APIs
  394. * to unshare ->b_data if a currently-committing transaction is
  395. * using it
  396. */
  397. BUFFER_TRACE(gd_bh, "get_write_access");
  398. err = ext4_journal_get_write_access(handle, gd_bh);
  399. if (err)
  400. goto error_return;
  401. /*
  402. * make sure we don't allow a parallel init on other groups in the
  403. * same buddy cache
  404. */
  405. down_write(&grp->alloc_sem);
  406. for (i = 0, blocks_freed = 0; i < count; i++) {
  407. BUFFER_TRACE(bitmap_bh, "clear bit");
  408. if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
  409. bit + i, bitmap_bh->b_data)) {
  410. ext4_error(sb, __func__,
  411. "bit already cleared for block %llu",
  412. (ext4_fsblk_t)(block + i));
  413. BUFFER_TRACE(bitmap_bh, "bit already cleared");
  414. } else {
  415. blocks_freed++;
  416. }
  417. }
  418. spin_lock(sb_bgl_lock(sbi, block_group));
  419. blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
  420. ext4_free_blks_set(sb, desc, blk_free_count);
  421. desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
  422. spin_unlock(sb_bgl_lock(sbi, block_group));
  423. percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
  424. if (sbi->s_log_groups_per_flex) {
  425. ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
  426. spin_lock(sb_bgl_lock(sbi, flex_group));
  427. sbi->s_flex_groups[flex_group].free_blocks += blocks_freed;
  428. spin_unlock(sb_bgl_lock(sbi, flex_group));
  429. }
  430. /*
  431. * request to reload the buddy with the
  432. * new bitmap information
  433. */
  434. set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
  435. ext4_mb_update_group_info(grp, blocks_freed);
  436. up_write(&grp->alloc_sem);
  437. /* We dirtied the bitmap block */
  438. BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
  439. err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
  440. /* And the group descriptor block */
  441. BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
  442. ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
  443. if (!err)
  444. err = ret;
  445. sb->s_dirt = 1;
  446. error_return:
  447. brelse(bitmap_bh);
  448. ext4_std_error(sb, err);
  449. return;
  450. }
  451. /**
  452. * ext4_free_blocks() -- Free given blocks and update quota
  453. * @handle: handle for this transaction
  454. * @inode: inode
  455. * @block: start physical block to free
  456. * @count: number of blocks to count
  457. * @metadata: Are these metadata blocks
  458. */
  459. void ext4_free_blocks(handle_t *handle, struct inode *inode,
  460. ext4_fsblk_t block, unsigned long count,
  461. int metadata)
  462. {
  463. struct super_block *sb;
  464. unsigned long dquot_freed_blocks;
  465. /* this isn't the right place to decide whether block is metadata
  466. * inode.c/extents.c knows better, but for safety ... */
  467. if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
  468. metadata = 1;
  469. /* We need to make sure we don't reuse
  470. * block released untill the transaction commit.
  471. * writeback mode have weak data consistency so
  472. * don't force data as metadata when freeing block
  473. * for writeback mode.
  474. */
  475. if (metadata == 0 && !ext4_should_writeback_data(inode))
  476. metadata = 1;
  477. sb = inode->i_sb;
  478. ext4_mb_free_blocks(handle, inode, block, count,
  479. metadata, &dquot_freed_blocks);
  480. if (dquot_freed_blocks)
  481. DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
  482. return;
  483. }
  484. /**
  485. * ext4_has_free_blocks()
  486. * @sbi: in-core super block structure.
  487. * @nblocks: number of needed blocks
  488. *
  489. * Check if filesystem has nblocks free & available for allocation.
  490. * On success return 1, return 0 on failure.
  491. */
  492. int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
  493. {
  494. s64 free_blocks, dirty_blocks, root_blocks;
  495. struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
  496. struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
  497. free_blocks = percpu_counter_read_positive(fbc);
  498. dirty_blocks = percpu_counter_read_positive(dbc);
  499. root_blocks = ext4_r_blocks_count(sbi->s_es);
  500. if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
  501. EXT4_FREEBLOCKS_WATERMARK) {
  502. free_blocks = percpu_counter_sum_positive(fbc);
  503. dirty_blocks = percpu_counter_sum_positive(dbc);
  504. if (dirty_blocks < 0) {
  505. printk(KERN_CRIT "Dirty block accounting "
  506. "went wrong %lld\n",
  507. (long long)dirty_blocks);
  508. }
  509. }
  510. /* Check whether we have space after
  511. * accounting for current dirty blocks & root reserved blocks.
  512. */
  513. if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
  514. return 1;
  515. /* Hm, nope. Are (enough) root reserved blocks available? */
  516. if (sbi->s_resuid == current_fsuid() ||
  517. ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
  518. capable(CAP_SYS_RESOURCE)) {
  519. if (free_blocks >= (nblocks + dirty_blocks))
  520. return 1;
  521. }
  522. return 0;
  523. }
  524. int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
  525. s64 nblocks)
  526. {
  527. if (ext4_has_free_blocks(sbi, nblocks)) {
  528. percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
  529. return 0;
  530. } else
  531. return -ENOSPC;
  532. }
  533. /**
  534. * ext4_should_retry_alloc()
  535. * @sb: super block
  536. * @retries number of attemps has been made
  537. *
  538. * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
  539. * it is profitable to retry the operation, this function will wait
  540. * for the current or commiting transaction to complete, and then
  541. * return TRUE.
  542. *
  543. * if the total number of retries exceed three times, return FALSE.
  544. */
  545. int ext4_should_retry_alloc(struct super_block *sb, int *retries)
  546. {
  547. if (!ext4_has_free_blocks(EXT4_SB(sb), 1) ||
  548. (*retries)++ > 3 ||
  549. !EXT4_SB(sb)->s_journal)
  550. return 0;
  551. jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
  552. return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
  553. }
  554. /*
  555. * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
  556. *
  557. * @handle: handle to this transaction
  558. * @inode: file inode
  559. * @goal: given target block(filesystem wide)
  560. * @count: pointer to total number of blocks needed
  561. * @errp: error code
  562. *
  563. * Return 1st allocated block number on success, *count stores total account
  564. * error stores in errp pointer
  565. */
  566. ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
  567. ext4_fsblk_t goal, unsigned long *count, int *errp)
  568. {
  569. struct ext4_allocation_request ar;
  570. ext4_fsblk_t ret;
  571. memset(&ar, 0, sizeof(ar));
  572. /* Fill with neighbour allocated blocks */
  573. ar.inode = inode;
  574. ar.goal = goal;
  575. ar.len = count ? *count : 1;
  576. ret = ext4_mb_new_blocks(handle, &ar, errp);
  577. if (count)
  578. *count = ar.len;
  579. /*
  580. * Account for the allocated meta blocks
  581. */
  582. if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
  583. spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
  584. EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
  585. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  586. }
  587. return ret;
  588. }
  589. /**
  590. * ext4_count_free_blocks() -- count filesystem free blocks
  591. * @sb: superblock
  592. *
  593. * Adds up the number of free blocks from each block group.
  594. */
  595. ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
  596. {
  597. ext4_fsblk_t desc_count;
  598. struct ext4_group_desc *gdp;
  599. ext4_group_t i;
  600. ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
  601. #ifdef EXT4FS_DEBUG
  602. struct ext4_super_block *es;
  603. ext4_fsblk_t bitmap_count;
  604. unsigned int x;
  605. struct buffer_head *bitmap_bh = NULL;
  606. es = EXT4_SB(sb)->s_es;
  607. desc_count = 0;
  608. bitmap_count = 0;
  609. gdp = NULL;
  610. smp_rmb();
  611. for (i = 0; i < ngroups; i++) {
  612. gdp = ext4_get_group_desc(sb, i, NULL);
  613. if (!gdp)
  614. continue;
  615. desc_count += ext4_free_blks_count(sb, gdp);
  616. brelse(bitmap_bh);
  617. bitmap_bh = ext4_read_block_bitmap(sb, i);
  618. if (bitmap_bh == NULL)
  619. continue;
  620. x = ext4_count_free(bitmap_bh, sb->s_blocksize);
  621. printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
  622. i, ext4_free_blks_count(sb, gdp), x);
  623. bitmap_count += x;
  624. }
  625. brelse(bitmap_bh);
  626. printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
  627. ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
  628. desc_count, bitmap_count);
  629. return bitmap_count;
  630. #else
  631. desc_count = 0;
  632. smp_rmb();
  633. for (i = 0; i < ngroups; i++) {
  634. gdp = ext4_get_group_desc(sb, i, NULL);
  635. if (!gdp)
  636. continue;
  637. desc_count += ext4_free_blks_count(sb, gdp);
  638. }
  639. return desc_count;
  640. #endif
  641. }
  642. static inline int test_root(ext4_group_t a, int b)
  643. {
  644. int num = b;
  645. while (a > num)
  646. num *= b;
  647. return num == a;
  648. }
  649. static int ext4_group_sparse(ext4_group_t group)
  650. {
  651. if (group <= 1)
  652. return 1;
  653. if (!(group & 1))
  654. return 0;
  655. return (test_root(group, 7) || test_root(group, 5) ||
  656. test_root(group, 3));
  657. }
  658. /**
  659. * ext4_bg_has_super - number of blocks used by the superblock in group
  660. * @sb: superblock for filesystem
  661. * @group: group number to check
  662. *
  663. * Return the number of blocks used by the superblock (primary or backup)
  664. * in this group. Currently this will be only 0 or 1.
  665. */
  666. int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
  667. {
  668. if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
  669. EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
  670. !ext4_group_sparse(group))
  671. return 0;
  672. return 1;
  673. }
  674. static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
  675. ext4_group_t group)
  676. {
  677. unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
  678. ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
  679. ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
  680. if (group == first || group == first + 1 || group == last)
  681. return 1;
  682. return 0;
  683. }
  684. static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
  685. ext4_group_t group)
  686. {
  687. return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0;
  688. }
  689. /**
  690. * ext4_bg_num_gdb - number of blocks used by the group table in group
  691. * @sb: superblock for filesystem
  692. * @group: group number to check
  693. *
  694. * Return the number of blocks used by the group descriptor table
  695. * (primary or backup) in this group. In the future there may be a
  696. * different number of descriptor blocks in each group.
  697. */
  698. unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
  699. {
  700. unsigned long first_meta_bg =
  701. le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
  702. unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
  703. if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
  704. metagroup < first_meta_bg)
  705. return ext4_bg_num_gdb_nometa(sb, group);
  706. return ext4_bg_num_gdb_meta(sb,group);
  707. }