xref: /openbmc/linux/fs/ext4/balloc.c (revision 4a3fad70)
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 	struct ext4_group_info *grp;
189 
190 	J_ASSERT_BH(bh, buffer_locked(bh));
191 
192 	/* If checksum is bad mark all blocks used to prevent allocation
193 	 * essentially implementing a per-group read-only flag. */
194 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
195 		grp = ext4_get_group_info(sb, block_group);
196 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
197 			percpu_counter_sub(&sbi->s_freeclusters_counter,
198 					   grp->bb_free);
199 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
200 		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
201 			int count;
202 			count = ext4_free_inodes_count(sb, gdp);
203 			percpu_counter_sub(&sbi->s_freeinodes_counter,
204 					   count);
205 		}
206 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
207 		return -EFSBADCRC;
208 	}
209 	memset(bh->b_data, 0, sb->s_blocksize);
210 
211 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
212 	if ((bit_max >> 3) >= bh->b_size)
213 		return -EFSCORRUPTED;
214 
215 	for (bit = 0; bit < bit_max; bit++)
216 		ext4_set_bit(bit, bh->b_data);
217 
218 	start = ext4_group_first_block_no(sb, block_group);
219 
220 	if (ext4_has_feature_flex_bg(sb))
221 		flex_bg = 1;
222 
223 	/* Set bits for block and inode bitmaps, and inode table */
224 	tmp = ext4_block_bitmap(sb, gdp);
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 	tmp = ext4_inode_bitmap(sb, gdp);
229 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
230 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
231 
232 	tmp = ext4_inode_table(sb, gdp);
233 	for (; tmp < ext4_inode_table(sb, gdp) +
234 		     sbi->s_itb_per_group; tmp++) {
235 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
236 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
237 	}
238 
239 	/*
240 	 * Also if the number of blocks within the group is less than
241 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
242 	 * of the block bitmap to 1
243 	 */
244 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
245 			     sb->s_blocksize * 8, bh->b_data);
246 	ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
247 	ext4_group_desc_csum_set(sb, block_group, gdp);
248 	return 0;
249 }
250 
251 /* Return the number of free blocks in a block group.  It is used when
252  * the block bitmap is uninitialized, so we can't just count the bits
253  * in the bitmap. */
254 unsigned ext4_free_clusters_after_init(struct super_block *sb,
255 				       ext4_group_t block_group,
256 				       struct ext4_group_desc *gdp)
257 {
258 	return num_clusters_in_group(sb, block_group) -
259 		ext4_num_overhead_clusters(sb, block_group, gdp);
260 }
261 
262 /*
263  * The free blocks are managed by bitmaps.  A file system contains several
264  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
265  * block for inodes, N blocks for the inode table and data blocks.
266  *
267  * The file system contains group descriptors which are located after the
268  * super block.  Each descriptor contains the number of the bitmap block and
269  * the free blocks count in the block.  The descriptors are loaded in memory
270  * when a file system is mounted (see ext4_fill_super).
271  */
272 
273 /**
274  * ext4_get_group_desc() -- load group descriptor from disk
275  * @sb:			super block
276  * @block_group:	given block group
277  * @bh:			pointer to the buffer head to store the block
278  *			group descriptor
279  */
280 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
281 					     ext4_group_t block_group,
282 					     struct buffer_head **bh)
283 {
284 	unsigned int group_desc;
285 	unsigned int offset;
286 	ext4_group_t ngroups = ext4_get_groups_count(sb);
287 	struct ext4_group_desc *desc;
288 	struct ext4_sb_info *sbi = EXT4_SB(sb);
289 
290 	if (block_group >= ngroups) {
291 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
292 			   " groups_count = %u", block_group, ngroups);
293 
294 		return NULL;
295 	}
296 
297 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
298 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
299 	if (!sbi->s_group_desc[group_desc]) {
300 		ext4_error(sb, "Group descriptor not loaded - "
301 			   "block_group = %u, group_desc = %u, desc = %u",
302 			   block_group, group_desc, offset);
303 		return NULL;
304 	}
305 
306 	desc = (struct ext4_group_desc *)(
307 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
308 		offset * EXT4_DESC_SIZE(sb));
309 	if (bh)
310 		*bh = sbi->s_group_desc[group_desc];
311 	return desc;
312 }
313 
314 /*
315  * Return the block number which was discovered to be invalid, or 0 if
316  * the block bitmap is valid.
317  */
318 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
319 					    struct ext4_group_desc *desc,
320 					    ext4_group_t block_group,
321 					    struct buffer_head *bh)
322 {
323 	struct ext4_sb_info *sbi = EXT4_SB(sb);
324 	ext4_grpblk_t offset;
325 	ext4_grpblk_t next_zero_bit;
326 	ext4_fsblk_t blk;
327 	ext4_fsblk_t group_first_block;
328 
329 	if (ext4_has_feature_flex_bg(sb)) {
330 		/* with FLEX_BG, the inode/block bitmaps and itable
331 		 * blocks may not be in the group at all
332 		 * so the bitmap validation will be skipped for those groups
333 		 * or it has to also read the block group where the bitmaps
334 		 * are located to verify they are set.
335 		 */
336 		return 0;
337 	}
338 	group_first_block = ext4_group_first_block_no(sb, block_group);
339 
340 	/* check whether block bitmap block number is set */
341 	blk = ext4_block_bitmap(sb, desc);
342 	offset = blk - group_first_block;
343 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
344 		/* bad block bitmap */
345 		return blk;
346 
347 	/* check whether the inode bitmap block number is set */
348 	blk = ext4_inode_bitmap(sb, desc);
349 	offset = blk - group_first_block;
350 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
351 		/* bad block bitmap */
352 		return blk;
353 
354 	/* check whether the inode table block number is set */
355 	blk = ext4_inode_table(sb, desc);
356 	offset = blk - group_first_block;
357 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
358 			EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group),
359 			EXT4_B2C(sbi, offset));
360 	if (next_zero_bit <
361 	    EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group))
362 		/* bad bitmap for inode tables */
363 		return blk;
364 	return 0;
365 }
366 
367 static int ext4_validate_block_bitmap(struct super_block *sb,
368 				      struct ext4_group_desc *desc,
369 				      ext4_group_t block_group,
370 				      struct buffer_head *bh)
371 {
372 	ext4_fsblk_t	blk;
373 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
374 	struct ext4_sb_info *sbi = EXT4_SB(sb);
375 
376 	if (buffer_verified(bh))
377 		return 0;
378 	if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
379 		return -EFSCORRUPTED;
380 
381 	ext4_lock_group(sb, block_group);
382 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
383 			desc, bh))) {
384 		ext4_unlock_group(sb, block_group);
385 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
386 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
387 			percpu_counter_sub(&sbi->s_freeclusters_counter,
388 					   grp->bb_free);
389 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
390 		return -EFSBADCRC;
391 	}
392 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
393 	if (unlikely(blk != 0)) {
394 		ext4_unlock_group(sb, block_group);
395 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
396 			   block_group, blk);
397 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
398 			percpu_counter_sub(&sbi->s_freeclusters_counter,
399 					   grp->bb_free);
400 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
401 		return -EFSCORRUPTED;
402 	}
403 	set_buffer_verified(bh);
404 	ext4_unlock_group(sb, block_group);
405 	return 0;
406 }
407 
408 /**
409  * ext4_read_block_bitmap_nowait()
410  * @sb:			super block
411  * @block_group:	given block group
412  *
413  * Read the bitmap for a given block_group,and validate the
414  * bits for block/inode/inode tables are set in the bitmaps
415  *
416  * Return buffer_head on success or NULL in case of failure.
417  */
418 struct buffer_head *
419 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
420 {
421 	struct ext4_group_desc *desc;
422 	struct buffer_head *bh;
423 	ext4_fsblk_t bitmap_blk;
424 	int err;
425 
426 	desc = ext4_get_group_desc(sb, block_group, NULL);
427 	if (!desc)
428 		return ERR_PTR(-EFSCORRUPTED);
429 	bitmap_blk = ext4_block_bitmap(sb, desc);
430 	bh = sb_getblk(sb, bitmap_blk);
431 	if (unlikely(!bh)) {
432 		ext4_error(sb, "Cannot get buffer for block bitmap - "
433 			   "block_group = %u, block_bitmap = %llu",
434 			   block_group, bitmap_blk);
435 		return ERR_PTR(-ENOMEM);
436 	}
437 
438 	if (bitmap_uptodate(bh))
439 		goto verify;
440 
441 	lock_buffer(bh);
442 	if (bitmap_uptodate(bh)) {
443 		unlock_buffer(bh);
444 		goto verify;
445 	}
446 	ext4_lock_group(sb, block_group);
447 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
448 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
449 		set_bitmap_uptodate(bh);
450 		set_buffer_uptodate(bh);
451 		ext4_unlock_group(sb, block_group);
452 		unlock_buffer(bh);
453 		if (err) {
454 			ext4_error(sb, "Failed to init block bitmap for group "
455 				   "%u: %d", block_group, err);
456 			goto out;
457 		}
458 		goto verify;
459 	}
460 	ext4_unlock_group(sb, block_group);
461 	if (buffer_uptodate(bh)) {
462 		/*
463 		 * if not uninit if bh is uptodate,
464 		 * bitmap is also uptodate
465 		 */
466 		set_bitmap_uptodate(bh);
467 		unlock_buffer(bh);
468 		goto verify;
469 	}
470 	/*
471 	 * submit the buffer_head for reading
472 	 */
473 	set_buffer_new(bh);
474 	trace_ext4_read_block_bitmap_load(sb, block_group);
475 	bh->b_end_io = ext4_end_bitmap_read;
476 	get_bh(bh);
477 	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
478 	return bh;
479 verify:
480 	err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
481 	if (err)
482 		goto out;
483 	return bh;
484 out:
485 	put_bh(bh);
486 	return ERR_PTR(err);
487 }
488 
489 /* Returns 0 on success, 1 on error */
490 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
491 			   struct buffer_head *bh)
492 {
493 	struct ext4_group_desc *desc;
494 
495 	if (!buffer_new(bh))
496 		return 0;
497 	desc = ext4_get_group_desc(sb, block_group, NULL);
498 	if (!desc)
499 		return -EFSCORRUPTED;
500 	wait_on_buffer(bh);
501 	if (!buffer_uptodate(bh)) {
502 		ext4_error(sb, "Cannot read block bitmap - "
503 			   "block_group = %u, block_bitmap = %llu",
504 			   block_group, (unsigned long long) bh->b_blocknr);
505 		return -EIO;
506 	}
507 	clear_buffer_new(bh);
508 	/* Panic or remount fs read-only if block bitmap is invalid */
509 	return ext4_validate_block_bitmap(sb, desc, block_group, bh);
510 }
511 
512 struct buffer_head *
513 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
514 {
515 	struct buffer_head *bh;
516 	int err;
517 
518 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
519 	if (IS_ERR(bh))
520 		return bh;
521 	err = ext4_wait_block_bitmap(sb, block_group, bh);
522 	if (err) {
523 		put_bh(bh);
524 		return ERR_PTR(err);
525 	}
526 	return bh;
527 }
528 
529 /**
530  * ext4_has_free_clusters()
531  * @sbi:	in-core super block structure.
532  * @nclusters:	number of needed blocks
533  * @flags:	flags from ext4_mb_new_blocks()
534  *
535  * Check if filesystem has nclusters free & available for allocation.
536  * On success return 1, return 0 on failure.
537  */
538 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
539 				  s64 nclusters, unsigned int flags)
540 {
541 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
542 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
543 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
544 
545 	free_clusters  = percpu_counter_read_positive(fcc);
546 	dirty_clusters = percpu_counter_read_positive(dcc);
547 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
548 
549 	/*
550 	 * r_blocks_count should always be multiple of the cluster ratio so
551 	 * we are safe to do a plane bit shift only.
552 	 */
553 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
554 	      resv_clusters;
555 
556 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
557 					EXT4_FREECLUSTERS_WATERMARK) {
558 		free_clusters  = percpu_counter_sum_positive(fcc);
559 		dirty_clusters = percpu_counter_sum_positive(dcc);
560 	}
561 	/* Check whether we have space after accounting for current
562 	 * dirty clusters & root reserved clusters.
563 	 */
564 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
565 		return 1;
566 
567 	/* Hm, nope.  Are (enough) root reserved clusters available? */
568 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
569 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
570 	    capable(CAP_SYS_RESOURCE) ||
571 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
572 
573 		if (free_clusters >= (nclusters + dirty_clusters +
574 				      resv_clusters))
575 			return 1;
576 	}
577 	/* No free blocks. Let's see if we can dip into reserved pool */
578 	if (flags & EXT4_MB_USE_RESERVED) {
579 		if (free_clusters >= (nclusters + dirty_clusters))
580 			return 1;
581 	}
582 
583 	return 0;
584 }
585 
586 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
587 			     s64 nclusters, unsigned int flags)
588 {
589 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
590 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
591 		return 0;
592 	} else
593 		return -ENOSPC;
594 }
595 
596 /**
597  * ext4_should_retry_alloc()
598  * @sb:			super block
599  * @retries		number of attemps has been made
600  *
601  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
602  * it is profitable to retry the operation, this function will wait
603  * for the current or committing transaction to complete, and then
604  * return TRUE.  We will only retry once.
605  */
606 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
607 {
608 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
609 	    (*retries)++ > 1 ||
610 	    !EXT4_SB(sb)->s_journal)
611 		return 0;
612 
613 	smp_mb();
614 	if (EXT4_SB(sb)->s_mb_free_pending == 0)
615 		return 0;
616 
617 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
618 	jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
619 	return 1;
620 }
621 
622 /*
623  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
624  *
625  * @handle:             handle to this transaction
626  * @inode:              file inode
627  * @goal:               given target block(filesystem wide)
628  * @count:		pointer to total number of clusters needed
629  * @errp:               error code
630  *
631  * Return 1st allocated block number on success, *count stores total account
632  * error stores in errp pointer
633  */
634 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
635 				  ext4_fsblk_t goal, unsigned int flags,
636 				  unsigned long *count, int *errp)
637 {
638 	struct ext4_allocation_request ar;
639 	ext4_fsblk_t ret;
640 
641 	memset(&ar, 0, sizeof(ar));
642 	/* Fill with neighbour allocated blocks */
643 	ar.inode = inode;
644 	ar.goal = goal;
645 	ar.len = count ? *count : 1;
646 	ar.flags = flags;
647 
648 	ret = ext4_mb_new_blocks(handle, &ar, errp);
649 	if (count)
650 		*count = ar.len;
651 	/*
652 	 * Account for the allocated meta blocks.  We will never
653 	 * fail EDQUOT for metdata, but we do account for it.
654 	 */
655 	if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
656 		dquot_alloc_block_nofail(inode,
657 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
658 	}
659 	return ret;
660 }
661 
662 /**
663  * ext4_count_free_clusters() -- count filesystem free clusters
664  * @sb:		superblock
665  *
666  * Adds up the number of free clusters from each block group.
667  */
668 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
669 {
670 	ext4_fsblk_t desc_count;
671 	struct ext4_group_desc *gdp;
672 	ext4_group_t i;
673 	ext4_group_t ngroups = ext4_get_groups_count(sb);
674 	struct ext4_group_info *grp;
675 #ifdef EXT4FS_DEBUG
676 	struct ext4_super_block *es;
677 	ext4_fsblk_t bitmap_count;
678 	unsigned int x;
679 	struct buffer_head *bitmap_bh = NULL;
680 
681 	es = EXT4_SB(sb)->s_es;
682 	desc_count = 0;
683 	bitmap_count = 0;
684 	gdp = NULL;
685 
686 	for (i = 0; i < ngroups; i++) {
687 		gdp = ext4_get_group_desc(sb, i, NULL);
688 		if (!gdp)
689 			continue;
690 		grp = NULL;
691 		if (EXT4_SB(sb)->s_group_info)
692 			grp = ext4_get_group_info(sb, i);
693 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
694 			desc_count += ext4_free_group_clusters(sb, gdp);
695 		brelse(bitmap_bh);
696 		bitmap_bh = ext4_read_block_bitmap(sb, i);
697 		if (IS_ERR(bitmap_bh)) {
698 			bitmap_bh = NULL;
699 			continue;
700 		}
701 
702 		x = ext4_count_free(bitmap_bh->b_data,
703 				    EXT4_CLUSTERS_PER_GROUP(sb) / 8);
704 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
705 			i, ext4_free_group_clusters(sb, gdp), x);
706 		bitmap_count += x;
707 	}
708 	brelse(bitmap_bh);
709 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
710 	       ", computed = %llu, %llu\n",
711 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
712 	       desc_count, bitmap_count);
713 	return bitmap_count;
714 #else
715 	desc_count = 0;
716 	for (i = 0; i < ngroups; i++) {
717 		gdp = ext4_get_group_desc(sb, i, NULL);
718 		if (!gdp)
719 			continue;
720 		grp = NULL;
721 		if (EXT4_SB(sb)->s_group_info)
722 			grp = ext4_get_group_info(sb, i);
723 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
724 			desc_count += ext4_free_group_clusters(sb, gdp);
725 	}
726 
727 	return desc_count;
728 #endif
729 }
730 
731 static inline int test_root(ext4_group_t a, int b)
732 {
733 	while (1) {
734 		if (a < b)
735 			return 0;
736 		if (a == b)
737 			return 1;
738 		if ((a % b) != 0)
739 			return 0;
740 		a = a / b;
741 	}
742 }
743 
744 /**
745  *	ext4_bg_has_super - number of blocks used by the superblock in group
746  *	@sb: superblock for filesystem
747  *	@group: group number to check
748  *
749  *	Return the number of blocks used by the superblock (primary or backup)
750  *	in this group.  Currently this will be only 0 or 1.
751  */
752 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
753 {
754 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
755 
756 	if (group == 0)
757 		return 1;
758 	if (ext4_has_feature_sparse_super2(sb)) {
759 		if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
760 		    group == le32_to_cpu(es->s_backup_bgs[1]))
761 			return 1;
762 		return 0;
763 	}
764 	if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
765 		return 1;
766 	if (!(group & 1))
767 		return 0;
768 	if (test_root(group, 3) || (test_root(group, 5)) ||
769 	    test_root(group, 7))
770 		return 1;
771 
772 	return 0;
773 }
774 
775 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
776 					ext4_group_t group)
777 {
778 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
779 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
780 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
781 
782 	if (group == first || group == first + 1 || group == last)
783 		return 1;
784 	return 0;
785 }
786 
787 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
788 					ext4_group_t group)
789 {
790 	if (!ext4_bg_has_super(sb, group))
791 		return 0;
792 
793 	if (ext4_has_feature_meta_bg(sb))
794 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
795 	else
796 		return EXT4_SB(sb)->s_gdb_count;
797 }
798 
799 /**
800  *	ext4_bg_num_gdb - number of blocks used by the group table in group
801  *	@sb: superblock for filesystem
802  *	@group: group number to check
803  *
804  *	Return the number of blocks used by the group descriptor table
805  *	(primary or backup) in this group.  In the future there may be a
806  *	different number of descriptor blocks in each group.
807  */
808 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
809 {
810 	unsigned long first_meta_bg =
811 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
812 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
813 
814 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
815 		return ext4_bg_num_gdb_nometa(sb, group);
816 
817 	return ext4_bg_num_gdb_meta(sb,group);
818 
819 }
820 
821 /*
822  * This function returns the number of file system metadata clusters at
823  * the beginning of a block group, including the reserved gdt blocks.
824  */
825 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
826 				     ext4_group_t block_group)
827 {
828 	struct ext4_sb_info *sbi = EXT4_SB(sb);
829 	unsigned num;
830 
831 	/* Check for superblock and gdt backups in this group */
832 	num = ext4_bg_has_super(sb, block_group);
833 
834 	if (!ext4_has_feature_meta_bg(sb) ||
835 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
836 			  sbi->s_desc_per_block) {
837 		if (num) {
838 			num += ext4_bg_num_gdb(sb, block_group);
839 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
840 		}
841 	} else { /* For META_BG_BLOCK_GROUPS */
842 		num += ext4_bg_num_gdb(sb, block_group);
843 	}
844 	return EXT4_NUM_B2C(sbi, num);
845 }
846 /**
847  *	ext4_inode_to_goal_block - return a hint for block allocation
848  *	@inode: inode for block allocation
849  *
850  *	Return the ideal location to start allocating blocks for a
851  *	newly created inode.
852  */
853 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
854 {
855 	struct ext4_inode_info *ei = EXT4_I(inode);
856 	ext4_group_t block_group;
857 	ext4_grpblk_t colour;
858 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
859 	ext4_fsblk_t bg_start;
860 	ext4_fsblk_t last_block;
861 
862 	block_group = ei->i_block_group;
863 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
864 		/*
865 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
866 		 * block groups per flexgroup, reserve the first block
867 		 * group for directories and special files.  Regular
868 		 * files will start at the second block group.  This
869 		 * tends to speed up directory access and improves
870 		 * fsck times.
871 		 */
872 		block_group &= ~(flex_size-1);
873 		if (S_ISREG(inode->i_mode))
874 			block_group++;
875 	}
876 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
877 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
878 
879 	/*
880 	 * If we are doing delayed allocation, we don't need take
881 	 * colour into account.
882 	 */
883 	if (test_opt(inode->i_sb, DELALLOC))
884 		return bg_start;
885 
886 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
887 		colour = (current->pid % 16) *
888 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
889 	else
890 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
891 	return bg_start + colour;
892 }
893 
894