balloc.c 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790
  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. struct ext4_group_desc *gdp)
  53. {
  54. ext4_fsblk_t tmp;
  55. struct ext4_sb_info *sbi = EXT4_SB(sb);
  56. /* block bitmap, inode bitmap, and inode table blocks */
  57. int used_blocks = sbi->s_itb_per_group + 2;
  58. if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
  59. if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
  60. block_group))
  61. used_blocks--;
  62. if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
  63. block_group))
  64. used_blocks--;
  65. tmp = ext4_inode_table(sb, gdp);
  66. for (; tmp < ext4_inode_table(sb, gdp) +
  67. sbi->s_itb_per_group; tmp++) {
  68. if (!ext4_block_in_group(sb, tmp, block_group))
  69. used_blocks -= 1;
  70. }
  71. }
  72. return used_blocks;
  73. }
  74. /* Initializes an uninitialized block bitmap if given, and returns the
  75. * number of blocks free in the group. */
  76. unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
  77. ext4_group_t block_group, struct ext4_group_desc *gdp)
  78. {
  79. int bit, bit_max;
  80. ext4_group_t ngroups = ext4_get_groups_count(sb);
  81. unsigned free_blocks, group_blocks;
  82. struct ext4_sb_info *sbi = EXT4_SB(sb);
  83. if (bh) {
  84. J_ASSERT_BH(bh, buffer_locked(bh));
  85. /* If checksum is bad mark all blocks used to prevent allocation
  86. * essentially implementing a per-group read-only flag. */
  87. if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
  88. ext4_error(sb, __func__,
  89. "Checksum bad for group %u", block_group);
  90. ext4_free_blks_set(sb, gdp, 0);
  91. ext4_free_inodes_set(sb, gdp, 0);
  92. ext4_itable_unused_set(sb, gdp, 0);
  93. memset(bh->b_data, 0xff, sb->s_blocksize);
  94. return 0;
  95. }
  96. memset(bh->b_data, 0, sb->s_blocksize);
  97. }
  98. /* Check for superblock and gdt backups in this group */
  99. bit_max = ext4_bg_has_super(sb, block_group);
  100. if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
  101. block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
  102. sbi->s_desc_per_block) {
  103. if (bit_max) {
  104. bit_max += ext4_bg_num_gdb(sb, block_group);
  105. bit_max +=
  106. le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
  107. }
  108. } else { /* For META_BG_BLOCK_GROUPS */
  109. bit_max += ext4_bg_num_gdb(sb, block_group);
  110. }
  111. if (block_group == ngroups - 1) {
  112. /*
  113. * Even though mke2fs always initialize first and last group
  114. * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
  115. * to make sure we calculate the right free blocks
  116. */
  117. group_blocks = ext4_blocks_count(sbi->s_es) -
  118. le32_to_cpu(sbi->s_es->s_first_data_block) -
  119. (EXT4_BLOCKS_PER_GROUP(sb) * (ngroups - 1));
  120. } else {
  121. group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
  122. }
  123. free_blocks = group_blocks - bit_max;
  124. if (bh) {
  125. ext4_fsblk_t start, tmp;
  126. int flex_bg = 0;
  127. for (bit = 0; bit < bit_max; bit++)
  128. ext4_set_bit(bit, bh->b_data);
  129. start = ext4_group_first_block_no(sb, block_group);
  130. if (EXT4_HAS_INCOMPAT_FEATURE(sb,
  131. EXT4_FEATURE_INCOMPAT_FLEX_BG))
  132. flex_bg = 1;
  133. /* Set bits for block and inode bitmaps, and inode table */
  134. tmp = ext4_block_bitmap(sb, gdp);
  135. if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
  136. ext4_set_bit(tmp - start, bh->b_data);
  137. tmp = ext4_inode_bitmap(sb, gdp);
  138. if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
  139. ext4_set_bit(tmp - start, bh->b_data);
  140. tmp = ext4_inode_table(sb, gdp);
  141. for (; tmp < ext4_inode_table(sb, gdp) +
  142. sbi->s_itb_per_group; tmp++) {
  143. if (!flex_bg ||
  144. ext4_block_in_group(sb, tmp, block_group))
  145. ext4_set_bit(tmp - start, bh->b_data);
  146. }
  147. /*
  148. * Also if the number of blocks within the group is
  149. * less than the blocksize * 8 ( which is the size
  150. * of bitmap ), set rest of the block bitmap to 1
  151. */
  152. mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
  153. }
  154. return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp);
  155. }
  156. /*
  157. * The free blocks are managed by bitmaps. A file system contains several
  158. * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
  159. * block for inodes, N blocks for the inode table and data blocks.
  160. *
  161. * The file system contains group descriptors which are located after the
  162. * super block. Each descriptor contains the number of the bitmap block and
  163. * the free blocks count in the block. The descriptors are loaded in memory
  164. * when a file system is mounted (see ext4_fill_super).
  165. */
  166. #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
  167. /**
  168. * ext4_get_group_desc() -- load group descriptor from disk
  169. * @sb: super block
  170. * @block_group: given block group
  171. * @bh: pointer to the buffer head to store the block
  172. * group descriptor
  173. */
  174. struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
  175. ext4_group_t block_group,
  176. struct buffer_head **bh)
  177. {
  178. unsigned int group_desc;
  179. unsigned int offset;
  180. ext4_group_t ngroups = ext4_get_groups_count(sb);
  181. struct ext4_group_desc *desc;
  182. struct ext4_sb_info *sbi = EXT4_SB(sb);
  183. if (block_group >= ngroups) {
  184. ext4_error(sb, "ext4_get_group_desc",
  185. "block_group >= groups_count - "
  186. "block_group = %u, groups_count = %u",
  187. block_group, ngroups);
  188. return NULL;
  189. }
  190. group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
  191. offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
  192. if (!sbi->s_group_desc[group_desc]) {
  193. ext4_error(sb, "ext4_get_group_desc",
  194. "Group descriptor not loaded - "
  195. "block_group = %u, group_desc = %u, desc = %u",
  196. block_group, group_desc, offset);
  197. return NULL;
  198. }
  199. desc = (struct ext4_group_desc *)(
  200. (__u8 *)sbi->s_group_desc[group_desc]->b_data +
  201. offset * EXT4_DESC_SIZE(sb));
  202. if (bh)
  203. *bh = sbi->s_group_desc[group_desc];
  204. return desc;
  205. }
  206. static int ext4_valid_block_bitmap(struct super_block *sb,
  207. struct ext4_group_desc *desc,
  208. unsigned int block_group,
  209. struct buffer_head *bh)
  210. {
  211. ext4_grpblk_t offset;
  212. ext4_grpblk_t next_zero_bit;
  213. ext4_fsblk_t bitmap_blk;
  214. ext4_fsblk_t group_first_block;
  215. if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
  216. /* with FLEX_BG, the inode/block bitmaps and itable
  217. * blocks may not be in the group at all
  218. * so the bitmap validation will be skipped for those groups
  219. * or it has to also read the block group where the bitmaps
  220. * are located to verify they are set.
  221. */
  222. return 1;
  223. }
  224. group_first_block = ext4_group_first_block_no(sb, block_group);
  225. /* check whether block bitmap block number is set */
  226. bitmap_blk = ext4_block_bitmap(sb, desc);
  227. offset = bitmap_blk - group_first_block;
  228. if (!ext4_test_bit(offset, bh->b_data))
  229. /* bad block bitmap */
  230. goto err_out;
  231. /* check whether the inode bitmap block number is set */
  232. bitmap_blk = ext4_inode_bitmap(sb, desc);
  233. offset = bitmap_blk - group_first_block;
  234. if (!ext4_test_bit(offset, bh->b_data))
  235. /* bad block bitmap */
  236. goto err_out;
  237. /* check whether the inode table block number is set */
  238. bitmap_blk = ext4_inode_table(sb, desc);
  239. offset = bitmap_blk - group_first_block;
  240. next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
  241. offset + EXT4_SB(sb)->s_itb_per_group,
  242. offset);
  243. if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
  244. /* good bitmap for inode tables */
  245. return 1;
  246. err_out:
  247. ext4_error(sb, __func__,
  248. "Invalid block bitmap - "
  249. "block_group = %d, block = %llu",
  250. block_group, bitmap_blk);
  251. return 0;
  252. }
  253. /**
  254. * ext4_read_block_bitmap()
  255. * @sb: super block
  256. * @block_group: given block group
  257. *
  258. * Read the bitmap for a given block_group,and validate the
  259. * bits for block/inode/inode tables are set in the bitmaps
  260. *
  261. * Return buffer_head on success or NULL in case of failure.
  262. */
  263. struct buffer_head *
  264. ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
  265. {
  266. struct ext4_group_desc *desc;
  267. struct buffer_head *bh = NULL;
  268. ext4_fsblk_t bitmap_blk;
  269. desc = ext4_get_group_desc(sb, block_group, NULL);
  270. if (!desc)
  271. return NULL;
  272. bitmap_blk = ext4_block_bitmap(sb, desc);
  273. bh = sb_getblk(sb, bitmap_blk);
  274. if (unlikely(!bh)) {
  275. ext4_error(sb, __func__,
  276. "Cannot read block bitmap - "
  277. "block_group = %u, block_bitmap = %llu",
  278. block_group, bitmap_blk);
  279. return NULL;
  280. }
  281. if (bitmap_uptodate(bh))
  282. return bh;
  283. lock_buffer(bh);
  284. if (bitmap_uptodate(bh)) {
  285. unlock_buffer(bh);
  286. return bh;
  287. }
  288. spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
  289. if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
  290. ext4_init_block_bitmap(sb, bh, block_group, desc);
  291. set_bitmap_uptodate(bh);
  292. set_buffer_uptodate(bh);
  293. spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
  294. unlock_buffer(bh);
  295. return bh;
  296. }
  297. spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
  298. if (buffer_uptodate(bh)) {
  299. /*
  300. * if not uninit if bh is uptodate,
  301. * bitmap is also uptodate
  302. */
  303. set_bitmap_uptodate(bh);
  304. unlock_buffer(bh);
  305. return bh;
  306. }
  307. /*
  308. * submit the buffer_head for read. We can
  309. * safely mark the bitmap as uptodate now.
  310. * We do it here so the bitmap uptodate bit
  311. * get set with buffer lock held.
  312. */
  313. set_bitmap_uptodate(bh);
  314. if (bh_submit_read(bh) < 0) {
  315. put_bh(bh);
  316. ext4_error(sb, __func__,
  317. "Cannot read block bitmap - "
  318. "block_group = %u, block_bitmap = %llu",
  319. block_group, bitmap_blk);
  320. return NULL;
  321. }
  322. ext4_valid_block_bitmap(sb, desc, block_group, bh);
  323. /*
  324. * file system mounted not to panic on error,
  325. * continue with corrupt bitmap
  326. */
  327. return bh;
  328. }
  329. /**
  330. * ext4_add_groupblocks() -- Add given blocks to an existing group
  331. * @handle: handle to this transaction
  332. * @sb: super block
  333. * @block: start physcial block to add to the block group
  334. * @count: number of blocks to free
  335. *
  336. * This marks the blocks as free in the bitmap. We ask the
  337. * mballoc to reload the buddy after this by setting group
  338. * EXT4_GROUP_INFO_NEED_INIT_BIT flag
  339. */
  340. void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
  341. ext4_fsblk_t block, unsigned long count)
  342. {
  343. struct buffer_head *bitmap_bh = NULL;
  344. struct buffer_head *gd_bh;
  345. ext4_group_t block_group;
  346. ext4_grpblk_t bit;
  347. unsigned int i;
  348. struct ext4_group_desc *desc;
  349. struct ext4_super_block *es;
  350. struct ext4_sb_info *sbi;
  351. int err = 0, ret, blk_free_count;
  352. ext4_grpblk_t blocks_freed;
  353. struct ext4_group_info *grp;
  354. sbi = EXT4_SB(sb);
  355. es = sbi->s_es;
  356. ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
  357. ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
  358. grp = ext4_get_group_info(sb, block_group);
  359. /*
  360. * Check to see if we are freeing blocks across a group
  361. * boundary.
  362. */
  363. if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
  364. goto error_return;
  365. }
  366. bitmap_bh = ext4_read_block_bitmap(sb, block_group);
  367. if (!bitmap_bh)
  368. goto error_return;
  369. desc = ext4_get_group_desc(sb, block_group, &gd_bh);
  370. if (!desc)
  371. goto error_return;
  372. if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
  373. in_range(ext4_inode_bitmap(sb, desc), block, count) ||
  374. in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
  375. in_range(block + count - 1, ext4_inode_table(sb, desc),
  376. sbi->s_itb_per_group)) {
  377. ext4_error(sb, __func__,
  378. "Adding blocks in system zones - "
  379. "Block = %llu, count = %lu",
  380. block, count);
  381. goto error_return;
  382. }
  383. /*
  384. * We are about to add blocks to the bitmap,
  385. * so we need undo access.
  386. */
  387. BUFFER_TRACE(bitmap_bh, "getting undo access");
  388. err = ext4_journal_get_undo_access(handle, bitmap_bh);
  389. if (err)
  390. goto error_return;
  391. /*
  392. * We are about to modify some metadata. Call the journal APIs
  393. * to unshare ->b_data if a currently-committing transaction is
  394. * using it
  395. */
  396. BUFFER_TRACE(gd_bh, "get_write_access");
  397. err = ext4_journal_get_write_access(handle, gd_bh);
  398. if (err)
  399. goto error_return;
  400. /*
  401. * make sure we don't allow a parallel init on other groups in the
  402. * same buddy cache
  403. */
  404. down_write(&grp->alloc_sem);
  405. for (i = 0, blocks_freed = 0; i < count; i++) {
  406. BUFFER_TRACE(bitmap_bh, "clear bit");
  407. if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
  408. bit + i, bitmap_bh->b_data)) {
  409. ext4_error(sb, __func__,
  410. "bit already cleared for block %llu",
  411. (ext4_fsblk_t)(block + i));
  412. BUFFER_TRACE(bitmap_bh, "bit already cleared");
  413. } else {
  414. blocks_freed++;
  415. }
  416. }
  417. spin_lock(sb_bgl_lock(sbi, block_group));
  418. blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
  419. ext4_free_blks_set(sb, desc, blk_free_count);
  420. desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
  421. spin_unlock(sb_bgl_lock(sbi, block_group));
  422. percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
  423. if (sbi->s_log_groups_per_flex) {
  424. ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
  425. atomic_add(blocks_freed,
  426. &sbi->s_flex_groups[flex_group].free_blocks);
  427. }
  428. /*
  429. * request to reload the buddy with the
  430. * new bitmap information
  431. */
  432. set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
  433. ext4_mb_update_group_info(grp, blocks_freed);
  434. up_write(&grp->alloc_sem);
  435. /* We dirtied the bitmap block */
  436. BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
  437. err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
  438. /* And the group descriptor block */
  439. BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
  440. ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
  441. if (!err)
  442. err = ret;
  443. sb->s_dirt = 1;
  444. error_return:
  445. brelse(bitmap_bh);
  446. ext4_std_error(sb, err);
  447. return;
  448. }
  449. /**
  450. * ext4_free_blocks() -- Free given blocks and update quota
  451. * @handle: handle for this transaction
  452. * @inode: inode
  453. * @block: start physical block to free
  454. * @count: number of blocks to count
  455. * @metadata: Are these metadata blocks
  456. */
  457. void ext4_free_blocks(handle_t *handle, struct inode *inode,
  458. ext4_fsblk_t block, unsigned long count,
  459. int metadata)
  460. {
  461. struct super_block *sb;
  462. unsigned long dquot_freed_blocks;
  463. /* this isn't the right place to decide whether block is metadata
  464. * inode.c/extents.c knows better, but for safety ... */
  465. if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
  466. metadata = 1;
  467. /* We need to make sure we don't reuse
  468. * block released untill the transaction commit.
  469. * writeback mode have weak data consistency so
  470. * don't force data as metadata when freeing block
  471. * for writeback mode.
  472. */
  473. if (metadata == 0 && !ext4_should_writeback_data(inode))
  474. metadata = 1;
  475. sb = inode->i_sb;
  476. ext4_mb_free_blocks(handle, inode, block, count,
  477. metadata, &dquot_freed_blocks);
  478. if (dquot_freed_blocks)
  479. vfs_dq_free_block(inode, dquot_freed_blocks);
  480. return;
  481. }
  482. /**
  483. * ext4_has_free_blocks()
  484. * @sbi: in-core super block structure.
  485. * @nblocks: number of needed blocks
  486. *
  487. * Check if filesystem has nblocks free & available for allocation.
  488. * On success return 1, return 0 on failure.
  489. */
  490. int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
  491. {
  492. s64 free_blocks, dirty_blocks, root_blocks;
  493. struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
  494. struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
  495. free_blocks = percpu_counter_read_positive(fbc);
  496. dirty_blocks = percpu_counter_read_positive(dbc);
  497. root_blocks = ext4_r_blocks_count(sbi->s_es);
  498. if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
  499. EXT4_FREEBLOCKS_WATERMARK) {
  500. free_blocks = percpu_counter_sum_positive(fbc);
  501. dirty_blocks = percpu_counter_sum_positive(dbc);
  502. if (dirty_blocks < 0) {
  503. printk(KERN_CRIT "Dirty block accounting "
  504. "went wrong %lld\n",
  505. (long long)dirty_blocks);
  506. }
  507. }
  508. /* Check whether we have space after
  509. * accounting for current dirty blocks & root reserved blocks.
  510. */
  511. if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
  512. return 1;
  513. /* Hm, nope. Are (enough) root reserved blocks available? */
  514. if (sbi->s_resuid == current_fsuid() ||
  515. ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
  516. capable(CAP_SYS_RESOURCE)) {
  517. if (free_blocks >= (nblocks + dirty_blocks))
  518. return 1;
  519. }
  520. return 0;
  521. }
  522. int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
  523. s64 nblocks)
  524. {
  525. if (ext4_has_free_blocks(sbi, nblocks)) {
  526. percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
  527. return 0;
  528. } else
  529. return -ENOSPC;
  530. }
  531. /**
  532. * ext4_should_retry_alloc()
  533. * @sb: super block
  534. * @retries number of attemps has been made
  535. *
  536. * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
  537. * it is profitable to retry the operation, this function will wait
  538. * for the current or commiting transaction to complete, and then
  539. * return TRUE.
  540. *
  541. * if the total number of retries exceed three times, return FALSE.
  542. */
  543. int ext4_should_retry_alloc(struct super_block *sb, int *retries)
  544. {
  545. if (!ext4_has_free_blocks(EXT4_SB(sb), 1) ||
  546. (*retries)++ > 3 ||
  547. !EXT4_SB(sb)->s_journal)
  548. return 0;
  549. jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
  550. return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
  551. }
  552. /*
  553. * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
  554. *
  555. * @handle: handle to this transaction
  556. * @inode: file inode
  557. * @goal: given target block(filesystem wide)
  558. * @count: pointer to total number of blocks needed
  559. * @errp: error code
  560. *
  561. * Return 1st allocated block number on success, *count stores total account
  562. * error stores in errp pointer
  563. */
  564. ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
  565. ext4_fsblk_t goal, unsigned long *count, int *errp)
  566. {
  567. struct ext4_allocation_request ar;
  568. ext4_fsblk_t ret;
  569. memset(&ar, 0, sizeof(ar));
  570. /* Fill with neighbour allocated blocks */
  571. ar.inode = inode;
  572. ar.goal = goal;
  573. ar.len = count ? *count : 1;
  574. ret = ext4_mb_new_blocks(handle, &ar, errp);
  575. if (count)
  576. *count = ar.len;
  577. /*
  578. * Account for the allocated meta blocks
  579. */
  580. if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
  581. spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
  582. EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
  583. spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
  584. }
  585. return ret;
  586. }
  587. /**
  588. * ext4_count_free_blocks() -- count filesystem free blocks
  589. * @sb: superblock
  590. *
  591. * Adds up the number of free blocks from each block group.
  592. */
  593. ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
  594. {
  595. ext4_fsblk_t desc_count;
  596. struct ext4_group_desc *gdp;
  597. ext4_group_t i;
  598. ext4_group_t ngroups = ext4_get_groups_count(sb);
  599. #ifdef EXT4FS_DEBUG
  600. struct ext4_super_block *es;
  601. ext4_fsblk_t bitmap_count;
  602. unsigned int x;
  603. struct buffer_head *bitmap_bh = NULL;
  604. es = EXT4_SB(sb)->s_es;
  605. desc_count = 0;
  606. bitmap_count = 0;
  607. gdp = NULL;
  608. for (i = 0; i < ngroups; i++) {
  609. gdp = ext4_get_group_desc(sb, i, NULL);
  610. if (!gdp)
  611. continue;
  612. desc_count += ext4_free_blks_count(sb, gdp);
  613. brelse(bitmap_bh);
  614. bitmap_bh = ext4_read_block_bitmap(sb, i);
  615. if (bitmap_bh == NULL)
  616. continue;
  617. x = ext4_count_free(bitmap_bh, sb->s_blocksize);
  618. printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
  619. i, ext4_free_blks_count(sb, gdp), x);
  620. bitmap_count += x;
  621. }
  622. brelse(bitmap_bh);
  623. printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
  624. ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
  625. desc_count, bitmap_count);
  626. return bitmap_count;
  627. #else
  628. desc_count = 0;
  629. for (i = 0; i < ngroups; i++) {
  630. gdp = ext4_get_group_desc(sb, i, NULL);
  631. if (!gdp)
  632. continue;
  633. desc_count += ext4_free_blks_count(sb, gdp);
  634. }
  635. return desc_count;
  636. #endif
  637. }
  638. static inline int test_root(ext4_group_t a, int b)
  639. {
  640. int num = b;
  641. while (a > num)
  642. num *= b;
  643. return num == a;
  644. }
  645. static int ext4_group_sparse(ext4_group_t group)
  646. {
  647. if (group <= 1)
  648. return 1;
  649. if (!(group & 1))
  650. return 0;
  651. return (test_root(group, 7) || test_root(group, 5) ||
  652. test_root(group, 3));
  653. }
  654. /**
  655. * ext4_bg_has_super - number of blocks used by the superblock in group
  656. * @sb: superblock for filesystem
  657. * @group: group number to check
  658. *
  659. * Return the number of blocks used by the superblock (primary or backup)
  660. * in this group. Currently this will be only 0 or 1.
  661. */
  662. int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
  663. {
  664. if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
  665. EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
  666. !ext4_group_sparse(group))
  667. return 0;
  668. return 1;
  669. }
  670. static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
  671. ext4_group_t group)
  672. {
  673. unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
  674. ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
  675. ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
  676. if (group == first || group == first + 1 || group == last)
  677. return 1;
  678. return 0;
  679. }
  680. static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
  681. ext4_group_t group)
  682. {
  683. return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0;
  684. }
  685. /**
  686. * ext4_bg_num_gdb - number of blocks used by the group table in group
  687. * @sb: superblock for filesystem
  688. * @group: group number to check
  689. *
  690. * Return the number of blocks used by the group descriptor table
  691. * (primary or backup) in this group. In the future there may be a
  692. * different number of descriptor blocks in each group.
  693. */
  694. unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
  695. {
  696. unsigned long first_meta_bg =
  697. le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
  698. unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
  699. if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
  700. metagroup < first_meta_bg)
  701. return ext4_bg_num_gdb_nometa(sb, group);
  702. return ext4_bg_num_gdb_meta(sb,group);
  703. }