xref: /openbmc/linux/fs/ext4/balloc.c (revision 4800cd83)
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 
14 #include <linux/time.h>
15 #include <linux/capability.h>
16 #include <linux/fs.h>
17 #include <linux/jbd2.h>
18 #include <linux/quotaops.h>
19 #include <linux/buffer_head.h>
20 #include "ext4.h"
21 #include "ext4_jbd2.h"
22 #include "mballoc.h"
23 
24 /*
25  * balloc.c contains the blocks allocation and deallocation routines
26  */
27 
28 /*
29  * Calculate the block group number and offset, given a block number
30  */
31 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
32 		ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
33 {
34 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
35 	ext4_grpblk_t offset;
36 
37 	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
38 	offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
39 	if (offsetp)
40 		*offsetp = offset;
41 	if (blockgrpp)
42 		*blockgrpp = blocknr;
43 
44 }
45 
46 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
47 			ext4_group_t block_group)
48 {
49 	ext4_group_t actual_group;
50 	ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
51 	if (actual_group == block_group)
52 		return 1;
53 	return 0;
54 }
55 
56 static int ext4_group_used_meta_blocks(struct super_block *sb,
57 				       ext4_group_t block_group,
58 				       struct ext4_group_desc *gdp)
59 {
60 	ext4_fsblk_t tmp;
61 	struct ext4_sb_info *sbi = EXT4_SB(sb);
62 	/* block bitmap, inode bitmap, and inode table blocks */
63 	int used_blocks = sbi->s_itb_per_group + 2;
64 
65 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
66 		if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
67 					block_group))
68 			used_blocks--;
69 
70 		if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
71 					block_group))
72 			used_blocks--;
73 
74 		tmp = ext4_inode_table(sb, gdp);
75 		for (; tmp < ext4_inode_table(sb, gdp) +
76 				sbi->s_itb_per_group; tmp++) {
77 			if (!ext4_block_in_group(sb, tmp, block_group))
78 				used_blocks -= 1;
79 		}
80 	}
81 	return used_blocks;
82 }
83 
84 /* Initializes an uninitialized block bitmap if given, and returns the
85  * number of blocks free in the group. */
86 unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
87 		 ext4_group_t block_group, struct ext4_group_desc *gdp)
88 {
89 	int bit, bit_max;
90 	ext4_group_t ngroups = ext4_get_groups_count(sb);
91 	unsigned free_blocks, group_blocks;
92 	struct ext4_sb_info *sbi = EXT4_SB(sb);
93 
94 	if (bh) {
95 		J_ASSERT_BH(bh, buffer_locked(bh));
96 
97 		/* If checksum is bad mark all blocks used to prevent allocation
98 		 * essentially implementing a per-group read-only flag. */
99 		if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
100 			ext4_error(sb, "Checksum bad for group %u",
101 					block_group);
102 			ext4_free_blks_set(sb, gdp, 0);
103 			ext4_free_inodes_set(sb, gdp, 0);
104 			ext4_itable_unused_set(sb, gdp, 0);
105 			memset(bh->b_data, 0xff, sb->s_blocksize);
106 			return 0;
107 		}
108 		memset(bh->b_data, 0, sb->s_blocksize);
109 	}
110 
111 	/* Check for superblock and gdt backups in this group */
112 	bit_max = ext4_bg_has_super(sb, block_group);
113 
114 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
115 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
116 			  sbi->s_desc_per_block) {
117 		if (bit_max) {
118 			bit_max += ext4_bg_num_gdb(sb, block_group);
119 			bit_max +=
120 				le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
121 		}
122 	} else { /* For META_BG_BLOCK_GROUPS */
123 		bit_max += ext4_bg_num_gdb(sb, block_group);
124 	}
125 
126 	if (block_group == ngroups - 1) {
127 		/*
128 		 * Even though mke2fs always initialize first and last group
129 		 * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
130 		 * to make sure we calculate the right free blocks
131 		 */
132 		group_blocks = ext4_blocks_count(sbi->s_es) -
133 			ext4_group_first_block_no(sb, ngroups - 1);
134 	} else {
135 		group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
136 	}
137 
138 	free_blocks = group_blocks - bit_max;
139 
140 	if (bh) {
141 		ext4_fsblk_t start, tmp;
142 		int flex_bg = 0;
143 
144 		for (bit = 0; bit < bit_max; bit++)
145 			ext4_set_bit(bit, bh->b_data);
146 
147 		start = ext4_group_first_block_no(sb, block_group);
148 
149 		if (EXT4_HAS_INCOMPAT_FEATURE(sb,
150 					      EXT4_FEATURE_INCOMPAT_FLEX_BG))
151 			flex_bg = 1;
152 
153 		/* Set bits for block and inode bitmaps, and inode table */
154 		tmp = ext4_block_bitmap(sb, gdp);
155 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
156 			ext4_set_bit(tmp - start, bh->b_data);
157 
158 		tmp = ext4_inode_bitmap(sb, gdp);
159 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
160 			ext4_set_bit(tmp - start, bh->b_data);
161 
162 		tmp = ext4_inode_table(sb, gdp);
163 		for (; tmp < ext4_inode_table(sb, gdp) +
164 				sbi->s_itb_per_group; tmp++) {
165 			if (!flex_bg ||
166 				ext4_block_in_group(sb, tmp, block_group))
167 				ext4_set_bit(tmp - start, bh->b_data);
168 		}
169 		/*
170 		 * Also if the number of blocks within the group is
171 		 * less than the blocksize * 8 ( which is the size
172 		 * of bitmap ), set rest of the block bitmap to 1
173 		 */
174 		ext4_mark_bitmap_end(group_blocks, sb->s_blocksize * 8,
175 				     bh->b_data);
176 	}
177 	return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp);
178 }
179 
180 
181 /*
182  * The free blocks are managed by bitmaps.  A file system contains several
183  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
184  * block for inodes, N blocks for the inode table and data blocks.
185  *
186  * The file system contains group descriptors which are located after the
187  * super block.  Each descriptor contains the number of the bitmap block and
188  * the free blocks count in the block.  The descriptors are loaded in memory
189  * when a file system is mounted (see ext4_fill_super).
190  */
191 
192 /**
193  * ext4_get_group_desc() -- load group descriptor from disk
194  * @sb:			super block
195  * @block_group:	given block group
196  * @bh:			pointer to the buffer head to store the block
197  *			group descriptor
198  */
199 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
200 					     ext4_group_t block_group,
201 					     struct buffer_head **bh)
202 {
203 	unsigned int group_desc;
204 	unsigned int offset;
205 	ext4_group_t ngroups = ext4_get_groups_count(sb);
206 	struct ext4_group_desc *desc;
207 	struct ext4_sb_info *sbi = EXT4_SB(sb);
208 
209 	if (block_group >= ngroups) {
210 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
211 			   " groups_count = %u", block_group, ngroups);
212 
213 		return NULL;
214 	}
215 
216 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
217 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
218 	if (!sbi->s_group_desc[group_desc]) {
219 		ext4_error(sb, "Group descriptor not loaded - "
220 			   "block_group = %u, group_desc = %u, desc = %u",
221 			   block_group, group_desc, offset);
222 		return NULL;
223 	}
224 
225 	desc = (struct ext4_group_desc *)(
226 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
227 		offset * EXT4_DESC_SIZE(sb));
228 	if (bh)
229 		*bh = sbi->s_group_desc[group_desc];
230 	return desc;
231 }
232 
233 static int ext4_valid_block_bitmap(struct super_block *sb,
234 					struct ext4_group_desc *desc,
235 					unsigned int block_group,
236 					struct buffer_head *bh)
237 {
238 	ext4_grpblk_t offset;
239 	ext4_grpblk_t next_zero_bit;
240 	ext4_fsblk_t bitmap_blk;
241 	ext4_fsblk_t group_first_block;
242 
243 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
244 		/* with FLEX_BG, the inode/block bitmaps and itable
245 		 * blocks may not be in the group at all
246 		 * so the bitmap validation will be skipped for those groups
247 		 * or it has to also read the block group where the bitmaps
248 		 * are located to verify they are set.
249 		 */
250 		return 1;
251 	}
252 	group_first_block = ext4_group_first_block_no(sb, block_group);
253 
254 	/* check whether block bitmap block number is set */
255 	bitmap_blk = ext4_block_bitmap(sb, desc);
256 	offset = bitmap_blk - group_first_block;
257 	if (!ext4_test_bit(offset, bh->b_data))
258 		/* bad block bitmap */
259 		goto err_out;
260 
261 	/* check whether the inode bitmap block number is set */
262 	bitmap_blk = ext4_inode_bitmap(sb, desc);
263 	offset = bitmap_blk - group_first_block;
264 	if (!ext4_test_bit(offset, bh->b_data))
265 		/* bad block bitmap */
266 		goto err_out;
267 
268 	/* check whether the inode table block number is set */
269 	bitmap_blk = ext4_inode_table(sb, desc);
270 	offset = bitmap_blk - group_first_block;
271 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
272 				offset + EXT4_SB(sb)->s_itb_per_group,
273 				offset);
274 	if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
275 		/* good bitmap for inode tables */
276 		return 1;
277 
278 err_out:
279 	ext4_error(sb, "Invalid block bitmap - block_group = %d, block = %llu",
280 			block_group, bitmap_blk);
281 	return 0;
282 }
283 /**
284  * ext4_read_block_bitmap()
285  * @sb:			super block
286  * @block_group:	given block group
287  *
288  * Read the bitmap for a given block_group,and validate the
289  * bits for block/inode/inode tables are set in the bitmaps
290  *
291  * Return buffer_head on success or NULL in case of failure.
292  */
293 struct buffer_head *
294 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
295 {
296 	struct ext4_group_desc *desc;
297 	struct buffer_head *bh = NULL;
298 	ext4_fsblk_t bitmap_blk;
299 
300 	desc = ext4_get_group_desc(sb, block_group, NULL);
301 	if (!desc)
302 		return NULL;
303 	bitmap_blk = ext4_block_bitmap(sb, desc);
304 	bh = sb_getblk(sb, bitmap_blk);
305 	if (unlikely(!bh)) {
306 		ext4_error(sb, "Cannot read block bitmap - "
307 			    "block_group = %u, block_bitmap = %llu",
308 			    block_group, bitmap_blk);
309 		return NULL;
310 	}
311 
312 	if (bitmap_uptodate(bh))
313 		return bh;
314 
315 	lock_buffer(bh);
316 	if (bitmap_uptodate(bh)) {
317 		unlock_buffer(bh);
318 		return bh;
319 	}
320 	ext4_lock_group(sb, block_group);
321 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
322 		ext4_init_block_bitmap(sb, bh, block_group, desc);
323 		set_bitmap_uptodate(bh);
324 		set_buffer_uptodate(bh);
325 		ext4_unlock_group(sb, block_group);
326 		unlock_buffer(bh);
327 		return bh;
328 	}
329 	ext4_unlock_group(sb, block_group);
330 	if (buffer_uptodate(bh)) {
331 		/*
332 		 * if not uninit if bh is uptodate,
333 		 * bitmap is also uptodate
334 		 */
335 		set_bitmap_uptodate(bh);
336 		unlock_buffer(bh);
337 		return bh;
338 	}
339 	/*
340 	 * submit the buffer_head for read. We can
341 	 * safely mark the bitmap as uptodate now.
342 	 * We do it here so the bitmap uptodate bit
343 	 * get set with buffer lock held.
344 	 */
345 	set_bitmap_uptodate(bh);
346 	if (bh_submit_read(bh) < 0) {
347 		put_bh(bh);
348 		ext4_error(sb, "Cannot read block bitmap - "
349 			    "block_group = %u, block_bitmap = %llu",
350 			    block_group, bitmap_blk);
351 		return NULL;
352 	}
353 	ext4_valid_block_bitmap(sb, desc, block_group, bh);
354 	/*
355 	 * file system mounted not to panic on error,
356 	 * continue with corrupt bitmap
357 	 */
358 	return bh;
359 }
360 
361 /**
362  * ext4_add_groupblocks() -- Add given blocks to an existing group
363  * @handle:			handle to this transaction
364  * @sb:				super block
365  * @block:			start physcial block to add to the block group
366  * @count:			number of blocks to free
367  *
368  * This marks the blocks as free in the bitmap. We ask the
369  * mballoc to reload the buddy after this by setting group
370  * EXT4_GROUP_INFO_NEED_INIT_BIT flag
371  */
372 void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
373 			 ext4_fsblk_t block, unsigned long count)
374 {
375 	struct buffer_head *bitmap_bh = NULL;
376 	struct buffer_head *gd_bh;
377 	ext4_group_t block_group;
378 	ext4_grpblk_t bit;
379 	unsigned int i;
380 	struct ext4_group_desc *desc;
381 	struct ext4_sb_info *sbi = EXT4_SB(sb);
382 	int err = 0, ret, blk_free_count;
383 	ext4_grpblk_t blocks_freed;
384 	struct ext4_group_info *grp;
385 
386 	ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
387 
388 	ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
389 	grp = ext4_get_group_info(sb, block_group);
390 	/*
391 	 * Check to see if we are freeing blocks across a group
392 	 * boundary.
393 	 */
394 	if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
395 		goto error_return;
396 	}
397 	bitmap_bh = ext4_read_block_bitmap(sb, block_group);
398 	if (!bitmap_bh)
399 		goto error_return;
400 	desc = ext4_get_group_desc(sb, block_group, &gd_bh);
401 	if (!desc)
402 		goto error_return;
403 
404 	if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
405 	    in_range(ext4_inode_bitmap(sb, desc), block, count) ||
406 	    in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
407 	    in_range(block + count - 1, ext4_inode_table(sb, desc),
408 		     sbi->s_itb_per_group)) {
409 		ext4_error(sb, "Adding blocks in system zones - "
410 			   "Block = %llu, count = %lu",
411 			   block, count);
412 		goto error_return;
413 	}
414 
415 	/*
416 	 * We are about to add blocks to the bitmap,
417 	 * so we need undo access.
418 	 */
419 	BUFFER_TRACE(bitmap_bh, "getting undo access");
420 	err = ext4_journal_get_undo_access(handle, bitmap_bh);
421 	if (err)
422 		goto error_return;
423 
424 	/*
425 	 * We are about to modify some metadata.  Call the journal APIs
426 	 * to unshare ->b_data if a currently-committing transaction is
427 	 * using it
428 	 */
429 	BUFFER_TRACE(gd_bh, "get_write_access");
430 	err = ext4_journal_get_write_access(handle, gd_bh);
431 	if (err)
432 		goto error_return;
433 	/*
434 	 * make sure we don't allow a parallel init on other groups in the
435 	 * same buddy cache
436 	 */
437 	down_write(&grp->alloc_sem);
438 	for (i = 0, blocks_freed = 0; i < count; i++) {
439 		BUFFER_TRACE(bitmap_bh, "clear bit");
440 		if (!ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group),
441 						bit + i, bitmap_bh->b_data)) {
442 			ext4_error(sb, "bit already cleared for block %llu",
443 				   (ext4_fsblk_t)(block + i));
444 			BUFFER_TRACE(bitmap_bh, "bit already cleared");
445 		} else {
446 			blocks_freed++;
447 		}
448 	}
449 	ext4_lock_group(sb, block_group);
450 	blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
451 	ext4_free_blks_set(sb, desc, blk_free_count);
452 	desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
453 	ext4_unlock_group(sb, block_group);
454 	percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
455 
456 	if (sbi->s_log_groups_per_flex) {
457 		ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
458 		atomic_add(blocks_freed,
459 			   &sbi->s_flex_groups[flex_group].free_blocks);
460 	}
461 	/*
462 	 * request to reload the buddy with the
463 	 * new bitmap information
464 	 */
465 	set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
466 	grp->bb_free += blocks_freed;
467 	up_write(&grp->alloc_sem);
468 
469 	/* We dirtied the bitmap block */
470 	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
471 	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
472 
473 	/* And the group descriptor block */
474 	BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
475 	ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
476 	if (!err)
477 		err = ret;
478 
479 error_return:
480 	brelse(bitmap_bh);
481 	ext4_std_error(sb, err);
482 	return;
483 }
484 
485 /**
486  * ext4_has_free_blocks()
487  * @sbi:	in-core super block structure.
488  * @nblocks:	number of needed blocks
489  *
490  * Check if filesystem has nblocks free & available for allocation.
491  * On success return 1, return 0 on failure.
492  */
493 static int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
494 {
495 	s64 free_blocks, dirty_blocks, root_blocks;
496 	struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
497 	struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
498 
499 	free_blocks  = percpu_counter_read_positive(fbc);
500 	dirty_blocks = percpu_counter_read_positive(dbc);
501 	root_blocks = ext4_r_blocks_count(sbi->s_es);
502 
503 	if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
504 						EXT4_FREEBLOCKS_WATERMARK) {
505 		free_blocks  = percpu_counter_sum_positive(fbc);
506 		dirty_blocks = percpu_counter_sum_positive(dbc);
507 		if (dirty_blocks < 0) {
508 			printk(KERN_CRIT "Dirty block accounting "
509 					"went wrong %lld\n",
510 					(long long)dirty_blocks);
511 		}
512 	}
513 	/* Check whether we have space after
514 	 * accounting for current dirty blocks & root reserved blocks.
515 	 */
516 	if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
517 		return 1;
518 
519 	/* Hm, nope.  Are (enough) root reserved blocks available? */
520 	if (sbi->s_resuid == current_fsuid() ||
521 	    ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
522 	    capable(CAP_SYS_RESOURCE)) {
523 		if (free_blocks >= (nblocks + dirty_blocks))
524 			return 1;
525 	}
526 
527 	return 0;
528 }
529 
530 int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
531 						s64 nblocks)
532 {
533 	if (ext4_has_free_blocks(sbi, nblocks)) {
534 		percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
535 		return 0;
536 	} else
537 		return -ENOSPC;
538 }
539 
540 /**
541  * ext4_should_retry_alloc()
542  * @sb:			super block
543  * @retries		number of attemps has been made
544  *
545  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
546  * it is profitable to retry the operation, this function will wait
547  * for the current or commiting transaction to complete, and then
548  * return TRUE.
549  *
550  * if the total number of retries exceed three times, return FALSE.
551  */
552 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
553 {
554 	if (!ext4_has_free_blocks(EXT4_SB(sb), 1) ||
555 	    (*retries)++ > 3 ||
556 	    !EXT4_SB(sb)->s_journal)
557 		return 0;
558 
559 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
560 
561 	return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
562 }
563 
564 /*
565  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
566  *
567  * @handle:             handle to this transaction
568  * @inode:              file inode
569  * @goal:               given target block(filesystem wide)
570  * @count:		pointer to total number of blocks needed
571  * @errp:               error code
572  *
573  * Return 1st allocated block number on success, *count stores total account
574  * error stores in errp pointer
575  */
576 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
577 		ext4_fsblk_t goal, unsigned long *count, int *errp)
578 {
579 	struct ext4_allocation_request ar;
580 	ext4_fsblk_t ret;
581 
582 	memset(&ar, 0, sizeof(ar));
583 	/* Fill with neighbour allocated blocks */
584 	ar.inode = inode;
585 	ar.goal = goal;
586 	ar.len = count ? *count : 1;
587 
588 	ret = ext4_mb_new_blocks(handle, &ar, errp);
589 	if (count)
590 		*count = ar.len;
591 	/*
592 	 * Account for the allocated meta blocks.  We will never
593 	 * fail EDQUOT for metdata, but we do account for it.
594 	 */
595 	if (!(*errp) &&
596 	    ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
597 		spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
598 		EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
599 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
600 		dquot_alloc_block_nofail(inode, ar.len);
601 	}
602 	return ret;
603 }
604 
605 /**
606  * ext4_count_free_blocks() -- count filesystem free blocks
607  * @sb:		superblock
608  *
609  * Adds up the number of free blocks from each block group.
610  */
611 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
612 {
613 	ext4_fsblk_t desc_count;
614 	struct ext4_group_desc *gdp;
615 	ext4_group_t i;
616 	ext4_group_t ngroups = ext4_get_groups_count(sb);
617 #ifdef EXT4FS_DEBUG
618 	struct ext4_super_block *es;
619 	ext4_fsblk_t bitmap_count;
620 	unsigned int x;
621 	struct buffer_head *bitmap_bh = NULL;
622 
623 	es = EXT4_SB(sb)->s_es;
624 	desc_count = 0;
625 	bitmap_count = 0;
626 	gdp = NULL;
627 
628 	for (i = 0; i < ngroups; i++) {
629 		gdp = ext4_get_group_desc(sb, i, NULL);
630 		if (!gdp)
631 			continue;
632 		desc_count += ext4_free_blks_count(sb, gdp);
633 		brelse(bitmap_bh);
634 		bitmap_bh = ext4_read_block_bitmap(sb, i);
635 		if (bitmap_bh == NULL)
636 			continue;
637 
638 		x = ext4_count_free(bitmap_bh, sb->s_blocksize);
639 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
640 			i, ext4_free_blks_count(sb, gdp), x);
641 		bitmap_count += x;
642 	}
643 	brelse(bitmap_bh);
644 	printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
645 		", computed = %llu, %llu\n", ext4_free_blocks_count(es),
646 	       desc_count, bitmap_count);
647 	return bitmap_count;
648 #else
649 	desc_count = 0;
650 	for (i = 0; i < ngroups; i++) {
651 		gdp = ext4_get_group_desc(sb, i, NULL);
652 		if (!gdp)
653 			continue;
654 		desc_count += ext4_free_blks_count(sb, gdp);
655 	}
656 
657 	return desc_count;
658 #endif
659 }
660 
661 static inline int test_root(ext4_group_t a, int b)
662 {
663 	int num = b;
664 
665 	while (a > num)
666 		num *= b;
667 	return num == a;
668 }
669 
670 static int ext4_group_sparse(ext4_group_t group)
671 {
672 	if (group <= 1)
673 		return 1;
674 	if (!(group & 1))
675 		return 0;
676 	return (test_root(group, 7) || test_root(group, 5) ||
677 		test_root(group, 3));
678 }
679 
680 /**
681  *	ext4_bg_has_super - number of blocks used by the superblock in group
682  *	@sb: superblock for filesystem
683  *	@group: group number to check
684  *
685  *	Return the number of blocks used by the superblock (primary or backup)
686  *	in this group.  Currently this will be only 0 or 1.
687  */
688 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
689 {
690 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
691 				EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
692 			!ext4_group_sparse(group))
693 		return 0;
694 	return 1;
695 }
696 
697 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
698 					ext4_group_t group)
699 {
700 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
701 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
702 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
703 
704 	if (group == first || group == first + 1 || group == last)
705 		return 1;
706 	return 0;
707 }
708 
709 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
710 					ext4_group_t group)
711 {
712 	if (!ext4_bg_has_super(sb, group))
713 		return 0;
714 
715 	if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
716 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
717 	else
718 		return EXT4_SB(sb)->s_gdb_count;
719 }
720 
721 /**
722  *	ext4_bg_num_gdb - number of blocks used by the group table in group
723  *	@sb: superblock for filesystem
724  *	@group: group number to check
725  *
726  *	Return the number of blocks used by the group descriptor table
727  *	(primary or backup) in this group.  In the future there may be a
728  *	different number of descriptor blocks in each group.
729  */
730 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
731 {
732 	unsigned long first_meta_bg =
733 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
734 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
735 
736 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
737 			metagroup < first_meta_bg)
738 		return ext4_bg_num_gdb_nometa(sb, group);
739 
740 	return ext4_bg_num_gdb_meta(sb,group);
741 
742 }
743 
744