xref: /openbmc/linux/fs/ext4/balloc.c (revision 65ee8aeb)
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 -EIO;
207 	}
208 	memset(bh->b_data, 0, sb->s_blocksize);
209 
210 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
211 	for (bit = 0; bit < bit_max; bit++)
212 		ext4_set_bit(bit, bh->b_data);
213 
214 	start = ext4_group_first_block_no(sb, block_group);
215 
216 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
217 		flex_bg = 1;
218 
219 	/* Set bits for block and inode bitmaps, and inode table */
220 	tmp = ext4_block_bitmap(sb, gdp);
221 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
222 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
223 
224 	tmp = ext4_inode_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_table(sb, gdp);
229 	for (; tmp < ext4_inode_table(sb, gdp) +
230 		     sbi->s_itb_per_group; tmp++) {
231 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
232 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
233 	}
234 
235 	/*
236 	 * Also if the number of blocks within the group is less than
237 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
238 	 * of the block bitmap to 1
239 	 */
240 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
241 			     sb->s_blocksize * 8, bh->b_data);
242 	ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
243 	ext4_group_desc_csum_set(sb, block_group, gdp);
244 	return 0;
245 }
246 
247 /* Return the number of free blocks in a block group.  It is used when
248  * the block bitmap is uninitialized, so we can't just count the bits
249  * in the bitmap. */
250 unsigned ext4_free_clusters_after_init(struct super_block *sb,
251 				       ext4_group_t block_group,
252 				       struct ext4_group_desc *gdp)
253 {
254 	return num_clusters_in_group(sb, block_group) -
255 		ext4_num_overhead_clusters(sb, block_group, gdp);
256 }
257 
258 /*
259  * The free blocks are managed by bitmaps.  A file system contains several
260  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
261  * block for inodes, N blocks for the inode table and data blocks.
262  *
263  * The file system contains group descriptors which are located after the
264  * super block.  Each descriptor contains the number of the bitmap block and
265  * the free blocks count in the block.  The descriptors are loaded in memory
266  * when a file system is mounted (see ext4_fill_super).
267  */
268 
269 /**
270  * ext4_get_group_desc() -- load group descriptor from disk
271  * @sb:			super block
272  * @block_group:	given block group
273  * @bh:			pointer to the buffer head to store the block
274  *			group descriptor
275  */
276 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
277 					     ext4_group_t block_group,
278 					     struct buffer_head **bh)
279 {
280 	unsigned int group_desc;
281 	unsigned int offset;
282 	ext4_group_t ngroups = ext4_get_groups_count(sb);
283 	struct ext4_group_desc *desc;
284 	struct ext4_sb_info *sbi = EXT4_SB(sb);
285 
286 	if (block_group >= ngroups) {
287 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
288 			   " groups_count = %u", block_group, ngroups);
289 
290 		return NULL;
291 	}
292 
293 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
294 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
295 	if (!sbi->s_group_desc[group_desc]) {
296 		ext4_error(sb, "Group descriptor not loaded - "
297 			   "block_group = %u, group_desc = %u, desc = %u",
298 			   block_group, group_desc, offset);
299 		return NULL;
300 	}
301 
302 	desc = (struct ext4_group_desc *)(
303 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
304 		offset * EXT4_DESC_SIZE(sb));
305 	if (bh)
306 		*bh = sbi->s_group_desc[group_desc];
307 	return desc;
308 }
309 
310 /*
311  * Return the block number which was discovered to be invalid, or 0 if
312  * the block bitmap is valid.
313  */
314 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
315 					    struct ext4_group_desc *desc,
316 					    ext4_group_t block_group,
317 					    struct buffer_head *bh)
318 {
319 	struct ext4_sb_info *sbi = EXT4_SB(sb);
320 	ext4_grpblk_t offset;
321 	ext4_grpblk_t next_zero_bit;
322 	ext4_fsblk_t blk;
323 	ext4_fsblk_t group_first_block;
324 
325 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
326 		/* with FLEX_BG, the inode/block bitmaps and itable
327 		 * blocks may not be in the group at all
328 		 * so the bitmap validation will be skipped for those groups
329 		 * or it has to also read the block group where the bitmaps
330 		 * are located to verify they are set.
331 		 */
332 		return 0;
333 	}
334 	group_first_block = ext4_group_first_block_no(sb, block_group);
335 
336 	/* check whether block bitmap block number is set */
337 	blk = ext4_block_bitmap(sb, desc);
338 	offset = blk - group_first_block;
339 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
340 		/* bad block bitmap */
341 		return blk;
342 
343 	/* check whether the inode bitmap block number is set */
344 	blk = ext4_inode_bitmap(sb, desc);
345 	offset = blk - group_first_block;
346 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
347 		/* bad block bitmap */
348 		return blk;
349 
350 	/* check whether the inode table block number is set */
351 	blk = ext4_inode_table(sb, desc);
352 	offset = blk - group_first_block;
353 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
354 			EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group),
355 			EXT4_B2C(sbi, offset));
356 	if (next_zero_bit <
357 	    EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group))
358 		/* bad bitmap for inode tables */
359 		return blk;
360 	return 0;
361 }
362 
363 static void ext4_validate_block_bitmap(struct super_block *sb,
364 				       struct ext4_group_desc *desc,
365 				       ext4_group_t block_group,
366 				       struct buffer_head *bh)
367 {
368 	ext4_fsblk_t	blk;
369 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
370 	struct ext4_sb_info *sbi = EXT4_SB(sb);
371 
372 	if (buffer_verified(bh))
373 		return;
374 
375 	ext4_lock_group(sb, block_group);
376 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
377 	if (unlikely(blk != 0)) {
378 		ext4_unlock_group(sb, block_group);
379 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
380 			   block_group, blk);
381 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
382 			percpu_counter_sub(&sbi->s_freeclusters_counter,
383 					   grp->bb_free);
384 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
385 		return;
386 	}
387 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
388 			desc, bh))) {
389 		ext4_unlock_group(sb, block_group);
390 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
391 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
392 			percpu_counter_sub(&sbi->s_freeclusters_counter,
393 					   grp->bb_free);
394 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
395 		return;
396 	}
397 	set_buffer_verified(bh);
398 	ext4_unlock_group(sb, block_group);
399 }
400 
401 /**
402  * ext4_read_block_bitmap_nowait()
403  * @sb:			super block
404  * @block_group:	given block group
405  *
406  * Read the bitmap for a given block_group,and validate the
407  * bits for block/inode/inode tables are set in the bitmaps
408  *
409  * Return buffer_head on success or NULL in case of failure.
410  */
411 struct buffer_head *
412 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
413 {
414 	struct ext4_group_desc *desc;
415 	struct buffer_head *bh;
416 	ext4_fsblk_t bitmap_blk;
417 
418 	desc = ext4_get_group_desc(sb, block_group, NULL);
419 	if (!desc)
420 		return NULL;
421 	bitmap_blk = ext4_block_bitmap(sb, desc);
422 	bh = sb_getblk(sb, bitmap_blk);
423 	if (unlikely(!bh)) {
424 		ext4_error(sb, "Cannot get buffer for block bitmap - "
425 			   "block_group = %u, block_bitmap = %llu",
426 			   block_group, bitmap_blk);
427 		return NULL;
428 	}
429 
430 	if (bitmap_uptodate(bh))
431 		goto verify;
432 
433 	lock_buffer(bh);
434 	if (bitmap_uptodate(bh)) {
435 		unlock_buffer(bh);
436 		goto verify;
437 	}
438 	ext4_lock_group(sb, block_group);
439 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
440 		int err;
441 
442 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
443 		set_bitmap_uptodate(bh);
444 		set_buffer_uptodate(bh);
445 		ext4_unlock_group(sb, block_group);
446 		unlock_buffer(bh);
447 		if (err)
448 			ext4_error(sb, "Checksum bad for grp %u", block_group);
449 		return bh;
450 	}
451 	ext4_unlock_group(sb, block_group);
452 	if (buffer_uptodate(bh)) {
453 		/*
454 		 * if not uninit if bh is uptodate,
455 		 * bitmap is also uptodate
456 		 */
457 		set_bitmap_uptodate(bh);
458 		unlock_buffer(bh);
459 		goto verify;
460 	}
461 	/*
462 	 * submit the buffer_head for reading
463 	 */
464 	set_buffer_new(bh);
465 	trace_ext4_read_block_bitmap_load(sb, block_group);
466 	bh->b_end_io = ext4_end_bitmap_read;
467 	get_bh(bh);
468 	submit_bh(READ | REQ_META | REQ_PRIO, bh);
469 	return bh;
470 verify:
471 	ext4_validate_block_bitmap(sb, desc, block_group, bh);
472 	if (buffer_verified(bh))
473 		return bh;
474 	put_bh(bh);
475 	return NULL;
476 }
477 
478 /* Returns 0 on success, 1 on error */
479 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
480 			   struct buffer_head *bh)
481 {
482 	struct ext4_group_desc *desc;
483 
484 	if (!buffer_new(bh))
485 		return 0;
486 	desc = ext4_get_group_desc(sb, block_group, NULL);
487 	if (!desc)
488 		return 1;
489 	wait_on_buffer(bh);
490 	if (!buffer_uptodate(bh)) {
491 		ext4_error(sb, "Cannot read block bitmap - "
492 			   "block_group = %u, block_bitmap = %llu",
493 			   block_group, (unsigned long long) bh->b_blocknr);
494 		return 1;
495 	}
496 	clear_buffer_new(bh);
497 	/* Panic or remount fs read-only if block bitmap is invalid */
498 	ext4_validate_block_bitmap(sb, desc, block_group, bh);
499 	/* ...but check for error just in case errors=continue. */
500 	return !buffer_verified(bh);
501 }
502 
503 struct buffer_head *
504 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
505 {
506 	struct buffer_head *bh;
507 
508 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
509 	if (!bh)
510 		return NULL;
511 	if (ext4_wait_block_bitmap(sb, block_group, bh)) {
512 		put_bh(bh);
513 		return NULL;
514 	}
515 	return bh;
516 }
517 
518 /**
519  * ext4_has_free_clusters()
520  * @sbi:	in-core super block structure.
521  * @nclusters:	number of needed blocks
522  * @flags:	flags from ext4_mb_new_blocks()
523  *
524  * Check if filesystem has nclusters free & available for allocation.
525  * On success return 1, return 0 on failure.
526  */
527 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
528 				  s64 nclusters, unsigned int flags)
529 {
530 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
531 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
532 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
533 
534 	free_clusters  = percpu_counter_read_positive(fcc);
535 	dirty_clusters = percpu_counter_read_positive(dcc);
536 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
537 
538 	/*
539 	 * r_blocks_count should always be multiple of the cluster ratio so
540 	 * we are safe to do a plane bit shift only.
541 	 */
542 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
543 	      resv_clusters;
544 
545 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
546 					EXT4_FREECLUSTERS_WATERMARK) {
547 		free_clusters  = percpu_counter_sum_positive(fcc);
548 		dirty_clusters = percpu_counter_sum_positive(dcc);
549 	}
550 	/* Check whether we have space after accounting for current
551 	 * dirty clusters & root reserved clusters.
552 	 */
553 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
554 		return 1;
555 
556 	/* Hm, nope.  Are (enough) root reserved clusters available? */
557 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
558 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
559 	    capable(CAP_SYS_RESOURCE) ||
560 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
561 
562 		if (free_clusters >= (nclusters + dirty_clusters +
563 				      resv_clusters))
564 			return 1;
565 	}
566 	/* No free blocks. Let's see if we can dip into reserved pool */
567 	if (flags & EXT4_MB_USE_RESERVED) {
568 		if (free_clusters >= (nclusters + dirty_clusters))
569 			return 1;
570 	}
571 
572 	return 0;
573 }
574 
575 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
576 			     s64 nclusters, unsigned int flags)
577 {
578 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
579 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
580 		return 0;
581 	} else
582 		return -ENOSPC;
583 }
584 
585 /**
586  * ext4_should_retry_alloc()
587  * @sb:			super block
588  * @retries		number of attemps has been made
589  *
590  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
591  * it is profitable to retry the operation, this function will wait
592  * for the current or committing transaction to complete, and then
593  * return TRUE.
594  *
595  * if the total number of retries exceed three times, return FALSE.
596  */
597 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
598 {
599 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
600 	    (*retries)++ > 3 ||
601 	    !EXT4_SB(sb)->s_journal)
602 		return 0;
603 
604 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
605 
606 	return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
607 }
608 
609 /*
610  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
611  *
612  * @handle:             handle to this transaction
613  * @inode:              file inode
614  * @goal:               given target block(filesystem wide)
615  * @count:		pointer to total number of clusters needed
616  * @errp:               error code
617  *
618  * Return 1st allocated block number on success, *count stores total account
619  * error stores in errp pointer
620  */
621 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
622 				  ext4_fsblk_t goal, unsigned int flags,
623 				  unsigned long *count, int *errp)
624 {
625 	struct ext4_allocation_request ar;
626 	ext4_fsblk_t ret;
627 
628 	memset(&ar, 0, sizeof(ar));
629 	/* Fill with neighbour allocated blocks */
630 	ar.inode = inode;
631 	ar.goal = goal;
632 	ar.len = count ? *count : 1;
633 	ar.flags = flags;
634 
635 	ret = ext4_mb_new_blocks(handle, &ar, errp);
636 	if (count)
637 		*count = ar.len;
638 	/*
639 	 * Account for the allocated meta blocks.  We will never
640 	 * fail EDQUOT for metdata, but we do account for it.
641 	 */
642 	if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
643 		dquot_alloc_block_nofail(inode,
644 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
645 	}
646 	return ret;
647 }
648 
649 /**
650  * ext4_count_free_clusters() -- count filesystem free clusters
651  * @sb:		superblock
652  *
653  * Adds up the number of free clusters from each block group.
654  */
655 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
656 {
657 	ext4_fsblk_t desc_count;
658 	struct ext4_group_desc *gdp;
659 	ext4_group_t i;
660 	ext4_group_t ngroups = ext4_get_groups_count(sb);
661 	struct ext4_group_info *grp;
662 #ifdef EXT4FS_DEBUG
663 	struct ext4_super_block *es;
664 	ext4_fsblk_t bitmap_count;
665 	unsigned int x;
666 	struct buffer_head *bitmap_bh = NULL;
667 
668 	es = EXT4_SB(sb)->s_es;
669 	desc_count = 0;
670 	bitmap_count = 0;
671 	gdp = NULL;
672 
673 	for (i = 0; i < ngroups; i++) {
674 		gdp = ext4_get_group_desc(sb, i, NULL);
675 		if (!gdp)
676 			continue;
677 		grp = NULL;
678 		if (EXT4_SB(sb)->s_group_info)
679 			grp = ext4_get_group_info(sb, i);
680 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
681 			desc_count += ext4_free_group_clusters(sb, gdp);
682 		brelse(bitmap_bh);
683 		bitmap_bh = ext4_read_block_bitmap(sb, i);
684 		if (bitmap_bh == NULL)
685 			continue;
686 
687 		x = ext4_count_free(bitmap_bh->b_data,
688 				    EXT4_CLUSTERS_PER_GROUP(sb) / 8);
689 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
690 			i, ext4_free_group_clusters(sb, gdp), x);
691 		bitmap_count += x;
692 	}
693 	brelse(bitmap_bh);
694 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
695 	       ", computed = %llu, %llu\n",
696 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
697 	       desc_count, bitmap_count);
698 	return bitmap_count;
699 #else
700 	desc_count = 0;
701 	for (i = 0; i < ngroups; i++) {
702 		gdp = ext4_get_group_desc(sb, i, NULL);
703 		if (!gdp)
704 			continue;
705 		grp = NULL;
706 		if (EXT4_SB(sb)->s_group_info)
707 			grp = ext4_get_group_info(sb, i);
708 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
709 			desc_count += ext4_free_group_clusters(sb, gdp);
710 	}
711 
712 	return desc_count;
713 #endif
714 }
715 
716 static inline int test_root(ext4_group_t a, int b)
717 {
718 	while (1) {
719 		if (a < b)
720 			return 0;
721 		if (a == b)
722 			return 1;
723 		if ((a % b) != 0)
724 			return 0;
725 		a = a / b;
726 	}
727 }
728 
729 /**
730  *	ext4_bg_has_super - number of blocks used by the superblock in group
731  *	@sb: superblock for filesystem
732  *	@group: group number to check
733  *
734  *	Return the number of blocks used by the superblock (primary or backup)
735  *	in this group.  Currently this will be only 0 or 1.
736  */
737 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
738 {
739 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
740 
741 	if (group == 0)
742 		return 1;
743 	if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_SPARSE_SUPER2)) {
744 		if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
745 		    group == le32_to_cpu(es->s_backup_bgs[1]))
746 			return 1;
747 		return 0;
748 	}
749 	if ((group <= 1) || !EXT4_HAS_RO_COMPAT_FEATURE(sb,
750 					EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER))
751 		return 1;
752 	if (!(group & 1))
753 		return 0;
754 	if (test_root(group, 3) || (test_root(group, 5)) ||
755 	    test_root(group, 7))
756 		return 1;
757 
758 	return 0;
759 }
760 
761 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
762 					ext4_group_t group)
763 {
764 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
765 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
766 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
767 
768 	if (group == first || group == first + 1 || group == last)
769 		return 1;
770 	return 0;
771 }
772 
773 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
774 					ext4_group_t group)
775 {
776 	if (!ext4_bg_has_super(sb, group))
777 		return 0;
778 
779 	if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
780 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
781 	else
782 		return EXT4_SB(sb)->s_gdb_count;
783 }
784 
785 /**
786  *	ext4_bg_num_gdb - number of blocks used by the group table in group
787  *	@sb: superblock for filesystem
788  *	@group: group number to check
789  *
790  *	Return the number of blocks used by the group descriptor table
791  *	(primary or backup) in this group.  In the future there may be a
792  *	different number of descriptor blocks in each group.
793  */
794 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
795 {
796 	unsigned long first_meta_bg =
797 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
798 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
799 
800 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
801 			metagroup < first_meta_bg)
802 		return ext4_bg_num_gdb_nometa(sb, group);
803 
804 	return ext4_bg_num_gdb_meta(sb,group);
805 
806 }
807 
808 /*
809  * This function returns the number of file system metadata clusters at
810  * the beginning of a block group, including the reserved gdt blocks.
811  */
812 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
813 				     ext4_group_t block_group)
814 {
815 	struct ext4_sb_info *sbi = EXT4_SB(sb);
816 	unsigned num;
817 
818 	/* Check for superblock and gdt backups in this group */
819 	num = ext4_bg_has_super(sb, block_group);
820 
821 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
822 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
823 			  sbi->s_desc_per_block) {
824 		if (num) {
825 			num += ext4_bg_num_gdb(sb, block_group);
826 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
827 		}
828 	} else { /* For META_BG_BLOCK_GROUPS */
829 		num += ext4_bg_num_gdb(sb, block_group);
830 	}
831 	return EXT4_NUM_B2C(sbi, num);
832 }
833 /**
834  *	ext4_inode_to_goal_block - return a hint for block allocation
835  *	@inode: inode for block allocation
836  *
837  *	Return the ideal location to start allocating blocks for a
838  *	newly created inode.
839  */
840 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
841 {
842 	struct ext4_inode_info *ei = EXT4_I(inode);
843 	ext4_group_t block_group;
844 	ext4_grpblk_t colour;
845 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
846 	ext4_fsblk_t bg_start;
847 	ext4_fsblk_t last_block;
848 
849 	block_group = ei->i_block_group;
850 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
851 		/*
852 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
853 		 * block groups per flexgroup, reserve the first block
854 		 * group for directories and special files.  Regular
855 		 * files will start at the second block group.  This
856 		 * tends to speed up directory access and improves
857 		 * fsck times.
858 		 */
859 		block_group &= ~(flex_size-1);
860 		if (S_ISREG(inode->i_mode))
861 			block_group++;
862 	}
863 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
864 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
865 
866 	/*
867 	 * If we are doing delayed allocation, we don't need take
868 	 * colour into account.
869 	 */
870 	if (test_opt(inode->i_sb, DELALLOC))
871 		return bg_start;
872 
873 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
874 		colour = (current->pid % 16) *
875 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
876 	else
877 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
878 	return bg_start + colour;
879 }
880 
881