xref: /openbmc/linux/fs/ext4/balloc.c (revision 95e9fd10)
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 #include <trace/events/ext4.h>
25 
26 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
27 					    ext4_group_t block_group);
28 /*
29  * balloc.c contains the blocks allocation and deallocation routines
30  */
31 
32 /*
33  * Calculate the block group number and offset into the block/cluster
34  * allocation bitmap, given a block number
35  */
36 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
37 		ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
38 {
39 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
40 	ext4_grpblk_t offset;
41 
42 	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
43 	offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
44 		EXT4_SB(sb)->s_cluster_bits;
45 	if (offsetp)
46 		*offsetp = offset;
47 	if (blockgrpp)
48 		*blockgrpp = blocknr;
49 
50 }
51 
52 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
53 			ext4_group_t block_group)
54 {
55 	ext4_group_t actual_group;
56 	ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
57 	if (actual_group == block_group)
58 		return 1;
59 	return 0;
60 }
61 
62 /* Return the number of clusters used for file system metadata; this
63  * represents the overhead needed by the file system.
64  */
65 unsigned ext4_num_overhead_clusters(struct super_block *sb,
66 				    ext4_group_t block_group,
67 				    struct ext4_group_desc *gdp)
68 {
69 	unsigned num_clusters;
70 	int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
71 	ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
72 	ext4_fsblk_t itbl_blk;
73 	struct ext4_sb_info *sbi = EXT4_SB(sb);
74 
75 	/* This is the number of clusters used by the superblock,
76 	 * block group descriptors, and reserved block group
77 	 * descriptor blocks */
78 	num_clusters = ext4_num_base_meta_clusters(sb, block_group);
79 
80 	/*
81 	 * For the allocation bitmaps and inode table, we first need
82 	 * to check to see if the block is in the block group.  If it
83 	 * is, then check to see if the cluster is already accounted
84 	 * for in the clusters used for the base metadata cluster, or
85 	 * if we can increment the base metadata cluster to include
86 	 * that block.  Otherwise, we will have to track the cluster
87 	 * used for the allocation bitmap or inode table explicitly.
88 	 * Normally all of these blocks are contiguous, so the special
89 	 * case handling shouldn't be necessary except for *very*
90 	 * unusual file system layouts.
91 	 */
92 	if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
93 		block_cluster = EXT4_B2C(sbi,
94 					 ext4_block_bitmap(sb, gdp) - start);
95 		if (block_cluster < num_clusters)
96 			block_cluster = -1;
97 		else if (block_cluster == num_clusters) {
98 			num_clusters++;
99 			block_cluster = -1;
100 		}
101 	}
102 
103 	if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
104 		inode_cluster = EXT4_B2C(sbi,
105 					 ext4_inode_bitmap(sb, gdp) - start);
106 		if (inode_cluster < num_clusters)
107 			inode_cluster = -1;
108 		else if (inode_cluster == num_clusters) {
109 			num_clusters++;
110 			inode_cluster = -1;
111 		}
112 	}
113 
114 	itbl_blk = ext4_inode_table(sb, gdp);
115 	for (i = 0; i < sbi->s_itb_per_group; i++) {
116 		if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
117 			c = EXT4_B2C(sbi, itbl_blk + i - start);
118 			if ((c < num_clusters) || (c == inode_cluster) ||
119 			    (c == block_cluster) || (c == itbl_cluster))
120 				continue;
121 			if (c == num_clusters) {
122 				num_clusters++;
123 				continue;
124 			}
125 			num_clusters++;
126 			itbl_cluster = c;
127 		}
128 	}
129 
130 	if (block_cluster != -1)
131 		num_clusters++;
132 	if (inode_cluster != -1)
133 		num_clusters++;
134 
135 	return num_clusters;
136 }
137 
138 static unsigned int num_clusters_in_group(struct super_block *sb,
139 					  ext4_group_t block_group)
140 {
141 	unsigned int blocks;
142 
143 	if (block_group == ext4_get_groups_count(sb) - 1) {
144 		/*
145 		 * Even though mke2fs always initializes the first and
146 		 * last group, just in case some other tool was used,
147 		 * we need to make sure we calculate the right free
148 		 * blocks.
149 		 */
150 		blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
151 			ext4_group_first_block_no(sb, block_group);
152 	} else
153 		blocks = EXT4_BLOCKS_PER_GROUP(sb);
154 	return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
155 }
156 
157 /* Initializes an uninitialized block bitmap */
158 void ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
159 			    ext4_group_t block_group,
160 			    struct ext4_group_desc *gdp)
161 {
162 	unsigned int bit, bit_max;
163 	struct ext4_sb_info *sbi = EXT4_SB(sb);
164 	ext4_fsblk_t start, tmp;
165 	int flex_bg = 0;
166 
167 	J_ASSERT_BH(bh, buffer_locked(bh));
168 
169 	/* If checksum is bad mark all blocks used to prevent allocation
170 	 * essentially implementing a per-group read-only flag. */
171 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
172 		ext4_error(sb, "Checksum bad for group %u", block_group);
173 		ext4_free_group_clusters_set(sb, gdp, 0);
174 		ext4_free_inodes_set(sb, gdp, 0);
175 		ext4_itable_unused_set(sb, gdp, 0);
176 		memset(bh->b_data, 0xff, sb->s_blocksize);
177 		ext4_block_bitmap_csum_set(sb, block_group, gdp, bh,
178 					   EXT4_BLOCKS_PER_GROUP(sb) / 8);
179 		return;
180 	}
181 	memset(bh->b_data, 0, sb->s_blocksize);
182 
183 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
184 	for (bit = 0; bit < bit_max; bit++)
185 		ext4_set_bit(bit, bh->b_data);
186 
187 	start = ext4_group_first_block_no(sb, block_group);
188 
189 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
190 		flex_bg = 1;
191 
192 	/* Set bits for block and inode bitmaps, and inode table */
193 	tmp = ext4_block_bitmap(sb, gdp);
194 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
195 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
196 
197 	tmp = ext4_inode_bitmap(sb, gdp);
198 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
199 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
200 
201 	tmp = ext4_inode_table(sb, gdp);
202 	for (; tmp < ext4_inode_table(sb, gdp) +
203 		     sbi->s_itb_per_group; tmp++) {
204 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
205 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
206 	}
207 
208 	/*
209 	 * Also if the number of blocks within the group is less than
210 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
211 	 * of the block bitmap to 1
212 	 */
213 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
214 			     sb->s_blocksize * 8, bh->b_data);
215 	ext4_block_bitmap_csum_set(sb, block_group, gdp, bh,
216 				   EXT4_BLOCKS_PER_GROUP(sb) / 8);
217 	ext4_group_desc_csum_set(sb, block_group, gdp);
218 }
219 
220 /* Return the number of free blocks in a block group.  It is used when
221  * the block bitmap is uninitialized, so we can't just count the bits
222  * in the bitmap. */
223 unsigned ext4_free_clusters_after_init(struct super_block *sb,
224 				       ext4_group_t block_group,
225 				       struct ext4_group_desc *gdp)
226 {
227 	return num_clusters_in_group(sb, block_group) -
228 		ext4_num_overhead_clusters(sb, block_group, gdp);
229 }
230 
231 /*
232  * The free blocks are managed by bitmaps.  A file system contains several
233  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
234  * block for inodes, N blocks for the inode table and data blocks.
235  *
236  * The file system contains group descriptors which are located after the
237  * super block.  Each descriptor contains the number of the bitmap block and
238  * the free blocks count in the block.  The descriptors are loaded in memory
239  * when a file system is mounted (see ext4_fill_super).
240  */
241 
242 /**
243  * ext4_get_group_desc() -- load group descriptor from disk
244  * @sb:			super block
245  * @block_group:	given block group
246  * @bh:			pointer to the buffer head to store the block
247  *			group descriptor
248  */
249 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
250 					     ext4_group_t block_group,
251 					     struct buffer_head **bh)
252 {
253 	unsigned int group_desc;
254 	unsigned int offset;
255 	ext4_group_t ngroups = ext4_get_groups_count(sb);
256 	struct ext4_group_desc *desc;
257 	struct ext4_sb_info *sbi = EXT4_SB(sb);
258 
259 	if (block_group >= ngroups) {
260 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
261 			   " groups_count = %u", block_group, ngroups);
262 
263 		return NULL;
264 	}
265 
266 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
267 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
268 	if (!sbi->s_group_desc[group_desc]) {
269 		ext4_error(sb, "Group descriptor not loaded - "
270 			   "block_group = %u, group_desc = %u, desc = %u",
271 			   block_group, group_desc, offset);
272 		return NULL;
273 	}
274 
275 	desc = (struct ext4_group_desc *)(
276 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
277 		offset * EXT4_DESC_SIZE(sb));
278 	if (bh)
279 		*bh = sbi->s_group_desc[group_desc];
280 	return desc;
281 }
282 
283 /*
284  * Return the block number which was discovered to be invalid, or 0 if
285  * the block bitmap is valid.
286  */
287 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
288 					    struct ext4_group_desc *desc,
289 					    unsigned int block_group,
290 					    struct buffer_head *bh)
291 {
292 	ext4_grpblk_t offset;
293 	ext4_grpblk_t next_zero_bit;
294 	ext4_fsblk_t blk;
295 	ext4_fsblk_t group_first_block;
296 
297 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
298 		/* with FLEX_BG, the inode/block bitmaps and itable
299 		 * blocks may not be in the group at all
300 		 * so the bitmap validation will be skipped for those groups
301 		 * or it has to also read the block group where the bitmaps
302 		 * are located to verify they are set.
303 		 */
304 		return 0;
305 	}
306 	group_first_block = ext4_group_first_block_no(sb, block_group);
307 
308 	/* check whether block bitmap block number is set */
309 	blk = ext4_block_bitmap(sb, desc);
310 	offset = blk - group_first_block;
311 	if (!ext4_test_bit(offset, bh->b_data))
312 		/* bad block bitmap */
313 		return blk;
314 
315 	/* check whether the inode bitmap block number is set */
316 	blk = ext4_inode_bitmap(sb, desc);
317 	offset = blk - group_first_block;
318 	if (!ext4_test_bit(offset, bh->b_data))
319 		/* bad block bitmap */
320 		return blk;
321 
322 	/* check whether the inode table block number is set */
323 	blk = ext4_inode_table(sb, desc);
324 	offset = blk - group_first_block;
325 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
326 				offset + EXT4_SB(sb)->s_itb_per_group,
327 				offset);
328 	if (next_zero_bit < offset + EXT4_SB(sb)->s_itb_per_group)
329 		/* bad bitmap for inode tables */
330 		return blk;
331 	return 0;
332 }
333 
334 void ext4_validate_block_bitmap(struct super_block *sb,
335 			       struct ext4_group_desc *desc,
336 			       unsigned int block_group,
337 			       struct buffer_head *bh)
338 {
339 	ext4_fsblk_t	blk;
340 
341 	if (buffer_verified(bh))
342 		return;
343 
344 	ext4_lock_group(sb, block_group);
345 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
346 	if (unlikely(blk != 0)) {
347 		ext4_unlock_group(sb, block_group);
348 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
349 			   block_group, blk);
350 		return;
351 	}
352 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
353 			desc, bh, EXT4_BLOCKS_PER_GROUP(sb) / 8))) {
354 		ext4_unlock_group(sb, block_group);
355 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
356 		return;
357 	}
358 	set_buffer_verified(bh);
359 	ext4_unlock_group(sb, block_group);
360 }
361 
362 /**
363  * ext4_read_block_bitmap()
364  * @sb:			super block
365  * @block_group:	given block group
366  *
367  * Read the bitmap for a given block_group,and validate the
368  * bits for block/inode/inode tables are set in the bitmaps
369  *
370  * Return buffer_head on success or NULL in case of failure.
371  */
372 struct buffer_head *
373 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
374 {
375 	struct ext4_group_desc *desc;
376 	struct buffer_head *bh;
377 	ext4_fsblk_t bitmap_blk;
378 
379 	desc = ext4_get_group_desc(sb, block_group, NULL);
380 	if (!desc)
381 		return NULL;
382 	bitmap_blk = ext4_block_bitmap(sb, desc);
383 	bh = sb_getblk(sb, bitmap_blk);
384 	if (unlikely(!bh)) {
385 		ext4_error(sb, "Cannot get buffer for block bitmap - "
386 			   "block_group = %u, block_bitmap = %llu",
387 			   block_group, bitmap_blk);
388 		return NULL;
389 	}
390 
391 	if (bitmap_uptodate(bh))
392 		goto verify;
393 
394 	lock_buffer(bh);
395 	if (bitmap_uptodate(bh)) {
396 		unlock_buffer(bh);
397 		goto verify;
398 	}
399 	ext4_lock_group(sb, block_group);
400 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
401 		ext4_init_block_bitmap(sb, bh, block_group, desc);
402 		set_bitmap_uptodate(bh);
403 		set_buffer_uptodate(bh);
404 		ext4_unlock_group(sb, block_group);
405 		unlock_buffer(bh);
406 		return bh;
407 	}
408 	ext4_unlock_group(sb, block_group);
409 	if (buffer_uptodate(bh)) {
410 		/*
411 		 * if not uninit if bh is uptodate,
412 		 * bitmap is also uptodate
413 		 */
414 		set_bitmap_uptodate(bh);
415 		unlock_buffer(bh);
416 		goto verify;
417 	}
418 	/*
419 	 * submit the buffer_head for reading
420 	 */
421 	set_buffer_new(bh);
422 	trace_ext4_read_block_bitmap_load(sb, block_group);
423 	bh->b_end_io = ext4_end_bitmap_read;
424 	get_bh(bh);
425 	submit_bh(READ, bh);
426 	return bh;
427 verify:
428 	ext4_validate_block_bitmap(sb, desc, block_group, bh);
429 	return bh;
430 }
431 
432 /* Returns 0 on success, 1 on error */
433 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
434 			   struct buffer_head *bh)
435 {
436 	struct ext4_group_desc *desc;
437 
438 	if (!buffer_new(bh))
439 		return 0;
440 	desc = ext4_get_group_desc(sb, block_group, NULL);
441 	if (!desc)
442 		return 1;
443 	wait_on_buffer(bh);
444 	if (!buffer_uptodate(bh)) {
445 		ext4_error(sb, "Cannot read block bitmap - "
446 			   "block_group = %u, block_bitmap = %llu",
447 			   block_group, (unsigned long long) bh->b_blocknr);
448 		return 1;
449 	}
450 	clear_buffer_new(bh);
451 	/* Panic or remount fs read-only if block bitmap is invalid */
452 	ext4_validate_block_bitmap(sb, desc, block_group, bh);
453 	return 0;
454 }
455 
456 struct buffer_head *
457 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
458 {
459 	struct buffer_head *bh;
460 
461 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
462 	if (ext4_wait_block_bitmap(sb, block_group, bh)) {
463 		put_bh(bh);
464 		return NULL;
465 	}
466 	return bh;
467 }
468 
469 /**
470  * ext4_has_free_clusters()
471  * @sbi:	in-core super block structure.
472  * @nclusters:	number of needed blocks
473  * @flags:	flags from ext4_mb_new_blocks()
474  *
475  * Check if filesystem has nclusters free & available for allocation.
476  * On success return 1, return 0 on failure.
477  */
478 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
479 				  s64 nclusters, unsigned int flags)
480 {
481 	s64 free_clusters, dirty_clusters, root_clusters;
482 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
483 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
484 
485 	free_clusters  = percpu_counter_read_positive(fcc);
486 	dirty_clusters = percpu_counter_read_positive(dcc);
487 	root_clusters = EXT4_B2C(sbi, ext4_r_blocks_count(sbi->s_es));
488 
489 	if (free_clusters - (nclusters + root_clusters + dirty_clusters) <
490 					EXT4_FREECLUSTERS_WATERMARK) {
491 		free_clusters  = EXT4_C2B(sbi, percpu_counter_sum_positive(fcc));
492 		dirty_clusters = percpu_counter_sum_positive(dcc);
493 	}
494 	/* Check whether we have space after accounting for current
495 	 * dirty clusters & root reserved clusters.
496 	 */
497 	if (free_clusters >= ((root_clusters + nclusters) + dirty_clusters))
498 		return 1;
499 
500 	/* Hm, nope.  Are (enough) root reserved clusters available? */
501 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
502 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
503 	    capable(CAP_SYS_RESOURCE) ||
504 		(flags & EXT4_MB_USE_ROOT_BLOCKS)) {
505 
506 		if (free_clusters >= (nclusters + dirty_clusters))
507 			return 1;
508 	}
509 
510 	return 0;
511 }
512 
513 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
514 			     s64 nclusters, unsigned int flags)
515 {
516 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
517 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
518 		return 0;
519 	} else
520 		return -ENOSPC;
521 }
522 
523 /**
524  * ext4_should_retry_alloc()
525  * @sb:			super block
526  * @retries		number of attemps has been made
527  *
528  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
529  * it is profitable to retry the operation, this function will wait
530  * for the current or committing transaction to complete, and then
531  * return TRUE.
532  *
533  * if the total number of retries exceed three times, return FALSE.
534  */
535 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
536 {
537 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
538 	    (*retries)++ > 3 ||
539 	    !EXT4_SB(sb)->s_journal)
540 		return 0;
541 
542 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
543 
544 	return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
545 }
546 
547 /*
548  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
549  *
550  * @handle:             handle to this transaction
551  * @inode:              file inode
552  * @goal:               given target block(filesystem wide)
553  * @count:		pointer to total number of clusters needed
554  * @errp:               error code
555  *
556  * Return 1st allocated block number on success, *count stores total account
557  * error stores in errp pointer
558  */
559 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
560 				  ext4_fsblk_t goal, unsigned int flags,
561 				  unsigned long *count, int *errp)
562 {
563 	struct ext4_allocation_request ar;
564 	ext4_fsblk_t ret;
565 
566 	memset(&ar, 0, sizeof(ar));
567 	/* Fill with neighbour allocated blocks */
568 	ar.inode = inode;
569 	ar.goal = goal;
570 	ar.len = count ? *count : 1;
571 	ar.flags = flags;
572 
573 	ret = ext4_mb_new_blocks(handle, &ar, errp);
574 	if (count)
575 		*count = ar.len;
576 	/*
577 	 * Account for the allocated meta blocks.  We will never
578 	 * fail EDQUOT for metdata, but we do account for it.
579 	 */
580 	if (!(*errp) &&
581 	    ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
582 		spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
583 		EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
584 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
585 		dquot_alloc_block_nofail(inode,
586 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
587 	}
588 	return ret;
589 }
590 
591 /**
592  * ext4_count_free_clusters() -- count filesystem free clusters
593  * @sb:		superblock
594  *
595  * Adds up the number of free clusters from each block group.
596  */
597 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
598 {
599 	ext4_fsblk_t desc_count;
600 	struct ext4_group_desc *gdp;
601 	ext4_group_t i;
602 	ext4_group_t ngroups = ext4_get_groups_count(sb);
603 #ifdef EXT4FS_DEBUG
604 	struct ext4_super_block *es;
605 	ext4_fsblk_t bitmap_count;
606 	unsigned int x;
607 	struct buffer_head *bitmap_bh = NULL;
608 
609 	es = EXT4_SB(sb)->s_es;
610 	desc_count = 0;
611 	bitmap_count = 0;
612 	gdp = NULL;
613 
614 	for (i = 0; i < ngroups; i++) {
615 		gdp = ext4_get_group_desc(sb, i, NULL);
616 		if (!gdp)
617 			continue;
618 		desc_count += ext4_free_group_clusters(sb, gdp);
619 		brelse(bitmap_bh);
620 		bitmap_bh = ext4_read_block_bitmap(sb, i);
621 		if (bitmap_bh == NULL)
622 			continue;
623 
624 		x = ext4_count_free(bitmap_bh->b_data,
625 				    EXT4_BLOCKS_PER_GROUP(sb) / 8);
626 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
627 			i, ext4_free_group_clusters(sb, gdp), x);
628 		bitmap_count += x;
629 	}
630 	brelse(bitmap_bh);
631 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
632 	       ", computed = %llu, %llu\n",
633 	       EXT4_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
634 	       desc_count, bitmap_count);
635 	return bitmap_count;
636 #else
637 	desc_count = 0;
638 	for (i = 0; i < ngroups; i++) {
639 		gdp = ext4_get_group_desc(sb, i, NULL);
640 		if (!gdp)
641 			continue;
642 		desc_count += ext4_free_group_clusters(sb, gdp);
643 	}
644 
645 	return desc_count;
646 #endif
647 }
648 
649 static inline int test_root(ext4_group_t a, int b)
650 {
651 	int num = b;
652 
653 	while (a > num)
654 		num *= b;
655 	return num == a;
656 }
657 
658 static int ext4_group_sparse(ext4_group_t group)
659 {
660 	if (group <= 1)
661 		return 1;
662 	if (!(group & 1))
663 		return 0;
664 	return (test_root(group, 7) || test_root(group, 5) ||
665 		test_root(group, 3));
666 }
667 
668 /**
669  *	ext4_bg_has_super - number of blocks used by the superblock in group
670  *	@sb: superblock for filesystem
671  *	@group: group number to check
672  *
673  *	Return the number of blocks used by the superblock (primary or backup)
674  *	in this group.  Currently this will be only 0 or 1.
675  */
676 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
677 {
678 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
679 				EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
680 			!ext4_group_sparse(group))
681 		return 0;
682 	return 1;
683 }
684 
685 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
686 					ext4_group_t group)
687 {
688 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
689 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
690 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
691 
692 	if (group == first || group == first + 1 || group == last)
693 		return 1;
694 	return 0;
695 }
696 
697 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
698 					ext4_group_t group)
699 {
700 	if (!ext4_bg_has_super(sb, group))
701 		return 0;
702 
703 	if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
704 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
705 	else
706 		return EXT4_SB(sb)->s_gdb_count;
707 }
708 
709 /**
710  *	ext4_bg_num_gdb - number of blocks used by the group table in group
711  *	@sb: superblock for filesystem
712  *	@group: group number to check
713  *
714  *	Return the number of blocks used by the group descriptor table
715  *	(primary or backup) in this group.  In the future there may be a
716  *	different number of descriptor blocks in each group.
717  */
718 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
719 {
720 	unsigned long first_meta_bg =
721 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
722 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
723 
724 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
725 			metagroup < first_meta_bg)
726 		return ext4_bg_num_gdb_nometa(sb, group);
727 
728 	return ext4_bg_num_gdb_meta(sb,group);
729 
730 }
731 
732 /*
733  * This function returns the number of file system metadata clusters at
734  * the beginning of a block group, including the reserved gdt blocks.
735  */
736 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
737 				     ext4_group_t block_group)
738 {
739 	struct ext4_sb_info *sbi = EXT4_SB(sb);
740 	unsigned num;
741 
742 	/* Check for superblock and gdt backups in this group */
743 	num = ext4_bg_has_super(sb, block_group);
744 
745 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
746 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
747 			  sbi->s_desc_per_block) {
748 		if (num) {
749 			num += ext4_bg_num_gdb(sb, block_group);
750 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
751 		}
752 	} else { /* For META_BG_BLOCK_GROUPS */
753 		num += ext4_bg_num_gdb(sb, block_group);
754 	}
755 	return EXT4_NUM_B2C(sbi, num);
756 }
757 /**
758  *	ext4_inode_to_goal_block - return a hint for block allocation
759  *	@inode: inode for block allocation
760  *
761  *	Return the ideal location to start allocating blocks for a
762  *	newly created inode.
763  */
764 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
765 {
766 	struct ext4_inode_info *ei = EXT4_I(inode);
767 	ext4_group_t block_group;
768 	ext4_grpblk_t colour;
769 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
770 	ext4_fsblk_t bg_start;
771 	ext4_fsblk_t last_block;
772 
773 	block_group = ei->i_block_group;
774 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
775 		/*
776 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
777 		 * block groups per flexgroup, reserve the first block
778 		 * group for directories and special files.  Regular
779 		 * files will start at the second block group.  This
780 		 * tends to speed up directory access and improves
781 		 * fsck times.
782 		 */
783 		block_group &= ~(flex_size-1);
784 		if (S_ISREG(inode->i_mode))
785 			block_group++;
786 	}
787 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
788 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
789 
790 	/*
791 	 * If we are doing delayed allocation, we don't need take
792 	 * colour into account.
793 	 */
794 	if (test_opt(inode->i_sb, DELALLOC))
795 		return bg_start;
796 
797 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
798 		colour = (current->pid % 16) *
799 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
800 	else
801 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
802 	return bg_start + colour;
803 }
804 
805