xref: /openbmc/linux/fs/ext4/balloc.c (revision b627b4ed)
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/jbd2.h>
18 #include <linux/quotaops.h>
19 #include <linux/buffer_head.h>
20 #include "ext4.h"
21 #include "ext4_jbd2.h"
22 #include "group.h"
23 #include "mballoc.h"
24 
25 /*
26  * balloc.c contains the blocks allocation and deallocation routines
27  */
28 
29 /*
30  * Calculate the block group number and offset, given a block number
31  */
32 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
33 		ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
34 {
35 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
36 	ext4_grpblk_t offset;
37 
38 	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
39 	offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
40 	if (offsetp)
41 		*offsetp = offset;
42 	if (blockgrpp)
43 		*blockgrpp = blocknr;
44 
45 }
46 
47 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
48 			ext4_group_t block_group)
49 {
50 	ext4_group_t actual_group;
51 	ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
52 	if (actual_group == block_group)
53 		return 1;
54 	return 0;
55 }
56 
57 static int ext4_group_used_meta_blocks(struct super_block *sb,
58 				       ext4_group_t block_group,
59 				       struct ext4_group_desc *gdp)
60 {
61 	ext4_fsblk_t tmp;
62 	struct ext4_sb_info *sbi = EXT4_SB(sb);
63 	/* block bitmap, inode bitmap, and inode table blocks */
64 	int used_blocks = sbi->s_itb_per_group + 2;
65 
66 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
67 		if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
68 					block_group))
69 			used_blocks--;
70 
71 		if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
72 					block_group))
73 			used_blocks--;
74 
75 		tmp = ext4_inode_table(sb, gdp);
76 		for (; tmp < ext4_inode_table(sb, gdp) +
77 				sbi->s_itb_per_group; tmp++) {
78 			if (!ext4_block_in_group(sb, tmp, block_group))
79 				used_blocks -= 1;
80 		}
81 	}
82 	return used_blocks;
83 }
84 
85 /* Initializes an uninitialized block bitmap if given, and returns the
86  * number of blocks free in the group. */
87 unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
88 		 ext4_group_t block_group, struct ext4_group_desc *gdp)
89 {
90 	int bit, bit_max;
91 	unsigned free_blocks, group_blocks;
92 	struct ext4_sb_info *sbi = EXT4_SB(sb);
93 
94 	if (bh) {
95 		J_ASSERT_BH(bh, buffer_locked(bh));
96 
97 		/* If checksum is bad mark all blocks used to prevent allocation
98 		 * essentially implementing a per-group read-only flag. */
99 		if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
100 			ext4_error(sb, __func__,
101 				  "Checksum bad for group %u", block_group);
102 			ext4_free_blks_set(sb, gdp, 0);
103 			ext4_free_inodes_set(sb, gdp, 0);
104 			ext4_itable_unused_set(sb, gdp, 0);
105 			memset(bh->b_data, 0xff, sb->s_blocksize);
106 			return 0;
107 		}
108 		memset(bh->b_data, 0, sb->s_blocksize);
109 	}
110 
111 	/* Check for superblock and gdt backups in this group */
112 	bit_max = ext4_bg_has_super(sb, block_group);
113 
114 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
115 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
116 			  sbi->s_desc_per_block) {
117 		if (bit_max) {
118 			bit_max += ext4_bg_num_gdb(sb, block_group);
119 			bit_max +=
120 				le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
121 		}
122 	} else { /* For META_BG_BLOCK_GROUPS */
123 		bit_max += ext4_bg_num_gdb(sb, block_group);
124 	}
125 
126 	if (block_group == sbi->s_groups_count - 1) {
127 		/*
128 		 * Even though mke2fs always initialize first and last group
129 		 * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
130 		 * to make sure we calculate the right free blocks
131 		 */
132 		group_blocks = ext4_blocks_count(sbi->s_es) -
133 			le32_to_cpu(sbi->s_es->s_first_data_block) -
134 			(EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count - 1));
135 	} else {
136 		group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
137 	}
138 
139 	free_blocks = group_blocks - bit_max;
140 
141 	if (bh) {
142 		ext4_fsblk_t start, tmp;
143 		int flex_bg = 0;
144 
145 		for (bit = 0; bit < bit_max; bit++)
146 			ext4_set_bit(bit, bh->b_data);
147 
148 		start = ext4_group_first_block_no(sb, block_group);
149 
150 		if (EXT4_HAS_INCOMPAT_FEATURE(sb,
151 					      EXT4_FEATURE_INCOMPAT_FLEX_BG))
152 			flex_bg = 1;
153 
154 		/* Set bits for block and inode bitmaps, and inode table */
155 		tmp = ext4_block_bitmap(sb, gdp);
156 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
157 			ext4_set_bit(tmp - start, bh->b_data);
158 
159 		tmp = ext4_inode_bitmap(sb, gdp);
160 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
161 			ext4_set_bit(tmp - start, bh->b_data);
162 
163 		tmp = ext4_inode_table(sb, gdp);
164 		for (; tmp < ext4_inode_table(sb, gdp) +
165 				sbi->s_itb_per_group; tmp++) {
166 			if (!flex_bg ||
167 				ext4_block_in_group(sb, tmp, block_group))
168 				ext4_set_bit(tmp - start, bh->b_data);
169 		}
170 		/*
171 		 * Also if the number of blocks within the group is
172 		 * less than the blocksize * 8 ( which is the size
173 		 * of bitmap ), set rest of the block bitmap to 1
174 		 */
175 		mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
176 	}
177 	return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp);
178 }
179 
180 
181 /*
182  * The free blocks are managed by bitmaps.  A file system contains several
183  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
184  * block for inodes, N blocks for the inode table and data blocks.
185  *
186  * The file system contains group descriptors which are located after the
187  * super block.  Each descriptor contains the number of the bitmap block and
188  * the free blocks count in the block.  The descriptors are loaded in memory
189  * when a file system is mounted (see ext4_fill_super).
190  */
191 
192 
193 #define in_range(b, first, len)	((b) >= (first) && (b) <= (first) + (len) - 1)
194 
195 /**
196  * ext4_get_group_desc() -- load group descriptor from disk
197  * @sb:			super block
198  * @block_group:	given block group
199  * @bh:			pointer to the buffer head to store the block
200  *			group descriptor
201  */
202 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
203 					     ext4_group_t block_group,
204 					     struct buffer_head **bh)
205 {
206 	unsigned int group_desc;
207 	unsigned int offset;
208 	struct ext4_group_desc *desc;
209 	struct ext4_sb_info *sbi = EXT4_SB(sb);
210 
211 	if (block_group >= sbi->s_groups_count) {
212 		ext4_error(sb, "ext4_get_group_desc",
213 			   "block_group >= groups_count - "
214 			   "block_group = %u, groups_count = %u",
215 			   block_group, sbi->s_groups_count);
216 
217 		return NULL;
218 	}
219 	smp_rmb();
220 
221 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
222 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
223 	if (!sbi->s_group_desc[group_desc]) {
224 		ext4_error(sb, "ext4_get_group_desc",
225 			   "Group descriptor not loaded - "
226 			   "block_group = %u, group_desc = %u, desc = %u",
227 			   block_group, group_desc, offset);
228 		return NULL;
229 	}
230 
231 	desc = (struct ext4_group_desc *)(
232 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
233 		offset * EXT4_DESC_SIZE(sb));
234 	if (bh)
235 		*bh = sbi->s_group_desc[group_desc];
236 	return desc;
237 }
238 
239 static int ext4_valid_block_bitmap(struct super_block *sb,
240 					struct ext4_group_desc *desc,
241 					unsigned int block_group,
242 					struct buffer_head *bh)
243 {
244 	ext4_grpblk_t offset;
245 	ext4_grpblk_t next_zero_bit;
246 	ext4_fsblk_t bitmap_blk;
247 	ext4_fsblk_t group_first_block;
248 
249 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
250 		/* with FLEX_BG, the inode/block bitmaps and itable
251 		 * blocks may not be in the group at all
252 		 * so the bitmap validation will be skipped for those groups
253 		 * or it has to also read the block group where the bitmaps
254 		 * are located to verify they are set.
255 		 */
256 		return 1;
257 	}
258 	group_first_block = ext4_group_first_block_no(sb, block_group);
259 
260 	/* check whether block bitmap block number is set */
261 	bitmap_blk = ext4_block_bitmap(sb, desc);
262 	offset = bitmap_blk - group_first_block;
263 	if (!ext4_test_bit(offset, bh->b_data))
264 		/* bad block bitmap */
265 		goto err_out;
266 
267 	/* check whether the inode bitmap block number is set */
268 	bitmap_blk = ext4_inode_bitmap(sb, desc);
269 	offset = bitmap_blk - group_first_block;
270 	if (!ext4_test_bit(offset, bh->b_data))
271 		/* bad block bitmap */
272 		goto err_out;
273 
274 	/* check whether the inode table block number is set */
275 	bitmap_blk = ext4_inode_table(sb, desc);
276 	offset = bitmap_blk - group_first_block;
277 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
278 				offset + EXT4_SB(sb)->s_itb_per_group,
279 				offset);
280 	if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
281 		/* good bitmap for inode tables */
282 		return 1;
283 
284 err_out:
285 	ext4_error(sb, __func__,
286 			"Invalid block bitmap - "
287 			"block_group = %d, block = %llu",
288 			block_group, bitmap_blk);
289 	return 0;
290 }
291 /**
292  * ext4_read_block_bitmap()
293  * @sb:			super block
294  * @block_group:	given block group
295  *
296  * Read the bitmap for a given block_group,and validate the
297  * bits for block/inode/inode tables are set in the bitmaps
298  *
299  * Return buffer_head on success or NULL in case of failure.
300  */
301 struct buffer_head *
302 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
303 {
304 	struct ext4_group_desc *desc;
305 	struct buffer_head *bh = NULL;
306 	ext4_fsblk_t bitmap_blk;
307 
308 	desc = ext4_get_group_desc(sb, block_group, NULL);
309 	if (!desc)
310 		return NULL;
311 	bitmap_blk = ext4_block_bitmap(sb, desc);
312 	bh = sb_getblk(sb, bitmap_blk);
313 	if (unlikely(!bh)) {
314 		ext4_error(sb, __func__,
315 			    "Cannot read block bitmap - "
316 			    "block_group = %u, block_bitmap = %llu",
317 			    block_group, bitmap_blk);
318 		return NULL;
319 	}
320 
321 	if (bitmap_uptodate(bh))
322 		return bh;
323 
324 	lock_buffer(bh);
325 	if (bitmap_uptodate(bh)) {
326 		unlock_buffer(bh);
327 		return bh;
328 	}
329 	spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
330 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
331 		ext4_init_block_bitmap(sb, bh, block_group, desc);
332 		set_bitmap_uptodate(bh);
333 		set_buffer_uptodate(bh);
334 		spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
335 		unlock_buffer(bh);
336 		return bh;
337 	}
338 	spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
339 	if (buffer_uptodate(bh)) {
340 		/*
341 		 * if not uninit if bh is uptodate,
342 		 * bitmap is also uptodate
343 		 */
344 		set_bitmap_uptodate(bh);
345 		unlock_buffer(bh);
346 		return bh;
347 	}
348 	/*
349 	 * submit the buffer_head for read. We can
350 	 * safely mark the bitmap as uptodate now.
351 	 * We do it here so the bitmap uptodate bit
352 	 * get set with buffer lock held.
353 	 */
354 	set_bitmap_uptodate(bh);
355 	if (bh_submit_read(bh) < 0) {
356 		put_bh(bh);
357 		ext4_error(sb, __func__,
358 			    "Cannot read block bitmap - "
359 			    "block_group = %u, block_bitmap = %llu",
360 			    block_group, bitmap_blk);
361 		return NULL;
362 	}
363 	ext4_valid_block_bitmap(sb, desc, block_group, bh);
364 	/*
365 	 * file system mounted not to panic on error,
366 	 * continue with corrupt bitmap
367 	 */
368 	return bh;
369 }
370 
371 /**
372  * ext4_add_groupblocks() -- Add given blocks to an existing group
373  * @handle:			handle to this transaction
374  * @sb:				super block
375  * @block:			start physcial block to add to the block group
376  * @count:			number of blocks to free
377  *
378  * This marks the blocks as free in the bitmap. We ask the
379  * mballoc to reload the buddy after this by setting group
380  * EXT4_GROUP_INFO_NEED_INIT_BIT flag
381  */
382 void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
383 			 ext4_fsblk_t block, unsigned long count)
384 {
385 	struct buffer_head *bitmap_bh = NULL;
386 	struct buffer_head *gd_bh;
387 	ext4_group_t block_group;
388 	ext4_grpblk_t bit;
389 	unsigned int i;
390 	struct ext4_group_desc *desc;
391 	struct ext4_super_block *es;
392 	struct ext4_sb_info *sbi;
393 	int err = 0, ret, blk_free_count;
394 	ext4_grpblk_t blocks_freed;
395 	struct ext4_group_info *grp;
396 
397 	sbi = EXT4_SB(sb);
398 	es = sbi->s_es;
399 	ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
400 
401 	ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
402 	grp = ext4_get_group_info(sb, block_group);
403 	/*
404 	 * Check to see if we are freeing blocks across a group
405 	 * boundary.
406 	 */
407 	if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
408 		goto error_return;
409 	}
410 	bitmap_bh = ext4_read_block_bitmap(sb, block_group);
411 	if (!bitmap_bh)
412 		goto error_return;
413 	desc = ext4_get_group_desc(sb, block_group, &gd_bh);
414 	if (!desc)
415 		goto error_return;
416 
417 	if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
418 	    in_range(ext4_inode_bitmap(sb, desc), block, count) ||
419 	    in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
420 	    in_range(block + count - 1, ext4_inode_table(sb, desc),
421 		     sbi->s_itb_per_group)) {
422 		ext4_error(sb, __func__,
423 			   "Adding blocks in system zones - "
424 			   "Block = %llu, count = %lu",
425 			   block, count);
426 		goto error_return;
427 	}
428 
429 	/*
430 	 * We are about to add blocks to the bitmap,
431 	 * so we need undo access.
432 	 */
433 	BUFFER_TRACE(bitmap_bh, "getting undo access");
434 	err = ext4_journal_get_undo_access(handle, bitmap_bh);
435 	if (err)
436 		goto error_return;
437 
438 	/*
439 	 * We are about to modify some metadata.  Call the journal APIs
440 	 * to unshare ->b_data if a currently-committing transaction is
441 	 * using it
442 	 */
443 	BUFFER_TRACE(gd_bh, "get_write_access");
444 	err = ext4_journal_get_write_access(handle, gd_bh);
445 	if (err)
446 		goto error_return;
447 	/*
448 	 * make sure we don't allow a parallel init on other groups in the
449 	 * same buddy cache
450 	 */
451 	down_write(&grp->alloc_sem);
452 	for (i = 0, blocks_freed = 0; i < count; i++) {
453 		BUFFER_TRACE(bitmap_bh, "clear bit");
454 		if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
455 						bit + i, bitmap_bh->b_data)) {
456 			ext4_error(sb, __func__,
457 				   "bit already cleared for block %llu",
458 				   (ext4_fsblk_t)(block + i));
459 			BUFFER_TRACE(bitmap_bh, "bit already cleared");
460 		} else {
461 			blocks_freed++;
462 		}
463 	}
464 	spin_lock(sb_bgl_lock(sbi, block_group));
465 	blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
466 	ext4_free_blks_set(sb, desc, blk_free_count);
467 	desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
468 	spin_unlock(sb_bgl_lock(sbi, block_group));
469 	percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
470 
471 	if (sbi->s_log_groups_per_flex) {
472 		ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
473 		atomic_add(blocks_freed,
474 			   &sbi->s_flex_groups[flex_group].free_blocks);
475 	}
476 	/*
477 	 * request to reload the buddy with the
478 	 * new bitmap information
479 	 */
480 	set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
481 	ext4_mb_update_group_info(grp, blocks_freed);
482 	up_write(&grp->alloc_sem);
483 
484 	/* We dirtied the bitmap block */
485 	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
486 	err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
487 
488 	/* And the group descriptor block */
489 	BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
490 	ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
491 	if (!err)
492 		err = ret;
493 	sb->s_dirt = 1;
494 
495 error_return:
496 	brelse(bitmap_bh);
497 	ext4_std_error(sb, err);
498 	return;
499 }
500 
501 /**
502  * ext4_free_blocks() -- Free given blocks and update quota
503  * @handle:		handle for this transaction
504  * @inode:		inode
505  * @block:		start physical block to free
506  * @count:		number of blocks to count
507  * @metadata: 		Are these metadata blocks
508  */
509 void ext4_free_blocks(handle_t *handle, struct inode *inode,
510 			ext4_fsblk_t block, unsigned long count,
511 			int metadata)
512 {
513 	struct super_block *sb;
514 	unsigned long dquot_freed_blocks;
515 
516 	/* this isn't the right place to decide whether block is metadata
517 	 * inode.c/extents.c knows better, but for safety ... */
518 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
519 		metadata = 1;
520 
521 	/* We need to make sure we don't reuse
522 	 * block released untill the transaction commit.
523 	 * writeback mode have weak data consistency so
524 	 * don't force data as metadata when freeing block
525 	 * for writeback mode.
526 	 */
527 	if (metadata == 0 && !ext4_should_writeback_data(inode))
528 		metadata = 1;
529 
530 	sb = inode->i_sb;
531 
532 	ext4_mb_free_blocks(handle, inode, block, count,
533 			    metadata, &dquot_freed_blocks);
534 	if (dquot_freed_blocks)
535 		vfs_dq_free_block(inode, dquot_freed_blocks);
536 	return;
537 }
538 
539 /**
540  * ext4_has_free_blocks()
541  * @sbi:	in-core super block structure.
542  * @nblocks:	number of needed blocks
543  *
544  * Check if filesystem has nblocks free & available for allocation.
545  * On success return 1, return 0 on failure.
546  */
547 int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
548 {
549 	s64 free_blocks, dirty_blocks, root_blocks;
550 	struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
551 	struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
552 
553 	free_blocks  = percpu_counter_read_positive(fbc);
554 	dirty_blocks = percpu_counter_read_positive(dbc);
555 	root_blocks = ext4_r_blocks_count(sbi->s_es);
556 
557 	if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
558 						EXT4_FREEBLOCKS_WATERMARK) {
559 		free_blocks  = percpu_counter_sum_positive(fbc);
560 		dirty_blocks = percpu_counter_sum_positive(dbc);
561 		if (dirty_blocks < 0) {
562 			printk(KERN_CRIT "Dirty block accounting "
563 					"went wrong %lld\n",
564 					(long long)dirty_blocks);
565 		}
566 	}
567 	/* Check whether we have space after
568 	 * accounting for current dirty blocks & root reserved blocks.
569 	 */
570 	if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
571 		return 1;
572 
573 	/* Hm, nope.  Are (enough) root reserved blocks available? */
574 	if (sbi->s_resuid == current_fsuid() ||
575 	    ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
576 	    capable(CAP_SYS_RESOURCE)) {
577 		if (free_blocks >= (nblocks + dirty_blocks))
578 			return 1;
579 	}
580 
581 	return 0;
582 }
583 
584 int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
585 						s64 nblocks)
586 {
587 	if (ext4_has_free_blocks(sbi, nblocks)) {
588 		percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
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 commiting transaction to complete, and then
602  * return TRUE.
603  *
604  * if the total number of retries exceed three times, return FALSE.
605  */
606 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
607 {
608 	if (!ext4_has_free_blocks(EXT4_SB(sb), 1) ||
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 	return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
616 }
617 
618 /*
619  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
620  *
621  * @handle:             handle to this transaction
622  * @inode:              file inode
623  * @goal:               given target block(filesystem wide)
624  * @count:		pointer to total number of blocks needed
625  * @errp:               error code
626  *
627  * Return 1st allocated block number on success, *count stores total account
628  * error stores in errp pointer
629  */
630 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
631 		ext4_fsblk_t goal, unsigned long *count, int *errp)
632 {
633 	struct ext4_allocation_request ar;
634 	ext4_fsblk_t ret;
635 
636 	memset(&ar, 0, sizeof(ar));
637 	/* Fill with neighbour allocated blocks */
638 	ar.inode = inode;
639 	ar.goal = goal;
640 	ar.len = count ? *count : 1;
641 
642 	ret = ext4_mb_new_blocks(handle, &ar, errp);
643 	if (count)
644 		*count = ar.len;
645 
646 	/*
647 	 * Account for the allocated meta blocks
648 	 */
649 	if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
650 		spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
651 		EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
652 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
653 	}
654 	return ret;
655 }
656 
657 /**
658  * ext4_count_free_blocks() -- count filesystem free blocks
659  * @sb:		superblock
660  *
661  * Adds up the number of free blocks from each block group.
662  */
663 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
664 {
665 	ext4_fsblk_t desc_count;
666 	struct ext4_group_desc *gdp;
667 	ext4_group_t i;
668 	ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
669 #ifdef EXT4FS_DEBUG
670 	struct ext4_super_block *es;
671 	ext4_fsblk_t bitmap_count;
672 	unsigned int x;
673 	struct buffer_head *bitmap_bh = NULL;
674 
675 	es = EXT4_SB(sb)->s_es;
676 	desc_count = 0;
677 	bitmap_count = 0;
678 	gdp = NULL;
679 
680 	smp_rmb();
681 	for (i = 0; i < ngroups; i++) {
682 		gdp = ext4_get_group_desc(sb, i, NULL);
683 		if (!gdp)
684 			continue;
685 		desc_count += ext4_free_blks_count(sb, gdp);
686 		brelse(bitmap_bh);
687 		bitmap_bh = ext4_read_block_bitmap(sb, i);
688 		if (bitmap_bh == NULL)
689 			continue;
690 
691 		x = ext4_count_free(bitmap_bh, sb->s_blocksize);
692 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
693 			i, ext4_free_blks_count(sb, gdp), x);
694 		bitmap_count += x;
695 	}
696 	brelse(bitmap_bh);
697 	printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
698 		", computed = %llu, %llu\n", ext4_free_blocks_count(es),
699 	       desc_count, bitmap_count);
700 	return bitmap_count;
701 #else
702 	desc_count = 0;
703 	smp_rmb();
704 	for (i = 0; i < ngroups; i++) {
705 		gdp = ext4_get_group_desc(sb, i, NULL);
706 		if (!gdp)
707 			continue;
708 		desc_count += ext4_free_blks_count(sb, gdp);
709 	}
710 
711 	return desc_count;
712 #endif
713 }
714 
715 static inline int test_root(ext4_group_t a, int b)
716 {
717 	int num = b;
718 
719 	while (a > num)
720 		num *= b;
721 	return num == a;
722 }
723 
724 static int ext4_group_sparse(ext4_group_t group)
725 {
726 	if (group <= 1)
727 		return 1;
728 	if (!(group & 1))
729 		return 0;
730 	return (test_root(group, 7) || test_root(group, 5) ||
731 		test_root(group, 3));
732 }
733 
734 /**
735  *	ext4_bg_has_super - number of blocks used by the superblock in group
736  *	@sb: superblock for filesystem
737  *	@group: group number to check
738  *
739  *	Return the number of blocks used by the superblock (primary or backup)
740  *	in this group.  Currently this will be only 0 or 1.
741  */
742 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
743 {
744 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
745 				EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
746 			!ext4_group_sparse(group))
747 		return 0;
748 	return 1;
749 }
750 
751 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
752 					ext4_group_t group)
753 {
754 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
755 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
756 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
757 
758 	if (group == first || group == first + 1 || group == last)
759 		return 1;
760 	return 0;
761 }
762 
763 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
764 					ext4_group_t group)
765 {
766 	return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0;
767 }
768 
769 /**
770  *	ext4_bg_num_gdb - number of blocks used by the group table in group
771  *	@sb: superblock for filesystem
772  *	@group: group number to check
773  *
774  *	Return the number of blocks used by the group descriptor table
775  *	(primary or backup) in this group.  In the future there may be a
776  *	different number of descriptor blocks in each group.
777  */
778 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
779 {
780 	unsigned long first_meta_bg =
781 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
782 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
783 
784 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
785 			metagroup < first_meta_bg)
786 		return ext4_bg_num_gdb_nometa(sb, group);
787 
788 	return ext4_bg_num_gdb_meta(sb,group);
789 
790 }
791 
792