xref: /linux/fs/ext4/balloc.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
1 /*
2  *  linux/fs/ext4/balloc.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
10  *  Big-endian to little-endian byte-swapping/bitmaps by
11  *        David S. Miller (davem@caip.rutgers.edu), 1995
12  */
13 
14 #include <linux/time.h>
15 #include <linux/capability.h>
16 #include <linux/fs.h>
17 #include <linux/jbd2.h>
18 #include <linux/quotaops.h>
19 #include <linux/buffer_head.h>
20 #include "ext4.h"
21 #include "ext4_jbd2.h"
22 #include "mballoc.h"
23 
24 #include <trace/events/ext4.h>
25 
26 /*
27  * balloc.c contains the blocks allocation and deallocation routines
28  */
29 
30 /*
31  * Calculate the block group number and offset into the block/cluster
32  * allocation bitmap, given a block number
33  */
34 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
35 		ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
36 {
37 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
38 	ext4_grpblk_t offset;
39 
40 	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
41 	offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
42 		EXT4_SB(sb)->s_cluster_bits;
43 	if (offsetp)
44 		*offsetp = offset;
45 	if (blockgrpp)
46 		*blockgrpp = blocknr;
47 
48 }
49 
50 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
51 			ext4_group_t block_group)
52 {
53 	ext4_group_t actual_group;
54 	ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
55 	if (actual_group == block_group)
56 		return 1;
57 	return 0;
58 }
59 
60 /* Return the number of clusters used for file system metadata; this
61  * represents the overhead needed by the file system.
62  */
63 unsigned ext4_num_overhead_clusters(struct super_block *sb,
64 				    ext4_group_t block_group,
65 				    struct ext4_group_desc *gdp)
66 {
67 	unsigned num_clusters;
68 	int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
69 	ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
70 	ext4_fsblk_t itbl_blk;
71 	struct ext4_sb_info *sbi = EXT4_SB(sb);
72 
73 	/* This is the number of clusters used by the superblock,
74 	 * block group descriptors, and reserved block group
75 	 * descriptor blocks */
76 	num_clusters = ext4_num_base_meta_clusters(sb, block_group);
77 
78 	/*
79 	 * For the allocation bitmaps and inode table, we first need
80 	 * to check to see if the block is in the block group.  If it
81 	 * is, then check to see if the cluster is already accounted
82 	 * for in the clusters used for the base metadata cluster, or
83 	 * if we can increment the base metadata cluster to include
84 	 * that block.  Otherwise, we will have to track the cluster
85 	 * used for the allocation bitmap or inode table explicitly.
86 	 * Normally all of these blocks are contiguous, so the special
87 	 * case handling shouldn't be necessary except for *very*
88 	 * unusual file system layouts.
89 	 */
90 	if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
91 		block_cluster = EXT4_B2C(sbi, (start -
92 					       ext4_block_bitmap(sb, gdp)));
93 		if (block_cluster < num_clusters)
94 			block_cluster = -1;
95 		else if (block_cluster == num_clusters) {
96 			num_clusters++;
97 			block_cluster = -1;
98 		}
99 	}
100 
101 	if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
102 		inode_cluster = EXT4_B2C(sbi,
103 					 start - ext4_inode_bitmap(sb, gdp));
104 		if (inode_cluster < num_clusters)
105 			inode_cluster = -1;
106 		else if (inode_cluster == num_clusters) {
107 			num_clusters++;
108 			inode_cluster = -1;
109 		}
110 	}
111 
112 	itbl_blk = ext4_inode_table(sb, gdp);
113 	for (i = 0; i < sbi->s_itb_per_group; i++) {
114 		if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
115 			c = EXT4_B2C(sbi, start - itbl_blk + i);
116 			if ((c < num_clusters) || (c == inode_cluster) ||
117 			    (c == block_cluster) || (c == itbl_cluster))
118 				continue;
119 			if (c == num_clusters) {
120 				num_clusters++;
121 				continue;
122 			}
123 			num_clusters++;
124 			itbl_cluster = c;
125 		}
126 	}
127 
128 	if (block_cluster != -1)
129 		num_clusters++;
130 	if (inode_cluster != -1)
131 		num_clusters++;
132 
133 	return num_clusters;
134 }
135 
136 static unsigned int num_clusters_in_group(struct super_block *sb,
137 					  ext4_group_t block_group)
138 {
139 	unsigned int blocks;
140 
141 	if (block_group == ext4_get_groups_count(sb) - 1) {
142 		/*
143 		 * Even though mke2fs always initializes the first and
144 		 * last group, just in case some other tool was used,
145 		 * we need to make sure we calculate the right free
146 		 * blocks.
147 		 */
148 		blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
149 			ext4_group_first_block_no(sb, block_group);
150 	} else
151 		blocks = EXT4_BLOCKS_PER_GROUP(sb);
152 	return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
153 }
154 
155 /* Initializes an uninitialized block bitmap */
156 void ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
157 			    ext4_group_t block_group,
158 			    struct ext4_group_desc *gdp)
159 {
160 	unsigned int bit, bit_max;
161 	struct ext4_sb_info *sbi = EXT4_SB(sb);
162 	ext4_fsblk_t start, tmp;
163 	int flex_bg = 0;
164 
165 	J_ASSERT_BH(bh, buffer_locked(bh));
166 
167 	/* If checksum is bad mark all blocks used to prevent allocation
168 	 * essentially implementing a per-group read-only flag. */
169 	if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
170 		ext4_error(sb, "Checksum bad for group %u", block_group);
171 		ext4_free_group_clusters_set(sb, gdp, 0);
172 		ext4_free_inodes_set(sb, gdp, 0);
173 		ext4_itable_unused_set(sb, gdp, 0);
174 		memset(bh->b_data, 0xff, sb->s_blocksize);
175 		return;
176 	}
177 	memset(bh->b_data, 0, sb->s_blocksize);
178 
179 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
180 	for (bit = 0; bit < bit_max; bit++)
181 		ext4_set_bit(bit, bh->b_data);
182 
183 	start = ext4_group_first_block_no(sb, block_group);
184 
185 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
186 		flex_bg = 1;
187 
188 	/* Set bits for block and inode bitmaps, and inode table */
189 	tmp = ext4_block_bitmap(sb, gdp);
190 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
191 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
192 
193 	tmp = ext4_inode_bitmap(sb, gdp);
194 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
195 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
196 
197 	tmp = ext4_inode_table(sb, gdp);
198 	for (; tmp < ext4_inode_table(sb, gdp) +
199 		     sbi->s_itb_per_group; tmp++) {
200 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
201 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
202 	}
203 
204 	/*
205 	 * Also if the number of blocks within the group is less than
206 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
207 	 * of the block bitmap to 1
208 	 */
209 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
210 			     sb->s_blocksize * 8, bh->b_data);
211 }
212 
213 /* Return the number of free blocks in a block group.  It is used when
214  * the block bitmap is uninitialized, so we can't just count the bits
215  * in the bitmap. */
216 unsigned ext4_free_clusters_after_init(struct super_block *sb,
217 				       ext4_group_t block_group,
218 				       struct ext4_group_desc *gdp)
219 {
220 	return num_clusters_in_group(sb, block_group) -
221 		ext4_num_overhead_clusters(sb, block_group, gdp);
222 }
223 
224 /*
225  * The free blocks are managed by bitmaps.  A file system contains several
226  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
227  * block for inodes, N blocks for the inode table and data blocks.
228  *
229  * The file system contains group descriptors which are located after the
230  * super block.  Each descriptor contains the number of the bitmap block and
231  * the free blocks count in the block.  The descriptors are loaded in memory
232  * when a file system is mounted (see ext4_fill_super).
233  */
234 
235 /**
236  * ext4_get_group_desc() -- load group descriptor from disk
237  * @sb:			super block
238  * @block_group:	given block group
239  * @bh:			pointer to the buffer head to store the block
240  *			group descriptor
241  */
242 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
243 					     ext4_group_t block_group,
244 					     struct buffer_head **bh)
245 {
246 	unsigned int group_desc;
247 	unsigned int offset;
248 	ext4_group_t ngroups = ext4_get_groups_count(sb);
249 	struct ext4_group_desc *desc;
250 	struct ext4_sb_info *sbi = EXT4_SB(sb);
251 
252 	if (block_group >= ngroups) {
253 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
254 			   " groups_count = %u", block_group, ngroups);
255 
256 		return NULL;
257 	}
258 
259 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
260 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
261 	if (!sbi->s_group_desc[group_desc]) {
262 		ext4_error(sb, "Group descriptor not loaded - "
263 			   "block_group = %u, group_desc = %u, desc = %u",
264 			   block_group, group_desc, offset);
265 		return NULL;
266 	}
267 
268 	desc = (struct ext4_group_desc *)(
269 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
270 		offset * EXT4_DESC_SIZE(sb));
271 	if (bh)
272 		*bh = sbi->s_group_desc[group_desc];
273 	return desc;
274 }
275 
276 static int ext4_valid_block_bitmap(struct super_block *sb,
277 					struct ext4_group_desc *desc,
278 					unsigned int block_group,
279 					struct buffer_head *bh)
280 {
281 	ext4_grpblk_t offset;
282 	ext4_grpblk_t next_zero_bit;
283 	ext4_fsblk_t bitmap_blk;
284 	ext4_fsblk_t group_first_block;
285 
286 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
287 		/* with FLEX_BG, the inode/block bitmaps and itable
288 		 * blocks may not be in the group at all
289 		 * so the bitmap validation will be skipped for those groups
290 		 * or it has to also read the block group where the bitmaps
291 		 * are located to verify they are set.
292 		 */
293 		return 1;
294 	}
295 	group_first_block = ext4_group_first_block_no(sb, block_group);
296 
297 	/* check whether block bitmap block number is set */
298 	bitmap_blk = ext4_block_bitmap(sb, desc);
299 	offset = bitmap_blk - group_first_block;
300 	if (!ext4_test_bit(offset, bh->b_data))
301 		/* bad block bitmap */
302 		goto err_out;
303 
304 	/* check whether the inode bitmap block number is set */
305 	bitmap_blk = ext4_inode_bitmap(sb, desc);
306 	offset = bitmap_blk - group_first_block;
307 	if (!ext4_test_bit(offset, bh->b_data))
308 		/* bad block bitmap */
309 		goto err_out;
310 
311 	/* check whether the inode table block number is set */
312 	bitmap_blk = ext4_inode_table(sb, desc);
313 	offset = bitmap_blk - group_first_block;
314 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
315 				offset + EXT4_SB(sb)->s_itb_per_group,
316 				offset);
317 	if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
318 		/* good bitmap for inode tables */
319 		return 1;
320 
321 err_out:
322 	ext4_error(sb, "Invalid block bitmap - block_group = %d, block = %llu",
323 			block_group, bitmap_blk);
324 	return 0;
325 }
326 /**
327  * ext4_read_block_bitmap()
328  * @sb:			super block
329  * @block_group:	given block group
330  *
331  * Read the bitmap for a given block_group,and validate the
332  * bits for block/inode/inode tables are set in the bitmaps
333  *
334  * Return buffer_head on success or NULL in case of failure.
335  */
336 struct buffer_head *
337 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
338 {
339 	struct ext4_group_desc *desc;
340 	struct buffer_head *bh = NULL;
341 	ext4_fsblk_t bitmap_blk;
342 
343 	desc = ext4_get_group_desc(sb, block_group, NULL);
344 	if (!desc)
345 		return NULL;
346 	bitmap_blk = ext4_block_bitmap(sb, desc);
347 	bh = sb_getblk(sb, bitmap_blk);
348 	if (unlikely(!bh)) {
349 		ext4_error(sb, "Cannot read block bitmap - "
350 			    "block_group = %u, block_bitmap = %llu",
351 			    block_group, bitmap_blk);
352 		return NULL;
353 	}
354 
355 	if (bitmap_uptodate(bh))
356 		return bh;
357 
358 	lock_buffer(bh);
359 	if (bitmap_uptodate(bh)) {
360 		unlock_buffer(bh);
361 		return bh;
362 	}
363 	ext4_lock_group(sb, block_group);
364 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
365 		ext4_init_block_bitmap(sb, bh, block_group, desc);
366 		set_bitmap_uptodate(bh);
367 		set_buffer_uptodate(bh);
368 		ext4_unlock_group(sb, block_group);
369 		unlock_buffer(bh);
370 		return bh;
371 	}
372 	ext4_unlock_group(sb, block_group);
373 	if (buffer_uptodate(bh)) {
374 		/*
375 		 * if not uninit if bh is uptodate,
376 		 * bitmap is also uptodate
377 		 */
378 		set_bitmap_uptodate(bh);
379 		unlock_buffer(bh);
380 		return bh;
381 	}
382 	/*
383 	 * submit the buffer_head for read. We can
384 	 * safely mark the bitmap as uptodate now.
385 	 * We do it here so the bitmap uptodate bit
386 	 * get set with buffer lock held.
387 	 */
388 	trace_ext4_read_block_bitmap_load(sb, block_group);
389 	set_bitmap_uptodate(bh);
390 	if (bh_submit_read(bh) < 0) {
391 		put_bh(bh);
392 		ext4_error(sb, "Cannot read block bitmap - "
393 			    "block_group = %u, block_bitmap = %llu",
394 			    block_group, bitmap_blk);
395 		return NULL;
396 	}
397 	ext4_valid_block_bitmap(sb, desc, block_group, bh);
398 	/*
399 	 * file system mounted not to panic on error,
400 	 * continue with corrupt bitmap
401 	 */
402 	return bh;
403 }
404 
405 /**
406  * ext4_has_free_clusters()
407  * @sbi:	in-core super block structure.
408  * @nclusters:	number of needed blocks
409  * @flags:	flags from ext4_mb_new_blocks()
410  *
411  * Check if filesystem has nclusters free & available for allocation.
412  * On success return 1, return 0 on failure.
413  */
414 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
415 				  s64 nclusters, unsigned int flags)
416 {
417 	s64 free_clusters, dirty_clusters, root_clusters;
418 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
419 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
420 
421 	free_clusters  = percpu_counter_read_positive(fcc);
422 	dirty_clusters = percpu_counter_read_positive(dcc);
423 	root_clusters = EXT4_B2C(sbi, ext4_r_blocks_count(sbi->s_es));
424 
425 	if (free_clusters - (nclusters + root_clusters + dirty_clusters) <
426 					EXT4_FREECLUSTERS_WATERMARK) {
427 		free_clusters  = EXT4_C2B(sbi, percpu_counter_sum_positive(fcc));
428 		dirty_clusters = percpu_counter_sum_positive(dcc);
429 	}
430 	/* Check whether we have space after accounting for current
431 	 * dirty clusters & root reserved clusters.
432 	 */
433 	if (free_clusters >= ((root_clusters + nclusters) + dirty_clusters))
434 		return 1;
435 
436 	/* Hm, nope.  Are (enough) root reserved clusters available? */
437 	if (sbi->s_resuid == current_fsuid() ||
438 	    ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
439 	    capable(CAP_SYS_RESOURCE) ||
440 		(flags & EXT4_MB_USE_ROOT_BLOCKS)) {
441 
442 		if (free_clusters >= (nclusters + dirty_clusters))
443 			return 1;
444 	}
445 
446 	return 0;
447 }
448 
449 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
450 			     s64 nclusters, unsigned int flags)
451 {
452 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
453 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
454 		return 0;
455 	} else
456 		return -ENOSPC;
457 }
458 
459 /**
460  * ext4_should_retry_alloc()
461  * @sb:			super block
462  * @retries		number of attemps has been made
463  *
464  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
465  * it is profitable to retry the operation, this function will wait
466  * for the current or committing transaction to complete, and then
467  * return TRUE.
468  *
469  * if the total number of retries exceed three times, return FALSE.
470  */
471 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
472 {
473 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
474 	    (*retries)++ > 3 ||
475 	    !EXT4_SB(sb)->s_journal)
476 		return 0;
477 
478 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
479 
480 	return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
481 }
482 
483 /*
484  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
485  *
486  * @handle:             handle to this transaction
487  * @inode:              file inode
488  * @goal:               given target block(filesystem wide)
489  * @count:		pointer to total number of clusters needed
490  * @errp:               error code
491  *
492  * Return 1st allocated block number on success, *count stores total account
493  * error stores in errp pointer
494  */
495 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
496 				  ext4_fsblk_t goal, unsigned int flags,
497 				  unsigned long *count, int *errp)
498 {
499 	struct ext4_allocation_request ar;
500 	ext4_fsblk_t ret;
501 
502 	memset(&ar, 0, sizeof(ar));
503 	/* Fill with neighbour allocated blocks */
504 	ar.inode = inode;
505 	ar.goal = goal;
506 	ar.len = count ? *count : 1;
507 	ar.flags = flags;
508 
509 	ret = ext4_mb_new_blocks(handle, &ar, errp);
510 	if (count)
511 		*count = ar.len;
512 	/*
513 	 * Account for the allocated meta blocks.  We will never
514 	 * fail EDQUOT for metdata, but we do account for it.
515 	 */
516 	if (!(*errp) &&
517 	    ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
518 		spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
519 		EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
520 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
521 		dquot_alloc_block_nofail(inode,
522 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
523 	}
524 	return ret;
525 }
526 
527 /**
528  * ext4_count_free_clusters() -- count filesystem free clusters
529  * @sb:		superblock
530  *
531  * Adds up the number of free clusters from each block group.
532  */
533 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
534 {
535 	ext4_fsblk_t desc_count;
536 	struct ext4_group_desc *gdp;
537 	ext4_group_t i;
538 	ext4_group_t ngroups = ext4_get_groups_count(sb);
539 #ifdef EXT4FS_DEBUG
540 	struct ext4_super_block *es;
541 	ext4_fsblk_t bitmap_count;
542 	unsigned int x;
543 	struct buffer_head *bitmap_bh = NULL;
544 
545 	es = EXT4_SB(sb)->s_es;
546 	desc_count = 0;
547 	bitmap_count = 0;
548 	gdp = NULL;
549 
550 	for (i = 0; i < ngroups; i++) {
551 		gdp = ext4_get_group_desc(sb, i, NULL);
552 		if (!gdp)
553 			continue;
554 		desc_count += ext4_free_group_clusters(sb, gdp);
555 		brelse(bitmap_bh);
556 		bitmap_bh = ext4_read_block_bitmap(sb, i);
557 		if (bitmap_bh == NULL)
558 			continue;
559 
560 		x = ext4_count_free(bitmap_bh, sb->s_blocksize);
561 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
562 			i, ext4_free_group_clusters(sb, gdp), x);
563 		bitmap_count += x;
564 	}
565 	brelse(bitmap_bh);
566 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
567 	       ", computed = %llu, %llu\n",
568 	       EXT4_B2C(sbi, ext4_free_blocks_count(es)),
569 	       desc_count, bitmap_count);
570 	return bitmap_count;
571 #else
572 	desc_count = 0;
573 	for (i = 0; i < ngroups; i++) {
574 		gdp = ext4_get_group_desc(sb, i, NULL);
575 		if (!gdp)
576 			continue;
577 		desc_count += ext4_free_group_clusters(sb, gdp);
578 	}
579 
580 	return desc_count;
581 #endif
582 }
583 
584 static inline int test_root(ext4_group_t a, int b)
585 {
586 	int num = b;
587 
588 	while (a > num)
589 		num *= b;
590 	return num == a;
591 }
592 
593 static int ext4_group_sparse(ext4_group_t group)
594 {
595 	if (group <= 1)
596 		return 1;
597 	if (!(group & 1))
598 		return 0;
599 	return (test_root(group, 7) || test_root(group, 5) ||
600 		test_root(group, 3));
601 }
602 
603 /**
604  *	ext4_bg_has_super - number of blocks used by the superblock in group
605  *	@sb: superblock for filesystem
606  *	@group: group number to check
607  *
608  *	Return the number of blocks used by the superblock (primary or backup)
609  *	in this group.  Currently this will be only 0 or 1.
610  */
611 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
612 {
613 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
614 				EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
615 			!ext4_group_sparse(group))
616 		return 0;
617 	return 1;
618 }
619 
620 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
621 					ext4_group_t group)
622 {
623 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
624 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
625 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
626 
627 	if (group == first || group == first + 1 || group == last)
628 		return 1;
629 	return 0;
630 }
631 
632 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
633 					ext4_group_t group)
634 {
635 	if (!ext4_bg_has_super(sb, group))
636 		return 0;
637 
638 	if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
639 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
640 	else
641 		return EXT4_SB(sb)->s_gdb_count;
642 }
643 
644 /**
645  *	ext4_bg_num_gdb - number of blocks used by the group table in group
646  *	@sb: superblock for filesystem
647  *	@group: group number to check
648  *
649  *	Return the number of blocks used by the group descriptor table
650  *	(primary or backup) in this group.  In the future there may be a
651  *	different number of descriptor blocks in each group.
652  */
653 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
654 {
655 	unsigned long first_meta_bg =
656 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
657 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
658 
659 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
660 			metagroup < first_meta_bg)
661 		return ext4_bg_num_gdb_nometa(sb, group);
662 
663 	return ext4_bg_num_gdb_meta(sb,group);
664 
665 }
666 
667 /*
668  * This function returns the number of file system metadata clusters at
669  * the beginning of a block group, including the reserved gdt blocks.
670  */
671 unsigned ext4_num_base_meta_clusters(struct super_block *sb,
672 				     ext4_group_t block_group)
673 {
674 	struct ext4_sb_info *sbi = EXT4_SB(sb);
675 	unsigned num;
676 
677 	/* Check for superblock and gdt backups in this group */
678 	num = ext4_bg_has_super(sb, block_group);
679 
680 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
681 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
682 			  sbi->s_desc_per_block) {
683 		if (num) {
684 			num += ext4_bg_num_gdb(sb, block_group);
685 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
686 		}
687 	} else { /* For META_BG_BLOCK_GROUPS */
688 		num += ext4_bg_num_gdb(sb, block_group);
689 	}
690 	return EXT4_NUM_B2C(sbi, num);
691 }
692 /**
693  *	ext4_inode_to_goal_block - return a hint for block allocation
694  *	@inode: inode for block allocation
695  *
696  *	Return the ideal location to start allocating blocks for a
697  *	newly created inode.
698  */
699 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
700 {
701 	struct ext4_inode_info *ei = EXT4_I(inode);
702 	ext4_group_t block_group;
703 	ext4_grpblk_t colour;
704 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
705 	ext4_fsblk_t bg_start;
706 	ext4_fsblk_t last_block;
707 
708 	block_group = ei->i_block_group;
709 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
710 		/*
711 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
712 		 * block groups per flexgroup, reserve the first block
713 		 * group for directories and special files.  Regular
714 		 * files will start at the second block group.  This
715 		 * tends to speed up directory access and improves
716 		 * fsck times.
717 		 */
718 		block_group &= ~(flex_size-1);
719 		if (S_ISREG(inode->i_mode))
720 			block_group++;
721 	}
722 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
723 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
724 
725 	/*
726 	 * If we are doing delayed allocation, we don't need take
727 	 * colour into account.
728 	 */
729 	if (test_opt(inode->i_sb, DELALLOC))
730 		return bg_start;
731 
732 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
733 		colour = (current->pid % 16) *
734 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
735 	else
736 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
737 	return bg_start + colour;
738 }
739 
740