xref: /linux/fs/ext4/balloc.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 #include <kunit/static_stub.h>
26 
27 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
28 					    ext4_group_t block_group);
29 /*
30  * balloc.c contains the blocks allocation and deallocation routines
31  */
32 
33 /*
34  * Calculate block group number for a given block number
35  */
36 ext4_group_t ext4_get_group_number(struct super_block *sb,
37 				   ext4_fsblk_t block)
38 {
39 	ext4_group_t group;
40 
41 	if (test_opt2(sb, STD_GROUP_SIZE))
42 		group = (block -
43 			 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >>
44 			(EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3);
45 	else
46 		ext4_get_group_no_and_offset(sb, block, &group, NULL);
47 	return group;
48 }
49 
50 /*
51  * Calculate the block group number and offset into the block/cluster
52  * allocation bitmap, given a block number
53  */
54 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
55 		ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
56 {
57 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
58 	ext4_grpblk_t offset;
59 
60 	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
61 	offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
62 		EXT4_SB(sb)->s_cluster_bits;
63 	if (offsetp)
64 		*offsetp = offset;
65 	if (blockgrpp)
66 		*blockgrpp = blocknr;
67 
68 }
69 
70 /*
71  * Check whether the 'block' lives within the 'block_group'. Returns 1 if so
72  * and 0 otherwise.
73  */
74 static inline int ext4_block_in_group(struct super_block *sb,
75 				      ext4_fsblk_t block,
76 				      ext4_group_t block_group)
77 {
78 	ext4_group_t actual_group;
79 
80 	actual_group = ext4_get_group_number(sb, block);
81 	return (actual_group == block_group) ? 1 : 0;
82 }
83 
84 /*
85  * Return the number of clusters used for file system metadata; this
86  * represents the overhead needed by the file system.
87  */
88 static unsigned ext4_num_overhead_clusters(struct super_block *sb,
89 					   ext4_group_t block_group,
90 					   struct ext4_group_desc *gdp)
91 {
92 	unsigned base_clusters, num_clusters;
93 	int block_cluster = -1, inode_cluster;
94 	int itbl_cluster_start = -1, itbl_cluster_end = -1;
95 	ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
96 	ext4_fsblk_t end = start + EXT4_BLOCKS_PER_GROUP(sb) - 1;
97 	ext4_fsblk_t itbl_blk_start, itbl_blk_end;
98 	struct ext4_sb_info *sbi = EXT4_SB(sb);
99 
100 	/* This is the number of clusters used by the superblock,
101 	 * block group descriptors, and reserved block group
102 	 * descriptor blocks */
103 	base_clusters = ext4_num_base_meta_clusters(sb, block_group);
104 	num_clusters = base_clusters;
105 
106 	/*
107 	 * Account and record inode table clusters if any cluster
108 	 * is in the block group, or inode table cluster range is
109 	 * [-1, -1] and won't overlap with block/inode bitmap cluster
110 	 * accounted below.
111 	 */
112 	itbl_blk_start = ext4_inode_table(sb, gdp);
113 	itbl_blk_end = itbl_blk_start + sbi->s_itb_per_group - 1;
114 	if (itbl_blk_start <= end && itbl_blk_end >= start) {
115 		itbl_blk_start = max(itbl_blk_start, start);
116 		itbl_blk_end = min(itbl_blk_end, end);
117 
118 		itbl_cluster_start = EXT4_B2C(sbi, itbl_blk_start - start);
119 		itbl_cluster_end = EXT4_B2C(sbi, itbl_blk_end - start);
120 
121 		num_clusters += itbl_cluster_end - itbl_cluster_start + 1;
122 		/* check if border cluster is overlapped */
123 		if (itbl_cluster_start == base_clusters - 1)
124 			num_clusters--;
125 	}
126 
127 	/*
128 	 * For the allocation bitmaps, we first need to check to see
129 	 * if the block is in the block group.  If it is, then check
130 	 * to see if the cluster is already accounted for in the clusters
131 	 * used for the base metadata cluster and inode tables cluster.
132 	 * Normally all of these blocks are contiguous, so the special
133 	 * case handling shouldn't be necessary except for *very*
134 	 * unusual file system layouts.
135 	 */
136 	if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
137 		block_cluster = EXT4_B2C(sbi,
138 					 ext4_block_bitmap(sb, gdp) - start);
139 		if (block_cluster >= base_clusters &&
140 		    (block_cluster < itbl_cluster_start ||
141 		    block_cluster > itbl_cluster_end))
142 			num_clusters++;
143 	}
144 
145 	if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
146 		inode_cluster = EXT4_B2C(sbi,
147 					 ext4_inode_bitmap(sb, gdp) - start);
148 		/*
149 		 * Additional check if inode bitmap is in just accounted
150 		 * block_cluster
151 		 */
152 		if (inode_cluster != block_cluster &&
153 		    inode_cluster >= base_clusters &&
154 		    (inode_cluster < itbl_cluster_start ||
155 		    inode_cluster > itbl_cluster_end))
156 			num_clusters++;
157 	}
158 
159 	return num_clusters;
160 }
161 
162 static unsigned int num_clusters_in_group(struct super_block *sb,
163 					  ext4_group_t block_group)
164 {
165 	unsigned int blocks;
166 
167 	if (block_group == ext4_get_groups_count(sb) - 1) {
168 		/*
169 		 * Even though mke2fs always initializes the first and
170 		 * last group, just in case some other tool was used,
171 		 * we need to make sure we calculate the right free
172 		 * blocks.
173 		 */
174 		blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
175 			ext4_group_first_block_no(sb, block_group);
176 	} else
177 		blocks = EXT4_BLOCKS_PER_GROUP(sb);
178 	return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
179 }
180 
181 /* Initializes an uninitialized block bitmap */
182 static int ext4_init_block_bitmap(struct super_block *sb,
183 				   struct buffer_head *bh,
184 				   ext4_group_t block_group,
185 				   struct ext4_group_desc *gdp)
186 {
187 	unsigned int bit, bit_max;
188 	struct ext4_sb_info *sbi = EXT4_SB(sb);
189 	ext4_fsblk_t start, tmp;
190 
191 	ASSERT(buffer_locked(bh));
192 
193 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
194 		ext4_mark_group_bitmap_corrupted(sb, block_group,
195 					EXT4_GROUP_INFO_BBITMAP_CORRUPT |
196 					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
197 		return -EFSBADCRC;
198 	}
199 	memset(bh->b_data, 0, sb->s_blocksize);
200 
201 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
202 	if ((bit_max >> 3) >= bh->b_size)
203 		return -EFSCORRUPTED;
204 
205 	for (bit = 0; bit < bit_max; bit++)
206 		ext4_set_bit(bit, bh->b_data);
207 
208 	start = ext4_group_first_block_no(sb, block_group);
209 
210 	/* Set bits for block and inode bitmaps, and inode table */
211 	tmp = ext4_block_bitmap(sb, gdp);
212 	if (ext4_block_in_group(sb, tmp, block_group))
213 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
214 
215 	tmp = ext4_inode_bitmap(sb, gdp);
216 	if (ext4_block_in_group(sb, tmp, block_group))
217 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
218 
219 	tmp = ext4_inode_table(sb, gdp);
220 	for (; tmp < ext4_inode_table(sb, gdp) +
221 		     sbi->s_itb_per_group; tmp++) {
222 		if (ext4_block_in_group(sb, tmp, block_group))
223 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
224 	}
225 
226 	/*
227 	 * Also if the number of blocks within the group is less than
228 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
229 	 * of the block bitmap to 1
230 	 */
231 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
232 			     sb->s_blocksize * 8, bh->b_data);
233 	return 0;
234 }
235 
236 /* Return the number of free blocks in a block group.  It is used when
237  * the block bitmap is uninitialized, so we can't just count the bits
238  * in the bitmap. */
239 unsigned ext4_free_clusters_after_init(struct super_block *sb,
240 				       ext4_group_t block_group,
241 				       struct ext4_group_desc *gdp)
242 {
243 	return num_clusters_in_group(sb, block_group) -
244 		ext4_num_overhead_clusters(sb, block_group, gdp);
245 }
246 
247 /*
248  * The free blocks are managed by bitmaps.  A file system contains several
249  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
250  * block for inodes, N blocks for the inode table and data blocks.
251  *
252  * The file system contains group descriptors which are located after the
253  * super block.  Each descriptor contains the number of the bitmap block and
254  * the free blocks count in the block.  The descriptors are loaded in memory
255  * when a file system is mounted (see ext4_fill_super).
256  */
257 
258 /**
259  * ext4_get_group_desc() -- load group descriptor from disk
260  * @sb:			super block
261  * @block_group:	given block group
262  * @bh:			pointer to the buffer head to store the block
263  *			group descriptor
264  */
265 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
266 					     ext4_group_t block_group,
267 					     struct buffer_head **bh)
268 {
269 	unsigned int group_desc;
270 	unsigned int offset;
271 	ext4_group_t ngroups = ext4_get_groups_count(sb);
272 	struct ext4_group_desc *desc;
273 	struct ext4_sb_info *sbi = EXT4_SB(sb);
274 	struct buffer_head *bh_p;
275 
276 	KUNIT_STATIC_STUB_REDIRECT(ext4_get_group_desc,
277 				   sb, block_group, bh);
278 
279 	if (block_group >= ngroups) {
280 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
281 			   " groups_count = %u", block_group, ngroups);
282 
283 		return NULL;
284 	}
285 
286 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
287 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
288 	bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc);
289 	/*
290 	 * sbi_array_rcu_deref returns with rcu unlocked, this is ok since
291 	 * the pointer being dereferenced won't be dereferenced again. By
292 	 * looking at the usage in add_new_gdb() the value isn't modified,
293 	 * just the pointer, and so it remains valid.
294 	 */
295 	if (!bh_p) {
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 *)bh_p->b_data +
304 		offset * EXT4_DESC_SIZE(sb));
305 	if (bh)
306 		*bh = bh_p;
307 	return desc;
308 }
309 
310 static ext4_fsblk_t ext4_valid_block_bitmap_padding(struct super_block *sb,
311 						    ext4_group_t block_group,
312 						    struct buffer_head *bh)
313 {
314 	ext4_grpblk_t next_zero_bit;
315 	unsigned long bitmap_size = sb->s_blocksize * 8;
316 	unsigned int offset = num_clusters_in_group(sb, block_group);
317 
318 	if (bitmap_size <= offset)
319 		return 0;
320 
321 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data, bitmap_size, offset);
322 
323 	return (next_zero_bit < bitmap_size ? next_zero_bit : 0);
324 }
325 
326 struct ext4_group_info *ext4_get_group_info(struct super_block *sb,
327 					    ext4_group_t group)
328 {
329 	struct ext4_group_info **grp_info;
330 	long indexv, indexh;
331 
332 	if (unlikely(group >= EXT4_SB(sb)->s_groups_count))
333 		return NULL;
334 	indexv = group >> (EXT4_DESC_PER_BLOCK_BITS(sb));
335 	indexh = group & ((EXT4_DESC_PER_BLOCK(sb)) - 1);
336 	grp_info = sbi_array_rcu_deref(EXT4_SB(sb), s_group_info, indexv);
337 	return grp_info[indexh];
338 }
339 
340 /*
341  * Return the block number which was discovered to be invalid, or 0 if
342  * the block bitmap is valid.
343  */
344 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
345 					    struct ext4_group_desc *desc,
346 					    ext4_group_t block_group,
347 					    struct buffer_head *bh)
348 {
349 	struct ext4_sb_info *sbi = EXT4_SB(sb);
350 	ext4_grpblk_t offset;
351 	ext4_grpblk_t next_zero_bit;
352 	ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
353 	ext4_fsblk_t blk;
354 	ext4_fsblk_t group_first_block;
355 
356 	if (ext4_has_feature_flex_bg(sb)) {
357 		/* with FLEX_BG, the inode/block bitmaps and itable
358 		 * blocks may not be in the group at all
359 		 * so the bitmap validation will be skipped for those groups
360 		 * or it has to also read the block group where the bitmaps
361 		 * are located to verify they are set.
362 		 */
363 		return 0;
364 	}
365 	group_first_block = ext4_group_first_block_no(sb, block_group);
366 
367 	/* check whether block bitmap block number is set */
368 	blk = ext4_block_bitmap(sb, desc);
369 	offset = blk - group_first_block;
370 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
371 	    !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
372 		/* bad block bitmap */
373 		return blk;
374 
375 	/* check whether the inode bitmap block number is set */
376 	blk = ext4_inode_bitmap(sb, desc);
377 	offset = blk - group_first_block;
378 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
379 	    !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
380 		/* bad block bitmap */
381 		return blk;
382 
383 	/* check whether the inode table block number is set */
384 	blk = ext4_inode_table(sb, desc);
385 	offset = blk - group_first_block;
386 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
387 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) >= max_bit)
388 		return blk;
389 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
390 			EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) + 1,
391 			EXT4_B2C(sbi, offset));
392 	if (next_zero_bit <
393 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) + 1)
394 		/* bad bitmap for inode tables */
395 		return blk;
396 	return 0;
397 }
398 
399 static int ext4_validate_block_bitmap(struct super_block *sb,
400 				      struct ext4_group_desc *desc,
401 				      ext4_group_t block_group,
402 				      struct buffer_head *bh)
403 {
404 	ext4_fsblk_t	blk;
405 	struct ext4_group_info *grp;
406 
407 	if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
408 		return 0;
409 
410 	grp = ext4_get_group_info(sb, block_group);
411 
412 	if (buffer_verified(bh))
413 		return 0;
414 	if (!grp || EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
415 		return -EFSCORRUPTED;
416 
417 	ext4_lock_group(sb, block_group);
418 	if (buffer_verified(bh))
419 		goto verified;
420 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, desc, bh) ||
421 		     ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) {
422 		ext4_unlock_group(sb, block_group);
423 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
424 		ext4_mark_group_bitmap_corrupted(sb, block_group,
425 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
426 		return -EFSBADCRC;
427 	}
428 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
429 	if (unlikely(blk != 0)) {
430 		ext4_unlock_group(sb, block_group);
431 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
432 			   block_group, blk);
433 		ext4_mark_group_bitmap_corrupted(sb, block_group,
434 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
435 		return -EFSCORRUPTED;
436 	}
437 	blk = ext4_valid_block_bitmap_padding(sb, block_group, bh);
438 	if (unlikely(blk != 0)) {
439 		ext4_unlock_group(sb, block_group);
440 		ext4_error(sb, "bg %u: block %llu: padding at end of block bitmap is not set",
441 			   block_group, blk);
442 		ext4_mark_group_bitmap_corrupted(sb, block_group,
443 						 EXT4_GROUP_INFO_BBITMAP_CORRUPT);
444 		return -EFSCORRUPTED;
445 	}
446 	set_buffer_verified(bh);
447 verified:
448 	ext4_unlock_group(sb, block_group);
449 	return 0;
450 }
451 
452 /**
453  * ext4_read_block_bitmap_nowait()
454  * @sb:			super block
455  * @block_group:	given block group
456  * @ignore_locked:	ignore locked buffers
457  *
458  * Read the bitmap for a given block_group,and validate the
459  * bits for block/inode/inode tables are set in the bitmaps
460  *
461  * Return buffer_head on success or an ERR_PTR in case of failure.
462  */
463 struct buffer_head *
464 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group,
465 			      bool ignore_locked)
466 {
467 	struct ext4_group_desc *desc;
468 	struct ext4_sb_info *sbi = EXT4_SB(sb);
469 	struct buffer_head *bh;
470 	ext4_fsblk_t bitmap_blk;
471 	int err;
472 
473 	KUNIT_STATIC_STUB_REDIRECT(ext4_read_block_bitmap_nowait,
474 				   sb, block_group, ignore_locked);
475 
476 	desc = ext4_get_group_desc(sb, block_group, NULL);
477 	if (!desc)
478 		return ERR_PTR(-EFSCORRUPTED);
479 	bitmap_blk = ext4_block_bitmap(sb, desc);
480 	if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
481 	    (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
482 		ext4_error(sb, "Invalid block bitmap block %llu in "
483 			   "block_group %u", bitmap_blk, block_group);
484 		ext4_mark_group_bitmap_corrupted(sb, block_group,
485 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
486 		return ERR_PTR(-EFSCORRUPTED);
487 	}
488 	bh = sb_getblk(sb, bitmap_blk);
489 	if (unlikely(!bh)) {
490 		ext4_warning(sb, "Cannot get buffer for block bitmap - "
491 			     "block_group = %u, block_bitmap = %llu",
492 			     block_group, bitmap_blk);
493 		return ERR_PTR(-ENOMEM);
494 	}
495 
496 	if (ignore_locked && buffer_locked(bh)) {
497 		/* buffer under IO already, return if called for prefetching */
498 		put_bh(bh);
499 		return NULL;
500 	}
501 
502 	if (bitmap_uptodate(bh))
503 		goto verify;
504 
505 	lock_buffer(bh);
506 	if (bitmap_uptodate(bh)) {
507 		unlock_buffer(bh);
508 		goto verify;
509 	}
510 	ext4_lock_group(sb, block_group);
511 	if (ext4_has_group_desc_csum(sb) &&
512 	    (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
513 		if (block_group == 0) {
514 			ext4_unlock_group(sb, block_group);
515 			unlock_buffer(bh);
516 			ext4_error(sb, "Block bitmap for bg 0 marked "
517 				   "uninitialized");
518 			err = -EFSCORRUPTED;
519 			goto out;
520 		}
521 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
522 		if (err) {
523 			ext4_unlock_group(sb, block_group);
524 			unlock_buffer(bh);
525 			ext4_error(sb, "Failed to init block bitmap for group "
526 				   "%u: %d", block_group, err);
527 			goto out;
528 		}
529 		set_bitmap_uptodate(bh);
530 		set_buffer_uptodate(bh);
531 		set_buffer_verified(bh);
532 		ext4_unlock_group(sb, block_group);
533 		unlock_buffer(bh);
534 		return bh;
535 	}
536 	ext4_unlock_group(sb, block_group);
537 	if (buffer_uptodate(bh)) {
538 		/*
539 		 * if not uninit if bh is uptodate,
540 		 * bitmap is also uptodate
541 		 */
542 		set_bitmap_uptodate(bh);
543 		unlock_buffer(bh);
544 		goto verify;
545 	}
546 	/*
547 	 * submit the buffer_head for reading
548 	 */
549 	set_buffer_new(bh);
550 	trace_ext4_read_block_bitmap_load(sb, block_group, ignore_locked);
551 	ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO |
552 			    (ignore_locked ? REQ_RAHEAD : 0),
553 			    ext4_end_bitmap_read,
554 			    ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_EIO));
555 	return bh;
556 verify:
557 	err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
558 	if (err)
559 		goto out;
560 	return bh;
561 out:
562 	put_bh(bh);
563 	return ERR_PTR(err);
564 }
565 
566 /* Returns 0 on success, -errno on error */
567 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
568 			   struct buffer_head *bh)
569 {
570 	struct ext4_group_desc *desc;
571 
572 	KUNIT_STATIC_STUB_REDIRECT(ext4_wait_block_bitmap,
573 				   sb, block_group, bh);
574 
575 	if (!buffer_new(bh))
576 		return 0;
577 	desc = ext4_get_group_desc(sb, block_group, NULL);
578 	if (!desc)
579 		return -EFSCORRUPTED;
580 	wait_on_buffer(bh);
581 	if (!buffer_uptodate(bh)) {
582 		ext4_error_err(sb, EIO, "Cannot read block bitmap - "
583 			       "block_group = %u, block_bitmap = %llu",
584 			       block_group, (unsigned long long) bh->b_blocknr);
585 		ext4_mark_group_bitmap_corrupted(sb, block_group,
586 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
587 		return -EIO;
588 	}
589 	clear_buffer_new(bh);
590 	/* Panic or remount fs read-only if block bitmap is invalid */
591 	return ext4_validate_block_bitmap(sb, desc, block_group, bh);
592 }
593 
594 struct buffer_head *
595 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
596 {
597 	struct buffer_head *bh;
598 	int err;
599 
600 	bh = ext4_read_block_bitmap_nowait(sb, block_group, false);
601 	if (IS_ERR(bh))
602 		return bh;
603 	err = ext4_wait_block_bitmap(sb, block_group, bh);
604 	if (err) {
605 		put_bh(bh);
606 		return ERR_PTR(err);
607 	}
608 	return bh;
609 }
610 
611 /**
612  * ext4_has_free_clusters()
613  * @sbi:	in-core super block structure.
614  * @nclusters:	number of needed blocks
615  * @flags:	flags from ext4_mb_new_blocks()
616  *
617  * Check if filesystem has nclusters free & available for allocation.
618  * On success return 1, return 0 on failure.
619  */
620 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
621 				  s64 nclusters, unsigned int flags)
622 {
623 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
624 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
625 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
626 
627 	free_clusters  = percpu_counter_read_positive(fcc);
628 	dirty_clusters = percpu_counter_read_positive(dcc);
629 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
630 
631 	/*
632 	 * r_blocks_count should always be multiple of the cluster ratio so
633 	 * we are safe to do a plane bit shift only.
634 	 */
635 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
636 	      resv_clusters;
637 
638 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
639 					EXT4_FREECLUSTERS_WATERMARK) {
640 		free_clusters  = percpu_counter_sum_positive(fcc);
641 		dirty_clusters = percpu_counter_sum_positive(dcc);
642 	}
643 	/* Check whether we have space after accounting for current
644 	 * dirty clusters & root reserved clusters.
645 	 */
646 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
647 		return 1;
648 
649 	/* Hm, nope.  Are (enough) root reserved clusters available? */
650 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
651 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
652 	    capable(CAP_SYS_RESOURCE) ||
653 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
654 
655 		if (free_clusters >= (nclusters + dirty_clusters +
656 				      resv_clusters))
657 			return 1;
658 	}
659 	/* No free blocks. Let's see if we can dip into reserved pool */
660 	if (flags & EXT4_MB_USE_RESERVED) {
661 		if (free_clusters >= (nclusters + dirty_clusters))
662 			return 1;
663 	}
664 
665 	return 0;
666 }
667 
668 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
669 			     s64 nclusters, unsigned int flags)
670 {
671 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
672 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
673 		return 0;
674 	} else
675 		return -ENOSPC;
676 }
677 
678 /**
679  * ext4_should_retry_alloc() - check if a block allocation should be retried
680  * @sb:			superblock
681  * @retries:		number of retry attempts made so far
682  *
683  * ext4_should_retry_alloc() is called when ENOSPC is returned while
684  * attempting to allocate blocks.  If there's an indication that a pending
685  * journal transaction might free some space and allow another attempt to
686  * succeed, this function will wait for the current or committing transaction
687  * to complete and then return TRUE.
688  */
689 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
690 {
691 	struct ext4_sb_info *sbi = EXT4_SB(sb);
692 
693 	if (!sbi->s_journal)
694 		return 0;
695 
696 	if (++(*retries) > 3) {
697 		percpu_counter_inc(&sbi->s_sra_exceeded_retry_limit);
698 		return 0;
699 	}
700 
701 	/*
702 	 * if there's no indication that blocks are about to be freed it's
703 	 * possible we just missed a transaction commit that did so
704 	 */
705 	smp_mb();
706 	if (sbi->s_mb_free_pending == 0) {
707 		if (test_opt(sb, DISCARD)) {
708 			atomic_inc(&sbi->s_retry_alloc_pending);
709 			flush_work(&sbi->s_discard_work);
710 			atomic_dec(&sbi->s_retry_alloc_pending);
711 		}
712 		return ext4_has_free_clusters(sbi, 1, 0);
713 	}
714 
715 	/*
716 	 * it's possible we've just missed a transaction commit here,
717 	 * so ignore the returned status
718 	 */
719 	ext4_debug("%s: retrying operation after ENOSPC\n", sb->s_id);
720 	(void) jbd2_journal_force_commit_nested(sbi->s_journal);
721 	return 1;
722 }
723 
724 /*
725  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
726  *
727  * @handle:             handle to this transaction
728  * @inode:              file inode
729  * @goal:               given target block(filesystem wide)
730  * @count:		pointer to total number of clusters needed
731  * @errp:               error code
732  *
733  * Return 1st allocated block number on success, *count stores total account
734  * error stores in errp pointer
735  */
736 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
737 				  ext4_fsblk_t goal, unsigned int flags,
738 				  unsigned long *count, int *errp)
739 {
740 	struct ext4_allocation_request ar;
741 	ext4_fsblk_t ret;
742 
743 	memset(&ar, 0, sizeof(ar));
744 	/* Fill with neighbour allocated blocks */
745 	ar.inode = inode;
746 	ar.goal = goal;
747 	ar.len = count ? *count : 1;
748 	ar.flags = flags;
749 
750 	ret = ext4_mb_new_blocks(handle, &ar, errp);
751 	if (count)
752 		*count = ar.len;
753 	/*
754 	 * Account for the allocated meta blocks.  We will never
755 	 * fail EDQUOT for metdata, but we do account for it.
756 	 */
757 	if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
758 		dquot_alloc_block_nofail(inode,
759 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
760 	}
761 	return ret;
762 }
763 
764 /**
765  * ext4_count_free_clusters() -- count filesystem free clusters
766  * @sb:		superblock
767  *
768  * Adds up the number of free clusters from each block group.
769  */
770 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
771 {
772 	ext4_fsblk_t desc_count;
773 	struct ext4_group_desc *gdp;
774 	ext4_group_t i;
775 	ext4_group_t ngroups = ext4_get_groups_count(sb);
776 	struct ext4_group_info *grp;
777 #ifdef EXT4FS_DEBUG
778 	struct ext4_super_block *es;
779 	ext4_fsblk_t bitmap_count;
780 	unsigned int x;
781 	struct buffer_head *bitmap_bh = NULL;
782 
783 	es = EXT4_SB(sb)->s_es;
784 	desc_count = 0;
785 	bitmap_count = 0;
786 	gdp = NULL;
787 
788 	for (i = 0; i < ngroups; i++) {
789 		gdp = ext4_get_group_desc(sb, i, NULL);
790 		if (!gdp)
791 			continue;
792 		grp = NULL;
793 		if (EXT4_SB(sb)->s_group_info)
794 			grp = ext4_get_group_info(sb, i);
795 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
796 			desc_count += ext4_free_group_clusters(sb, gdp);
797 		brelse(bitmap_bh);
798 		bitmap_bh = ext4_read_block_bitmap(sb, i);
799 		if (IS_ERR(bitmap_bh)) {
800 			bitmap_bh = NULL;
801 			continue;
802 		}
803 
804 		x = ext4_count_free(bitmap_bh->b_data,
805 				    EXT4_CLUSTERS_PER_GROUP(sb) / 8);
806 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
807 			i, ext4_free_group_clusters(sb, gdp), x);
808 		bitmap_count += x;
809 	}
810 	brelse(bitmap_bh);
811 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
812 	       ", computed = %llu, %llu\n",
813 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
814 	       desc_count, bitmap_count);
815 	return bitmap_count;
816 #else
817 	desc_count = 0;
818 	for (i = 0; i < ngroups; i++) {
819 		gdp = ext4_get_group_desc(sb, i, NULL);
820 		if (!gdp)
821 			continue;
822 		grp = NULL;
823 		if (EXT4_SB(sb)->s_group_info)
824 			grp = ext4_get_group_info(sb, i);
825 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
826 			desc_count += ext4_free_group_clusters(sb, gdp);
827 	}
828 
829 	return desc_count;
830 #endif
831 }
832 
833 static inline int test_root(ext4_group_t a, int b)
834 {
835 	while (1) {
836 		if (a < b)
837 			return 0;
838 		if (a == b)
839 			return 1;
840 		if ((a % b) != 0)
841 			return 0;
842 		a = a / b;
843 	}
844 }
845 
846 /**
847  *	ext4_bg_has_super - number of blocks used by the superblock in group
848  *	@sb: superblock for filesystem
849  *	@group: group number to check
850  *
851  *	Return the number of blocks used by the superblock (primary or backup)
852  *	in this group.  Currently this will be only 0 or 1.
853  */
854 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
855 {
856 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
857 
858 	if (group == 0)
859 		return 1;
860 	if (ext4_has_feature_sparse_super2(sb)) {
861 		if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
862 		    group == le32_to_cpu(es->s_backup_bgs[1]))
863 			return 1;
864 		return 0;
865 	}
866 	if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
867 		return 1;
868 	if (!(group & 1))
869 		return 0;
870 	if (test_root(group, 3) || (test_root(group, 5)) ||
871 	    test_root(group, 7))
872 		return 1;
873 
874 	return 0;
875 }
876 
877 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
878 					ext4_group_t group)
879 {
880 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
881 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
882 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
883 
884 	if (group == first || group == first + 1 || group == last)
885 		return 1;
886 	return 0;
887 }
888 
889 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
890 					ext4_group_t group)
891 {
892 	if (!ext4_bg_has_super(sb, group))
893 		return 0;
894 
895 	if (ext4_has_feature_meta_bg(sb))
896 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
897 	else
898 		return EXT4_SB(sb)->s_gdb_count;
899 }
900 
901 /**
902  *	ext4_bg_num_gdb - number of blocks used by the group table in group
903  *	@sb: superblock for filesystem
904  *	@group: group number to check
905  *
906  *	Return the number of blocks used by the group descriptor table
907  *	(primary or backup) in this group.  In the future there may be a
908  *	different number of descriptor blocks in each group.
909  */
910 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
911 {
912 	unsigned long first_meta_bg =
913 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
914 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
915 
916 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
917 		return ext4_bg_num_gdb_nometa(sb, group);
918 
919 	return ext4_bg_num_gdb_meta(sb,group);
920 
921 }
922 
923 /*
924  * This function returns the number of file system metadata blocks at
925  * the beginning of a block group, including the reserved gdt blocks.
926  */
927 unsigned int ext4_num_base_meta_blocks(struct super_block *sb,
928 				       ext4_group_t block_group)
929 {
930 	struct ext4_sb_info *sbi = EXT4_SB(sb);
931 	unsigned num;
932 
933 	/* Check for superblock and gdt backups in this group */
934 	num = ext4_bg_has_super(sb, block_group);
935 
936 	if (!ext4_has_feature_meta_bg(sb) ||
937 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
938 			  sbi->s_desc_per_block) {
939 		if (num) {
940 			num += ext4_bg_num_gdb_nometa(sb, block_group);
941 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
942 		}
943 	} else { /* For META_BG_BLOCK_GROUPS */
944 		num += ext4_bg_num_gdb_meta(sb, block_group);
945 	}
946 	return num;
947 }
948 
949 static unsigned int ext4_num_base_meta_clusters(struct super_block *sb,
950 						ext4_group_t block_group)
951 {
952 	return EXT4_NUM_B2C(EXT4_SB(sb), ext4_num_base_meta_blocks(sb, block_group));
953 }
954 
955 /**
956  *	ext4_inode_to_goal_block - return a hint for block allocation
957  *	@inode: inode for block allocation
958  *
959  *	Return the ideal location to start allocating blocks for a
960  *	newly created inode.
961  */
962 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
963 {
964 	struct ext4_inode_info *ei = EXT4_I(inode);
965 	ext4_group_t block_group;
966 	ext4_grpblk_t colour;
967 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
968 	ext4_fsblk_t bg_start;
969 	ext4_fsblk_t last_block;
970 
971 	block_group = ei->i_block_group;
972 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
973 		/*
974 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
975 		 * block groups per flexgroup, reserve the first block
976 		 * group for directories and special files.  Regular
977 		 * files will start at the second block group.  This
978 		 * tends to speed up directory access and improves
979 		 * fsck times.
980 		 */
981 		block_group &= ~(flex_size-1);
982 		if (S_ISREG(inode->i_mode))
983 			block_group++;
984 	}
985 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
986 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
987 
988 	/*
989 	 * If we are doing delayed allocation, we don't need take
990 	 * colour into account.
991 	 */
992 	if (test_opt(inode->i_sb, DELALLOC))
993 		return bg_start;
994 
995 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
996 		colour = (task_pid_nr(current) % 16) *
997 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
998 	else
999 		colour = (task_pid_nr(current) % 16) *
1000 			((last_block - bg_start) / 16);
1001 	return bg_start + colour;
1002 }
1003 
1004