xref: /openbmc/linux/fs/ext4/balloc.c (revision dfc53baa)
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 
188 	J_ASSERT_BH(bh, buffer_locked(bh));
189 
190 	/* If checksum is bad mark all blocks used to prevent allocation
191 	 * essentially implementing a per-group read-only flag. */
192 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
193 		ext4_mark_group_bitmap_corrupted(sb, block_group,
194 					EXT4_GROUP_INFO_BBITMAP_CORRUPT |
195 					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
196 		return -EFSBADCRC;
197 	}
198 	memset(bh->b_data, 0, sb->s_blocksize);
199 
200 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
201 	if ((bit_max >> 3) >= bh->b_size)
202 		return -EFSCORRUPTED;
203 
204 	for (bit = 0; bit < bit_max; bit++)
205 		ext4_set_bit(bit, bh->b_data);
206 
207 	start = ext4_group_first_block_no(sb, block_group);
208 
209 	/* Set bits for block and inode bitmaps, and inode table */
210 	tmp = ext4_block_bitmap(sb, gdp);
211 	if (ext4_block_in_group(sb, tmp, block_group))
212 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
213 
214 	tmp = ext4_inode_bitmap(sb, gdp);
215 	if (ext4_block_in_group(sb, tmp, block_group))
216 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
217 
218 	tmp = ext4_inode_table(sb, gdp);
219 	for (; tmp < ext4_inode_table(sb, gdp) +
220 		     sbi->s_itb_per_group; tmp++) {
221 		if (ext4_block_in_group(sb, tmp, block_group))
222 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
223 	}
224 
225 	/*
226 	 * Also if the number of blocks within the group is less than
227 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
228 	 * of the block bitmap to 1
229 	 */
230 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
231 			     sb->s_blocksize * 8, bh->b_data);
232 	return 0;
233 }
234 
235 /* Return the number of free blocks in a block group.  It is used when
236  * the block bitmap is uninitialized, so we can't just count the bits
237  * in the bitmap. */
238 unsigned ext4_free_clusters_after_init(struct super_block *sb,
239 				       ext4_group_t block_group,
240 				       struct ext4_group_desc *gdp)
241 {
242 	return num_clusters_in_group(sb, block_group) -
243 		ext4_num_overhead_clusters(sb, block_group, gdp);
244 }
245 
246 /*
247  * The free blocks are managed by bitmaps.  A file system contains several
248  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
249  * block for inodes, N blocks for the inode table and data blocks.
250  *
251  * The file system contains group descriptors which are located after the
252  * super block.  Each descriptor contains the number of the bitmap block and
253  * the free blocks count in the block.  The descriptors are loaded in memory
254  * when a file system is mounted (see ext4_fill_super).
255  */
256 
257 /**
258  * ext4_get_group_desc() -- load group descriptor from disk
259  * @sb:			super block
260  * @block_group:	given block group
261  * @bh:			pointer to the buffer head to store the block
262  *			group descriptor
263  */
264 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
265 					     ext4_group_t block_group,
266 					     struct buffer_head **bh)
267 {
268 	unsigned int group_desc;
269 	unsigned int offset;
270 	ext4_group_t ngroups = ext4_get_groups_count(sb);
271 	struct ext4_group_desc *desc;
272 	struct ext4_sb_info *sbi = EXT4_SB(sb);
273 	struct buffer_head *bh_p;
274 
275 	if (block_group >= ngroups) {
276 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
277 			   " groups_count = %u", block_group, ngroups);
278 
279 		return NULL;
280 	}
281 
282 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
283 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
284 	bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc);
285 	/*
286 	 * sbi_array_rcu_deref returns with rcu unlocked, this is ok since
287 	 * the pointer being dereferenced won't be dereferenced again. By
288 	 * looking at the usage in add_new_gdb() the value isn't modified,
289 	 * just the pointer, and so it remains valid.
290 	 */
291 	if (!bh_p) {
292 		ext4_error(sb, "Group descriptor not loaded - "
293 			   "block_group = %u, group_desc = %u, desc = %u",
294 			   block_group, group_desc, offset);
295 		return NULL;
296 	}
297 
298 	desc = (struct ext4_group_desc *)(
299 		(__u8 *)bh_p->b_data +
300 		offset * EXT4_DESC_SIZE(sb));
301 	if (bh)
302 		*bh = bh_p;
303 	return desc;
304 }
305 
306 /*
307  * Return the block number which was discovered to be invalid, or 0 if
308  * the block bitmap is valid.
309  */
310 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
311 					    struct ext4_group_desc *desc,
312 					    ext4_group_t block_group,
313 					    struct buffer_head *bh)
314 {
315 	struct ext4_sb_info *sbi = EXT4_SB(sb);
316 	ext4_grpblk_t offset;
317 	ext4_grpblk_t next_zero_bit;
318 	ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
319 	ext4_fsblk_t blk;
320 	ext4_fsblk_t group_first_block;
321 
322 	if (ext4_has_feature_flex_bg(sb)) {
323 		/* with FLEX_BG, the inode/block bitmaps and itable
324 		 * blocks may not be in the group at all
325 		 * so the bitmap validation will be skipped for those groups
326 		 * or it has to also read the block group where the bitmaps
327 		 * are located to verify they are set.
328 		 */
329 		return 0;
330 	}
331 	group_first_block = ext4_group_first_block_no(sb, block_group);
332 
333 	/* check whether block bitmap block number is set */
334 	blk = ext4_block_bitmap(sb, desc);
335 	offset = blk - group_first_block;
336 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
337 	    !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
338 		/* bad block bitmap */
339 		return blk;
340 
341 	/* check whether the inode bitmap block number is set */
342 	blk = ext4_inode_bitmap(sb, desc);
343 	offset = blk - group_first_block;
344 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
345 	    !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
346 		/* bad block bitmap */
347 		return blk;
348 
349 	/* check whether the inode table block number is set */
350 	blk = ext4_inode_table(sb, desc);
351 	offset = blk - group_first_block;
352 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
353 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
354 		return blk;
355 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
356 			EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
357 			EXT4_B2C(sbi, offset));
358 	if (next_zero_bit <
359 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
360 		/* bad bitmap for inode tables */
361 		return blk;
362 	return 0;
363 }
364 
365 static int ext4_validate_block_bitmap(struct super_block *sb,
366 				      struct ext4_group_desc *desc,
367 				      ext4_group_t block_group,
368 				      struct buffer_head *bh)
369 {
370 	ext4_fsblk_t	blk;
371 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
372 
373 	if (buffer_verified(bh))
374 		return 0;
375 	if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
376 		return -EFSCORRUPTED;
377 
378 	ext4_lock_group(sb, block_group);
379 	if (buffer_verified(bh))
380 		goto verified;
381 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
382 						    desc, bh) ||
383 		     ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) {
384 		ext4_unlock_group(sb, block_group);
385 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
386 		ext4_mark_group_bitmap_corrupted(sb, block_group,
387 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
388 		return -EFSBADCRC;
389 	}
390 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
391 	if (unlikely(blk != 0)) {
392 		ext4_unlock_group(sb, block_group);
393 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
394 			   block_group, blk);
395 		ext4_mark_group_bitmap_corrupted(sb, block_group,
396 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
397 		return -EFSCORRUPTED;
398 	}
399 	set_buffer_verified(bh);
400 verified:
401 	ext4_unlock_group(sb, block_group);
402 	return 0;
403 }
404 
405 /**
406  * ext4_read_block_bitmap_nowait()
407  * @sb:			super block
408  * @block_group:	given block group
409  *
410  * Read the bitmap for a given block_group,and validate the
411  * bits for block/inode/inode tables are set in the bitmaps
412  *
413  * Return buffer_head on success or an ERR_PTR in case of failure.
414  */
415 struct buffer_head *
416 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group,
417 			      bool ignore_locked)
418 {
419 	struct ext4_group_desc *desc;
420 	struct ext4_sb_info *sbi = EXT4_SB(sb);
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 	if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
430 	    (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
431 		ext4_error(sb, "Invalid block bitmap block %llu in "
432 			   "block_group %u", bitmap_blk, block_group);
433 		ext4_mark_group_bitmap_corrupted(sb, block_group,
434 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
435 		return ERR_PTR(-EFSCORRUPTED);
436 	}
437 	bh = sb_getblk(sb, bitmap_blk);
438 	if (unlikely(!bh)) {
439 		ext4_warning(sb, "Cannot get buffer for block bitmap - "
440 			     "block_group = %u, block_bitmap = %llu",
441 			     block_group, bitmap_blk);
442 		return ERR_PTR(-ENOMEM);
443 	}
444 
445 	if (ignore_locked && buffer_locked(bh)) {
446 		/* buffer under IO already, return if called for prefetching */
447 		put_bh(bh);
448 		return NULL;
449 	}
450 
451 	if (bitmap_uptodate(bh))
452 		goto verify;
453 
454 	lock_buffer(bh);
455 	if (bitmap_uptodate(bh)) {
456 		unlock_buffer(bh);
457 		goto verify;
458 	}
459 	ext4_lock_group(sb, block_group);
460 	if (ext4_has_group_desc_csum(sb) &&
461 	    (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
462 		if (block_group == 0) {
463 			ext4_unlock_group(sb, block_group);
464 			unlock_buffer(bh);
465 			ext4_error(sb, "Block bitmap for bg 0 marked "
466 				   "uninitialized");
467 			err = -EFSCORRUPTED;
468 			goto out;
469 		}
470 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
471 		set_bitmap_uptodate(bh);
472 		set_buffer_uptodate(bh);
473 		set_buffer_verified(bh);
474 		ext4_unlock_group(sb, block_group);
475 		unlock_buffer(bh);
476 		if (err) {
477 			ext4_error(sb, "Failed to init block bitmap for group "
478 				   "%u: %d", block_group, err);
479 			goto out;
480 		}
481 		goto verify;
482 	}
483 	ext4_unlock_group(sb, block_group);
484 	if (buffer_uptodate(bh)) {
485 		/*
486 		 * if not uninit if bh is uptodate,
487 		 * bitmap is also uptodate
488 		 */
489 		set_bitmap_uptodate(bh);
490 		unlock_buffer(bh);
491 		goto verify;
492 	}
493 	/*
494 	 * submit the buffer_head for reading
495 	 */
496 	set_buffer_new(bh);
497 	trace_ext4_read_block_bitmap_load(sb, block_group, ignore_locked);
498 	bh->b_end_io = ext4_end_bitmap_read;
499 	get_bh(bh);
500 	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO |
501 		  (ignore_locked ? REQ_RAHEAD : 0), bh);
502 	return bh;
503 verify:
504 	err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
505 	if (err)
506 		goto out;
507 	return bh;
508 out:
509 	put_bh(bh);
510 	return ERR_PTR(err);
511 }
512 
513 /* Returns 0 on success, -errno on error */
514 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
515 			   struct buffer_head *bh)
516 {
517 	struct ext4_group_desc *desc;
518 
519 	if (!buffer_new(bh))
520 		return 0;
521 	desc = ext4_get_group_desc(sb, block_group, NULL);
522 	if (!desc)
523 		return -EFSCORRUPTED;
524 	wait_on_buffer(bh);
525 	ext4_simulate_fail_bh(sb, bh, EXT4_SIM_BBITMAP_EIO);
526 	if (!buffer_uptodate(bh)) {
527 		ext4_error_err(sb, EIO, "Cannot read block bitmap - "
528 			       "block_group = %u, block_bitmap = %llu",
529 			       block_group, (unsigned long long) bh->b_blocknr);
530 		ext4_mark_group_bitmap_corrupted(sb, block_group,
531 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
532 		return -EIO;
533 	}
534 	clear_buffer_new(bh);
535 	/* Panic or remount fs read-only if block bitmap is invalid */
536 	return ext4_validate_block_bitmap(sb, desc, block_group, bh);
537 }
538 
539 struct buffer_head *
540 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
541 {
542 	struct buffer_head *bh;
543 	int err;
544 
545 	bh = ext4_read_block_bitmap_nowait(sb, block_group, false);
546 	if (IS_ERR(bh))
547 		return bh;
548 	err = ext4_wait_block_bitmap(sb, block_group, bh);
549 	if (err) {
550 		put_bh(bh);
551 		return ERR_PTR(err);
552 	}
553 	return bh;
554 }
555 
556 /**
557  * ext4_has_free_clusters()
558  * @sbi:	in-core super block structure.
559  * @nclusters:	number of needed blocks
560  * @flags:	flags from ext4_mb_new_blocks()
561  *
562  * Check if filesystem has nclusters free & available for allocation.
563  * On success return 1, return 0 on failure.
564  */
565 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
566 				  s64 nclusters, unsigned int flags)
567 {
568 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
569 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
570 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
571 
572 	free_clusters  = percpu_counter_read_positive(fcc);
573 	dirty_clusters = percpu_counter_read_positive(dcc);
574 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
575 
576 	/*
577 	 * r_blocks_count should always be multiple of the cluster ratio so
578 	 * we are safe to do a plane bit shift only.
579 	 */
580 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
581 	      resv_clusters;
582 
583 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
584 					EXT4_FREECLUSTERS_WATERMARK) {
585 		free_clusters  = percpu_counter_sum_positive(fcc);
586 		dirty_clusters = percpu_counter_sum_positive(dcc);
587 	}
588 	/* Check whether we have space after accounting for current
589 	 * dirty clusters & root reserved clusters.
590 	 */
591 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
592 		return 1;
593 
594 	/* Hm, nope.  Are (enough) root reserved clusters available? */
595 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
596 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
597 	    capable(CAP_SYS_RESOURCE) ||
598 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
599 
600 		if (free_clusters >= (nclusters + dirty_clusters +
601 				      resv_clusters))
602 			return 1;
603 	}
604 	/* No free blocks. Let's see if we can dip into reserved pool */
605 	if (flags & EXT4_MB_USE_RESERVED) {
606 		if (free_clusters >= (nclusters + dirty_clusters))
607 			return 1;
608 	}
609 
610 	return 0;
611 }
612 
613 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
614 			     s64 nclusters, unsigned int flags)
615 {
616 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
617 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
618 		return 0;
619 	} else
620 		return -ENOSPC;
621 }
622 
623 /**
624  * ext4_should_retry_alloc() - check if a block allocation should be retried
625  * @sb:			super block
626  * @retries:		number of attemps has been made
627  *
628  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
629  * it is profitable to retry the operation, this function will wait
630  * for the current or committing transaction to complete, and then
631  * return TRUE.  We will only retry once.
632  */
633 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
634 {
635 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
636 	    (*retries)++ > 1 ||
637 	    !EXT4_SB(sb)->s_journal)
638 		return 0;
639 
640 	smp_mb();
641 	if (EXT4_SB(sb)->s_mb_free_pending == 0)
642 		return 0;
643 
644 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
645 	jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
646 	return 1;
647 }
648 
649 /*
650  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
651  *
652  * @handle:             handle to this transaction
653  * @inode:              file inode
654  * @goal:               given target block(filesystem wide)
655  * @count:		pointer to total number of clusters needed
656  * @errp:               error code
657  *
658  * Return 1st allocated block number on success, *count stores total account
659  * error stores in errp pointer
660  */
661 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
662 				  ext4_fsblk_t goal, unsigned int flags,
663 				  unsigned long *count, int *errp)
664 {
665 	struct ext4_allocation_request ar;
666 	ext4_fsblk_t ret;
667 
668 	memset(&ar, 0, sizeof(ar));
669 	/* Fill with neighbour allocated blocks */
670 	ar.inode = inode;
671 	ar.goal = goal;
672 	ar.len = count ? *count : 1;
673 	ar.flags = flags;
674 
675 	ret = ext4_mb_new_blocks(handle, &ar, errp);
676 	if (count)
677 		*count = ar.len;
678 	/*
679 	 * Account for the allocated meta blocks.  We will never
680 	 * fail EDQUOT for metdata, but we do account for it.
681 	 */
682 	if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
683 		dquot_alloc_block_nofail(inode,
684 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
685 	}
686 	return ret;
687 }
688 
689 /**
690  * ext4_count_free_clusters() -- count filesystem free clusters
691  * @sb:		superblock
692  *
693  * Adds up the number of free clusters from each block group.
694  */
695 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
696 {
697 	ext4_fsblk_t desc_count;
698 	struct ext4_group_desc *gdp;
699 	ext4_group_t i;
700 	ext4_group_t ngroups = ext4_get_groups_count(sb);
701 	struct ext4_group_info *grp;
702 #ifdef EXT4FS_DEBUG
703 	struct ext4_super_block *es;
704 	ext4_fsblk_t bitmap_count;
705 	unsigned int x;
706 	struct buffer_head *bitmap_bh = NULL;
707 
708 	es = EXT4_SB(sb)->s_es;
709 	desc_count = 0;
710 	bitmap_count = 0;
711 	gdp = NULL;
712 
713 	for (i = 0; i < ngroups; i++) {
714 		gdp = ext4_get_group_desc(sb, i, NULL);
715 		if (!gdp)
716 			continue;
717 		grp = NULL;
718 		if (EXT4_SB(sb)->s_group_info)
719 			grp = ext4_get_group_info(sb, i);
720 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
721 			desc_count += ext4_free_group_clusters(sb, gdp);
722 		brelse(bitmap_bh);
723 		bitmap_bh = ext4_read_block_bitmap(sb, i);
724 		if (IS_ERR(bitmap_bh)) {
725 			bitmap_bh = NULL;
726 			continue;
727 		}
728 
729 		x = ext4_count_free(bitmap_bh->b_data,
730 				    EXT4_CLUSTERS_PER_GROUP(sb) / 8);
731 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
732 			i, ext4_free_group_clusters(sb, gdp), x);
733 		bitmap_count += x;
734 	}
735 	brelse(bitmap_bh);
736 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
737 	       ", computed = %llu, %llu\n",
738 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
739 	       desc_count, bitmap_count);
740 	return bitmap_count;
741 #else
742 	desc_count = 0;
743 	for (i = 0; i < ngroups; i++) {
744 		gdp = ext4_get_group_desc(sb, i, NULL);
745 		if (!gdp)
746 			continue;
747 		grp = NULL;
748 		if (EXT4_SB(sb)->s_group_info)
749 			grp = ext4_get_group_info(sb, i);
750 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
751 			desc_count += ext4_free_group_clusters(sb, gdp);
752 	}
753 
754 	return desc_count;
755 #endif
756 }
757 
758 static inline int test_root(ext4_group_t a, int b)
759 {
760 	while (1) {
761 		if (a < b)
762 			return 0;
763 		if (a == b)
764 			return 1;
765 		if ((a % b) != 0)
766 			return 0;
767 		a = a / b;
768 	}
769 }
770 
771 /**
772  *	ext4_bg_has_super - number of blocks used by the superblock in group
773  *	@sb: superblock for filesystem
774  *	@group: group number to check
775  *
776  *	Return the number of blocks used by the superblock (primary or backup)
777  *	in this group.  Currently this will be only 0 or 1.
778  */
779 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
780 {
781 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
782 
783 	if (group == 0)
784 		return 1;
785 	if (ext4_has_feature_sparse_super2(sb)) {
786 		if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
787 		    group == le32_to_cpu(es->s_backup_bgs[1]))
788 			return 1;
789 		return 0;
790 	}
791 	if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
792 		return 1;
793 	if (!(group & 1))
794 		return 0;
795 	if (test_root(group, 3) || (test_root(group, 5)) ||
796 	    test_root(group, 7))
797 		return 1;
798 
799 	return 0;
800 }
801 
802 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
803 					ext4_group_t group)
804 {
805 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
806 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
807 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
808 
809 	if (group == first || group == first + 1 || group == last)
810 		return 1;
811 	return 0;
812 }
813 
814 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
815 					ext4_group_t group)
816 {
817 	if (!ext4_bg_has_super(sb, group))
818 		return 0;
819 
820 	if (ext4_has_feature_meta_bg(sb))
821 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
822 	else
823 		return EXT4_SB(sb)->s_gdb_count;
824 }
825 
826 /**
827  *	ext4_bg_num_gdb - number of blocks used by the group table in group
828  *	@sb: superblock for filesystem
829  *	@group: group number to check
830  *
831  *	Return the number of blocks used by the group descriptor table
832  *	(primary or backup) in this group.  In the future there may be a
833  *	different number of descriptor blocks in each group.
834  */
835 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
836 {
837 	unsigned long first_meta_bg =
838 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
839 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
840 
841 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
842 		return ext4_bg_num_gdb_nometa(sb, group);
843 
844 	return ext4_bg_num_gdb_meta(sb,group);
845 
846 }
847 
848 /*
849  * This function returns the number of file system metadata clusters at
850  * the beginning of a block group, including the reserved gdt blocks.
851  */
852 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
853 				     ext4_group_t block_group)
854 {
855 	struct ext4_sb_info *sbi = EXT4_SB(sb);
856 	unsigned num;
857 
858 	/* Check for superblock and gdt backups in this group */
859 	num = ext4_bg_has_super(sb, block_group);
860 
861 	if (!ext4_has_feature_meta_bg(sb) ||
862 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
863 			  sbi->s_desc_per_block) {
864 		if (num) {
865 			num += ext4_bg_num_gdb(sb, block_group);
866 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
867 		}
868 	} else { /* For META_BG_BLOCK_GROUPS */
869 		num += ext4_bg_num_gdb(sb, block_group);
870 	}
871 	return EXT4_NUM_B2C(sbi, num);
872 }
873 /**
874  *	ext4_inode_to_goal_block - return a hint for block allocation
875  *	@inode: inode for block allocation
876  *
877  *	Return the ideal location to start allocating blocks for a
878  *	newly created inode.
879  */
880 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
881 {
882 	struct ext4_inode_info *ei = EXT4_I(inode);
883 	ext4_group_t block_group;
884 	ext4_grpblk_t colour;
885 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
886 	ext4_fsblk_t bg_start;
887 	ext4_fsblk_t last_block;
888 
889 	block_group = ei->i_block_group;
890 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
891 		/*
892 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
893 		 * block groups per flexgroup, reserve the first block
894 		 * group for directories and special files.  Regular
895 		 * files will start at the second block group.  This
896 		 * tends to speed up directory access and improves
897 		 * fsck times.
898 		 */
899 		block_group &= ~(flex_size-1);
900 		if (S_ISREG(inode->i_mode))
901 			block_group++;
902 	}
903 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
904 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
905 
906 	/*
907 	 * If we are doing delayed allocation, we don't need take
908 	 * colour into account.
909 	 */
910 	if (test_opt(inode->i_sb, DELALLOC))
911 		return bg_start;
912 
913 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
914 		colour = (task_pid_nr(current) % 16) *
915 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
916 	else
917 		colour = (task_pid_nr(current) % 16) *
918 			((last_block - bg_start) / 16);
919 	return bg_start + colour;
920 }
921 
922