xref: /openbmc/linux/fs/ext4/balloc.c (revision 7e6f7d24)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/balloc.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
11  *  Big-endian to little-endian byte-swapping/bitmaps by
12  *        David S. Miller (davem@caip.rutgers.edu), 1995
13  */
14 
15 #include <linux/time.h>
16 #include <linux/capability.h>
17 #include <linux/fs.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 static 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 static int ext4_init_block_bitmap(struct super_block *sb,
180 				   struct buffer_head *bh,
181 				   ext4_group_t block_group,
182 				   struct ext4_group_desc *gdp)
183 {
184 	unsigned int bit, bit_max;
185 	struct ext4_sb_info *sbi = EXT4_SB(sb);
186 	ext4_fsblk_t start, tmp;
187 	int flex_bg = 0;
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_mark_group_bitmap_corrupted(sb, block_group,
195 					EXT4_GROUP_INFO_BBITMAP_CORRUPT |
196 					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
197 		return -EFSBADCRC;
198 	}
199 	memset(bh->b_data, 0, sb->s_blocksize);
200 
201 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
202 	if ((bit_max >> 3) >= bh->b_size)
203 		return -EFSCORRUPTED;
204 
205 	for (bit = 0; bit < bit_max; bit++)
206 		ext4_set_bit(bit, bh->b_data);
207 
208 	start = ext4_group_first_block_no(sb, block_group);
209 
210 	if (ext4_has_feature_flex_bg(sb))
211 		flex_bg = 1;
212 
213 	/* Set bits for block and inode bitmaps, and inode table */
214 	tmp = ext4_block_bitmap(sb, gdp);
215 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
216 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
217 
218 	tmp = ext4_inode_bitmap(sb, gdp);
219 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
220 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
221 
222 	tmp = ext4_inode_table(sb, gdp);
223 	for (; tmp < ext4_inode_table(sb, gdp) +
224 		     sbi->s_itb_per_group; tmp++) {
225 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
226 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
227 	}
228 
229 	/*
230 	 * Also if the number of blocks within the group is less than
231 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
232 	 * of the block bitmap to 1
233 	 */
234 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
235 			     sb->s_blocksize * 8, bh->b_data);
236 	return 0;
237 }
238 
239 /* Return the number of free blocks in a block group.  It is used when
240  * the block bitmap is uninitialized, so we can't just count the bits
241  * in the bitmap. */
242 unsigned ext4_free_clusters_after_init(struct super_block *sb,
243 				       ext4_group_t block_group,
244 				       struct ext4_group_desc *gdp)
245 {
246 	return num_clusters_in_group(sb, block_group) -
247 		ext4_num_overhead_clusters(sb, block_group, gdp);
248 }
249 
250 /*
251  * The free blocks are managed by bitmaps.  A file system contains several
252  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
253  * block for inodes, N blocks for the inode table and data blocks.
254  *
255  * The file system contains group descriptors which are located after the
256  * super block.  Each descriptor contains the number of the bitmap block and
257  * the free blocks count in the block.  The descriptors are loaded in memory
258  * when a file system is mounted (see ext4_fill_super).
259  */
260 
261 /**
262  * ext4_get_group_desc() -- load group descriptor from disk
263  * @sb:			super block
264  * @block_group:	given block group
265  * @bh:			pointer to the buffer head to store the block
266  *			group descriptor
267  */
268 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
269 					     ext4_group_t block_group,
270 					     struct buffer_head **bh)
271 {
272 	unsigned int group_desc;
273 	unsigned int offset;
274 	ext4_group_t ngroups = ext4_get_groups_count(sb);
275 	struct ext4_group_desc *desc;
276 	struct ext4_sb_info *sbi = EXT4_SB(sb);
277 
278 	if (block_group >= ngroups) {
279 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
280 			   " groups_count = %u", block_group, ngroups);
281 
282 		return NULL;
283 	}
284 
285 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
286 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
287 	if (!sbi->s_group_desc[group_desc]) {
288 		ext4_error(sb, "Group descriptor not loaded - "
289 			   "block_group = %u, group_desc = %u, desc = %u",
290 			   block_group, group_desc, offset);
291 		return NULL;
292 	}
293 
294 	desc = (struct ext4_group_desc *)(
295 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
296 		offset * EXT4_DESC_SIZE(sb));
297 	if (bh)
298 		*bh = sbi->s_group_desc[group_desc];
299 	return desc;
300 }
301 
302 /*
303  * Return the block number which was discovered to be invalid, or 0 if
304  * the block bitmap is valid.
305  */
306 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
307 					    struct ext4_group_desc *desc,
308 					    ext4_group_t block_group,
309 					    struct buffer_head *bh)
310 {
311 	struct ext4_sb_info *sbi = EXT4_SB(sb);
312 	ext4_grpblk_t offset;
313 	ext4_grpblk_t next_zero_bit;
314 	ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
315 	ext4_fsblk_t blk;
316 	ext4_fsblk_t group_first_block;
317 
318 	if (ext4_has_feature_flex_bg(sb)) {
319 		/* with FLEX_BG, the inode/block bitmaps and itable
320 		 * blocks may not be in the group at all
321 		 * so the bitmap validation will be skipped for those groups
322 		 * or it has to also read the block group where the bitmaps
323 		 * are located to verify they are set.
324 		 */
325 		return 0;
326 	}
327 	group_first_block = ext4_group_first_block_no(sb, block_group);
328 
329 	/* check whether block bitmap block number is set */
330 	blk = ext4_block_bitmap(sb, desc);
331 	offset = blk - group_first_block;
332 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
333 	    !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
334 		/* bad block bitmap */
335 		return blk;
336 
337 	/* check whether the inode bitmap block number is set */
338 	blk = ext4_inode_bitmap(sb, desc);
339 	offset = blk - group_first_block;
340 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
341 	    !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
342 		/* bad block bitmap */
343 		return blk;
344 
345 	/* check whether the inode table block number is set */
346 	blk = ext4_inode_table(sb, desc);
347 	offset = blk - group_first_block;
348 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
349 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
350 		return blk;
351 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
352 			EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
353 			EXT4_B2C(sbi, offset));
354 	if (next_zero_bit <
355 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
356 		/* bad bitmap for inode tables */
357 		return blk;
358 	return 0;
359 }
360 
361 static int ext4_validate_block_bitmap(struct super_block *sb,
362 				      struct ext4_group_desc *desc,
363 				      ext4_group_t block_group,
364 				      struct buffer_head *bh)
365 {
366 	ext4_fsblk_t	blk;
367 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
368 
369 	if (buffer_verified(bh))
370 		return 0;
371 	if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
372 		return -EFSCORRUPTED;
373 
374 	ext4_lock_group(sb, block_group);
375 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
376 			desc, bh))) {
377 		ext4_unlock_group(sb, block_group);
378 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
379 		ext4_mark_group_bitmap_corrupted(sb, block_group,
380 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
381 		return -EFSBADCRC;
382 	}
383 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
384 	if (unlikely(blk != 0)) {
385 		ext4_unlock_group(sb, block_group);
386 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
387 			   block_group, blk);
388 		ext4_mark_group_bitmap_corrupted(sb, block_group,
389 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
390 		return -EFSCORRUPTED;
391 	}
392 	set_buffer_verified(bh);
393 	ext4_unlock_group(sb, block_group);
394 	return 0;
395 }
396 
397 /**
398  * ext4_read_block_bitmap_nowait()
399  * @sb:			super block
400  * @block_group:	given block group
401  *
402  * Read the bitmap for a given block_group,and validate the
403  * bits for block/inode/inode tables are set in the bitmaps
404  *
405  * Return buffer_head on success or NULL in case of failure.
406  */
407 struct buffer_head *
408 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
409 {
410 	struct ext4_group_desc *desc;
411 	struct ext4_sb_info *sbi = EXT4_SB(sb);
412 	struct buffer_head *bh;
413 	ext4_fsblk_t bitmap_blk;
414 	int err;
415 
416 	desc = ext4_get_group_desc(sb, block_group, NULL);
417 	if (!desc)
418 		return ERR_PTR(-EFSCORRUPTED);
419 	bitmap_blk = ext4_block_bitmap(sb, desc);
420 	if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
421 	    (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
422 		ext4_error(sb, "Invalid block bitmap block %llu in "
423 			   "block_group %u", bitmap_blk, block_group);
424 		ext4_mark_group_bitmap_corrupted(sb, block_group,
425 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
426 		return ERR_PTR(-EFSCORRUPTED);
427 	}
428 	bh = sb_getblk(sb, bitmap_blk);
429 	if (unlikely(!bh)) {
430 		ext4_error(sb, "Cannot get buffer for block bitmap - "
431 			   "block_group = %u, block_bitmap = %llu",
432 			   block_group, bitmap_blk);
433 		return ERR_PTR(-ENOMEM);
434 	}
435 
436 	if (bitmap_uptodate(bh))
437 		goto verify;
438 
439 	lock_buffer(bh);
440 	if (bitmap_uptodate(bh)) {
441 		unlock_buffer(bh);
442 		goto verify;
443 	}
444 	ext4_lock_group(sb, block_group);
445 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
446 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
447 		set_bitmap_uptodate(bh);
448 		set_buffer_uptodate(bh);
449 		set_buffer_verified(bh);
450 		ext4_unlock_group(sb, block_group);
451 		unlock_buffer(bh);
452 		if (err) {
453 			ext4_error(sb, "Failed to init block bitmap for group "
454 				   "%u: %d", block_group, err);
455 			goto out;
456 		}
457 		goto verify;
458 	}
459 	ext4_unlock_group(sb, block_group);
460 	if (buffer_uptodate(bh)) {
461 		/*
462 		 * if not uninit if bh is uptodate,
463 		 * bitmap is also uptodate
464 		 */
465 		set_bitmap_uptodate(bh);
466 		unlock_buffer(bh);
467 		goto verify;
468 	}
469 	/*
470 	 * submit the buffer_head for reading
471 	 */
472 	set_buffer_new(bh);
473 	trace_ext4_read_block_bitmap_load(sb, block_group);
474 	bh->b_end_io = ext4_end_bitmap_read;
475 	get_bh(bh);
476 	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
477 	return bh;
478 verify:
479 	err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
480 	if (err)
481 		goto out;
482 	return bh;
483 out:
484 	put_bh(bh);
485 	return ERR_PTR(err);
486 }
487 
488 /* Returns 0 on success, 1 on error */
489 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
490 			   struct buffer_head *bh)
491 {
492 	struct ext4_group_desc *desc;
493 
494 	if (!buffer_new(bh))
495 		return 0;
496 	desc = ext4_get_group_desc(sb, block_group, NULL);
497 	if (!desc)
498 		return -EFSCORRUPTED;
499 	wait_on_buffer(bh);
500 	if (!buffer_uptodate(bh)) {
501 		ext4_error(sb, "Cannot read block bitmap - "
502 			   "block_group = %u, block_bitmap = %llu",
503 			   block_group, (unsigned long long) bh->b_blocknr);
504 		ext4_mark_group_bitmap_corrupted(sb, block_group,
505 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
506 		return -EIO;
507 	}
508 	clear_buffer_new(bh);
509 	/* Panic or remount fs read-only if block bitmap is invalid */
510 	return ext4_validate_block_bitmap(sb, desc, block_group, bh);
511 }
512 
513 struct buffer_head *
514 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
515 {
516 	struct buffer_head *bh;
517 	int err;
518 
519 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
520 	if (IS_ERR(bh))
521 		return bh;
522 	err = ext4_wait_block_bitmap(sb, block_group, bh);
523 	if (err) {
524 		put_bh(bh);
525 		return ERR_PTR(err);
526 	}
527 	return bh;
528 }
529 
530 /**
531  * ext4_has_free_clusters()
532  * @sbi:	in-core super block structure.
533  * @nclusters:	number of needed blocks
534  * @flags:	flags from ext4_mb_new_blocks()
535  *
536  * Check if filesystem has nclusters free & available for allocation.
537  * On success return 1, return 0 on failure.
538  */
539 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
540 				  s64 nclusters, unsigned int flags)
541 {
542 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
543 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
544 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
545 
546 	free_clusters  = percpu_counter_read_positive(fcc);
547 	dirty_clusters = percpu_counter_read_positive(dcc);
548 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
549 
550 	/*
551 	 * r_blocks_count should always be multiple of the cluster ratio so
552 	 * we are safe to do a plane bit shift only.
553 	 */
554 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
555 	      resv_clusters;
556 
557 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
558 					EXT4_FREECLUSTERS_WATERMARK) {
559 		free_clusters  = percpu_counter_sum_positive(fcc);
560 		dirty_clusters = percpu_counter_sum_positive(dcc);
561 	}
562 	/* Check whether we have space after accounting for current
563 	 * dirty clusters & root reserved clusters.
564 	 */
565 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
566 		return 1;
567 
568 	/* Hm, nope.  Are (enough) root reserved clusters available? */
569 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
570 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
571 	    capable(CAP_SYS_RESOURCE) ||
572 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
573 
574 		if (free_clusters >= (nclusters + dirty_clusters +
575 				      resv_clusters))
576 			return 1;
577 	}
578 	/* No free blocks. Let's see if we can dip into reserved pool */
579 	if (flags & EXT4_MB_USE_RESERVED) {
580 		if (free_clusters >= (nclusters + dirty_clusters))
581 			return 1;
582 	}
583 
584 	return 0;
585 }
586 
587 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
588 			     s64 nclusters, unsigned int flags)
589 {
590 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
591 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
592 		return 0;
593 	} else
594 		return -ENOSPC;
595 }
596 
597 /**
598  * ext4_should_retry_alloc()
599  * @sb:			super block
600  * @retries		number of attemps has been made
601  *
602  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
603  * it is profitable to retry the operation, this function will wait
604  * for the current or committing transaction to complete, and then
605  * return TRUE.  We will only retry once.
606  */
607 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
608 {
609 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
610 	    (*retries)++ > 1 ||
611 	    !EXT4_SB(sb)->s_journal)
612 		return 0;
613 
614 	smp_mb();
615 	if (EXT4_SB(sb)->s_mb_free_pending == 0)
616 		return 0;
617 
618 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
619 	jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
620 	return 1;
621 }
622 
623 /*
624  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
625  *
626  * @handle:             handle to this transaction
627  * @inode:              file inode
628  * @goal:               given target block(filesystem wide)
629  * @count:		pointer to total number of clusters needed
630  * @errp:               error code
631  *
632  * Return 1st allocated block number on success, *count stores total account
633  * error stores in errp pointer
634  */
635 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
636 				  ext4_fsblk_t goal, unsigned int flags,
637 				  unsigned long *count, int *errp)
638 {
639 	struct ext4_allocation_request ar;
640 	ext4_fsblk_t ret;
641 
642 	memset(&ar, 0, sizeof(ar));
643 	/* Fill with neighbour allocated blocks */
644 	ar.inode = inode;
645 	ar.goal = goal;
646 	ar.len = count ? *count : 1;
647 	ar.flags = flags;
648 
649 	ret = ext4_mb_new_blocks(handle, &ar, errp);
650 	if (count)
651 		*count = ar.len;
652 	/*
653 	 * Account for the allocated meta blocks.  We will never
654 	 * fail EDQUOT for metdata, but we do account for it.
655 	 */
656 	if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
657 		dquot_alloc_block_nofail(inode,
658 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
659 	}
660 	return ret;
661 }
662 
663 /**
664  * ext4_count_free_clusters() -- count filesystem free clusters
665  * @sb:		superblock
666  *
667  * Adds up the number of free clusters from each block group.
668  */
669 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
670 {
671 	ext4_fsblk_t desc_count;
672 	struct ext4_group_desc *gdp;
673 	ext4_group_t i;
674 	ext4_group_t ngroups = ext4_get_groups_count(sb);
675 	struct ext4_group_info *grp;
676 #ifdef EXT4FS_DEBUG
677 	struct ext4_super_block *es;
678 	ext4_fsblk_t bitmap_count;
679 	unsigned int x;
680 	struct buffer_head *bitmap_bh = NULL;
681 
682 	es = EXT4_SB(sb)->s_es;
683 	desc_count = 0;
684 	bitmap_count = 0;
685 	gdp = NULL;
686 
687 	for (i = 0; i < ngroups; i++) {
688 		gdp = ext4_get_group_desc(sb, i, NULL);
689 		if (!gdp)
690 			continue;
691 		grp = NULL;
692 		if (EXT4_SB(sb)->s_group_info)
693 			grp = ext4_get_group_info(sb, i);
694 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
695 			desc_count += ext4_free_group_clusters(sb, gdp);
696 		brelse(bitmap_bh);
697 		bitmap_bh = ext4_read_block_bitmap(sb, i);
698 		if (IS_ERR(bitmap_bh)) {
699 			bitmap_bh = NULL;
700 			continue;
701 		}
702 
703 		x = ext4_count_free(bitmap_bh->b_data,
704 				    EXT4_CLUSTERS_PER_GROUP(sb) / 8);
705 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
706 			i, ext4_free_group_clusters(sb, gdp), x);
707 		bitmap_count += x;
708 	}
709 	brelse(bitmap_bh);
710 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
711 	       ", computed = %llu, %llu\n",
712 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
713 	       desc_count, bitmap_count);
714 	return bitmap_count;
715 #else
716 	desc_count = 0;
717 	for (i = 0; i < ngroups; i++) {
718 		gdp = ext4_get_group_desc(sb, i, NULL);
719 		if (!gdp)
720 			continue;
721 		grp = NULL;
722 		if (EXT4_SB(sb)->s_group_info)
723 			grp = ext4_get_group_info(sb, i);
724 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
725 			desc_count += ext4_free_group_clusters(sb, gdp);
726 	}
727 
728 	return desc_count;
729 #endif
730 }
731 
732 static inline int test_root(ext4_group_t a, int b)
733 {
734 	while (1) {
735 		if (a < b)
736 			return 0;
737 		if (a == b)
738 			return 1;
739 		if ((a % b) != 0)
740 			return 0;
741 		a = a / b;
742 	}
743 }
744 
745 /**
746  *	ext4_bg_has_super - number of blocks used by the superblock in group
747  *	@sb: superblock for filesystem
748  *	@group: group number to check
749  *
750  *	Return the number of blocks used by the superblock (primary or backup)
751  *	in this group.  Currently this will be only 0 or 1.
752  */
753 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
754 {
755 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
756 
757 	if (group == 0)
758 		return 1;
759 	if (ext4_has_feature_sparse_super2(sb)) {
760 		if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
761 		    group == le32_to_cpu(es->s_backup_bgs[1]))
762 			return 1;
763 		return 0;
764 	}
765 	if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
766 		return 1;
767 	if (!(group & 1))
768 		return 0;
769 	if (test_root(group, 3) || (test_root(group, 5)) ||
770 	    test_root(group, 7))
771 		return 1;
772 
773 	return 0;
774 }
775 
776 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
777 					ext4_group_t group)
778 {
779 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
780 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
781 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
782 
783 	if (group == first || group == first + 1 || group == last)
784 		return 1;
785 	return 0;
786 }
787 
788 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
789 					ext4_group_t group)
790 {
791 	if (!ext4_bg_has_super(sb, group))
792 		return 0;
793 
794 	if (ext4_has_feature_meta_bg(sb))
795 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
796 	else
797 		return EXT4_SB(sb)->s_gdb_count;
798 }
799 
800 /**
801  *	ext4_bg_num_gdb - number of blocks used by the group table in group
802  *	@sb: superblock for filesystem
803  *	@group: group number to check
804  *
805  *	Return the number of blocks used by the group descriptor table
806  *	(primary or backup) in this group.  In the future there may be a
807  *	different number of descriptor blocks in each group.
808  */
809 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
810 {
811 	unsigned long first_meta_bg =
812 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
813 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
814 
815 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
816 		return ext4_bg_num_gdb_nometa(sb, group);
817 
818 	return ext4_bg_num_gdb_meta(sb,group);
819 
820 }
821 
822 /*
823  * This function returns the number of file system metadata clusters at
824  * the beginning of a block group, including the reserved gdt blocks.
825  */
826 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
827 				     ext4_group_t block_group)
828 {
829 	struct ext4_sb_info *sbi = EXT4_SB(sb);
830 	unsigned num;
831 
832 	/* Check for superblock and gdt backups in this group */
833 	num = ext4_bg_has_super(sb, block_group);
834 
835 	if (!ext4_has_feature_meta_bg(sb) ||
836 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
837 			  sbi->s_desc_per_block) {
838 		if (num) {
839 			num += ext4_bg_num_gdb(sb, block_group);
840 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
841 		}
842 	} else { /* For META_BG_BLOCK_GROUPS */
843 		num += ext4_bg_num_gdb(sb, block_group);
844 	}
845 	return EXT4_NUM_B2C(sbi, num);
846 }
847 /**
848  *	ext4_inode_to_goal_block - return a hint for block allocation
849  *	@inode: inode for block allocation
850  *
851  *	Return the ideal location to start allocating blocks for a
852  *	newly created inode.
853  */
854 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
855 {
856 	struct ext4_inode_info *ei = EXT4_I(inode);
857 	ext4_group_t block_group;
858 	ext4_grpblk_t colour;
859 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
860 	ext4_fsblk_t bg_start;
861 	ext4_fsblk_t last_block;
862 
863 	block_group = ei->i_block_group;
864 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
865 		/*
866 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
867 		 * block groups per flexgroup, reserve the first block
868 		 * group for directories and special files.  Regular
869 		 * files will start at the second block group.  This
870 		 * tends to speed up directory access and improves
871 		 * fsck times.
872 		 */
873 		block_group &= ~(flex_size-1);
874 		if (S_ISREG(inode->i_mode))
875 			block_group++;
876 	}
877 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
878 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
879 
880 	/*
881 	 * If we are doing delayed allocation, we don't need take
882 	 * colour into account.
883 	 */
884 	if (test_opt(inode->i_sb, DELALLOC))
885 		return bg_start;
886 
887 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
888 		colour = (current->pid % 16) *
889 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
890 	else
891 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
892 	return bg_start + colour;
893 }
894 
895