• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *  linux/fs/ext4/balloc.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
10  *  Big-endian to little-endian byte-swapping/bitmaps by
11  *        David S. Miller (davem@caip.rutgers.edu), 1995
12  */
13 
14 #include <linux/time.h>
15 #include <linux/capability.h>
16 #include <linux/fs.h>
17 #include <linux/quotaops.h>
18 #include <linux/buffer_head.h>
19 #include "ext4.h"
20 #include "ext4_jbd2.h"
21 #include "mballoc.h"
22 
23 #include <trace/events/ext4.h>
24 
25 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
26 					    ext4_group_t block_group);
27 /*
28  * balloc.c contains the blocks allocation and deallocation routines
29  */
30 
31 /*
32  * Calculate block group number for a given block number
33  */
ext4_get_group_number(struct super_block * sb,ext4_fsblk_t block)34 ext4_group_t ext4_get_group_number(struct super_block *sb,
35 				   ext4_fsblk_t block)
36 {
37 	ext4_group_t group;
38 
39 	if (test_opt2(sb, STD_GROUP_SIZE))
40 		group = (block -
41 			 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >>
42 			(EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3);
43 	else
44 		ext4_get_group_no_and_offset(sb, block, &group, NULL);
45 	return group;
46 }
47 
48 /*
49  * Calculate the block group number and offset into the block/cluster
50  * allocation bitmap, given a block number
51  */
ext4_get_group_no_and_offset(struct super_block * sb,ext4_fsblk_t blocknr,ext4_group_t * blockgrpp,ext4_grpblk_t * offsetp)52 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
53 		ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
54 {
55 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
56 	ext4_grpblk_t offset;
57 
58 	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
59 	offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
60 		EXT4_SB(sb)->s_cluster_bits;
61 	if (offsetp)
62 		*offsetp = offset;
63 	if (blockgrpp)
64 		*blockgrpp = blocknr;
65 
66 }
67 
68 /*
69  * Check whether the 'block' lives within the 'block_group'. Returns 1 if so
70  * and 0 otherwise.
71  */
ext4_block_in_group(struct super_block * sb,ext4_fsblk_t block,ext4_group_t block_group)72 static inline int ext4_block_in_group(struct super_block *sb,
73 				      ext4_fsblk_t block,
74 				      ext4_group_t block_group)
75 {
76 	ext4_group_t actual_group;
77 
78 	actual_group = ext4_get_group_number(sb, block);
79 	return (actual_group == block_group) ? 1 : 0;
80 }
81 
82 /* Return the number of clusters used for file system metadata; this
83  * represents the overhead needed by the file system.
84  */
ext4_num_overhead_clusters(struct super_block * sb,ext4_group_t block_group,struct ext4_group_desc * gdp)85 static unsigned ext4_num_overhead_clusters(struct super_block *sb,
86 					   ext4_group_t block_group,
87 					   struct ext4_group_desc *gdp)
88 {
89 	unsigned num_clusters;
90 	int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
91 	ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
92 	ext4_fsblk_t itbl_blk;
93 	struct ext4_sb_info *sbi = EXT4_SB(sb);
94 
95 	/* This is the number of clusters used by the superblock,
96 	 * block group descriptors, and reserved block group
97 	 * descriptor blocks */
98 	num_clusters = ext4_num_base_meta_clusters(sb, block_group);
99 
100 	/*
101 	 * For the allocation bitmaps and inode table, we first need
102 	 * to check to see if the block is in the block group.  If it
103 	 * is, then check to see if the cluster is already accounted
104 	 * for in the clusters used for the base metadata cluster, or
105 	 * if we can increment the base metadata cluster to include
106 	 * that block.  Otherwise, we will have to track the cluster
107 	 * used for the allocation bitmap or inode table explicitly.
108 	 * Normally all of these blocks are contiguous, so the special
109 	 * case handling shouldn't be necessary except for *very*
110 	 * unusual file system layouts.
111 	 */
112 	if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
113 		block_cluster = EXT4_B2C(sbi,
114 					 ext4_block_bitmap(sb, gdp) - start);
115 		if (block_cluster < num_clusters)
116 			block_cluster = -1;
117 		else if (block_cluster == num_clusters) {
118 			num_clusters++;
119 			block_cluster = -1;
120 		}
121 	}
122 
123 	if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
124 		inode_cluster = EXT4_B2C(sbi,
125 					 ext4_inode_bitmap(sb, gdp) - start);
126 		if (inode_cluster < num_clusters)
127 			inode_cluster = -1;
128 		else if (inode_cluster == num_clusters) {
129 			num_clusters++;
130 			inode_cluster = -1;
131 		}
132 	}
133 
134 	itbl_blk = ext4_inode_table(sb, gdp);
135 	for (i = 0; i < sbi->s_itb_per_group; i++) {
136 		if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
137 			c = EXT4_B2C(sbi, itbl_blk + i - start);
138 			if ((c < num_clusters) || (c == inode_cluster) ||
139 			    (c == block_cluster) || (c == itbl_cluster))
140 				continue;
141 			if (c == num_clusters) {
142 				num_clusters++;
143 				continue;
144 			}
145 			num_clusters++;
146 			itbl_cluster = c;
147 		}
148 	}
149 
150 	if (block_cluster != -1)
151 		num_clusters++;
152 	if (inode_cluster != -1)
153 		num_clusters++;
154 
155 	return num_clusters;
156 }
157 
num_clusters_in_group(struct super_block * sb,ext4_group_t block_group)158 static unsigned int num_clusters_in_group(struct super_block *sb,
159 					  ext4_group_t block_group)
160 {
161 	unsigned int blocks;
162 
163 	if (block_group == ext4_get_groups_count(sb) - 1) {
164 		/*
165 		 * Even though mke2fs always initializes the first and
166 		 * last group, just in case some other tool was used,
167 		 * we need to make sure we calculate the right free
168 		 * blocks.
169 		 */
170 		blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
171 			ext4_group_first_block_no(sb, block_group);
172 	} else
173 		blocks = EXT4_BLOCKS_PER_GROUP(sb);
174 	return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
175 }
176 
177 /* Initializes an uninitialized block bitmap */
ext4_init_block_bitmap(struct super_block * sb,struct buffer_head * bh,ext4_group_t block_group,struct ext4_group_desc * gdp)178 static int ext4_init_block_bitmap(struct super_block *sb,
179 				   struct buffer_head *bh,
180 				   ext4_group_t block_group,
181 				   struct ext4_group_desc *gdp)
182 {
183 	unsigned int bit, bit_max;
184 	struct ext4_sb_info *sbi = EXT4_SB(sb);
185 	ext4_fsblk_t start, tmp;
186 	int flex_bg = 0;
187 	struct ext4_group_info *grp;
188 
189 	J_ASSERT_BH(bh, buffer_locked(bh));
190 
191 	/* If checksum is bad mark all blocks used to prevent allocation
192 	 * essentially implementing a per-group read-only flag. */
193 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
194 		grp = ext4_get_group_info(sb, block_group);
195 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
196 			percpu_counter_sub(&sbi->s_freeclusters_counter,
197 					   grp->bb_free);
198 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
199 		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
200 			int count;
201 			count = ext4_free_inodes_count(sb, gdp);
202 			percpu_counter_sub(&sbi->s_freeinodes_counter,
203 					   count);
204 		}
205 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
206 		return -EFSBADCRC;
207 	}
208 	memset(bh->b_data, 0, sb->s_blocksize);
209 
210 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
211 	if ((bit_max >> 3) >= bh->b_size)
212 		return -EFSCORRUPTED;
213 
214 	for (bit = 0; bit < bit_max; bit++)
215 		ext4_set_bit(bit, bh->b_data);
216 
217 	start = ext4_group_first_block_no(sb, block_group);
218 
219 	if (ext4_has_feature_flex_bg(sb))
220 		flex_bg = 1;
221 
222 	/* Set bits for block and inode bitmaps, and inode table */
223 	tmp = ext4_block_bitmap(sb, gdp);
224 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
225 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
226 
227 	tmp = ext4_inode_bitmap(sb, gdp);
228 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
229 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
230 
231 	tmp = ext4_inode_table(sb, gdp);
232 	for (; tmp < ext4_inode_table(sb, gdp) +
233 		     sbi->s_itb_per_group; tmp++) {
234 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
235 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
236 	}
237 
238 	/*
239 	 * Also if the number of blocks within the group is less than
240 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
241 	 * of the block bitmap to 1
242 	 */
243 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
244 			     sb->s_blocksize * 8, bh->b_data);
245 	return 0;
246 }
247 
248 /* Return the number of free blocks in a block group.  It is used when
249  * the block bitmap is uninitialized, so we can't just count the bits
250  * in the bitmap. */
ext4_free_clusters_after_init(struct super_block * sb,ext4_group_t block_group,struct ext4_group_desc * gdp)251 unsigned ext4_free_clusters_after_init(struct super_block *sb,
252 				       ext4_group_t block_group,
253 				       struct ext4_group_desc *gdp)
254 {
255 	return num_clusters_in_group(sb, block_group) -
256 		ext4_num_overhead_clusters(sb, block_group, gdp);
257 }
258 
259 /*
260  * The free blocks are managed by bitmaps.  A file system contains several
261  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
262  * block for inodes, N blocks for the inode table and data blocks.
263  *
264  * The file system contains group descriptors which are located after the
265  * super block.  Each descriptor contains the number of the bitmap block and
266  * the free blocks count in the block.  The descriptors are loaded in memory
267  * when a file system is mounted (see ext4_fill_super).
268  */
269 
270 /**
271  * ext4_get_group_desc() -- load group descriptor from disk
272  * @sb:			super block
273  * @block_group:	given block group
274  * @bh:			pointer to the buffer head to store the block
275  *			group descriptor
276  */
ext4_get_group_desc(struct super_block * sb,ext4_group_t block_group,struct buffer_head ** bh)277 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
278 					     ext4_group_t block_group,
279 					     struct buffer_head **bh)
280 {
281 	unsigned int group_desc;
282 	unsigned int offset;
283 	ext4_group_t ngroups = ext4_get_groups_count(sb);
284 	struct ext4_group_desc *desc;
285 	struct ext4_sb_info *sbi = EXT4_SB(sb);
286 
287 	if (block_group >= ngroups) {
288 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
289 			   " groups_count = %u", block_group, ngroups);
290 
291 		return NULL;
292 	}
293 
294 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
295 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
296 	if (!sbi->s_group_desc[group_desc]) {
297 		ext4_error(sb, "Group descriptor not loaded - "
298 			   "block_group = %u, group_desc = %u, desc = %u",
299 			   block_group, group_desc, offset);
300 		return NULL;
301 	}
302 
303 	desc = (struct ext4_group_desc *)(
304 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
305 		offset * EXT4_DESC_SIZE(sb));
306 	if (bh)
307 		*bh = sbi->s_group_desc[group_desc];
308 	return desc;
309 }
310 
311 /*
312  * Return the block number which was discovered to be invalid, or 0 if
313  * the block bitmap is valid.
314  */
ext4_valid_block_bitmap(struct super_block * sb,struct ext4_group_desc * desc,ext4_group_t block_group,struct buffer_head * bh)315 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
316 					    struct ext4_group_desc *desc,
317 					    ext4_group_t block_group,
318 					    struct buffer_head *bh)
319 {
320 	struct ext4_sb_info *sbi = EXT4_SB(sb);
321 	ext4_grpblk_t offset;
322 	ext4_grpblk_t next_zero_bit;
323 	ext4_fsblk_t blk;
324 	ext4_fsblk_t group_first_block;
325 
326 	if (ext4_has_feature_flex_bg(sb)) {
327 		/* with FLEX_BG, the inode/block bitmaps and itable
328 		 * blocks may not be in the group at all
329 		 * so the bitmap validation will be skipped for those groups
330 		 * or it has to also read the block group where the bitmaps
331 		 * are located to verify they are set.
332 		 */
333 		return 0;
334 	}
335 	group_first_block = ext4_group_first_block_no(sb, block_group);
336 
337 	/* check whether block bitmap block number is set */
338 	blk = ext4_block_bitmap(sb, desc);
339 	offset = blk - group_first_block;
340 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
341 		/* bad block bitmap */
342 		return blk;
343 
344 	/* check whether the inode bitmap block number is set */
345 	blk = ext4_inode_bitmap(sb, desc);
346 	offset = blk - group_first_block;
347 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
348 		/* bad block bitmap */
349 		return blk;
350 
351 	/* check whether the inode table block number is set */
352 	blk = ext4_inode_table(sb, desc);
353 	offset = blk - group_first_block;
354 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
355 			EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group),
356 			EXT4_B2C(sbi, offset));
357 	if (next_zero_bit <
358 	    EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group))
359 		/* bad bitmap for inode tables */
360 		return blk;
361 	return 0;
362 }
363 
ext4_validate_block_bitmap(struct super_block * sb,struct ext4_group_desc * desc,ext4_group_t block_group,struct buffer_head * bh)364 static int ext4_validate_block_bitmap(struct super_block *sb,
365 				      struct ext4_group_desc *desc,
366 				      ext4_group_t block_group,
367 				      struct buffer_head *bh)
368 {
369 	ext4_fsblk_t	blk;
370 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
371 	struct ext4_sb_info *sbi = EXT4_SB(sb);
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 (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
380 			desc, bh))) {
381 		ext4_unlock_group(sb, block_group);
382 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
383 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
384 			percpu_counter_sub(&sbi->s_freeclusters_counter,
385 					   grp->bb_free);
386 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
387 		return -EFSBADCRC;
388 	}
389 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
390 	if (unlikely(blk != 0)) {
391 		ext4_unlock_group(sb, block_group);
392 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
393 			   block_group, blk);
394 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
395 			percpu_counter_sub(&sbi->s_freeclusters_counter,
396 					   grp->bb_free);
397 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
398 		return -EFSCORRUPTED;
399 	}
400 	set_buffer_verified(bh);
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 NULL in case of failure.
414  */
415 struct buffer_head *
ext4_read_block_bitmap_nowait(struct super_block * sb,ext4_group_t block_group)416 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
417 {
418 	struct ext4_group_desc *desc;
419 	struct buffer_head *bh;
420 	ext4_fsblk_t bitmap_blk;
421 	int err;
422 
423 	desc = ext4_get_group_desc(sb, block_group, NULL);
424 	if (!desc)
425 		return ERR_PTR(-EFSCORRUPTED);
426 	bitmap_blk = ext4_block_bitmap(sb, desc);
427 	bh = sb_getblk(sb, bitmap_blk);
428 	if (unlikely(!bh)) {
429 		ext4_error(sb, "Cannot get buffer for block bitmap - "
430 			   "block_group = %u, block_bitmap = %llu",
431 			   block_group, bitmap_blk);
432 		return ERR_PTR(-ENOMEM);
433 	}
434 
435 	if (bitmap_uptodate(bh))
436 		goto verify;
437 
438 	lock_buffer(bh);
439 	if (bitmap_uptodate(bh)) {
440 		unlock_buffer(bh);
441 		goto verify;
442 	}
443 	ext4_lock_group(sb, block_group);
444 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
445 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
446 		set_bitmap_uptodate(bh);
447 		set_buffer_uptodate(bh);
448 		set_buffer_verified(bh);
449 		ext4_unlock_group(sb, block_group);
450 		unlock_buffer(bh);
451 		if (err) {
452 			ext4_error(sb, "Failed to init block bitmap for group "
453 				   "%u: %d", block_group, err);
454 			goto out;
455 		}
456 		goto verify;
457 	}
458 	ext4_unlock_group(sb, block_group);
459 	if (buffer_uptodate(bh)) {
460 		/*
461 		 * if not uninit if bh is uptodate,
462 		 * bitmap is also uptodate
463 		 */
464 		set_bitmap_uptodate(bh);
465 		unlock_buffer(bh);
466 		goto verify;
467 	}
468 	/*
469 	 * submit the buffer_head for reading
470 	 */
471 	set_buffer_new(bh);
472 	trace_ext4_read_block_bitmap_load(sb, block_group);
473 	bh->b_end_io = ext4_end_bitmap_read;
474 	get_bh(bh);
475 	submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
476 	return bh;
477 verify:
478 	err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
479 	if (err)
480 		goto out;
481 	return bh;
482 out:
483 	put_bh(bh);
484 	return ERR_PTR(err);
485 }
486 
487 /* Returns 0 on success, 1 on error */
ext4_wait_block_bitmap(struct super_block * sb,ext4_group_t block_group,struct buffer_head * bh)488 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
489 			   struct buffer_head *bh)
490 {
491 	struct ext4_group_desc *desc;
492 
493 	if (!buffer_new(bh))
494 		return 0;
495 	desc = ext4_get_group_desc(sb, block_group, NULL);
496 	if (!desc)
497 		return -EFSCORRUPTED;
498 	wait_on_buffer(bh);
499 	if (!buffer_uptodate(bh)) {
500 		ext4_error(sb, "Cannot read block bitmap - "
501 			   "block_group = %u, block_bitmap = %llu",
502 			   block_group, (unsigned long long) bh->b_blocknr);
503 		return -EIO;
504 	}
505 	clear_buffer_new(bh);
506 	/* Panic or remount fs read-only if block bitmap is invalid */
507 	return ext4_validate_block_bitmap(sb, desc, block_group, bh);
508 }
509 
510 struct buffer_head *
ext4_read_block_bitmap(struct super_block * sb,ext4_group_t block_group)511 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
512 {
513 	struct buffer_head *bh;
514 	int err;
515 
516 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
517 	if (IS_ERR(bh))
518 		return bh;
519 	err = ext4_wait_block_bitmap(sb, block_group, bh);
520 	if (err) {
521 		put_bh(bh);
522 		return ERR_PTR(err);
523 	}
524 	return bh;
525 }
526 
527 /**
528  * ext4_has_free_clusters()
529  * @sbi:	in-core super block structure.
530  * @nclusters:	number of needed blocks
531  * @flags:	flags from ext4_mb_new_blocks()
532  *
533  * Check if filesystem has nclusters free & available for allocation.
534  * On success return 1, return 0 on failure.
535  */
ext4_has_free_clusters(struct ext4_sb_info * sbi,s64 nclusters,unsigned int flags)536 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
537 				  s64 nclusters, unsigned int flags)
538 {
539 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
540 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
541 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
542 
543 	free_clusters  = percpu_counter_read_positive(fcc);
544 	dirty_clusters = percpu_counter_read_positive(dcc);
545 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
546 
547 	/*
548 	 * r_blocks_count should always be multiple of the cluster ratio so
549 	 * we are safe to do a plane bit shift only.
550 	 */
551 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
552 	      resv_clusters;
553 
554 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
555 					EXT4_FREECLUSTERS_WATERMARK) {
556 		free_clusters  = percpu_counter_sum_positive(fcc);
557 		dirty_clusters = percpu_counter_sum_positive(dcc);
558 	}
559 	/* Check whether we have space after accounting for current
560 	 * dirty clusters & root reserved clusters.
561 	 */
562 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
563 		return 1;
564 
565 	/* Hm, nope.  Are (enough) root reserved clusters available? */
566 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
567 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
568 	    capable(CAP_SYS_RESOURCE) ||
569 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
570 
571 		if (free_clusters >= (nclusters + dirty_clusters +
572 				      resv_clusters))
573 			return 1;
574 	}
575 	/* No free blocks. Let's see if we can dip into reserved pool */
576 	if (flags & EXT4_MB_USE_RESERVED) {
577 		if (free_clusters >= (nclusters + dirty_clusters))
578 			return 1;
579 	}
580 
581 	return 0;
582 }
583 
ext4_claim_free_clusters(struct ext4_sb_info * sbi,s64 nclusters,unsigned int flags)584 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
585 			     s64 nclusters, unsigned int flags)
586 {
587 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
588 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
589 		return 0;
590 	} else
591 		return -ENOSPC;
592 }
593 
594 /**
595  * ext4_should_retry_alloc()
596  * @sb:			super block
597  * @retries		number of attemps has been made
598  *
599  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
600  * it is profitable to retry the operation, this function will wait
601  * for the current or committing transaction to complete, and then
602  * return TRUE.
603  *
604  * if the total number of retries exceed three times, return FALSE.
605  */
ext4_should_retry_alloc(struct super_block * sb,int * retries)606 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
607 {
608 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
609 	    (*retries)++ > 3 ||
610 	    !EXT4_SB(sb)->s_journal)
611 		return 0;
612 
613 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
614 
615 	smp_mb();
616 	if (EXT4_SB(sb)->s_mb_free_pending)
617 		jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
618 	return 1;
619 }
620 
621 /*
622  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
623  *
624  * @handle:             handle to this transaction
625  * @inode:              file inode
626  * @goal:               given target block(filesystem wide)
627  * @count:		pointer to total number of clusters needed
628  * @errp:               error code
629  *
630  * Return 1st allocated block number on success, *count stores total account
631  * error stores in errp pointer
632  */
ext4_new_meta_blocks(handle_t * handle,struct inode * inode,ext4_fsblk_t goal,unsigned int flags,unsigned long * count,int * errp)633 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
634 				  ext4_fsblk_t goal, unsigned int flags,
635 				  unsigned long *count, int *errp)
636 {
637 	struct ext4_allocation_request ar;
638 	ext4_fsblk_t ret;
639 
640 	memset(&ar, 0, sizeof(ar));
641 	/* Fill with neighbour allocated blocks */
642 	ar.inode = inode;
643 	ar.goal = goal;
644 	ar.len = count ? *count : 1;
645 	ar.flags = flags;
646 
647 	ret = ext4_mb_new_blocks(handle, &ar, errp);
648 	if (count)
649 		*count = ar.len;
650 	/*
651 	 * Account for the allocated meta blocks.  We will never
652 	 * fail EDQUOT for metdata, but we do account for it.
653 	 */
654 	if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
655 		dquot_alloc_block_nofail(inode,
656 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
657 	}
658 	return ret;
659 }
660 
661 /**
662  * ext4_count_free_clusters() -- count filesystem free clusters
663  * @sb:		superblock
664  *
665  * Adds up the number of free clusters from each block group.
666  */
ext4_count_free_clusters(struct super_block * sb)667 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
668 {
669 	ext4_fsblk_t desc_count;
670 	struct ext4_group_desc *gdp;
671 	ext4_group_t i;
672 	ext4_group_t ngroups = ext4_get_groups_count(sb);
673 	struct ext4_group_info *grp;
674 #ifdef EXT4FS_DEBUG
675 	struct ext4_super_block *es;
676 	ext4_fsblk_t bitmap_count;
677 	unsigned int x;
678 	struct buffer_head *bitmap_bh = NULL;
679 
680 	es = EXT4_SB(sb)->s_es;
681 	desc_count = 0;
682 	bitmap_count = 0;
683 	gdp = NULL;
684 
685 	for (i = 0; i < ngroups; i++) {
686 		gdp = ext4_get_group_desc(sb, i, NULL);
687 		if (!gdp)
688 			continue;
689 		grp = NULL;
690 		if (EXT4_SB(sb)->s_group_info)
691 			grp = ext4_get_group_info(sb, i);
692 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
693 			desc_count += ext4_free_group_clusters(sb, gdp);
694 		brelse(bitmap_bh);
695 		bitmap_bh = ext4_read_block_bitmap(sb, i);
696 		if (IS_ERR(bitmap_bh)) {
697 			bitmap_bh = NULL;
698 			continue;
699 		}
700 
701 		x = ext4_count_free(bitmap_bh->b_data,
702 				    EXT4_CLUSTERS_PER_GROUP(sb) / 8);
703 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
704 			i, ext4_free_group_clusters(sb, gdp), x);
705 		bitmap_count += x;
706 	}
707 	brelse(bitmap_bh);
708 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
709 	       ", computed = %llu, %llu\n",
710 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
711 	       desc_count, bitmap_count);
712 	return bitmap_count;
713 #else
714 	desc_count = 0;
715 	for (i = 0; i < ngroups; i++) {
716 		gdp = ext4_get_group_desc(sb, i, NULL);
717 		if (!gdp)
718 			continue;
719 		grp = NULL;
720 		if (EXT4_SB(sb)->s_group_info)
721 			grp = ext4_get_group_info(sb, i);
722 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
723 			desc_count += ext4_free_group_clusters(sb, gdp);
724 	}
725 
726 	return desc_count;
727 #endif
728 }
729 
test_root(ext4_group_t a,int b)730 static inline int test_root(ext4_group_t a, int b)
731 {
732 	while (1) {
733 		if (a < b)
734 			return 0;
735 		if (a == b)
736 			return 1;
737 		if ((a % b) != 0)
738 			return 0;
739 		a = a / b;
740 	}
741 }
742 
743 /**
744  *	ext4_bg_has_super - number of blocks used by the superblock in group
745  *	@sb: superblock for filesystem
746  *	@group: group number to check
747  *
748  *	Return the number of blocks used by the superblock (primary or backup)
749  *	in this group.  Currently this will be only 0 or 1.
750  */
ext4_bg_has_super(struct super_block * sb,ext4_group_t group)751 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
752 {
753 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
754 
755 	if (group == 0)
756 		return 1;
757 	if (ext4_has_feature_sparse_super2(sb)) {
758 		if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
759 		    group == le32_to_cpu(es->s_backup_bgs[1]))
760 			return 1;
761 		return 0;
762 	}
763 	if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
764 		return 1;
765 	if (!(group & 1))
766 		return 0;
767 	if (test_root(group, 3) || (test_root(group, 5)) ||
768 	    test_root(group, 7))
769 		return 1;
770 
771 	return 0;
772 }
773 
ext4_bg_num_gdb_meta(struct super_block * sb,ext4_group_t group)774 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
775 					ext4_group_t group)
776 {
777 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
778 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
779 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
780 
781 	if (group == first || group == first + 1 || group == last)
782 		return 1;
783 	return 0;
784 }
785 
ext4_bg_num_gdb_nometa(struct super_block * sb,ext4_group_t group)786 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
787 					ext4_group_t group)
788 {
789 	if (!ext4_bg_has_super(sb, group))
790 		return 0;
791 
792 	if (ext4_has_feature_meta_bg(sb))
793 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
794 	else
795 		return EXT4_SB(sb)->s_gdb_count;
796 }
797 
798 /**
799  *	ext4_bg_num_gdb - number of blocks used by the group table in group
800  *	@sb: superblock for filesystem
801  *	@group: group number to check
802  *
803  *	Return the number of blocks used by the group descriptor table
804  *	(primary or backup) in this group.  In the future there may be a
805  *	different number of descriptor blocks in each group.
806  */
ext4_bg_num_gdb(struct super_block * sb,ext4_group_t group)807 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
808 {
809 	unsigned long first_meta_bg =
810 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
811 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
812 
813 	if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
814 		return ext4_bg_num_gdb_nometa(sb, group);
815 
816 	return ext4_bg_num_gdb_meta(sb,group);
817 
818 }
819 
820 /*
821  * This function returns the number of file system metadata clusters at
822  * the beginning of a block group, including the reserved gdt blocks.
823  */
ext4_num_base_meta_clusters(struct super_block * sb,ext4_group_t block_group)824 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
825 				     ext4_group_t block_group)
826 {
827 	struct ext4_sb_info *sbi = EXT4_SB(sb);
828 	unsigned num;
829 
830 	/* Check for superblock and gdt backups in this group */
831 	num = ext4_bg_has_super(sb, block_group);
832 
833 	if (!ext4_has_feature_meta_bg(sb) ||
834 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
835 			  sbi->s_desc_per_block) {
836 		if (num) {
837 			num += ext4_bg_num_gdb(sb, block_group);
838 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
839 		}
840 	} else { /* For META_BG_BLOCK_GROUPS */
841 		num += ext4_bg_num_gdb(sb, block_group);
842 	}
843 	return EXT4_NUM_B2C(sbi, num);
844 }
845 /**
846  *	ext4_inode_to_goal_block - return a hint for block allocation
847  *	@inode: inode for block allocation
848  *
849  *	Return the ideal location to start allocating blocks for a
850  *	newly created inode.
851  */
ext4_inode_to_goal_block(struct inode * inode)852 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
853 {
854 	struct ext4_inode_info *ei = EXT4_I(inode);
855 	ext4_group_t block_group;
856 	ext4_grpblk_t colour;
857 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
858 	ext4_fsblk_t bg_start;
859 	ext4_fsblk_t last_block;
860 
861 	block_group = ei->i_block_group;
862 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
863 		/*
864 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
865 		 * block groups per flexgroup, reserve the first block
866 		 * group for directories and special files.  Regular
867 		 * files will start at the second block group.  This
868 		 * tends to speed up directory access and improves
869 		 * fsck times.
870 		 */
871 		block_group &= ~(flex_size-1);
872 		if (S_ISREG(inode->i_mode))
873 			block_group++;
874 	}
875 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
876 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
877 
878 	/*
879 	 * If we are doing delayed allocation, we don't need take
880 	 * colour into account.
881 	 */
882 	if (test_opt(inode->i_sb, DELALLOC))
883 		return bg_start;
884 
885 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
886 		colour = (current->pid % 16) *
887 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
888 	else
889 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
890 	return bg_start + colour;
891 }
892 
893