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 "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 the block group number and offset into the block/cluster
34  * allocation bitmap, given a block number
35  */
ext4_get_group_no_and_offset(struct super_block * sb,ext4_fsblk_t blocknr,ext4_group_t * blockgrpp,ext4_grpblk_t * offsetp)36 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
37 		ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
38 {
39 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
40 	ext4_grpblk_t offset;
41 
42 	blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
43 	offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
44 		EXT4_SB(sb)->s_cluster_bits;
45 	if (offsetp)
46 		*offsetp = offset;
47 	if (blockgrpp)
48 		*blockgrpp = blocknr;
49 
50 }
51 
ext4_block_in_group(struct super_block * sb,ext4_fsblk_t block,ext4_group_t block_group)52 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
53 			ext4_group_t block_group)
54 {
55 	ext4_group_t actual_group;
56 	ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
57 	if (actual_group == block_group)
58 		return 1;
59 	return 0;
60 }
61 
62 /* Return the number of clusters used for file system metadata; this
63  * represents the overhead needed by the file system.
64  */
ext4_num_overhead_clusters(struct super_block * sb,ext4_group_t block_group,struct ext4_group_desc * gdp)65 unsigned ext4_num_overhead_clusters(struct super_block *sb,
66 				    ext4_group_t block_group,
67 				    struct ext4_group_desc *gdp)
68 {
69 	unsigned num_clusters;
70 	int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
71 	ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
72 	ext4_fsblk_t itbl_blk;
73 	struct ext4_sb_info *sbi = EXT4_SB(sb);
74 
75 	/* This is the number of clusters used by the superblock,
76 	 * block group descriptors, and reserved block group
77 	 * descriptor blocks */
78 	num_clusters = ext4_num_base_meta_clusters(sb, block_group);
79 
80 	/*
81 	 * For the allocation bitmaps and inode table, we first need
82 	 * to check to see if the block is in the block group.  If it
83 	 * is, then check to see if the cluster is already accounted
84 	 * for in the clusters used for the base metadata cluster, or
85 	 * if we can increment the base metadata cluster to include
86 	 * that block.  Otherwise, we will have to track the cluster
87 	 * used for the allocation bitmap or inode table explicitly.
88 	 * Normally all of these blocks are contiguous, so the special
89 	 * case handling shouldn't be necessary except for *very*
90 	 * unusual file system layouts.
91 	 */
92 	if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
93 		block_cluster = EXT4_B2C(sbi,
94 					 ext4_block_bitmap(sb, gdp) - start);
95 		if (block_cluster < num_clusters)
96 			block_cluster = -1;
97 		else if (block_cluster == num_clusters) {
98 			num_clusters++;
99 			block_cluster = -1;
100 		}
101 	}
102 
103 	if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
104 		inode_cluster = EXT4_B2C(sbi,
105 					 ext4_inode_bitmap(sb, gdp) - start);
106 		if (inode_cluster < num_clusters)
107 			inode_cluster = -1;
108 		else if (inode_cluster == num_clusters) {
109 			num_clusters++;
110 			inode_cluster = -1;
111 		}
112 	}
113 
114 	itbl_blk = ext4_inode_table(sb, gdp);
115 	for (i = 0; i < sbi->s_itb_per_group; i++) {
116 		if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
117 			c = EXT4_B2C(sbi, itbl_blk + i - start);
118 			if ((c < num_clusters) || (c == inode_cluster) ||
119 			    (c == block_cluster) || (c == itbl_cluster))
120 				continue;
121 			if (c == num_clusters) {
122 				num_clusters++;
123 				continue;
124 			}
125 			num_clusters++;
126 			itbl_cluster = c;
127 		}
128 	}
129 
130 	if (block_cluster != -1)
131 		num_clusters++;
132 	if (inode_cluster != -1)
133 		num_clusters++;
134 
135 	return num_clusters;
136 }
137 
num_clusters_in_group(struct super_block * sb,ext4_group_t block_group)138 static unsigned int num_clusters_in_group(struct super_block *sb,
139 					  ext4_group_t block_group)
140 {
141 	unsigned int blocks;
142 
143 	if (block_group == ext4_get_groups_count(sb) - 1) {
144 		/*
145 		 * Even though mke2fs always initializes the first and
146 		 * last group, just in case some other tool was used,
147 		 * we need to make sure we calculate the right free
148 		 * blocks.
149 		 */
150 		blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
151 			ext4_group_first_block_no(sb, block_group);
152 	} else
153 		blocks = EXT4_BLOCKS_PER_GROUP(sb);
154 	return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
155 }
156 
157 /* 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)158 void ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
159 			    ext4_group_t block_group,
160 			    struct ext4_group_desc *gdp)
161 {
162 	unsigned int bit, bit_max;
163 	struct ext4_sb_info *sbi = EXT4_SB(sb);
164 	ext4_fsblk_t start, tmp;
165 	int flex_bg = 0;
166 
167 	J_ASSERT_BH(bh, buffer_locked(bh));
168 
169 	/* If checksum is bad mark all blocks used to prevent allocation
170 	 * essentially implementing a per-group read-only flag. */
171 	if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
172 		ext4_error(sb, "Checksum bad for group %u", block_group);
173 		ext4_free_group_clusters_set(sb, gdp, 0);
174 		ext4_free_inodes_set(sb, gdp, 0);
175 		ext4_itable_unused_set(sb, gdp, 0);
176 		memset(bh->b_data, 0xff, sb->s_blocksize);
177 		return;
178 	}
179 	memset(bh->b_data, 0, sb->s_blocksize);
180 
181 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
182 	for (bit = 0; bit < bit_max; bit++)
183 		ext4_set_bit(bit, bh->b_data);
184 
185 	start = ext4_group_first_block_no(sb, block_group);
186 
187 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
188 		flex_bg = 1;
189 
190 	/* Set bits for block and inode bitmaps, and inode table */
191 	tmp = ext4_block_bitmap(sb, gdp);
192 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
193 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
194 
195 	tmp = ext4_inode_bitmap(sb, gdp);
196 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
197 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
198 
199 	tmp = ext4_inode_table(sb, gdp);
200 	for (; tmp < ext4_inode_table(sb, gdp) +
201 		     sbi->s_itb_per_group; tmp++) {
202 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
203 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
204 	}
205 
206 	/*
207 	 * Also if the number of blocks within the group is less than
208 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
209 	 * of the block bitmap to 1
210 	 */
211 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
212 			     sb->s_blocksize * 8, bh->b_data);
213 }
214 
215 /* Return the number of free blocks in a block group.  It is used when
216  * the block bitmap is uninitialized, so we can't just count the bits
217  * in the bitmap. */
ext4_free_clusters_after_init(struct super_block * sb,ext4_group_t block_group,struct ext4_group_desc * gdp)218 unsigned ext4_free_clusters_after_init(struct super_block *sb,
219 				       ext4_group_t block_group,
220 				       struct ext4_group_desc *gdp)
221 {
222 	return num_clusters_in_group(sb, block_group) -
223 		ext4_num_overhead_clusters(sb, block_group, gdp);
224 }
225 
226 /*
227  * The free blocks are managed by bitmaps.  A file system contains several
228  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
229  * block for inodes, N blocks for the inode table and data blocks.
230  *
231  * The file system contains group descriptors which are located after the
232  * super block.  Each descriptor contains the number of the bitmap block and
233  * the free blocks count in the block.  The descriptors are loaded in memory
234  * when a file system is mounted (see ext4_fill_super).
235  */
236 
237 /**
238  * ext4_get_group_desc() -- load group descriptor from disk
239  * @sb:			super block
240  * @block_group:	given block group
241  * @bh:			pointer to the buffer head to store the block
242  *			group descriptor
243  */
ext4_get_group_desc(struct super_block * sb,ext4_group_t block_group,struct buffer_head ** bh)244 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
245 					     ext4_group_t block_group,
246 					     struct buffer_head **bh)
247 {
248 	unsigned int group_desc;
249 	unsigned int offset;
250 	ext4_group_t ngroups = ext4_get_groups_count(sb);
251 	struct ext4_group_desc *desc;
252 	struct ext4_sb_info *sbi = EXT4_SB(sb);
253 
254 	if (block_group >= ngroups) {
255 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
256 			   " groups_count = %u", block_group, ngroups);
257 
258 		return NULL;
259 	}
260 
261 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
262 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
263 	if (!sbi->s_group_desc[group_desc]) {
264 		ext4_error(sb, "Group descriptor not loaded - "
265 			   "block_group = %u, group_desc = %u, desc = %u",
266 			   block_group, group_desc, offset);
267 		return NULL;
268 	}
269 
270 	desc = (struct ext4_group_desc *)(
271 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
272 		offset * EXT4_DESC_SIZE(sb));
273 	if (bh)
274 		*bh = sbi->s_group_desc[group_desc];
275 	return desc;
276 }
277 
ext4_valid_block_bitmap(struct super_block * sb,struct ext4_group_desc * desc,unsigned int block_group,struct buffer_head * bh)278 static int ext4_valid_block_bitmap(struct super_block *sb,
279 					struct ext4_group_desc *desc,
280 					unsigned int block_group,
281 					struct buffer_head *bh)
282 {
283 	ext4_grpblk_t offset;
284 	ext4_grpblk_t next_zero_bit;
285 	ext4_fsblk_t bitmap_blk;
286 	ext4_fsblk_t group_first_block;
287 
288 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
289 		/* with FLEX_BG, the inode/block bitmaps and itable
290 		 * blocks may not be in the group at all
291 		 * so the bitmap validation will be skipped for those groups
292 		 * or it has to also read the block group where the bitmaps
293 		 * are located to verify they are set.
294 		 */
295 		return 1;
296 	}
297 	group_first_block = ext4_group_first_block_no(sb, block_group);
298 
299 	/* check whether block bitmap block number is set */
300 	bitmap_blk = ext4_block_bitmap(sb, desc);
301 	offset = bitmap_blk - group_first_block;
302 	if (!ext4_test_bit(offset, bh->b_data))
303 		/* bad block bitmap */
304 		goto err_out;
305 
306 	/* check whether the inode bitmap block number is set */
307 	bitmap_blk = ext4_inode_bitmap(sb, desc);
308 	offset = bitmap_blk - group_first_block;
309 	if (!ext4_test_bit(offset, bh->b_data))
310 		/* bad block bitmap */
311 		goto err_out;
312 
313 	/* check whether the inode table block number is set */
314 	bitmap_blk = ext4_inode_table(sb, desc);
315 	offset = bitmap_blk - group_first_block;
316 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
317 				offset + EXT4_SB(sb)->s_itb_per_group,
318 				offset);
319 	if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
320 		/* good bitmap for inode tables */
321 		return 1;
322 
323 err_out:
324 	ext4_error(sb, "Invalid block bitmap - block_group = %d, block = %llu",
325 			block_group, bitmap_blk);
326 	return 0;
327 }
328 /**
329  * ext4_read_block_bitmap_nowait()
330  * @sb:			super block
331  * @block_group:	given block group
332  *
333  * Read the bitmap for a given block_group,and validate the
334  * bits for block/inode/inode tables are set in the bitmaps
335  *
336  * Return buffer_head on success or NULL in case of failure.
337  */
338 struct buffer_head *
ext4_read_block_bitmap_nowait(struct super_block * sb,ext4_group_t block_group)339 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
340 {
341 	struct ext4_group_desc *desc;
342 	struct buffer_head *bh;
343 	ext4_fsblk_t bitmap_blk;
344 
345 	desc = ext4_get_group_desc(sb, block_group, NULL);
346 	if (!desc)
347 		return NULL;
348 	bitmap_blk = ext4_block_bitmap(sb, desc);
349 	bh = sb_getblk(sb, bitmap_blk);
350 	if (unlikely(!bh)) {
351 		ext4_error(sb, "Cannot get buffer for block bitmap - "
352 			   "block_group = %u, block_bitmap = %llu",
353 			   block_group, bitmap_blk);
354 		return NULL;
355 	}
356 
357 	if (bitmap_uptodate(bh))
358 		return bh;
359 
360 	lock_buffer(bh);
361 	if (bitmap_uptodate(bh)) {
362 		unlock_buffer(bh);
363 		return bh;
364 	}
365 	ext4_lock_group(sb, block_group);
366 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
367 		ext4_init_block_bitmap(sb, bh, block_group, desc);
368 		set_bitmap_uptodate(bh);
369 		set_buffer_uptodate(bh);
370 		ext4_unlock_group(sb, block_group);
371 		unlock_buffer(bh);
372 		return bh;
373 	}
374 	ext4_unlock_group(sb, block_group);
375 	if (buffer_uptodate(bh)) {
376 		/*
377 		 * if not uninit if bh is uptodate,
378 		 * bitmap is also uptodate
379 		 */
380 		set_bitmap_uptodate(bh);
381 		unlock_buffer(bh);
382 		return bh;
383 	}
384 	/*
385 	 * submit the buffer_head for reading
386 	 */
387 	set_buffer_new(bh);
388 	trace_ext4_read_block_bitmap_load(sb, block_group);
389 	bh->b_end_io = ext4_end_bitmap_read;
390 	get_bh(bh);
391 	submit_bh(READ, bh);
392 	return bh;
393 }
394 
395 /* Returns 0 on success, 1 on error */
ext4_wait_block_bitmap(struct super_block * sb,ext4_group_t block_group,struct buffer_head * bh)396 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
397 			   struct buffer_head *bh)
398 {
399 	struct ext4_group_desc *desc;
400 
401 	if (!buffer_new(bh))
402 		return 0;
403 	desc = ext4_get_group_desc(sb, block_group, NULL);
404 	if (!desc)
405 		return 1;
406 	wait_on_buffer(bh);
407 	if (!buffer_uptodate(bh)) {
408 		ext4_error(sb, "Cannot read block bitmap - "
409 			   "block_group = %u, block_bitmap = %llu",
410 			   block_group, (unsigned long long) bh->b_blocknr);
411 		return 1;
412 	}
413 	clear_buffer_new(bh);
414 	/* Panic or remount fs read-only if block bitmap is invalid */
415 	ext4_valid_block_bitmap(sb, desc, block_group, bh);
416 	return 0;
417 }
418 
419 struct buffer_head *
ext4_read_block_bitmap(struct super_block * sb,ext4_group_t block_group)420 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
421 {
422 	struct buffer_head *bh;
423 
424 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
425 	if (!bh)
426 		return NULL;
427 	if (ext4_wait_block_bitmap(sb, block_group, bh)) {
428 		put_bh(bh);
429 		return NULL;
430 	}
431 	return bh;
432 }
433 
434 /**
435  * ext4_has_free_clusters()
436  * @sbi:	in-core super block structure.
437  * @nclusters:	number of needed blocks
438  * @flags:	flags from ext4_mb_new_blocks()
439  *
440  * Check if filesystem has nclusters free & available for allocation.
441  * On success return 1, return 0 on failure.
442  */
ext4_has_free_clusters(struct ext4_sb_info * sbi,s64 nclusters,unsigned int flags)443 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
444 				  s64 nclusters, unsigned int flags)
445 {
446 	s64 free_clusters, dirty_clusters, root_clusters;
447 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
448 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
449 
450 	free_clusters  = percpu_counter_read_positive(fcc);
451 	dirty_clusters = percpu_counter_read_positive(dcc);
452 
453 	/*
454 	 * r_blocks_count should always be multiple of the cluster ratio so
455 	 * we are safe to do a plane bit shift only.
456 	 */
457 	root_clusters = ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits;
458 
459 	if (free_clusters - (nclusters + root_clusters + dirty_clusters) <
460 					EXT4_FREECLUSTERS_WATERMARK) {
461 		free_clusters  = percpu_counter_sum_positive(fcc);
462 		dirty_clusters = percpu_counter_sum_positive(dcc);
463 	}
464 	/* Check whether we have space after accounting for current
465 	 * dirty clusters & root reserved clusters.
466 	 */
467 	if (free_clusters >= ((root_clusters + nclusters) + dirty_clusters))
468 		return 1;
469 
470 	/* Hm, nope.  Are (enough) root reserved clusters available? */
471 	if (sbi->s_resuid == current_fsuid() ||
472 	    ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
473 	    capable(CAP_SYS_RESOURCE) ||
474 		(flags & EXT4_MB_USE_ROOT_BLOCKS)) {
475 
476 		if (free_clusters >= (nclusters + dirty_clusters))
477 			return 1;
478 	}
479 
480 	return 0;
481 }
482 
ext4_claim_free_clusters(struct ext4_sb_info * sbi,s64 nclusters,unsigned int flags)483 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
484 			     s64 nclusters, unsigned int flags)
485 {
486 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
487 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
488 		return 0;
489 	} else
490 		return -ENOSPC;
491 }
492 
493 /**
494  * ext4_should_retry_alloc()
495  * @sb:			super block
496  * @retries		number of attemps has been made
497  *
498  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
499  * it is profitable to retry the operation, this function will wait
500  * for the current or committing transaction to complete, and then
501  * return TRUE.
502  *
503  * if the total number of retries exceed three times, return FALSE.
504  */
ext4_should_retry_alloc(struct super_block * sb,int * retries)505 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
506 {
507 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
508 	    (*retries)++ > 3 ||
509 	    !EXT4_SB(sb)->s_journal)
510 		return 0;
511 
512 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
513 
514 	return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
515 }
516 
517 /*
518  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
519  *
520  * @handle:             handle to this transaction
521  * @inode:              file inode
522  * @goal:               given target block(filesystem wide)
523  * @count:		pointer to total number of clusters needed
524  * @errp:               error code
525  *
526  * Return 1st allocated block number on success, *count stores total account
527  * error stores in errp pointer
528  */
ext4_new_meta_blocks(handle_t * handle,struct inode * inode,ext4_fsblk_t goal,unsigned int flags,unsigned long * count,int * errp)529 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
530 				  ext4_fsblk_t goal, unsigned int flags,
531 				  unsigned long *count, int *errp)
532 {
533 	struct ext4_allocation_request ar;
534 	ext4_fsblk_t ret;
535 
536 	memset(&ar, 0, sizeof(ar));
537 	/* Fill with neighbour allocated blocks */
538 	ar.inode = inode;
539 	ar.goal = goal;
540 	ar.len = count ? *count : 1;
541 	ar.flags = flags;
542 
543 	ret = ext4_mb_new_blocks(handle, &ar, errp);
544 	if (count)
545 		*count = ar.len;
546 	/*
547 	 * Account for the allocated meta blocks.  We will never
548 	 * fail EDQUOT for metdata, but we do account for it.
549 	 */
550 	if (!(*errp) &&
551 	    ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
552 		spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
553 		EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
554 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
555 		dquot_alloc_block_nofail(inode,
556 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
557 	}
558 	return ret;
559 }
560 
561 /**
562  * ext4_count_free_clusters() -- count filesystem free clusters
563  * @sb:		superblock
564  *
565  * Adds up the number of free clusters from each block group.
566  */
ext4_count_free_clusters(struct super_block * sb)567 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
568 {
569 	ext4_fsblk_t desc_count;
570 	struct ext4_group_desc *gdp;
571 	ext4_group_t i;
572 	ext4_group_t ngroups = ext4_get_groups_count(sb);
573 #ifdef EXT4FS_DEBUG
574 	struct ext4_super_block *es;
575 	ext4_fsblk_t bitmap_count;
576 	unsigned int x;
577 	struct buffer_head *bitmap_bh = NULL;
578 
579 	es = EXT4_SB(sb)->s_es;
580 	desc_count = 0;
581 	bitmap_count = 0;
582 	gdp = NULL;
583 
584 	for (i = 0; i < ngroups; i++) {
585 		gdp = ext4_get_group_desc(sb, i, NULL);
586 		if (!gdp)
587 			continue;
588 		desc_count += ext4_free_group_clusters(sb, gdp);
589 		brelse(bitmap_bh);
590 		bitmap_bh = ext4_read_block_bitmap(sb, i);
591 		if (bitmap_bh == NULL)
592 			continue;
593 
594 		x = ext4_count_free(bitmap_bh->b_data,
595 				    EXT4_BLOCKS_PER_GROUP(sb) / 8);
596 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
597 			i, ext4_free_group_clusters(sb, gdp), x);
598 		bitmap_count += x;
599 	}
600 	brelse(bitmap_bh);
601 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
602 	       ", computed = %llu, %llu\n",
603 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
604 	       desc_count, bitmap_count);
605 	return bitmap_count;
606 #else
607 	desc_count = 0;
608 	for (i = 0; i < ngroups; i++) {
609 		gdp = ext4_get_group_desc(sb, i, NULL);
610 		if (!gdp)
611 			continue;
612 		desc_count += ext4_free_group_clusters(sb, gdp);
613 	}
614 
615 	return desc_count;
616 #endif
617 }
618 
test_root(ext4_group_t a,int b)619 static inline int test_root(ext4_group_t a, int b)
620 {
621 	int num = b;
622 
623 	while (a > num)
624 		num *= b;
625 	return num == a;
626 }
627 
ext4_group_sparse(ext4_group_t group)628 static int ext4_group_sparse(ext4_group_t group)
629 {
630 	if (group <= 1)
631 		return 1;
632 	if (!(group & 1))
633 		return 0;
634 	return (test_root(group, 7) || test_root(group, 5) ||
635 		test_root(group, 3));
636 }
637 
638 /**
639  *	ext4_bg_has_super - number of blocks used by the superblock in group
640  *	@sb: superblock for filesystem
641  *	@group: group number to check
642  *
643  *	Return the number of blocks used by the superblock (primary or backup)
644  *	in this group.  Currently this will be only 0 or 1.
645  */
ext4_bg_has_super(struct super_block * sb,ext4_group_t group)646 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
647 {
648 	if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
649 				EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
650 			!ext4_group_sparse(group))
651 		return 0;
652 	return 1;
653 }
654 
ext4_bg_num_gdb_meta(struct super_block * sb,ext4_group_t group)655 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
656 					ext4_group_t group)
657 {
658 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
659 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
660 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
661 
662 	if (group == first || group == first + 1 || group == last)
663 		return 1;
664 	return 0;
665 }
666 
ext4_bg_num_gdb_nometa(struct super_block * sb,ext4_group_t group)667 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
668 					ext4_group_t group)
669 {
670 	if (!ext4_bg_has_super(sb, group))
671 		return 0;
672 
673 	if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
674 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
675 	else
676 		return EXT4_SB(sb)->s_gdb_count;
677 }
678 
679 /**
680  *	ext4_bg_num_gdb - number of blocks used by the group table in group
681  *	@sb: superblock for filesystem
682  *	@group: group number to check
683  *
684  *	Return the number of blocks used by the group descriptor table
685  *	(primary or backup) in this group.  In the future there may be a
686  *	different number of descriptor blocks in each group.
687  */
ext4_bg_num_gdb(struct super_block * sb,ext4_group_t group)688 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
689 {
690 	unsigned long first_meta_bg =
691 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
692 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
693 
694 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
695 			metagroup < first_meta_bg)
696 		return ext4_bg_num_gdb_nometa(sb, group);
697 
698 	return ext4_bg_num_gdb_meta(sb,group);
699 
700 }
701 
702 /*
703  * This function returns the number of file system metadata clusters at
704  * the beginning of a block group, including the reserved gdt blocks.
705  */
ext4_num_base_meta_clusters(struct super_block * sb,ext4_group_t block_group)706 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
707 				     ext4_group_t block_group)
708 {
709 	struct ext4_sb_info *sbi = EXT4_SB(sb);
710 	unsigned num;
711 
712 	/* Check for superblock and gdt backups in this group */
713 	num = ext4_bg_has_super(sb, block_group);
714 
715 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
716 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
717 			  sbi->s_desc_per_block) {
718 		if (num) {
719 			num += ext4_bg_num_gdb(sb, block_group);
720 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
721 		}
722 	} else { /* For META_BG_BLOCK_GROUPS */
723 		num += ext4_bg_num_gdb(sb, block_group);
724 	}
725 	return EXT4_NUM_B2C(sbi, num);
726 }
727 /**
728  *	ext4_inode_to_goal_block - return a hint for block allocation
729  *	@inode: inode for block allocation
730  *
731  *	Return the ideal location to start allocating blocks for a
732  *	newly created inode.
733  */
ext4_inode_to_goal_block(struct inode * inode)734 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
735 {
736 	struct ext4_inode_info *ei = EXT4_I(inode);
737 	ext4_group_t block_group;
738 	ext4_grpblk_t colour;
739 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
740 	ext4_fsblk_t bg_start;
741 	ext4_fsblk_t last_block;
742 
743 	block_group = ei->i_block_group;
744 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
745 		/*
746 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
747 		 * block groups per flexgroup, reserve the first block
748 		 * group for directories and special files.  Regular
749 		 * files will start at the second block group.  This
750 		 * tends to speed up directory access and improves
751 		 * fsck times.
752 		 */
753 		block_group &= ~(flex_size-1);
754 		if (S_ISREG(inode->i_mode))
755 			block_group++;
756 	}
757 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
758 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
759 
760 	/*
761 	 * If we are doing delayed allocation, we don't need take
762 	 * colour into account.
763 	 */
764 	if (test_opt(inode->i_sb, DELALLOC))
765 		return bg_start;
766 
767 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
768 		colour = (current->pid % 16) *
769 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
770 	else
771 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
772 	return bg_start + colour;
773 }
774 
775