xref: /linux/fs/ext4/balloc.c (revision c98be0c96db00e9b6b02d31e0fa7590c54cdaaac)
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 block group number for a given block number
34  */
35 ext4_group_t ext4_get_group_number(struct super_block *sb,
36 				   ext4_fsblk_t block)
37 {
38 	ext4_group_t group;
39 
40 	if (test_opt2(sb, STD_GROUP_SIZE))
41 		group = (block -
42 			 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >>
43 			(EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3);
44 	else
45 		ext4_get_group_no_and_offset(sb, block, &group, NULL);
46 	return group;
47 }
48 
49 /*
50  * Calculate the block group number and offset into the block/cluster
51  * allocation bitmap, given a block number
52  */
53 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
54 		ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
55 {
56 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
57 	ext4_grpblk_t offset;
58 
59 	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
60 	offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
61 		EXT4_SB(sb)->s_cluster_bits;
62 	if (offsetp)
63 		*offsetp = offset;
64 	if (blockgrpp)
65 		*blockgrpp = blocknr;
66 
67 }
68 
69 /*
70  * Check whether the 'block' lives within the 'block_group'. Returns 1 if so
71  * and 0 otherwise.
72  */
73 static inline int ext4_block_in_group(struct super_block *sb,
74 				      ext4_fsblk_t block,
75 				      ext4_group_t block_group)
76 {
77 	ext4_group_t actual_group;
78 
79 	actual_group = ext4_get_group_number(sb, block);
80 	return (actual_group == block_group) ? 1 : 0;
81 }
82 
83 /* Return the number of clusters used for file system metadata; this
84  * represents the overhead needed by the file system.
85  */
86 unsigned ext4_num_overhead_clusters(struct super_block *sb,
87 				    ext4_group_t block_group,
88 				    struct ext4_group_desc *gdp)
89 {
90 	unsigned num_clusters;
91 	int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
92 	ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
93 	ext4_fsblk_t itbl_blk;
94 	struct ext4_sb_info *sbi = EXT4_SB(sb);
95 
96 	/* This is the number of clusters used by the superblock,
97 	 * block group descriptors, and reserved block group
98 	 * descriptor blocks */
99 	num_clusters = ext4_num_base_meta_clusters(sb, block_group);
100 
101 	/*
102 	 * For the allocation bitmaps and inode table, we first need
103 	 * to check to see if the block is in the block group.  If it
104 	 * is, then check to see if the cluster is already accounted
105 	 * for in the clusters used for the base metadata cluster, or
106 	 * if we can increment the base metadata cluster to include
107 	 * that block.  Otherwise, we will have to track the cluster
108 	 * used for the allocation bitmap or inode table explicitly.
109 	 * Normally all of these blocks are contiguous, so the special
110 	 * case handling shouldn't be necessary except for *very*
111 	 * unusual file system layouts.
112 	 */
113 	if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
114 		block_cluster = EXT4_B2C(sbi,
115 					 ext4_block_bitmap(sb, gdp) - start);
116 		if (block_cluster < num_clusters)
117 			block_cluster = -1;
118 		else if (block_cluster == num_clusters) {
119 			num_clusters++;
120 			block_cluster = -1;
121 		}
122 	}
123 
124 	if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
125 		inode_cluster = EXT4_B2C(sbi,
126 					 ext4_inode_bitmap(sb, gdp) - start);
127 		if (inode_cluster < num_clusters)
128 			inode_cluster = -1;
129 		else if (inode_cluster == num_clusters) {
130 			num_clusters++;
131 			inode_cluster = -1;
132 		}
133 	}
134 
135 	itbl_blk = ext4_inode_table(sb, gdp);
136 	for (i = 0; i < sbi->s_itb_per_group; i++) {
137 		if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
138 			c = EXT4_B2C(sbi, itbl_blk + i - start);
139 			if ((c < num_clusters) || (c == inode_cluster) ||
140 			    (c == block_cluster) || (c == itbl_cluster))
141 				continue;
142 			if (c == num_clusters) {
143 				num_clusters++;
144 				continue;
145 			}
146 			num_clusters++;
147 			itbl_cluster = c;
148 		}
149 	}
150 
151 	if (block_cluster != -1)
152 		num_clusters++;
153 	if (inode_cluster != -1)
154 		num_clusters++;
155 
156 	return num_clusters;
157 }
158 
159 static unsigned int num_clusters_in_group(struct super_block *sb,
160 					  ext4_group_t block_group)
161 {
162 	unsigned int blocks;
163 
164 	if (block_group == ext4_get_groups_count(sb) - 1) {
165 		/*
166 		 * Even though mke2fs always initializes the first and
167 		 * last group, just in case some other tool was used,
168 		 * we need to make sure we calculate the right free
169 		 * blocks.
170 		 */
171 		blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
172 			ext4_group_first_block_no(sb, block_group);
173 	} else
174 		blocks = EXT4_BLOCKS_PER_GROUP(sb);
175 	return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
176 }
177 
178 /* Initializes an uninitialized block bitmap */
179 void ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
180 			    ext4_group_t block_group,
181 			    struct ext4_group_desc *gdp)
182 {
183 	unsigned int bit, bit_max;
184 	struct ext4_sb_info *sbi = EXT4_SB(sb);
185 	ext4_fsblk_t start, tmp;
186 	int flex_bg = 0;
187 	struct ext4_group_info *grp;
188 
189 	J_ASSERT_BH(bh, buffer_locked(bh));
190 
191 	/* If checksum is bad mark all blocks used to prevent allocation
192 	 * essentially implementing a per-group read-only flag. */
193 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
194 		ext4_error(sb, "Checksum bad for group %u", block_group);
195 		grp = ext4_get_group_info(sb, block_group);
196 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
197 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
198 		return;
199 	}
200 	memset(bh->b_data, 0, sb->s_blocksize);
201 
202 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
203 	for (bit = 0; bit < bit_max; bit++)
204 		ext4_set_bit(bit, bh->b_data);
205 
206 	start = ext4_group_first_block_no(sb, block_group);
207 
208 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
209 		flex_bg = 1;
210 
211 	/* Set bits for block and inode bitmaps, and inode table */
212 	tmp = ext4_block_bitmap(sb, gdp);
213 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
214 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
215 
216 	tmp = ext4_inode_bitmap(sb, gdp);
217 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
218 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
219 
220 	tmp = ext4_inode_table(sb, gdp);
221 	for (; tmp < ext4_inode_table(sb, gdp) +
222 		     sbi->s_itb_per_group; tmp++) {
223 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
224 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
225 	}
226 
227 	/*
228 	 * Also if the number of blocks within the group is less than
229 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
230 	 * of the block bitmap to 1
231 	 */
232 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
233 			     sb->s_blocksize * 8, bh->b_data);
234 	ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
235 	ext4_group_desc_csum_set(sb, block_group, gdp);
236 }
237 
238 /* Return the number of free blocks in a block group.  It is used when
239  * the block bitmap is uninitialized, so we can't just count the bits
240  * in the bitmap. */
241 unsigned ext4_free_clusters_after_init(struct super_block *sb,
242 				       ext4_group_t block_group,
243 				       struct ext4_group_desc *gdp)
244 {
245 	return num_clusters_in_group(sb, block_group) -
246 		ext4_num_overhead_clusters(sb, block_group, gdp);
247 }
248 
249 /*
250  * The free blocks are managed by bitmaps.  A file system contains several
251  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
252  * block for inodes, N blocks for the inode table and data blocks.
253  *
254  * The file system contains group descriptors which are located after the
255  * super block.  Each descriptor contains the number of the bitmap block and
256  * the free blocks count in the block.  The descriptors are loaded in memory
257  * when a file system is mounted (see ext4_fill_super).
258  */
259 
260 /**
261  * ext4_get_group_desc() -- load group descriptor from disk
262  * @sb:			super block
263  * @block_group:	given block group
264  * @bh:			pointer to the buffer head to store the block
265  *			group descriptor
266  */
267 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
268 					     ext4_group_t block_group,
269 					     struct buffer_head **bh)
270 {
271 	unsigned int group_desc;
272 	unsigned int offset;
273 	ext4_group_t ngroups = ext4_get_groups_count(sb);
274 	struct ext4_group_desc *desc;
275 	struct ext4_sb_info *sbi = EXT4_SB(sb);
276 
277 	if (block_group >= ngroups) {
278 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
279 			   " groups_count = %u", block_group, ngroups);
280 
281 		return NULL;
282 	}
283 
284 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
285 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
286 	if (!sbi->s_group_desc[group_desc]) {
287 		ext4_error(sb, "Group descriptor not loaded - "
288 			   "block_group = %u, group_desc = %u, desc = %u",
289 			   block_group, group_desc, offset);
290 		return NULL;
291 	}
292 
293 	desc = (struct ext4_group_desc *)(
294 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
295 		offset * EXT4_DESC_SIZE(sb));
296 	if (bh)
297 		*bh = sbi->s_group_desc[group_desc];
298 	return desc;
299 }
300 
301 /*
302  * Return the block number which was discovered to be invalid, or 0 if
303  * the block bitmap is valid.
304  */
305 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
306 					    struct ext4_group_desc *desc,
307 					    ext4_group_t block_group,
308 					    struct buffer_head *bh)
309 {
310 	ext4_grpblk_t offset;
311 	ext4_grpblk_t next_zero_bit;
312 	ext4_fsblk_t blk;
313 	ext4_fsblk_t group_first_block;
314 
315 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
316 		/* with FLEX_BG, the inode/block bitmaps and itable
317 		 * blocks may not be in the group at all
318 		 * so the bitmap validation will be skipped for those groups
319 		 * or it has to also read the block group where the bitmaps
320 		 * are located to verify they are set.
321 		 */
322 		return 0;
323 	}
324 	group_first_block = ext4_group_first_block_no(sb, block_group);
325 
326 	/* check whether block bitmap block number is set */
327 	blk = ext4_block_bitmap(sb, desc);
328 	offset = blk - group_first_block;
329 	if (!ext4_test_bit(offset, bh->b_data))
330 		/* bad block bitmap */
331 		return blk;
332 
333 	/* check whether the inode bitmap block number is set */
334 	blk = ext4_inode_bitmap(sb, desc);
335 	offset = blk - group_first_block;
336 	if (!ext4_test_bit(offset, bh->b_data))
337 		/* bad block bitmap */
338 		return blk;
339 
340 	/* check whether the inode table block number is set */
341 	blk = ext4_inode_table(sb, desc);
342 	offset = blk - group_first_block;
343 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
344 				offset + EXT4_SB(sb)->s_itb_per_group,
345 				offset);
346 	if (next_zero_bit < offset + EXT4_SB(sb)->s_itb_per_group)
347 		/* bad bitmap for inode tables */
348 		return blk;
349 	return 0;
350 }
351 
352 void ext4_validate_block_bitmap(struct super_block *sb,
353 			       struct ext4_group_desc *desc,
354 			       ext4_group_t block_group,
355 			       struct buffer_head *bh)
356 {
357 	ext4_fsblk_t	blk;
358 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
359 
360 	if (buffer_verified(bh))
361 		return;
362 
363 	ext4_lock_group(sb, block_group);
364 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
365 	if (unlikely(blk != 0)) {
366 		ext4_unlock_group(sb, block_group);
367 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
368 			   block_group, blk);
369 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
370 		return;
371 	}
372 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
373 			desc, bh))) {
374 		ext4_unlock_group(sb, block_group);
375 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
376 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
377 		return;
378 	}
379 	set_buffer_verified(bh);
380 	ext4_unlock_group(sb, block_group);
381 }
382 
383 /**
384  * ext4_read_block_bitmap_nowait()
385  * @sb:			super block
386  * @block_group:	given block group
387  *
388  * Read the bitmap for a given block_group,and validate the
389  * bits for block/inode/inode tables are set in the bitmaps
390  *
391  * Return buffer_head on success or NULL in case of failure.
392  */
393 struct buffer_head *
394 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
395 {
396 	struct ext4_group_desc *desc;
397 	struct buffer_head *bh;
398 	ext4_fsblk_t bitmap_blk;
399 
400 	desc = ext4_get_group_desc(sb, block_group, NULL);
401 	if (!desc)
402 		return NULL;
403 	bitmap_blk = ext4_block_bitmap(sb, desc);
404 	bh = sb_getblk(sb, bitmap_blk);
405 	if (unlikely(!bh)) {
406 		ext4_error(sb, "Cannot get buffer for block bitmap - "
407 			   "block_group = %u, block_bitmap = %llu",
408 			   block_group, bitmap_blk);
409 		return NULL;
410 	}
411 
412 	if (bitmap_uptodate(bh))
413 		goto verify;
414 
415 	lock_buffer(bh);
416 	if (bitmap_uptodate(bh)) {
417 		unlock_buffer(bh);
418 		goto verify;
419 	}
420 	ext4_lock_group(sb, block_group);
421 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
422 		ext4_init_block_bitmap(sb, bh, block_group, desc);
423 		set_bitmap_uptodate(bh);
424 		set_buffer_uptodate(bh);
425 		ext4_unlock_group(sb, block_group);
426 		unlock_buffer(bh);
427 		return bh;
428 	}
429 	ext4_unlock_group(sb, block_group);
430 	if (buffer_uptodate(bh)) {
431 		/*
432 		 * if not uninit if bh is uptodate,
433 		 * bitmap is also uptodate
434 		 */
435 		set_bitmap_uptodate(bh);
436 		unlock_buffer(bh);
437 		goto verify;
438 	}
439 	/*
440 	 * submit the buffer_head for reading
441 	 */
442 	set_buffer_new(bh);
443 	trace_ext4_read_block_bitmap_load(sb, block_group);
444 	bh->b_end_io = ext4_end_bitmap_read;
445 	get_bh(bh);
446 	submit_bh(READ | REQ_META | REQ_PRIO, bh);
447 	return bh;
448 verify:
449 	ext4_validate_block_bitmap(sb, desc, block_group, bh);
450 	if (buffer_verified(bh))
451 		return bh;
452 	put_bh(bh);
453 	return NULL;
454 }
455 
456 /* Returns 0 on success, 1 on error */
457 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
458 			   struct buffer_head *bh)
459 {
460 	struct ext4_group_desc *desc;
461 
462 	if (!buffer_new(bh))
463 		return 0;
464 	desc = ext4_get_group_desc(sb, block_group, NULL);
465 	if (!desc)
466 		return 1;
467 	wait_on_buffer(bh);
468 	if (!buffer_uptodate(bh)) {
469 		ext4_error(sb, "Cannot read block bitmap - "
470 			   "block_group = %u, block_bitmap = %llu",
471 			   block_group, (unsigned long long) bh->b_blocknr);
472 		return 1;
473 	}
474 	clear_buffer_new(bh);
475 	/* Panic or remount fs read-only if block bitmap is invalid */
476 	ext4_validate_block_bitmap(sb, desc, block_group, bh);
477 	/* ...but check for error just in case errors=continue. */
478 	return !buffer_verified(bh);
479 }
480 
481 struct buffer_head *
482 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
483 {
484 	struct buffer_head *bh;
485 
486 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
487 	if (!bh)
488 		return NULL;
489 	if (ext4_wait_block_bitmap(sb, block_group, bh)) {
490 		put_bh(bh);
491 		return NULL;
492 	}
493 	return bh;
494 }
495 
496 /**
497  * ext4_has_free_clusters()
498  * @sbi:	in-core super block structure.
499  * @nclusters:	number of needed blocks
500  * @flags:	flags from ext4_mb_new_blocks()
501  *
502  * Check if filesystem has nclusters free & available for allocation.
503  * On success return 1, return 0 on failure.
504  */
505 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
506 				  s64 nclusters, unsigned int flags)
507 {
508 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
509 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
510 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
511 
512 	free_clusters  = percpu_counter_read_positive(fcc);
513 	dirty_clusters = percpu_counter_read_positive(dcc);
514 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
515 
516 	/*
517 	 * r_blocks_count should always be multiple of the cluster ratio so
518 	 * we are safe to do a plane bit shift only.
519 	 */
520 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
521 	      resv_clusters;
522 
523 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
524 					EXT4_FREECLUSTERS_WATERMARK) {
525 		free_clusters  = percpu_counter_sum_positive(fcc);
526 		dirty_clusters = percpu_counter_sum_positive(dcc);
527 	}
528 	/* Check whether we have space after accounting for current
529 	 * dirty clusters & root reserved clusters.
530 	 */
531 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
532 		return 1;
533 
534 	/* Hm, nope.  Are (enough) root reserved clusters available? */
535 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
536 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
537 	    capable(CAP_SYS_RESOURCE) ||
538 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
539 
540 		if (free_clusters >= (nclusters + dirty_clusters +
541 				      resv_clusters))
542 			return 1;
543 	}
544 	/* No free blocks. Let's see if we can dip into reserved pool */
545 	if (flags & EXT4_MB_USE_RESERVED) {
546 		if (free_clusters >= (nclusters + dirty_clusters))
547 			return 1;
548 	}
549 
550 	return 0;
551 }
552 
553 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
554 			     s64 nclusters, unsigned int flags)
555 {
556 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
557 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
558 		return 0;
559 	} else
560 		return -ENOSPC;
561 }
562 
563 /**
564  * ext4_should_retry_alloc()
565  * @sb:			super block
566  * @retries		number of attemps has been made
567  *
568  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
569  * it is profitable to retry the operation, this function will wait
570  * for the current or committing transaction to complete, and then
571  * return TRUE.
572  *
573  * if the total number of retries exceed three times, return FALSE.
574  */
575 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
576 {
577 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
578 	    (*retries)++ > 3 ||
579 	    !EXT4_SB(sb)->s_journal)
580 		return 0;
581 
582 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
583 
584 	return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
585 }
586 
587 /*
588  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
589  *
590  * @handle:             handle to this transaction
591  * @inode:              file inode
592  * @goal:               given target block(filesystem wide)
593  * @count:		pointer to total number of clusters needed
594  * @errp:               error code
595  *
596  * Return 1st allocated block number on success, *count stores total account
597  * error stores in errp pointer
598  */
599 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
600 				  ext4_fsblk_t goal, unsigned int flags,
601 				  unsigned long *count, int *errp)
602 {
603 	struct ext4_allocation_request ar;
604 	ext4_fsblk_t ret;
605 
606 	memset(&ar, 0, sizeof(ar));
607 	/* Fill with neighbour allocated blocks */
608 	ar.inode = inode;
609 	ar.goal = goal;
610 	ar.len = count ? *count : 1;
611 	ar.flags = flags;
612 
613 	ret = ext4_mb_new_blocks(handle, &ar, errp);
614 	if (count)
615 		*count = ar.len;
616 	/*
617 	 * Account for the allocated meta blocks.  We will never
618 	 * fail EDQUOT for metdata, but we do account for it.
619 	 */
620 	if (!(*errp) &&
621 	    ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
622 		spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
623 		EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
624 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
625 		dquot_alloc_block_nofail(inode,
626 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
627 	}
628 	return ret;
629 }
630 
631 /**
632  * ext4_count_free_clusters() -- count filesystem free clusters
633  * @sb:		superblock
634  *
635  * Adds up the number of free clusters from each block group.
636  */
637 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
638 {
639 	ext4_fsblk_t desc_count;
640 	struct ext4_group_desc *gdp;
641 	ext4_group_t i;
642 	ext4_group_t ngroups = ext4_get_groups_count(sb);
643 	struct ext4_group_info *grp;
644 #ifdef EXT4FS_DEBUG
645 	struct ext4_super_block *es;
646 	ext4_fsblk_t bitmap_count;
647 	unsigned int x;
648 	struct buffer_head *bitmap_bh = NULL;
649 
650 	es = EXT4_SB(sb)->s_es;
651 	desc_count = 0;
652 	bitmap_count = 0;
653 	gdp = NULL;
654 
655 	for (i = 0; i < ngroups; i++) {
656 		gdp = ext4_get_group_desc(sb, i, NULL);
657 		if (!gdp)
658 			continue;
659 		grp = NULL;
660 		if (EXT4_SB(sb)->s_group_info)
661 			grp = ext4_get_group_info(sb, i);
662 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
663 			desc_count += ext4_free_group_clusters(sb, gdp);
664 		brelse(bitmap_bh);
665 		bitmap_bh = ext4_read_block_bitmap(sb, i);
666 		if (bitmap_bh == NULL)
667 			continue;
668 
669 		x = ext4_count_free(bitmap_bh->b_data,
670 				    EXT4_BLOCKS_PER_GROUP(sb) / 8);
671 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
672 			i, ext4_free_group_clusters(sb, gdp), x);
673 		bitmap_count += x;
674 	}
675 	brelse(bitmap_bh);
676 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
677 	       ", computed = %llu, %llu\n",
678 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
679 	       desc_count, bitmap_count);
680 	return bitmap_count;
681 #else
682 	desc_count = 0;
683 	for (i = 0; i < ngroups; i++) {
684 		gdp = ext4_get_group_desc(sb, i, NULL);
685 		if (!gdp)
686 			continue;
687 		grp = NULL;
688 		if (EXT4_SB(sb)->s_group_info)
689 			grp = ext4_get_group_info(sb, i);
690 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
691 			desc_count += ext4_free_group_clusters(sb, gdp);
692 	}
693 
694 	return desc_count;
695 #endif
696 }
697 
698 static inline int test_root(ext4_group_t a, int b)
699 {
700 	while (1) {
701 		if (a < b)
702 			return 0;
703 		if (a == b)
704 			return 1;
705 		if ((a % b) != 0)
706 			return 0;
707 		a = a / b;
708 	}
709 }
710 
711 static int ext4_group_sparse(ext4_group_t group)
712 {
713 	if (group <= 1)
714 		return 1;
715 	if (!(group & 1))
716 		return 0;
717 	return (test_root(group, 7) || test_root(group, 5) ||
718 		test_root(group, 3));
719 }
720 
721 /**
722  *	ext4_bg_has_super - number of blocks used by the superblock in group
723  *	@sb: superblock for filesystem
724  *	@group: group number to check
725  *
726  *	Return the number of blocks used by the superblock (primary or backup)
727  *	in this group.  Currently this will be only 0 or 1.
728  */
729 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
730 {
731 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
732 				EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
733 			!ext4_group_sparse(group))
734 		return 0;
735 	return 1;
736 }
737 
738 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
739 					ext4_group_t group)
740 {
741 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
742 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
743 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
744 
745 	if (group == first || group == first + 1 || group == last)
746 		return 1;
747 	return 0;
748 }
749 
750 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
751 					ext4_group_t group)
752 {
753 	if (!ext4_bg_has_super(sb, group))
754 		return 0;
755 
756 	if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
757 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
758 	else
759 		return EXT4_SB(sb)->s_gdb_count;
760 }
761 
762 /**
763  *	ext4_bg_num_gdb - number of blocks used by the group table in group
764  *	@sb: superblock for filesystem
765  *	@group: group number to check
766  *
767  *	Return the number of blocks used by the group descriptor table
768  *	(primary or backup) in this group.  In the future there may be a
769  *	different number of descriptor blocks in each group.
770  */
771 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
772 {
773 	unsigned long first_meta_bg =
774 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
775 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
776 
777 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
778 			metagroup < first_meta_bg)
779 		return ext4_bg_num_gdb_nometa(sb, group);
780 
781 	return ext4_bg_num_gdb_meta(sb,group);
782 
783 }
784 
785 /*
786  * This function returns the number of file system metadata clusters at
787  * the beginning of a block group, including the reserved gdt blocks.
788  */
789 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
790 				     ext4_group_t block_group)
791 {
792 	struct ext4_sb_info *sbi = EXT4_SB(sb);
793 	unsigned num;
794 
795 	/* Check for superblock and gdt backups in this group */
796 	num = ext4_bg_has_super(sb, block_group);
797 
798 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
799 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
800 			  sbi->s_desc_per_block) {
801 		if (num) {
802 			num += ext4_bg_num_gdb(sb, block_group);
803 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
804 		}
805 	} else { /* For META_BG_BLOCK_GROUPS */
806 		num += ext4_bg_num_gdb(sb, block_group);
807 	}
808 	return EXT4_NUM_B2C(sbi, num);
809 }
810 /**
811  *	ext4_inode_to_goal_block - return a hint for block allocation
812  *	@inode: inode for block allocation
813  *
814  *	Return the ideal location to start allocating blocks for a
815  *	newly created inode.
816  */
817 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
818 {
819 	struct ext4_inode_info *ei = EXT4_I(inode);
820 	ext4_group_t block_group;
821 	ext4_grpblk_t colour;
822 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
823 	ext4_fsblk_t bg_start;
824 	ext4_fsblk_t last_block;
825 
826 	block_group = ei->i_block_group;
827 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
828 		/*
829 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
830 		 * block groups per flexgroup, reserve the first block
831 		 * group for directories and special files.  Regular
832 		 * files will start at the second block group.  This
833 		 * tends to speed up directory access and improves
834 		 * fsck times.
835 		 */
836 		block_group &= ~(flex_size-1);
837 		if (S_ISREG(inode->i_mode))
838 			block_group++;
839 	}
840 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
841 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
842 
843 	/*
844 	 * If we are doing delayed allocation, we don't need take
845 	 * colour into account.
846 	 */
847 	if (test_opt(inode->i_sb, DELALLOC))
848 		return bg_start;
849 
850 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
851 		colour = (current->pid % 16) *
852 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
853 	else
854 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
855 	return bg_start + colour;
856 }
857 
858