xref: /openbmc/linux/fs/ext4/balloc.c (revision d2ba09c1)
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 	return 0;
247 }
248 
249 /* Return the number of free blocks in a block group.  It is used when
250  * the block bitmap is uninitialized, so we can't just count the bits
251  * in the bitmap. */
252 unsigned ext4_free_clusters_after_init(struct super_block *sb,
253 				       ext4_group_t block_group,
254 				       struct ext4_group_desc *gdp)
255 {
256 	return num_clusters_in_group(sb, block_group) -
257 		ext4_num_overhead_clusters(sb, block_group, gdp);
258 }
259 
260 /*
261  * The free blocks are managed by bitmaps.  A file system contains several
262  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
263  * block for inodes, N blocks for the inode table and data blocks.
264  *
265  * The file system contains group descriptors which are located after the
266  * super block.  Each descriptor contains the number of the bitmap block and
267  * the free blocks count in the block.  The descriptors are loaded in memory
268  * when a file system is mounted (see ext4_fill_super).
269  */
270 
271 /**
272  * ext4_get_group_desc() -- load group descriptor from disk
273  * @sb:			super block
274  * @block_group:	given block group
275  * @bh:			pointer to the buffer head to store the block
276  *			group descriptor
277  */
278 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
279 					     ext4_group_t block_group,
280 					     struct buffer_head **bh)
281 {
282 	unsigned int group_desc;
283 	unsigned int offset;
284 	ext4_group_t ngroups = ext4_get_groups_count(sb);
285 	struct ext4_group_desc *desc;
286 	struct ext4_sb_info *sbi = EXT4_SB(sb);
287 
288 	if (block_group >= ngroups) {
289 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
290 			   " groups_count = %u", block_group, ngroups);
291 
292 		return NULL;
293 	}
294 
295 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
296 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
297 	if (!sbi->s_group_desc[group_desc]) {
298 		ext4_error(sb, "Group descriptor not loaded - "
299 			   "block_group = %u, group_desc = %u, desc = %u",
300 			   block_group, group_desc, offset);
301 		return NULL;
302 	}
303 
304 	desc = (struct ext4_group_desc *)(
305 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
306 		offset * EXT4_DESC_SIZE(sb));
307 	if (bh)
308 		*bh = sbi->s_group_desc[group_desc];
309 	return desc;
310 }
311 
312 /*
313  * Return the block number which was discovered to be invalid, or 0 if
314  * the block bitmap is valid.
315  */
316 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
317 					    struct ext4_group_desc *desc,
318 					    ext4_group_t block_group,
319 					    struct buffer_head *bh)
320 {
321 	struct ext4_sb_info *sbi = EXT4_SB(sb);
322 	ext4_grpblk_t offset;
323 	ext4_grpblk_t next_zero_bit;
324 	ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
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 (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
343 	    !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 (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
351 	    !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
352 		/* bad block bitmap */
353 		return blk;
354 
355 	/* check whether the inode table block number is set */
356 	blk = ext4_inode_table(sb, desc);
357 	offset = blk - group_first_block;
358 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
359 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
360 		return blk;
361 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
362 			EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
363 			EXT4_B2C(sbi, offset));
364 	if (next_zero_bit <
365 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
366 		/* bad bitmap for inode tables */
367 		return blk;
368 	return 0;
369 }
370 
371 static int ext4_validate_block_bitmap(struct super_block *sb,
372 				      struct ext4_group_desc *desc,
373 				      ext4_group_t block_group,
374 				      struct buffer_head *bh)
375 {
376 	ext4_fsblk_t	blk;
377 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
378 	struct ext4_sb_info *sbi = EXT4_SB(sb);
379 
380 	if (buffer_verified(bh))
381 		return 0;
382 	if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
383 		return -EFSCORRUPTED;
384 
385 	ext4_lock_group(sb, block_group);
386 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
387 			desc, bh))) {
388 		ext4_unlock_group(sb, block_group);
389 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
390 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
391 			percpu_counter_sub(&sbi->s_freeclusters_counter,
392 					   grp->bb_free);
393 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
394 		return -EFSBADCRC;
395 	}
396 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
397 	if (unlikely(blk != 0)) {
398 		ext4_unlock_group(sb, block_group);
399 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
400 			   block_group, blk);
401 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
402 			percpu_counter_sub(&sbi->s_freeclusters_counter,
403 					   grp->bb_free);
404 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
405 		return -EFSCORRUPTED;
406 	}
407 	set_buffer_verified(bh);
408 	ext4_unlock_group(sb, block_group);
409 	return 0;
410 }
411 
412 /**
413  * ext4_read_block_bitmap_nowait()
414  * @sb:			super block
415  * @block_group:	given block group
416  *
417  * Read the bitmap for a given block_group,and validate the
418  * bits for block/inode/inode tables are set in the bitmaps
419  *
420  * Return buffer_head on success or NULL in case of failure.
421  */
422 struct buffer_head *
423 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
424 {
425 	struct ext4_group_desc *desc;
426 	struct ext4_sb_info *sbi = EXT4_SB(sb);
427 	struct buffer_head *bh;
428 	ext4_fsblk_t bitmap_blk;
429 	int err;
430 
431 	desc = ext4_get_group_desc(sb, block_group, NULL);
432 	if (!desc)
433 		return ERR_PTR(-EFSCORRUPTED);
434 	bitmap_blk = ext4_block_bitmap(sb, desc);
435 	if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
436 	    (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
437 		ext4_error(sb, "Invalid block bitmap block %llu in "
438 			   "block_group %u", bitmap_blk, block_group);
439 		return ERR_PTR(-EFSCORRUPTED);
440 	}
441 	bh = sb_getblk(sb, bitmap_blk);
442 	if (unlikely(!bh)) {
443 		ext4_error(sb, "Cannot get buffer for block bitmap - "
444 			   "block_group = %u, block_bitmap = %llu",
445 			   block_group, bitmap_blk);
446 		return ERR_PTR(-ENOMEM);
447 	}
448 
449 	if (bitmap_uptodate(bh))
450 		goto verify;
451 
452 	lock_buffer(bh);
453 	if (bitmap_uptodate(bh)) {
454 		unlock_buffer(bh);
455 		goto verify;
456 	}
457 	ext4_lock_group(sb, block_group);
458 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
459 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
460 		set_bitmap_uptodate(bh);
461 		set_buffer_uptodate(bh);
462 		set_buffer_verified(bh);
463 		ext4_unlock_group(sb, block_group);
464 		unlock_buffer(bh);
465 		if (err) {
466 			ext4_error(sb, "Failed to init block bitmap for group "
467 				   "%u: %d", block_group, err);
468 			goto out;
469 		}
470 		goto verify;
471 	}
472 	ext4_unlock_group(sb, block_group);
473 	if (buffer_uptodate(bh)) {
474 		/*
475 		 * if not uninit if bh is uptodate,
476 		 * bitmap is also uptodate
477 		 */
478 		set_bitmap_uptodate(bh);
479 		unlock_buffer(bh);
480 		goto verify;
481 	}
482 	/*
483 	 * submit the buffer_head for reading
484 	 */
485 	set_buffer_new(bh);
486 	trace_ext4_read_block_bitmap_load(sb, block_group);
487 	bh->b_end_io = ext4_end_bitmap_read;
488 	get_bh(bh);
489 	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
490 	return bh;
491 verify:
492 	err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
493 	if (err)
494 		goto out;
495 	return bh;
496 out:
497 	put_bh(bh);
498 	return ERR_PTR(err);
499 }
500 
501 /* Returns 0 on success, 1 on error */
502 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
503 			   struct buffer_head *bh)
504 {
505 	struct ext4_group_desc *desc;
506 
507 	if (!buffer_new(bh))
508 		return 0;
509 	desc = ext4_get_group_desc(sb, block_group, NULL);
510 	if (!desc)
511 		return -EFSCORRUPTED;
512 	wait_on_buffer(bh);
513 	if (!buffer_uptodate(bh)) {
514 		ext4_error(sb, "Cannot read block bitmap - "
515 			   "block_group = %u, block_bitmap = %llu",
516 			   block_group, (unsigned long long) bh->b_blocknr);
517 		return -EIO;
518 	}
519 	clear_buffer_new(bh);
520 	/* Panic or remount fs read-only if block bitmap is invalid */
521 	return ext4_validate_block_bitmap(sb, desc, block_group, bh);
522 }
523 
524 struct buffer_head *
525 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
526 {
527 	struct buffer_head *bh;
528 	int err;
529 
530 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
531 	if (IS_ERR(bh))
532 		return bh;
533 	err = ext4_wait_block_bitmap(sb, block_group, bh);
534 	if (err) {
535 		put_bh(bh);
536 		return ERR_PTR(err);
537 	}
538 	return bh;
539 }
540 
541 /**
542  * ext4_has_free_clusters()
543  * @sbi:	in-core super block structure.
544  * @nclusters:	number of needed blocks
545  * @flags:	flags from ext4_mb_new_blocks()
546  *
547  * Check if filesystem has nclusters free & available for allocation.
548  * On success return 1, return 0 on failure.
549  */
550 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
551 				  s64 nclusters, unsigned int flags)
552 {
553 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
554 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
555 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
556 
557 	free_clusters  = percpu_counter_read_positive(fcc);
558 	dirty_clusters = percpu_counter_read_positive(dcc);
559 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
560 
561 	/*
562 	 * r_blocks_count should always be multiple of the cluster ratio so
563 	 * we are safe to do a plane bit shift only.
564 	 */
565 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
566 	      resv_clusters;
567 
568 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
569 					EXT4_FREECLUSTERS_WATERMARK) {
570 		free_clusters  = percpu_counter_sum_positive(fcc);
571 		dirty_clusters = percpu_counter_sum_positive(dcc);
572 	}
573 	/* Check whether we have space after accounting for current
574 	 * dirty clusters & root reserved clusters.
575 	 */
576 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
577 		return 1;
578 
579 	/* Hm, nope.  Are (enough) root reserved clusters available? */
580 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
581 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
582 	    capable(CAP_SYS_RESOURCE) ||
583 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
584 
585 		if (free_clusters >= (nclusters + dirty_clusters +
586 				      resv_clusters))
587 			return 1;
588 	}
589 	/* No free blocks. Let's see if we can dip into reserved pool */
590 	if (flags & EXT4_MB_USE_RESERVED) {
591 		if (free_clusters >= (nclusters + dirty_clusters))
592 			return 1;
593 	}
594 
595 	return 0;
596 }
597 
598 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
599 			     s64 nclusters, unsigned int flags)
600 {
601 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
602 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
603 		return 0;
604 	} else
605 		return -ENOSPC;
606 }
607 
608 /**
609  * ext4_should_retry_alloc()
610  * @sb:			super block
611  * @retries		number of attemps has been made
612  *
613  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
614  * it is profitable to retry the operation, this function will wait
615  * for the current or committing transaction to complete, and then
616  * return TRUE.  We will only retry once.
617  */
618 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
619 {
620 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
621 	    (*retries)++ > 1 ||
622 	    !EXT4_SB(sb)->s_journal)
623 		return 0;
624 
625 	smp_mb();
626 	if (EXT4_SB(sb)->s_mb_free_pending == 0)
627 		return 0;
628 
629 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
630 	jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
631 	return 1;
632 }
633 
634 /*
635  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
636  *
637  * @handle:             handle to this transaction
638  * @inode:              file inode
639  * @goal:               given target block(filesystem wide)
640  * @count:		pointer to total number of clusters needed
641  * @errp:               error code
642  *
643  * Return 1st allocated block number on success, *count stores total account
644  * error stores in errp pointer
645  */
646 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
647 				  ext4_fsblk_t goal, unsigned int flags,
648 				  unsigned long *count, int *errp)
649 {
650 	struct ext4_allocation_request ar;
651 	ext4_fsblk_t ret;
652 
653 	memset(&ar, 0, sizeof(ar));
654 	/* Fill with neighbour allocated blocks */
655 	ar.inode = inode;
656 	ar.goal = goal;
657 	ar.len = count ? *count : 1;
658 	ar.flags = flags;
659 
660 	ret = ext4_mb_new_blocks(handle, &ar, errp);
661 	if (count)
662 		*count = ar.len;
663 	/*
664 	 * Account for the allocated meta blocks.  We will never
665 	 * fail EDQUOT for metdata, but we do account for it.
666 	 */
667 	if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
668 		dquot_alloc_block_nofail(inode,
669 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
670 	}
671 	return ret;
672 }
673 
674 /**
675  * ext4_count_free_clusters() -- count filesystem free clusters
676  * @sb:		superblock
677  *
678  * Adds up the number of free clusters from each block group.
679  */
680 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
681 {
682 	ext4_fsblk_t desc_count;
683 	struct ext4_group_desc *gdp;
684 	ext4_group_t i;
685 	ext4_group_t ngroups = ext4_get_groups_count(sb);
686 	struct ext4_group_info *grp;
687 #ifdef EXT4FS_DEBUG
688 	struct ext4_super_block *es;
689 	ext4_fsblk_t bitmap_count;
690 	unsigned int x;
691 	struct buffer_head *bitmap_bh = NULL;
692 
693 	es = EXT4_SB(sb)->s_es;
694 	desc_count = 0;
695 	bitmap_count = 0;
696 	gdp = NULL;
697 
698 	for (i = 0; i < ngroups; i++) {
699 		gdp = ext4_get_group_desc(sb, i, NULL);
700 		if (!gdp)
701 			continue;
702 		grp = NULL;
703 		if (EXT4_SB(sb)->s_group_info)
704 			grp = ext4_get_group_info(sb, i);
705 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
706 			desc_count += ext4_free_group_clusters(sb, gdp);
707 		brelse(bitmap_bh);
708 		bitmap_bh = ext4_read_block_bitmap(sb, i);
709 		if (IS_ERR(bitmap_bh)) {
710 			bitmap_bh = NULL;
711 			continue;
712 		}
713 
714 		x = ext4_count_free(bitmap_bh->b_data,
715 				    EXT4_CLUSTERS_PER_GROUP(sb) / 8);
716 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
717 			i, ext4_free_group_clusters(sb, gdp), x);
718 		bitmap_count += x;
719 	}
720 	brelse(bitmap_bh);
721 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
722 	       ", computed = %llu, %llu\n",
723 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
724 	       desc_count, bitmap_count);
725 	return bitmap_count;
726 #else
727 	desc_count = 0;
728 	for (i = 0; i < ngroups; i++) {
729 		gdp = ext4_get_group_desc(sb, i, NULL);
730 		if (!gdp)
731 			continue;
732 		grp = NULL;
733 		if (EXT4_SB(sb)->s_group_info)
734 			grp = ext4_get_group_info(sb, i);
735 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
736 			desc_count += ext4_free_group_clusters(sb, gdp);
737 	}
738 
739 	return desc_count;
740 #endif
741 }
742 
743 static inline int test_root(ext4_group_t a, int b)
744 {
745 	while (1) {
746 		if (a < b)
747 			return 0;
748 		if (a == b)
749 			return 1;
750 		if ((a % b) != 0)
751 			return 0;
752 		a = a / b;
753 	}
754 }
755 
756 /**
757  *	ext4_bg_has_super - number of blocks used by the superblock in group
758  *	@sb: superblock for filesystem
759  *	@group: group number to check
760  *
761  *	Return the number of blocks used by the superblock (primary or backup)
762  *	in this group.  Currently this will be only 0 or 1.
763  */
764 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
765 {
766 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
767 
768 	if (group == 0)
769 		return 1;
770 	if (ext4_has_feature_sparse_super2(sb)) {
771 		if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
772 		    group == le32_to_cpu(es->s_backup_bgs[1]))
773 			return 1;
774 		return 0;
775 	}
776 	if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
777 		return 1;
778 	if (!(group & 1))
779 		return 0;
780 	if (test_root(group, 3) || (test_root(group, 5)) ||
781 	    test_root(group, 7))
782 		return 1;
783 
784 	return 0;
785 }
786 
787 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
788 					ext4_group_t group)
789 {
790 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
791 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
792 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
793 
794 	if (group == first || group == first + 1 || group == last)
795 		return 1;
796 	return 0;
797 }
798 
799 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
800 					ext4_group_t group)
801 {
802 	if (!ext4_bg_has_super(sb, group))
803 		return 0;
804 
805 	if (ext4_has_feature_meta_bg(sb))
806 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
807 	else
808 		return EXT4_SB(sb)->s_gdb_count;
809 }
810 
811 /**
812  *	ext4_bg_num_gdb - number of blocks used by the group table in group
813  *	@sb: superblock for filesystem
814  *	@group: group number to check
815  *
816  *	Return the number of blocks used by the group descriptor table
817  *	(primary or backup) in this group.  In the future there may be a
818  *	different number of descriptor blocks in each group.
819  */
820 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
821 {
822 	unsigned long first_meta_bg =
823 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
824 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
825 
826 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
827 		return ext4_bg_num_gdb_nometa(sb, group);
828 
829 	return ext4_bg_num_gdb_meta(sb,group);
830 
831 }
832 
833 /*
834  * This function returns the number of file system metadata clusters at
835  * the beginning of a block group, including the reserved gdt blocks.
836  */
837 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
838 				     ext4_group_t block_group)
839 {
840 	struct ext4_sb_info *sbi = EXT4_SB(sb);
841 	unsigned num;
842 
843 	/* Check for superblock and gdt backups in this group */
844 	num = ext4_bg_has_super(sb, block_group);
845 
846 	if (!ext4_has_feature_meta_bg(sb) ||
847 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
848 			  sbi->s_desc_per_block) {
849 		if (num) {
850 			num += ext4_bg_num_gdb(sb, block_group);
851 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
852 		}
853 	} else { /* For META_BG_BLOCK_GROUPS */
854 		num += ext4_bg_num_gdb(sb, block_group);
855 	}
856 	return EXT4_NUM_B2C(sbi, num);
857 }
858 /**
859  *	ext4_inode_to_goal_block - return a hint for block allocation
860  *	@inode: inode for block allocation
861  *
862  *	Return the ideal location to start allocating blocks for a
863  *	newly created inode.
864  */
865 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
866 {
867 	struct ext4_inode_info *ei = EXT4_I(inode);
868 	ext4_group_t block_group;
869 	ext4_grpblk_t colour;
870 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
871 	ext4_fsblk_t bg_start;
872 	ext4_fsblk_t last_block;
873 
874 	block_group = ei->i_block_group;
875 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
876 		/*
877 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
878 		 * block groups per flexgroup, reserve the first block
879 		 * group for directories and special files.  Regular
880 		 * files will start at the second block group.  This
881 		 * tends to speed up directory access and improves
882 		 * fsck times.
883 		 */
884 		block_group &= ~(flex_size-1);
885 		if (S_ISREG(inode->i_mode))
886 			block_group++;
887 	}
888 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
889 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
890 
891 	/*
892 	 * If we are doing delayed allocation, we don't need take
893 	 * colour into account.
894 	 */
895 	if (test_opt(inode->i_sb, DELALLOC))
896 		return bg_start;
897 
898 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
899 		colour = (current->pid % 16) *
900 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
901 	else
902 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
903 	return bg_start + colour;
904 }
905 
906