xref: /linux/fs/ext4/balloc.c (revision 0804f3bec9e9a3e9f3b5431ac9a11417041bc4c2)
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 /*
84  * Return the number of clusters used for file system metadata; this
85  * represents the overhead needed by the file system.
86  */
87 static unsigned ext4_num_overhead_clusters(struct super_block *sb,
88 					   ext4_group_t block_group,
89 					   struct ext4_group_desc *gdp)
90 {
91 	unsigned base_clusters, num_clusters;
92 	int block_cluster = -1, inode_cluster;
93 	int itbl_cluster_start = -1, itbl_cluster_end = -1;
94 	ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
95 	ext4_fsblk_t end = start + EXT4_BLOCKS_PER_GROUP(sb) - 1;
96 	ext4_fsblk_t itbl_blk_start, itbl_blk_end;
97 	struct ext4_sb_info *sbi = EXT4_SB(sb);
98 
99 	/* This is the number of clusters used by the superblock,
100 	 * block group descriptors, and reserved block group
101 	 * descriptor blocks */
102 	base_clusters = ext4_num_base_meta_clusters(sb, block_group);
103 	num_clusters = base_clusters;
104 
105 	/*
106 	 * Account and record inode table clusters if any cluster
107 	 * is in the block group, or inode table cluster range is
108 	 * [-1, -1] and won't overlap with block/inode bitmap cluster
109 	 * accounted below.
110 	 */
111 	itbl_blk_start = ext4_inode_table(sb, gdp);
112 	itbl_blk_end = itbl_blk_start + sbi->s_itb_per_group - 1;
113 	if (itbl_blk_start <= end && itbl_blk_end >= start) {
114 		itbl_blk_start = itbl_blk_start >= start ?
115 			itbl_blk_start : start;
116 		itbl_blk_end = itbl_blk_end <= end ?
117 			itbl_blk_end : end;
118 
119 		itbl_cluster_start = EXT4_B2C(sbi, itbl_blk_start - start);
120 		itbl_cluster_end = EXT4_B2C(sbi, itbl_blk_end - start);
121 
122 		num_clusters += itbl_cluster_end - itbl_cluster_start + 1;
123 		/* check if border cluster is overlapped */
124 		if (itbl_cluster_start == base_clusters - 1)
125 			num_clusters--;
126 	}
127 
128 	/*
129 	 * For the allocation bitmaps, we first need to check to see
130 	 * if the block is in the block group.  If it is, then check
131 	 * to see if the cluster is already accounted for in the clusters
132 	 * used for the base metadata cluster and inode tables cluster.
133 	 * Normally all of these blocks are contiguous, so the special
134 	 * case handling shouldn't be necessary except for *very*
135 	 * unusual file system layouts.
136 	 */
137 	if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
138 		block_cluster = EXT4_B2C(sbi,
139 					 ext4_block_bitmap(sb, gdp) - start);
140 		if (block_cluster >= base_clusters &&
141 		    (block_cluster < itbl_cluster_start ||
142 		    block_cluster > itbl_cluster_end))
143 			num_clusters++;
144 	}
145 
146 	if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
147 		inode_cluster = EXT4_B2C(sbi,
148 					 ext4_inode_bitmap(sb, gdp) - start);
149 		/*
150 		 * Additional check if inode bitmap is in just accounted
151 		 * block_cluster
152 		 */
153 		if (inode_cluster != block_cluster &&
154 		    inode_cluster >= base_clusters &&
155 		    (inode_cluster < itbl_cluster_start ||
156 		    inode_cluster > itbl_cluster_end))
157 			num_clusters++;
158 	}
159 
160 	return num_clusters;
161 }
162 
163 static unsigned int num_clusters_in_group(struct super_block *sb,
164 					  ext4_group_t block_group)
165 {
166 	unsigned int blocks;
167 
168 	if (block_group == ext4_get_groups_count(sb) - 1) {
169 		/*
170 		 * Even though mke2fs always initializes the first and
171 		 * last group, just in case some other tool was used,
172 		 * we need to make sure we calculate the right free
173 		 * blocks.
174 		 */
175 		blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
176 			ext4_group_first_block_no(sb, block_group);
177 	} else
178 		blocks = EXT4_BLOCKS_PER_GROUP(sb);
179 	return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
180 }
181 
182 /* Initializes an uninitialized block bitmap */
183 static int ext4_init_block_bitmap(struct super_block *sb,
184 				   struct buffer_head *bh,
185 				   ext4_group_t block_group,
186 				   struct ext4_group_desc *gdp)
187 {
188 	unsigned int bit, bit_max;
189 	struct ext4_sb_info *sbi = EXT4_SB(sb);
190 	ext4_fsblk_t start, tmp;
191 
192 	ASSERT(buffer_locked(bh));
193 
194 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
195 		ext4_mark_group_bitmap_corrupted(sb, block_group,
196 					EXT4_GROUP_INFO_BBITMAP_CORRUPT |
197 					EXT4_GROUP_INFO_IBITMAP_CORRUPT);
198 		return -EFSBADCRC;
199 	}
200 	memset(bh->b_data, 0, sb->s_blocksize);
201 
202 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
203 	if ((bit_max >> 3) >= bh->b_size)
204 		return -EFSCORRUPTED;
205 
206 	for (bit = 0; bit < bit_max; bit++)
207 		ext4_set_bit(bit, bh->b_data);
208 
209 	start = ext4_group_first_block_no(sb, block_group);
210 
211 	/* Set bits for block and inode bitmaps, and inode table */
212 	tmp = ext4_block_bitmap(sb, gdp);
213 	if (ext4_block_in_group(sb, tmp, block_group))
214 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
215 
216 	tmp = ext4_inode_bitmap(sb, gdp);
217 	if (ext4_block_in_group(sb, tmp, block_group))
218 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
219 
220 	tmp = ext4_inode_table(sb, gdp);
221 	for (; tmp < ext4_inode_table(sb, gdp) +
222 		     sbi->s_itb_per_group; tmp++) {
223 		if (ext4_block_in_group(sb, tmp, block_group))
224 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
225 	}
226 
227 	/*
228 	 * Also if the number of blocks within the group is less than
229 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
230 	 * of the block bitmap to 1
231 	 */
232 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
233 			     sb->s_blocksize * 8, bh->b_data);
234 	return 0;
235 }
236 
237 /* Return the number of free blocks in a block group.  It is used when
238  * the block bitmap is uninitialized, so we can't just count the bits
239  * in the bitmap. */
240 unsigned ext4_free_clusters_after_init(struct super_block *sb,
241 				       ext4_group_t block_group,
242 				       struct ext4_group_desc *gdp)
243 {
244 	return num_clusters_in_group(sb, block_group) -
245 		ext4_num_overhead_clusters(sb, block_group, gdp);
246 }
247 
248 /*
249  * The free blocks are managed by bitmaps.  A file system contains several
250  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
251  * block for inodes, N blocks for the inode table and data blocks.
252  *
253  * The file system contains group descriptors which are located after the
254  * super block.  Each descriptor contains the number of the bitmap block and
255  * the free blocks count in the block.  The descriptors are loaded in memory
256  * when a file system is mounted (see ext4_fill_super).
257  */
258 
259 /**
260  * ext4_get_group_desc() -- load group descriptor from disk
261  * @sb:			super block
262  * @block_group:	given block group
263  * @bh:			pointer to the buffer head to store the block
264  *			group descriptor
265  */
266 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
267 					     ext4_group_t block_group,
268 					     struct buffer_head **bh)
269 {
270 	unsigned int group_desc;
271 	unsigned int offset;
272 	ext4_group_t ngroups = ext4_get_groups_count(sb);
273 	struct ext4_group_desc *desc;
274 	struct ext4_sb_info *sbi = EXT4_SB(sb);
275 	struct buffer_head *bh_p;
276 
277 	if (block_group >= ngroups) {
278 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
279 			   " groups_count = %u", block_group, ngroups);
280 
281 		return NULL;
282 	}
283 
284 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
285 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
286 	bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc);
287 	/*
288 	 * sbi_array_rcu_deref returns with rcu unlocked, this is ok since
289 	 * the pointer being dereferenced won't be dereferenced again. By
290 	 * looking at the usage in add_new_gdb() the value isn't modified,
291 	 * just the pointer, and so it remains valid.
292 	 */
293 	if (!bh_p) {
294 		ext4_error(sb, "Group descriptor not loaded - "
295 			   "block_group = %u, group_desc = %u, desc = %u",
296 			   block_group, group_desc, offset);
297 		return NULL;
298 	}
299 
300 	desc = (struct ext4_group_desc *)(
301 		(__u8 *)bh_p->b_data +
302 		offset * EXT4_DESC_SIZE(sb));
303 	if (bh)
304 		*bh = bh_p;
305 	return desc;
306 }
307 
308 static ext4_fsblk_t ext4_valid_block_bitmap_padding(struct super_block *sb,
309 						    ext4_group_t block_group,
310 						    struct buffer_head *bh)
311 {
312 	ext4_grpblk_t next_zero_bit;
313 	unsigned long bitmap_size = sb->s_blocksize * 8;
314 	unsigned int offset = num_clusters_in_group(sb, block_group);
315 
316 	if (bitmap_size <= offset)
317 		return 0;
318 
319 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data, bitmap_size, offset);
320 
321 	return (next_zero_bit < bitmap_size ? next_zero_bit : 0);
322 }
323 
324 struct ext4_group_info *ext4_get_group_info(struct super_block *sb,
325 					    ext4_group_t group)
326 {
327 	struct ext4_group_info **grp_info;
328 	long indexv, indexh;
329 
330 	if (unlikely(group >= EXT4_SB(sb)->s_groups_count))
331 		return NULL;
332 	indexv = group >> (EXT4_DESC_PER_BLOCK_BITS(sb));
333 	indexh = group & ((EXT4_DESC_PER_BLOCK(sb)) - 1);
334 	grp_info = sbi_array_rcu_deref(EXT4_SB(sb), s_group_info, indexv);
335 	return grp_info[indexh];
336 }
337 
338 /*
339  * Return the block number which was discovered to be invalid, or 0 if
340  * the block bitmap is valid.
341  */
342 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
343 					    struct ext4_group_desc *desc,
344 					    ext4_group_t block_group,
345 					    struct buffer_head *bh)
346 {
347 	struct ext4_sb_info *sbi = EXT4_SB(sb);
348 	ext4_grpblk_t offset;
349 	ext4_grpblk_t next_zero_bit;
350 	ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
351 	ext4_fsblk_t blk;
352 	ext4_fsblk_t group_first_block;
353 
354 	if (ext4_has_feature_flex_bg(sb)) {
355 		/* with FLEX_BG, the inode/block bitmaps and itable
356 		 * blocks may not be in the group at all
357 		 * so the bitmap validation will be skipped for those groups
358 		 * or it has to also read the block group where the bitmaps
359 		 * are located to verify they are set.
360 		 */
361 		return 0;
362 	}
363 	group_first_block = ext4_group_first_block_no(sb, block_group);
364 
365 	/* check whether block bitmap block number is set */
366 	blk = ext4_block_bitmap(sb, desc);
367 	offset = blk - group_first_block;
368 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
369 	    !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
370 		/* bad block bitmap */
371 		return blk;
372 
373 	/* check whether the inode bitmap block number is set */
374 	blk = ext4_inode_bitmap(sb, desc);
375 	offset = blk - group_first_block;
376 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
377 	    !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
378 		/* bad block bitmap */
379 		return blk;
380 
381 	/* check whether the inode table block number is set */
382 	blk = ext4_inode_table(sb, desc);
383 	offset = blk - group_first_block;
384 	if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
385 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) >= max_bit)
386 		return blk;
387 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
388 			EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) + 1,
389 			EXT4_B2C(sbi, offset));
390 	if (next_zero_bit <
391 	    EXT4_B2C(sbi, offset + sbi->s_itb_per_group - 1) + 1)
392 		/* bad bitmap for inode tables */
393 		return blk;
394 	return 0;
395 }
396 
397 static int ext4_validate_block_bitmap(struct super_block *sb,
398 				      struct ext4_group_desc *desc,
399 				      ext4_group_t block_group,
400 				      struct buffer_head *bh)
401 {
402 	ext4_fsblk_t	blk;
403 	struct ext4_group_info *grp;
404 
405 	if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
406 		return 0;
407 
408 	grp = ext4_get_group_info(sb, block_group);
409 
410 	if (buffer_verified(bh))
411 		return 0;
412 	if (!grp || EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
413 		return -EFSCORRUPTED;
414 
415 	ext4_lock_group(sb, block_group);
416 	if (buffer_verified(bh))
417 		goto verified;
418 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, desc, bh) ||
419 		     ext4_simulate_fail(sb, EXT4_SIM_BBITMAP_CRC))) {
420 		ext4_unlock_group(sb, block_group);
421 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
422 		ext4_mark_group_bitmap_corrupted(sb, block_group,
423 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
424 		return -EFSBADCRC;
425 	}
426 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
427 	if (unlikely(blk != 0)) {
428 		ext4_unlock_group(sb, block_group);
429 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
430 			   block_group, blk);
431 		ext4_mark_group_bitmap_corrupted(sb, block_group,
432 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
433 		return -EFSCORRUPTED;
434 	}
435 	blk = ext4_valid_block_bitmap_padding(sb, block_group, bh);
436 	if (unlikely(blk != 0)) {
437 		ext4_unlock_group(sb, block_group);
438 		ext4_error(sb, "bg %u: block %llu: padding at end of block bitmap is not set",
439 			   block_group, blk);
440 		ext4_mark_group_bitmap_corrupted(sb, block_group,
441 						 EXT4_GROUP_INFO_BBITMAP_CORRUPT);
442 		return -EFSCORRUPTED;
443 	}
444 	set_buffer_verified(bh);
445 verified:
446 	ext4_unlock_group(sb, block_group);
447 	return 0;
448 }
449 
450 /**
451  * ext4_read_block_bitmap_nowait()
452  * @sb:			super block
453  * @block_group:	given block group
454  * @ignore_locked:	ignore locked buffers
455  *
456  * Read the bitmap for a given block_group,and validate the
457  * bits for block/inode/inode tables are set in the bitmaps
458  *
459  * Return buffer_head on success or an ERR_PTR in case of failure.
460  */
461 struct buffer_head *
462 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group,
463 			      bool ignore_locked)
464 {
465 	struct ext4_group_desc *desc;
466 	struct ext4_sb_info *sbi = EXT4_SB(sb);
467 	struct buffer_head *bh;
468 	ext4_fsblk_t bitmap_blk;
469 	int err;
470 
471 	desc = ext4_get_group_desc(sb, block_group, NULL);
472 	if (!desc)
473 		return ERR_PTR(-EFSCORRUPTED);
474 	bitmap_blk = ext4_block_bitmap(sb, desc);
475 	if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
476 	    (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
477 		ext4_error(sb, "Invalid block bitmap block %llu in "
478 			   "block_group %u", bitmap_blk, block_group);
479 		ext4_mark_group_bitmap_corrupted(sb, block_group,
480 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
481 		return ERR_PTR(-EFSCORRUPTED);
482 	}
483 	bh = sb_getblk(sb, bitmap_blk);
484 	if (unlikely(!bh)) {
485 		ext4_warning(sb, "Cannot get buffer for block bitmap - "
486 			     "block_group = %u, block_bitmap = %llu",
487 			     block_group, bitmap_blk);
488 		return ERR_PTR(-ENOMEM);
489 	}
490 
491 	if (ignore_locked && buffer_locked(bh)) {
492 		/* buffer under IO already, return if called for prefetching */
493 		put_bh(bh);
494 		return NULL;
495 	}
496 
497 	if (bitmap_uptodate(bh))
498 		goto verify;
499 
500 	lock_buffer(bh);
501 	if (bitmap_uptodate(bh)) {
502 		unlock_buffer(bh);
503 		goto verify;
504 	}
505 	ext4_lock_group(sb, block_group);
506 	if (ext4_has_group_desc_csum(sb) &&
507 	    (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
508 		if (block_group == 0) {
509 			ext4_unlock_group(sb, block_group);
510 			unlock_buffer(bh);
511 			ext4_error(sb, "Block bitmap for bg 0 marked "
512 				   "uninitialized");
513 			err = -EFSCORRUPTED;
514 			goto out;
515 		}
516 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
517 		if (err) {
518 			ext4_unlock_group(sb, block_group);
519 			unlock_buffer(bh);
520 			ext4_error(sb, "Failed to init block bitmap for group "
521 				   "%u: %d", block_group, err);
522 			goto out;
523 		}
524 		set_bitmap_uptodate(bh);
525 		set_buffer_uptodate(bh);
526 		set_buffer_verified(bh);
527 		ext4_unlock_group(sb, block_group);
528 		unlock_buffer(bh);
529 		return bh;
530 	}
531 	ext4_unlock_group(sb, block_group);
532 	if (buffer_uptodate(bh)) {
533 		/*
534 		 * if not uninit if bh is uptodate,
535 		 * bitmap is also uptodate
536 		 */
537 		set_bitmap_uptodate(bh);
538 		unlock_buffer(bh);
539 		goto verify;
540 	}
541 	/*
542 	 * submit the buffer_head for reading
543 	 */
544 	set_buffer_new(bh);
545 	trace_ext4_read_block_bitmap_load(sb, block_group, ignore_locked);
546 	ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO |
547 			    (ignore_locked ? REQ_RAHEAD : 0),
548 			    ext4_end_bitmap_read);
549 	return bh;
550 verify:
551 	err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
552 	if (err)
553 		goto out;
554 	return bh;
555 out:
556 	put_bh(bh);
557 	return ERR_PTR(err);
558 }
559 
560 /* Returns 0 on success, -errno on error */
561 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
562 			   struct buffer_head *bh)
563 {
564 	struct ext4_group_desc *desc;
565 
566 	if (!buffer_new(bh))
567 		return 0;
568 	desc = ext4_get_group_desc(sb, block_group, NULL);
569 	if (!desc)
570 		return -EFSCORRUPTED;
571 	wait_on_buffer(bh);
572 	ext4_simulate_fail_bh(sb, bh, EXT4_SIM_BBITMAP_EIO);
573 	if (!buffer_uptodate(bh)) {
574 		ext4_error_err(sb, EIO, "Cannot read block bitmap - "
575 			       "block_group = %u, block_bitmap = %llu",
576 			       block_group, (unsigned long long) bh->b_blocknr);
577 		ext4_mark_group_bitmap_corrupted(sb, block_group,
578 					EXT4_GROUP_INFO_BBITMAP_CORRUPT);
579 		return -EIO;
580 	}
581 	clear_buffer_new(bh);
582 	/* Panic or remount fs read-only if block bitmap is invalid */
583 	return ext4_validate_block_bitmap(sb, desc, block_group, bh);
584 }
585 
586 struct buffer_head *
587 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
588 {
589 	struct buffer_head *bh;
590 	int err;
591 
592 	bh = ext4_read_block_bitmap_nowait(sb, block_group, false);
593 	if (IS_ERR(bh))
594 		return bh;
595 	err = ext4_wait_block_bitmap(sb, block_group, bh);
596 	if (err) {
597 		put_bh(bh);
598 		return ERR_PTR(err);
599 	}
600 	return bh;
601 }
602 
603 /**
604  * ext4_has_free_clusters()
605  * @sbi:	in-core super block structure.
606  * @nclusters:	number of needed blocks
607  * @flags:	flags from ext4_mb_new_blocks()
608  *
609  * Check if filesystem has nclusters free & available for allocation.
610  * On success return 1, return 0 on failure.
611  */
612 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
613 				  s64 nclusters, unsigned int flags)
614 {
615 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
616 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
617 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
618 
619 	free_clusters  = percpu_counter_read_positive(fcc);
620 	dirty_clusters = percpu_counter_read_positive(dcc);
621 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
622 
623 	/*
624 	 * r_blocks_count should always be multiple of the cluster ratio so
625 	 * we are safe to do a plane bit shift only.
626 	 */
627 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
628 	      resv_clusters;
629 
630 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
631 					EXT4_FREECLUSTERS_WATERMARK) {
632 		free_clusters  = percpu_counter_sum_positive(fcc);
633 		dirty_clusters = percpu_counter_sum_positive(dcc);
634 	}
635 	/* Check whether we have space after accounting for current
636 	 * dirty clusters & root reserved clusters.
637 	 */
638 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
639 		return 1;
640 
641 	/* Hm, nope.  Are (enough) root reserved clusters available? */
642 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
643 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
644 	    capable(CAP_SYS_RESOURCE) ||
645 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
646 
647 		if (free_clusters >= (nclusters + dirty_clusters +
648 				      resv_clusters))
649 			return 1;
650 	}
651 	/* No free blocks. Let's see if we can dip into reserved pool */
652 	if (flags & EXT4_MB_USE_RESERVED) {
653 		if (free_clusters >= (nclusters + dirty_clusters))
654 			return 1;
655 	}
656 
657 	return 0;
658 }
659 
660 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
661 			     s64 nclusters, unsigned int flags)
662 {
663 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
664 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
665 		return 0;
666 	} else
667 		return -ENOSPC;
668 }
669 
670 /**
671  * ext4_should_retry_alloc() - check if a block allocation should be retried
672  * @sb:			superblock
673  * @retries:		number of retry attempts made so far
674  *
675  * ext4_should_retry_alloc() is called when ENOSPC is returned while
676  * attempting to allocate blocks.  If there's an indication that a pending
677  * journal transaction might free some space and allow another attempt to
678  * succeed, this function will wait for the current or committing transaction
679  * to complete and then return TRUE.
680  */
681 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
682 {
683 	struct ext4_sb_info *sbi = EXT4_SB(sb);
684 
685 	if (!sbi->s_journal)
686 		return 0;
687 
688 	if (++(*retries) > 3) {
689 		percpu_counter_inc(&sbi->s_sra_exceeded_retry_limit);
690 		return 0;
691 	}
692 
693 	/*
694 	 * if there's no indication that blocks are about to be freed it's
695 	 * possible we just missed a transaction commit that did so
696 	 */
697 	smp_mb();
698 	if (sbi->s_mb_free_pending == 0) {
699 		if (test_opt(sb, DISCARD)) {
700 			atomic_inc(&sbi->s_retry_alloc_pending);
701 			flush_work(&sbi->s_discard_work);
702 			atomic_dec(&sbi->s_retry_alloc_pending);
703 		}
704 		return ext4_has_free_clusters(sbi, 1, 0);
705 	}
706 
707 	/*
708 	 * it's possible we've just missed a transaction commit here,
709 	 * so ignore the returned status
710 	 */
711 	ext4_debug("%s: retrying operation after ENOSPC\n", sb->s_id);
712 	(void) jbd2_journal_force_commit_nested(sbi->s_journal);
713 	return 1;
714 }
715 
716 /*
717  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
718  *
719  * @handle:             handle to this transaction
720  * @inode:              file inode
721  * @goal:               given target block(filesystem wide)
722  * @count:		pointer to total number of clusters needed
723  * @errp:               error code
724  *
725  * Return 1st allocated block number on success, *count stores total account
726  * error stores in errp pointer
727  */
728 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
729 				  ext4_fsblk_t goal, unsigned int flags,
730 				  unsigned long *count, int *errp)
731 {
732 	struct ext4_allocation_request ar;
733 	ext4_fsblk_t ret;
734 
735 	memset(&ar, 0, sizeof(ar));
736 	/* Fill with neighbour allocated blocks */
737 	ar.inode = inode;
738 	ar.goal = goal;
739 	ar.len = count ? *count : 1;
740 	ar.flags = flags;
741 
742 	ret = ext4_mb_new_blocks(handle, &ar, errp);
743 	if (count)
744 		*count = ar.len;
745 	/*
746 	 * Account for the allocated meta blocks.  We will never
747 	 * fail EDQUOT for metdata, but we do account for it.
748 	 */
749 	if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
750 		dquot_alloc_block_nofail(inode,
751 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
752 	}
753 	return ret;
754 }
755 
756 /**
757  * ext4_count_free_clusters() -- count filesystem free clusters
758  * @sb:		superblock
759  *
760  * Adds up the number of free clusters from each block group.
761  */
762 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
763 {
764 	ext4_fsblk_t desc_count;
765 	struct ext4_group_desc *gdp;
766 	ext4_group_t i;
767 	ext4_group_t ngroups = ext4_get_groups_count(sb);
768 	struct ext4_group_info *grp;
769 #ifdef EXT4FS_DEBUG
770 	struct ext4_super_block *es;
771 	ext4_fsblk_t bitmap_count;
772 	unsigned int x;
773 	struct buffer_head *bitmap_bh = NULL;
774 
775 	es = EXT4_SB(sb)->s_es;
776 	desc_count = 0;
777 	bitmap_count = 0;
778 	gdp = NULL;
779 
780 	for (i = 0; i < ngroups; i++) {
781 		gdp = ext4_get_group_desc(sb, i, NULL);
782 		if (!gdp)
783 			continue;
784 		grp = NULL;
785 		if (EXT4_SB(sb)->s_group_info)
786 			grp = ext4_get_group_info(sb, i);
787 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
788 			desc_count += ext4_free_group_clusters(sb, gdp);
789 		brelse(bitmap_bh);
790 		bitmap_bh = ext4_read_block_bitmap(sb, i);
791 		if (IS_ERR(bitmap_bh)) {
792 			bitmap_bh = NULL;
793 			continue;
794 		}
795 
796 		x = ext4_count_free(bitmap_bh->b_data,
797 				    EXT4_CLUSTERS_PER_GROUP(sb) / 8);
798 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
799 			i, ext4_free_group_clusters(sb, gdp), x);
800 		bitmap_count += x;
801 	}
802 	brelse(bitmap_bh);
803 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
804 	       ", computed = %llu, %llu\n",
805 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
806 	       desc_count, bitmap_count);
807 	return bitmap_count;
808 #else
809 	desc_count = 0;
810 	for (i = 0; i < ngroups; i++) {
811 		gdp = ext4_get_group_desc(sb, i, NULL);
812 		if (!gdp)
813 			continue;
814 		grp = NULL;
815 		if (EXT4_SB(sb)->s_group_info)
816 			grp = ext4_get_group_info(sb, i);
817 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
818 			desc_count += ext4_free_group_clusters(sb, gdp);
819 	}
820 
821 	return desc_count;
822 #endif
823 }
824 
825 static inline int test_root(ext4_group_t a, int b)
826 {
827 	while (1) {
828 		if (a < b)
829 			return 0;
830 		if (a == b)
831 			return 1;
832 		if ((a % b) != 0)
833 			return 0;
834 		a = a / b;
835 	}
836 }
837 
838 /**
839  *	ext4_bg_has_super - number of blocks used by the superblock in group
840  *	@sb: superblock for filesystem
841  *	@group: group number to check
842  *
843  *	Return the number of blocks used by the superblock (primary or backup)
844  *	in this group.  Currently this will be only 0 or 1.
845  */
846 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
847 {
848 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
849 
850 	if (group == 0)
851 		return 1;
852 	if (ext4_has_feature_sparse_super2(sb)) {
853 		if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
854 		    group == le32_to_cpu(es->s_backup_bgs[1]))
855 			return 1;
856 		return 0;
857 	}
858 	if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
859 		return 1;
860 	if (!(group & 1))
861 		return 0;
862 	if (test_root(group, 3) || (test_root(group, 5)) ||
863 	    test_root(group, 7))
864 		return 1;
865 
866 	return 0;
867 }
868 
869 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
870 					ext4_group_t group)
871 {
872 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
873 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
874 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
875 
876 	if (group == first || group == first + 1 || group == last)
877 		return 1;
878 	return 0;
879 }
880 
881 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
882 					ext4_group_t group)
883 {
884 	if (!ext4_bg_has_super(sb, group))
885 		return 0;
886 
887 	if (ext4_has_feature_meta_bg(sb))
888 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
889 	else
890 		return EXT4_SB(sb)->s_gdb_count;
891 }
892 
893 /**
894  *	ext4_bg_num_gdb - number of blocks used by the group table in group
895  *	@sb: superblock for filesystem
896  *	@group: group number to check
897  *
898  *	Return the number of blocks used by the group descriptor table
899  *	(primary or backup) in this group.  In the future there may be a
900  *	different number of descriptor blocks in each group.
901  */
902 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
903 {
904 	unsigned long first_meta_bg =
905 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
906 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
907 
908 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
909 		return ext4_bg_num_gdb_nometa(sb, group);
910 
911 	return ext4_bg_num_gdb_meta(sb,group);
912 
913 }
914 
915 /*
916  * This function returns the number of file system metadata blocks at
917  * the beginning of a block group, including the reserved gdt blocks.
918  */
919 unsigned int ext4_num_base_meta_blocks(struct super_block *sb,
920 				       ext4_group_t block_group)
921 {
922 	struct ext4_sb_info *sbi = EXT4_SB(sb);
923 	unsigned num;
924 
925 	/* Check for superblock and gdt backups in this group */
926 	num = ext4_bg_has_super(sb, block_group);
927 
928 	if (!ext4_has_feature_meta_bg(sb) ||
929 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
930 			  sbi->s_desc_per_block) {
931 		if (num) {
932 			num += ext4_bg_num_gdb_nometa(sb, block_group);
933 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
934 		}
935 	} else { /* For META_BG_BLOCK_GROUPS */
936 		num += ext4_bg_num_gdb_meta(sb, block_group);
937 	}
938 	return num;
939 }
940 
941 static unsigned int ext4_num_base_meta_clusters(struct super_block *sb,
942 						ext4_group_t block_group)
943 {
944 	return EXT4_NUM_B2C(EXT4_SB(sb), ext4_num_base_meta_blocks(sb, block_group));
945 }
946 
947 /**
948  *	ext4_inode_to_goal_block - return a hint for block allocation
949  *	@inode: inode for block allocation
950  *
951  *	Return the ideal location to start allocating blocks for a
952  *	newly created inode.
953  */
954 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
955 {
956 	struct ext4_inode_info *ei = EXT4_I(inode);
957 	ext4_group_t block_group;
958 	ext4_grpblk_t colour;
959 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
960 	ext4_fsblk_t bg_start;
961 	ext4_fsblk_t last_block;
962 
963 	block_group = ei->i_block_group;
964 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
965 		/*
966 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
967 		 * block groups per flexgroup, reserve the first block
968 		 * group for directories and special files.  Regular
969 		 * files will start at the second block group.  This
970 		 * tends to speed up directory access and improves
971 		 * fsck times.
972 		 */
973 		block_group &= ~(flex_size-1);
974 		if (S_ISREG(inode->i_mode))
975 			block_group++;
976 	}
977 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
978 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
979 
980 	/*
981 	 * If we are doing delayed allocation, we don't need take
982 	 * colour into account.
983 	 */
984 	if (test_opt(inode->i_sb, DELALLOC))
985 		return bg_start;
986 
987 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
988 		colour = (task_pid_nr(current) % 16) *
989 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
990 	else
991 		colour = (task_pid_nr(current) % 16) *
992 			((last_block - bg_start) / 16);
993 	return bg_start + colour;
994 }
995 
996