xref: /openbmc/linux/fs/ext4/balloc.c (revision 80483c3a)
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/quotaops.h>
18 #include <linux/buffer_head.h>
19 #include "ext4.h"
20 #include "ext4_jbd2.h"
21 #include "mballoc.h"
22 
23 #include <trace/events/ext4.h>
24 
25 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
26 					    ext4_group_t block_group);
27 /*
28  * balloc.c contains the blocks allocation and deallocation routines
29  */
30 
31 /*
32  * Calculate block group number for a given block number
33  */
34 ext4_group_t ext4_get_group_number(struct super_block *sb,
35 				   ext4_fsblk_t block)
36 {
37 	ext4_group_t group;
38 
39 	if (test_opt2(sb, STD_GROUP_SIZE))
40 		group = (block -
41 			 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >>
42 			(EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3);
43 	else
44 		ext4_get_group_no_and_offset(sb, block, &group, NULL);
45 	return group;
46 }
47 
48 /*
49  * Calculate the block group number and offset into the block/cluster
50  * allocation bitmap, given a block number
51  */
52 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
53 		ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
54 {
55 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
56 	ext4_grpblk_t offset;
57 
58 	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
59 	offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
60 		EXT4_SB(sb)->s_cluster_bits;
61 	if (offsetp)
62 		*offsetp = offset;
63 	if (blockgrpp)
64 		*blockgrpp = blocknr;
65 
66 }
67 
68 /*
69  * Check whether the 'block' lives within the 'block_group'. Returns 1 if so
70  * and 0 otherwise.
71  */
72 static inline int ext4_block_in_group(struct super_block *sb,
73 				      ext4_fsblk_t block,
74 				      ext4_group_t block_group)
75 {
76 	ext4_group_t actual_group;
77 
78 	actual_group = ext4_get_group_number(sb, block);
79 	return (actual_group == block_group) ? 1 : 0;
80 }
81 
82 /* Return the number of clusters used for file system metadata; this
83  * represents the overhead needed by the file system.
84  */
85 static unsigned ext4_num_overhead_clusters(struct super_block *sb,
86 					   ext4_group_t block_group,
87 					   struct ext4_group_desc *gdp)
88 {
89 	unsigned num_clusters;
90 	int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
91 	ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
92 	ext4_fsblk_t itbl_blk;
93 	struct ext4_sb_info *sbi = EXT4_SB(sb);
94 
95 	/* This is the number of clusters used by the superblock,
96 	 * block group descriptors, and reserved block group
97 	 * descriptor blocks */
98 	num_clusters = ext4_num_base_meta_clusters(sb, block_group);
99 
100 	/*
101 	 * For the allocation bitmaps and inode table, we first need
102 	 * to check to see if the block is in the block group.  If it
103 	 * is, then check to see if the cluster is already accounted
104 	 * for in the clusters used for the base metadata cluster, or
105 	 * if we can increment the base metadata cluster to include
106 	 * that block.  Otherwise, we will have to track the cluster
107 	 * used for the allocation bitmap or inode table explicitly.
108 	 * Normally all of these blocks are contiguous, so the special
109 	 * case handling shouldn't be necessary except for *very*
110 	 * unusual file system layouts.
111 	 */
112 	if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
113 		block_cluster = EXT4_B2C(sbi,
114 					 ext4_block_bitmap(sb, gdp) - start);
115 		if (block_cluster < num_clusters)
116 			block_cluster = -1;
117 		else if (block_cluster == num_clusters) {
118 			num_clusters++;
119 			block_cluster = -1;
120 		}
121 	}
122 
123 	if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
124 		inode_cluster = EXT4_B2C(sbi,
125 					 ext4_inode_bitmap(sb, gdp) - start);
126 		if (inode_cluster < num_clusters)
127 			inode_cluster = -1;
128 		else if (inode_cluster == num_clusters) {
129 			num_clusters++;
130 			inode_cluster = -1;
131 		}
132 	}
133 
134 	itbl_blk = ext4_inode_table(sb, gdp);
135 	for (i = 0; i < sbi->s_itb_per_group; i++) {
136 		if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
137 			c = EXT4_B2C(sbi, itbl_blk + i - start);
138 			if ((c < num_clusters) || (c == inode_cluster) ||
139 			    (c == block_cluster) || (c == itbl_cluster))
140 				continue;
141 			if (c == num_clusters) {
142 				num_clusters++;
143 				continue;
144 			}
145 			num_clusters++;
146 			itbl_cluster = c;
147 		}
148 	}
149 
150 	if (block_cluster != -1)
151 		num_clusters++;
152 	if (inode_cluster != -1)
153 		num_clusters++;
154 
155 	return num_clusters;
156 }
157 
158 static unsigned int num_clusters_in_group(struct super_block *sb,
159 					  ext4_group_t block_group)
160 {
161 	unsigned int blocks;
162 
163 	if (block_group == ext4_get_groups_count(sb) - 1) {
164 		/*
165 		 * Even though mke2fs always initializes the first and
166 		 * last group, just in case some other tool was used,
167 		 * we need to make sure we calculate the right free
168 		 * blocks.
169 		 */
170 		blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
171 			ext4_group_first_block_no(sb, block_group);
172 	} else
173 		blocks = EXT4_BLOCKS_PER_GROUP(sb);
174 	return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
175 }
176 
177 /* Initializes an uninitialized block bitmap */
178 static int ext4_init_block_bitmap(struct super_block *sb,
179 				   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 		grp = ext4_get_group_info(sb, block_group);
195 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
196 			percpu_counter_sub(&sbi->s_freeclusters_counter,
197 					   grp->bb_free);
198 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
199 		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
200 			int count;
201 			count = ext4_free_inodes_count(sb, gdp);
202 			percpu_counter_sub(&sbi->s_freeinodes_counter,
203 					   count);
204 		}
205 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
206 		return -EFSBADCRC;
207 	}
208 	memset(bh->b_data, 0, sb->s_blocksize);
209 
210 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
211 	if ((bit_max >> 3) >= bh->b_size)
212 		return -EFSCORRUPTED;
213 
214 	for (bit = 0; bit < bit_max; bit++)
215 		ext4_set_bit(bit, bh->b_data);
216 
217 	start = ext4_group_first_block_no(sb, block_group);
218 
219 	if (ext4_has_feature_flex_bg(sb))
220 		flex_bg = 1;
221 
222 	/* Set bits for block and inode bitmaps, and inode table */
223 	tmp = ext4_block_bitmap(sb, gdp);
224 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
225 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
226 
227 	tmp = ext4_inode_bitmap(sb, gdp);
228 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
229 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
230 
231 	tmp = ext4_inode_table(sb, gdp);
232 	for (; tmp < ext4_inode_table(sb, gdp) +
233 		     sbi->s_itb_per_group; tmp++) {
234 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
235 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
236 	}
237 
238 	/*
239 	 * Also if the number of blocks within the group is less than
240 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
241 	 * of the block bitmap to 1
242 	 */
243 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
244 			     sb->s_blocksize * 8, bh->b_data);
245 	ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
246 	ext4_group_desc_csum_set(sb, block_group, gdp);
247 	return 0;
248 }
249 
250 /* Return the number of free blocks in a block group.  It is used when
251  * the block bitmap is uninitialized, so we can't just count the bits
252  * in the bitmap. */
253 unsigned ext4_free_clusters_after_init(struct super_block *sb,
254 				       ext4_group_t block_group,
255 				       struct ext4_group_desc *gdp)
256 {
257 	return num_clusters_in_group(sb, block_group) -
258 		ext4_num_overhead_clusters(sb, block_group, gdp);
259 }
260 
261 /*
262  * The free blocks are managed by bitmaps.  A file system contains several
263  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
264  * block for inodes, N blocks for the inode table and data blocks.
265  *
266  * The file system contains group descriptors which are located after the
267  * super block.  Each descriptor contains the number of the bitmap block and
268  * the free blocks count in the block.  The descriptors are loaded in memory
269  * when a file system is mounted (see ext4_fill_super).
270  */
271 
272 /**
273  * ext4_get_group_desc() -- load group descriptor from disk
274  * @sb:			super block
275  * @block_group:	given block group
276  * @bh:			pointer to the buffer head to store the block
277  *			group descriptor
278  */
279 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
280 					     ext4_group_t block_group,
281 					     struct buffer_head **bh)
282 {
283 	unsigned int group_desc;
284 	unsigned int offset;
285 	ext4_group_t ngroups = ext4_get_groups_count(sb);
286 	struct ext4_group_desc *desc;
287 	struct ext4_sb_info *sbi = EXT4_SB(sb);
288 
289 	if (block_group >= ngroups) {
290 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
291 			   " groups_count = %u", block_group, ngroups);
292 
293 		return NULL;
294 	}
295 
296 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
297 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
298 	if (!sbi->s_group_desc[group_desc]) {
299 		ext4_error(sb, "Group descriptor not loaded - "
300 			   "block_group = %u, group_desc = %u, desc = %u",
301 			   block_group, group_desc, offset);
302 		return NULL;
303 	}
304 
305 	desc = (struct ext4_group_desc *)(
306 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
307 		offset * EXT4_DESC_SIZE(sb));
308 	if (bh)
309 		*bh = sbi->s_group_desc[group_desc];
310 	return desc;
311 }
312 
313 /*
314  * Return the block number which was discovered to be invalid, or 0 if
315  * the block bitmap is valid.
316  */
317 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
318 					    struct ext4_group_desc *desc,
319 					    ext4_group_t block_group,
320 					    struct buffer_head *bh)
321 {
322 	struct ext4_sb_info *sbi = EXT4_SB(sb);
323 	ext4_grpblk_t offset;
324 	ext4_grpblk_t next_zero_bit;
325 	ext4_fsblk_t blk;
326 	ext4_fsblk_t group_first_block;
327 
328 	if (ext4_has_feature_flex_bg(sb)) {
329 		/* with FLEX_BG, the inode/block bitmaps and itable
330 		 * blocks may not be in the group at all
331 		 * so the bitmap validation will be skipped for those groups
332 		 * or it has to also read the block group where the bitmaps
333 		 * are located to verify they are set.
334 		 */
335 		return 0;
336 	}
337 	group_first_block = ext4_group_first_block_no(sb, block_group);
338 
339 	/* check whether block bitmap block number is set */
340 	blk = ext4_block_bitmap(sb, desc);
341 	offset = blk - group_first_block;
342 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
343 		/* bad block bitmap */
344 		return blk;
345 
346 	/* check whether the inode bitmap block number is set */
347 	blk = ext4_inode_bitmap(sb, desc);
348 	offset = blk - group_first_block;
349 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
350 		/* bad block bitmap */
351 		return blk;
352 
353 	/* check whether the inode table block number is set */
354 	blk = ext4_inode_table(sb, desc);
355 	offset = blk - group_first_block;
356 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
357 			EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group),
358 			EXT4_B2C(sbi, offset));
359 	if (next_zero_bit <
360 	    EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group))
361 		/* bad bitmap for inode tables */
362 		return blk;
363 	return 0;
364 }
365 
366 static int ext4_validate_block_bitmap(struct super_block *sb,
367 				      struct ext4_group_desc *desc,
368 				      ext4_group_t block_group,
369 				      struct buffer_head *bh)
370 {
371 	ext4_fsblk_t	blk;
372 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
373 	struct ext4_sb_info *sbi = EXT4_SB(sb);
374 
375 	if (buffer_verified(bh))
376 		return 0;
377 	if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
378 		return -EFSCORRUPTED;
379 
380 	ext4_lock_group(sb, block_group);
381 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
382 			desc, bh))) {
383 		ext4_unlock_group(sb, block_group);
384 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
385 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
386 			percpu_counter_sub(&sbi->s_freeclusters_counter,
387 					   grp->bb_free);
388 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
389 		return -EFSBADCRC;
390 	}
391 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
392 	if (unlikely(blk != 0)) {
393 		ext4_unlock_group(sb, block_group);
394 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
395 			   block_group, blk);
396 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
397 			percpu_counter_sub(&sbi->s_freeclusters_counter,
398 					   grp->bb_free);
399 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
400 		return -EFSCORRUPTED;
401 	}
402 	set_buffer_verified(bh);
403 	ext4_unlock_group(sb, block_group);
404 	return 0;
405 }
406 
407 /**
408  * ext4_read_block_bitmap_nowait()
409  * @sb:			super block
410  * @block_group:	given block group
411  *
412  * Read the bitmap for a given block_group,and validate the
413  * bits for block/inode/inode tables are set in the bitmaps
414  *
415  * Return buffer_head on success or NULL in case of failure.
416  */
417 struct buffer_head *
418 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
419 {
420 	struct ext4_group_desc *desc;
421 	struct buffer_head *bh;
422 	ext4_fsblk_t bitmap_blk;
423 	int err;
424 
425 	desc = ext4_get_group_desc(sb, block_group, NULL);
426 	if (!desc)
427 		return ERR_PTR(-EFSCORRUPTED);
428 	bitmap_blk = ext4_block_bitmap(sb, desc);
429 	bh = sb_getblk(sb, bitmap_blk);
430 	if (unlikely(!bh)) {
431 		ext4_error(sb, "Cannot get buffer for block bitmap - "
432 			   "block_group = %u, block_bitmap = %llu",
433 			   block_group, bitmap_blk);
434 		return ERR_PTR(-ENOMEM);
435 	}
436 
437 	if (bitmap_uptodate(bh))
438 		goto verify;
439 
440 	lock_buffer(bh);
441 	if (bitmap_uptodate(bh)) {
442 		unlock_buffer(bh);
443 		goto verify;
444 	}
445 	ext4_lock_group(sb, block_group);
446 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
447 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
448 		set_bitmap_uptodate(bh);
449 		set_buffer_uptodate(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 		return -EIO;
505 	}
506 	clear_buffer_new(bh);
507 	/* Panic or remount fs read-only if block bitmap is invalid */
508 	return ext4_validate_block_bitmap(sb, desc, block_group, bh);
509 }
510 
511 struct buffer_head *
512 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
513 {
514 	struct buffer_head *bh;
515 	int err;
516 
517 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
518 	if (IS_ERR(bh))
519 		return bh;
520 	err = ext4_wait_block_bitmap(sb, block_group, bh);
521 	if (err) {
522 		put_bh(bh);
523 		return ERR_PTR(err);
524 	}
525 	return bh;
526 }
527 
528 /**
529  * ext4_has_free_clusters()
530  * @sbi:	in-core super block structure.
531  * @nclusters:	number of needed blocks
532  * @flags:	flags from ext4_mb_new_blocks()
533  *
534  * Check if filesystem has nclusters free & available for allocation.
535  * On success return 1, return 0 on failure.
536  */
537 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
538 				  s64 nclusters, unsigned int flags)
539 {
540 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
541 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
542 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
543 
544 	free_clusters  = percpu_counter_read_positive(fcc);
545 	dirty_clusters = percpu_counter_read_positive(dcc);
546 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
547 
548 	/*
549 	 * r_blocks_count should always be multiple of the cluster ratio so
550 	 * we are safe to do a plane bit shift only.
551 	 */
552 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
553 	      resv_clusters;
554 
555 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
556 					EXT4_FREECLUSTERS_WATERMARK) {
557 		free_clusters  = percpu_counter_sum_positive(fcc);
558 		dirty_clusters = percpu_counter_sum_positive(dcc);
559 	}
560 	/* Check whether we have space after accounting for current
561 	 * dirty clusters & root reserved clusters.
562 	 */
563 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
564 		return 1;
565 
566 	/* Hm, nope.  Are (enough) root reserved clusters available? */
567 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
568 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
569 	    capable(CAP_SYS_RESOURCE) ||
570 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
571 
572 		if (free_clusters >= (nclusters + dirty_clusters +
573 				      resv_clusters))
574 			return 1;
575 	}
576 	/* No free blocks. Let's see if we can dip into reserved pool */
577 	if (flags & EXT4_MB_USE_RESERVED) {
578 		if (free_clusters >= (nclusters + dirty_clusters))
579 			return 1;
580 	}
581 
582 	return 0;
583 }
584 
585 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
586 			     s64 nclusters, unsigned int flags)
587 {
588 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
589 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
590 		return 0;
591 	} else
592 		return -ENOSPC;
593 }
594 
595 /**
596  * ext4_should_retry_alloc()
597  * @sb:			super block
598  * @retries		number of attemps has been made
599  *
600  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
601  * it is profitable to retry the operation, this function will wait
602  * for the current or committing transaction to complete, and then
603  * return TRUE.
604  *
605  * if the total number of retries exceed three times, return FALSE.
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)++ > 3 ||
611 	    !EXT4_SB(sb)->s_journal)
612 		return 0;
613 
614 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
615 
616 	smp_mb();
617 	if (EXT4_SB(sb)->s_mb_free_pending)
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