• 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/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 block group number for a given block number
34  */
ext4_get_group_number(struct super_block * sb,ext4_fsblk_t block)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  */
ext4_get_group_no_and_offset(struct super_block * sb,ext4_fsblk_t blocknr,ext4_group_t * blockgrpp,ext4_grpblk_t * offsetp)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  */
ext4_block_in_group(struct super_block * sb,ext4_fsblk_t block,ext4_group_t block_group)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 /* Return the number of clusters used for file system metadata; this
84  * represents the overhead needed by the file system.
85  */
ext4_num_overhead_clusters(struct super_block * sb,ext4_group_t block_group,struct ext4_group_desc * gdp)86 static unsigned ext4_num_overhead_clusters(struct super_block *sb,
87 					   ext4_group_t block_group,
88 					   struct ext4_group_desc *gdp)
89 {
90 	unsigned num_clusters;
91 	int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
92 	ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
93 	ext4_fsblk_t itbl_blk;
94 	struct ext4_sb_info *sbi = EXT4_SB(sb);
95 
96 	/* This is the number of clusters used by the superblock,
97 	 * block group descriptors, and reserved block group
98 	 * descriptor blocks */
99 	num_clusters = ext4_num_base_meta_clusters(sb, block_group);
100 
101 	/*
102 	 * For the allocation bitmaps and inode table, we first need
103 	 * to check to see if the block is in the block group.  If it
104 	 * is, then check to see if the cluster is already accounted
105 	 * for in the clusters used for the base metadata cluster, or
106 	 * if we can increment the base metadata cluster to include
107 	 * that block.  Otherwise, we will have to track the cluster
108 	 * used for the allocation bitmap or inode table explicitly.
109 	 * Normally all of these blocks are contiguous, so the special
110 	 * case handling shouldn't be necessary except for *very*
111 	 * unusual file system layouts.
112 	 */
113 	if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
114 		block_cluster = EXT4_B2C(sbi,
115 					 ext4_block_bitmap(sb, gdp) - start);
116 		if (block_cluster < num_clusters)
117 			block_cluster = -1;
118 		else if (block_cluster == num_clusters) {
119 			num_clusters++;
120 			block_cluster = -1;
121 		}
122 	}
123 
124 	if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
125 		inode_cluster = EXT4_B2C(sbi,
126 					 ext4_inode_bitmap(sb, gdp) - start);
127 		if (inode_cluster < num_clusters)
128 			inode_cluster = -1;
129 		else if (inode_cluster == num_clusters) {
130 			num_clusters++;
131 			inode_cluster = -1;
132 		}
133 	}
134 
135 	itbl_blk = ext4_inode_table(sb, gdp);
136 	for (i = 0; i < sbi->s_itb_per_group; i++) {
137 		if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
138 			c = EXT4_B2C(sbi, itbl_blk + i - start);
139 			if ((c < num_clusters) || (c == inode_cluster) ||
140 			    (c == block_cluster) || (c == itbl_cluster))
141 				continue;
142 			if (c == num_clusters) {
143 				num_clusters++;
144 				continue;
145 			}
146 			num_clusters++;
147 			itbl_cluster = c;
148 		}
149 	}
150 
151 	if (block_cluster != -1)
152 		num_clusters++;
153 	if (inode_cluster != -1)
154 		num_clusters++;
155 
156 	return num_clusters;
157 }
158 
num_clusters_in_group(struct super_block * sb,ext4_group_t block_group)159 static unsigned int num_clusters_in_group(struct super_block *sb,
160 					  ext4_group_t block_group)
161 {
162 	unsigned int blocks;
163 
164 	if (block_group == ext4_get_groups_count(sb) - 1) {
165 		/*
166 		 * Even though mke2fs always initializes the first and
167 		 * last group, just in case some other tool was used,
168 		 * we need to make sure we calculate the right free
169 		 * blocks.
170 		 */
171 		blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
172 			ext4_group_first_block_no(sb, block_group);
173 	} else
174 		blocks = EXT4_BLOCKS_PER_GROUP(sb);
175 	return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
176 }
177 
178 /* 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)179 static int ext4_init_block_bitmap(struct super_block *sb,
180 				   struct buffer_head *bh,
181 				   ext4_group_t block_group,
182 				   struct ext4_group_desc *gdp)
183 {
184 	unsigned int bit, bit_max;
185 	struct ext4_sb_info *sbi = EXT4_SB(sb);
186 	ext4_fsblk_t start, tmp;
187 	int flex_bg = 0;
188 	struct ext4_group_info *grp;
189 
190 	J_ASSERT_BH(bh, buffer_locked(bh));
191 
192 	/* If checksum is bad mark all blocks used to prevent allocation
193 	 * essentially implementing a per-group read-only flag. */
194 	if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
195 		grp = ext4_get_group_info(sb, block_group);
196 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
197 			percpu_counter_sub(&sbi->s_freeclusters_counter,
198 					   grp->bb_free);
199 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
200 		if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
201 			int count;
202 			count = ext4_free_inodes_count(sb, gdp);
203 			percpu_counter_sub(&sbi->s_freeinodes_counter,
204 					   count);
205 		}
206 		set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
207 		return -EIO;
208 	}
209 	memset(bh->b_data, 0, sb->s_blocksize);
210 
211 	bit_max = ext4_num_base_meta_clusters(sb, block_group);
212 	if ((bit_max >> 3) >= bh->b_size)
213 		return -EIO;
214 
215 	for (bit = 0; bit < bit_max; bit++)
216 		ext4_set_bit(bit, bh->b_data);
217 
218 	start = ext4_group_first_block_no(sb, block_group);
219 
220 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
221 		flex_bg = 1;
222 
223 	/* Set bits for block and inode bitmaps, and inode table */
224 	tmp = ext4_block_bitmap(sb, gdp);
225 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
226 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
227 
228 	tmp = ext4_inode_bitmap(sb, gdp);
229 	if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
230 		ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
231 
232 	tmp = ext4_inode_table(sb, gdp);
233 	for (; tmp < ext4_inode_table(sb, gdp) +
234 		     sbi->s_itb_per_group; tmp++) {
235 		if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
236 			ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
237 	}
238 
239 	/*
240 	 * Also if the number of blocks within the group is less than
241 	 * the blocksize * 8 ( which is the size of bitmap ), set rest
242 	 * of the block bitmap to 1
243 	 */
244 	ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
245 			     sb->s_blocksize * 8, bh->b_data);
246 	ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
247 	ext4_group_desc_csum_set(sb, block_group, gdp);
248 	return 0;
249 }
250 
251 /* Return the number of free blocks in a block group.  It is used when
252  * the block bitmap is uninitialized, so we can't just count the bits
253  * in the bitmap. */
ext4_free_clusters_after_init(struct super_block * sb,ext4_group_t block_group,struct ext4_group_desc * gdp)254 unsigned ext4_free_clusters_after_init(struct super_block *sb,
255 				       ext4_group_t block_group,
256 				       struct ext4_group_desc *gdp)
257 {
258 	return num_clusters_in_group(sb, block_group) -
259 		ext4_num_overhead_clusters(sb, block_group, gdp);
260 }
261 
262 /*
263  * The free blocks are managed by bitmaps.  A file system contains several
264  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
265  * block for inodes, N blocks for the inode table and data blocks.
266  *
267  * The file system contains group descriptors which are located after the
268  * super block.  Each descriptor contains the number of the bitmap block and
269  * the free blocks count in the block.  The descriptors are loaded in memory
270  * when a file system is mounted (see ext4_fill_super).
271  */
272 
273 /**
274  * ext4_get_group_desc() -- load group descriptor from disk
275  * @sb:			super block
276  * @block_group:	given block group
277  * @bh:			pointer to the buffer head to store the block
278  *			group descriptor
279  */
ext4_get_group_desc(struct super_block * sb,ext4_group_t block_group,struct buffer_head ** bh)280 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
281 					     ext4_group_t block_group,
282 					     struct buffer_head **bh)
283 {
284 	unsigned int group_desc;
285 	unsigned int offset;
286 	ext4_group_t ngroups = ext4_get_groups_count(sb);
287 	struct ext4_group_desc *desc;
288 	struct ext4_sb_info *sbi = EXT4_SB(sb);
289 
290 	if (block_group >= ngroups) {
291 		ext4_error(sb, "block_group >= groups_count - block_group = %u,"
292 			   " groups_count = %u", block_group, ngroups);
293 
294 		return NULL;
295 	}
296 
297 	group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
298 	offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
299 	if (!sbi->s_group_desc[group_desc]) {
300 		ext4_error(sb, "Group descriptor not loaded - "
301 			   "block_group = %u, group_desc = %u, desc = %u",
302 			   block_group, group_desc, offset);
303 		return NULL;
304 	}
305 
306 	desc = (struct ext4_group_desc *)(
307 		(__u8 *)sbi->s_group_desc[group_desc]->b_data +
308 		offset * EXT4_DESC_SIZE(sb));
309 	if (bh)
310 		*bh = sbi->s_group_desc[group_desc];
311 	return desc;
312 }
313 
314 /*
315  * Return the block number which was discovered to be invalid, or 0 if
316  * the block bitmap is valid.
317  */
ext4_valid_block_bitmap(struct super_block * sb,struct ext4_group_desc * desc,ext4_group_t block_group,struct buffer_head * bh)318 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
319 					    struct ext4_group_desc *desc,
320 					    ext4_group_t block_group,
321 					    struct buffer_head *bh)
322 {
323 	struct ext4_sb_info *sbi = EXT4_SB(sb);
324 	ext4_grpblk_t offset;
325 	ext4_grpblk_t next_zero_bit;
326 	ext4_fsblk_t blk;
327 	ext4_fsblk_t group_first_block;
328 
329 	if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
330 		/* with FLEX_BG, the inode/block bitmaps and itable
331 		 * blocks may not be in the group at all
332 		 * so the bitmap validation will be skipped for those groups
333 		 * or it has to also read the block group where the bitmaps
334 		 * are located to verify they are set.
335 		 */
336 		return 0;
337 	}
338 	group_first_block = ext4_group_first_block_no(sb, block_group);
339 
340 	/* check whether block bitmap block number is set */
341 	blk = ext4_block_bitmap(sb, desc);
342 	offset = blk - group_first_block;
343 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
344 		/* bad block bitmap */
345 		return blk;
346 
347 	/* check whether the inode bitmap block number is set */
348 	blk = ext4_inode_bitmap(sb, desc);
349 	offset = blk - group_first_block;
350 	if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
351 		/* bad block bitmap */
352 		return blk;
353 
354 	/* check whether the inode table block number is set */
355 	blk = ext4_inode_table(sb, desc);
356 	offset = blk - group_first_block;
357 	next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
358 			EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group),
359 			EXT4_B2C(sbi, offset));
360 	if (next_zero_bit <
361 	    EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group))
362 		/* bad bitmap for inode tables */
363 		return blk;
364 	return 0;
365 }
366 
ext4_validate_block_bitmap(struct super_block * sb,struct ext4_group_desc * desc,ext4_group_t block_group,struct buffer_head * bh)367 static void ext4_validate_block_bitmap(struct super_block *sb,
368 				       struct ext4_group_desc *desc,
369 				       ext4_group_t block_group,
370 				       struct buffer_head *bh)
371 {
372 	ext4_fsblk_t	blk;
373 	struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
374 	struct ext4_sb_info *sbi = EXT4_SB(sb);
375 
376 	if (buffer_verified(bh))
377 		return;
378 
379 	ext4_lock_group(sb, block_group);
380 	blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
381 	if (unlikely(blk != 0)) {
382 		ext4_unlock_group(sb, block_group);
383 		ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
384 			   block_group, blk);
385 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
386 			percpu_counter_sub(&sbi->s_freeclusters_counter,
387 					   grp->bb_free);
388 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
389 		return;
390 	}
391 	if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
392 			desc, bh))) {
393 		ext4_unlock_group(sb, block_group);
394 		ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
395 		if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
396 			percpu_counter_sub(&sbi->s_freeclusters_counter,
397 					   grp->bb_free);
398 		set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
399 		return;
400 	}
401 	set_buffer_verified(bh);
402 	ext4_unlock_group(sb, block_group);
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 
422 	desc = ext4_get_group_desc(sb, block_group, NULL);
423 	if (!desc)
424 		return NULL;
425 	bitmap_blk = ext4_block_bitmap(sb, desc);
426 	bh = sb_getblk(sb, bitmap_blk);
427 	if (unlikely(!bh)) {
428 		ext4_error(sb, "Cannot get buffer for block bitmap - "
429 			   "block_group = %u, block_bitmap = %llu",
430 			   block_group, bitmap_blk);
431 		return NULL;
432 	}
433 
434 	if (bitmap_uptodate(bh))
435 		goto verify;
436 
437 	lock_buffer(bh);
438 	if (bitmap_uptodate(bh)) {
439 		unlock_buffer(bh);
440 		goto verify;
441 	}
442 	ext4_lock_group(sb, block_group);
443 	if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
444 		int err;
445 
446 		err = ext4_init_block_bitmap(sb, bh, block_group, desc);
447 		set_bitmap_uptodate(bh);
448 		set_buffer_uptodate(bh);
449 		ext4_unlock_group(sb, block_group);
450 		unlock_buffer(bh);
451 		if (err)
452 			ext4_error(sb, "Checksum bad for grp %u", block_group);
453 		return bh;
454 	}
455 	ext4_unlock_group(sb, block_group);
456 	if (buffer_uptodate(bh)) {
457 		/*
458 		 * if not uninit if bh is uptodate,
459 		 * bitmap is also uptodate
460 		 */
461 		set_bitmap_uptodate(bh);
462 		unlock_buffer(bh);
463 		goto verify;
464 	}
465 	/*
466 	 * submit the buffer_head for reading
467 	 */
468 	set_buffer_new(bh);
469 	trace_ext4_read_block_bitmap_load(sb, block_group);
470 	bh->b_end_io = ext4_end_bitmap_read;
471 	get_bh(bh);
472 	submit_bh(READ | REQ_META | REQ_PRIO, bh);
473 	return bh;
474 verify:
475 	ext4_validate_block_bitmap(sb, desc, block_group, bh);
476 	if (buffer_verified(bh))
477 		return bh;
478 	put_bh(bh);
479 	return NULL;
480 }
481 
482 /* Returns 0 on success, 1 on error */
ext4_wait_block_bitmap(struct super_block * sb,ext4_group_t block_group,struct buffer_head * bh)483 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
484 			   struct buffer_head *bh)
485 {
486 	struct ext4_group_desc *desc;
487 
488 	if (!buffer_new(bh))
489 		return 0;
490 	desc = ext4_get_group_desc(sb, block_group, NULL);
491 	if (!desc)
492 		return 1;
493 	wait_on_buffer(bh);
494 	if (!buffer_uptodate(bh)) {
495 		ext4_error(sb, "Cannot read block bitmap - "
496 			   "block_group = %u, block_bitmap = %llu",
497 			   block_group, (unsigned long long) bh->b_blocknr);
498 		return 1;
499 	}
500 	clear_buffer_new(bh);
501 	/* Panic or remount fs read-only if block bitmap is invalid */
502 	ext4_validate_block_bitmap(sb, desc, block_group, bh);
503 	/* ...but check for error just in case errors=continue. */
504 	return !buffer_verified(bh);
505 }
506 
507 struct buffer_head *
ext4_read_block_bitmap(struct super_block * sb,ext4_group_t block_group)508 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
509 {
510 	struct buffer_head *bh;
511 
512 	bh = ext4_read_block_bitmap_nowait(sb, block_group);
513 	if (!bh)
514 		return NULL;
515 	if (ext4_wait_block_bitmap(sb, block_group, bh)) {
516 		put_bh(bh);
517 		return NULL;
518 	}
519 	return bh;
520 }
521 
522 /**
523  * ext4_has_free_clusters()
524  * @sbi:	in-core super block structure.
525  * @nclusters:	number of needed blocks
526  * @flags:	flags from ext4_mb_new_blocks()
527  *
528  * Check if filesystem has nclusters free & available for allocation.
529  * On success return 1, return 0 on failure.
530  */
ext4_has_free_clusters(struct ext4_sb_info * sbi,s64 nclusters,unsigned int flags)531 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
532 				  s64 nclusters, unsigned int flags)
533 {
534 	s64 free_clusters, dirty_clusters, rsv, resv_clusters;
535 	struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
536 	struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
537 
538 	free_clusters  = percpu_counter_read_positive(fcc);
539 	dirty_clusters = percpu_counter_read_positive(dcc);
540 	resv_clusters = atomic64_read(&sbi->s_resv_clusters);
541 
542 	/*
543 	 * r_blocks_count should always be multiple of the cluster ratio so
544 	 * we are safe to do a plane bit shift only.
545 	 */
546 	rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
547 	      resv_clusters;
548 
549 	if (free_clusters - (nclusters + rsv + dirty_clusters) <
550 					EXT4_FREECLUSTERS_WATERMARK) {
551 		free_clusters  = percpu_counter_sum_positive(fcc);
552 		dirty_clusters = percpu_counter_sum_positive(dcc);
553 	}
554 	/* Check whether we have space after accounting for current
555 	 * dirty clusters & root reserved clusters.
556 	 */
557 	if (free_clusters >= (rsv + nclusters + dirty_clusters))
558 		return 1;
559 
560 	/* Hm, nope.  Are (enough) root reserved clusters available? */
561 	if (uid_eq(sbi->s_resuid, current_fsuid()) ||
562 	    (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
563 	    capable(CAP_SYS_RESOURCE) ||
564 	    (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
565 
566 		if (free_clusters >= (nclusters + dirty_clusters +
567 				      resv_clusters))
568 			return 1;
569 	}
570 	/* No free blocks. Let's see if we can dip into reserved pool */
571 	if (flags & EXT4_MB_USE_RESERVED) {
572 		if (free_clusters >= (nclusters + dirty_clusters))
573 			return 1;
574 	}
575 
576 	return 0;
577 }
578 
ext4_claim_free_clusters(struct ext4_sb_info * sbi,s64 nclusters,unsigned int flags)579 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
580 			     s64 nclusters, unsigned int flags)
581 {
582 	if (ext4_has_free_clusters(sbi, nclusters, flags)) {
583 		percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
584 		return 0;
585 	} else
586 		return -ENOSPC;
587 }
588 
589 /**
590  * ext4_should_retry_alloc()
591  * @sb:			super block
592  * @retries		number of attemps has been made
593  *
594  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
595  * it is profitable to retry the operation, this function will wait
596  * for the current or committing transaction to complete, and then
597  * return TRUE.
598  *
599  * if the total number of retries exceed three times, return FALSE.
600  */
ext4_should_retry_alloc(struct super_block * sb,int * retries)601 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
602 {
603 	if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
604 	    (*retries)++ > 3 ||
605 	    !EXT4_SB(sb)->s_journal)
606 		return 0;
607 
608 	jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
609 
610 	return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
611 }
612 
613 /*
614  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
615  *
616  * @handle:             handle to this transaction
617  * @inode:              file inode
618  * @goal:               given target block(filesystem wide)
619  * @count:		pointer to total number of clusters needed
620  * @errp:               error code
621  *
622  * Return 1st allocated block number on success, *count stores total account
623  * error stores in errp pointer
624  */
ext4_new_meta_blocks(handle_t * handle,struct inode * inode,ext4_fsblk_t goal,unsigned int flags,unsigned long * count,int * errp)625 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
626 				  ext4_fsblk_t goal, unsigned int flags,
627 				  unsigned long *count, int *errp)
628 {
629 	struct ext4_allocation_request ar;
630 	ext4_fsblk_t ret;
631 
632 	memset(&ar, 0, sizeof(ar));
633 	/* Fill with neighbour allocated blocks */
634 	ar.inode = inode;
635 	ar.goal = goal;
636 	ar.len = count ? *count : 1;
637 	ar.flags = flags;
638 
639 	ret = ext4_mb_new_blocks(handle, &ar, errp);
640 	if (count)
641 		*count = ar.len;
642 	/*
643 	 * Account for the allocated meta blocks.  We will never
644 	 * fail EDQUOT for metdata, but we do account for it.
645 	 */
646 	if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
647 		spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
648 		spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
649 		dquot_alloc_block_nofail(inode,
650 				EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
651 	}
652 	return ret;
653 }
654 
655 /**
656  * ext4_count_free_clusters() -- count filesystem free clusters
657  * @sb:		superblock
658  *
659  * Adds up the number of free clusters from each block group.
660  */
ext4_count_free_clusters(struct super_block * sb)661 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
662 {
663 	ext4_fsblk_t desc_count;
664 	struct ext4_group_desc *gdp;
665 	ext4_group_t i;
666 	ext4_group_t ngroups = ext4_get_groups_count(sb);
667 	struct ext4_group_info *grp;
668 #ifdef EXT4FS_DEBUG
669 	struct ext4_super_block *es;
670 	ext4_fsblk_t bitmap_count;
671 	unsigned int x;
672 	struct buffer_head *bitmap_bh = NULL;
673 
674 	es = EXT4_SB(sb)->s_es;
675 	desc_count = 0;
676 	bitmap_count = 0;
677 	gdp = NULL;
678 
679 	for (i = 0; i < ngroups; i++) {
680 		gdp = ext4_get_group_desc(sb, i, NULL);
681 		if (!gdp)
682 			continue;
683 		grp = NULL;
684 		if (EXT4_SB(sb)->s_group_info)
685 			grp = ext4_get_group_info(sb, i);
686 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
687 			desc_count += ext4_free_group_clusters(sb, gdp);
688 		brelse(bitmap_bh);
689 		bitmap_bh = ext4_read_block_bitmap(sb, i);
690 		if (bitmap_bh == NULL)
691 			continue;
692 
693 		x = ext4_count_free(bitmap_bh->b_data,
694 				    EXT4_CLUSTERS_PER_GROUP(sb) / 8);
695 		printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
696 			i, ext4_free_group_clusters(sb, gdp), x);
697 		bitmap_count += x;
698 	}
699 	brelse(bitmap_bh);
700 	printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
701 	       ", computed = %llu, %llu\n",
702 	       EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
703 	       desc_count, bitmap_count);
704 	return bitmap_count;
705 #else
706 	desc_count = 0;
707 	for (i = 0; i < ngroups; i++) {
708 		gdp = ext4_get_group_desc(sb, i, NULL);
709 		if (!gdp)
710 			continue;
711 		grp = NULL;
712 		if (EXT4_SB(sb)->s_group_info)
713 			grp = ext4_get_group_info(sb, i);
714 		if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
715 			desc_count += ext4_free_group_clusters(sb, gdp);
716 	}
717 
718 	return desc_count;
719 #endif
720 }
721 
test_root(ext4_group_t a,int b)722 static inline int test_root(ext4_group_t a, int b)
723 {
724 	while (1) {
725 		if (a < b)
726 			return 0;
727 		if (a == b)
728 			return 1;
729 		if ((a % b) != 0)
730 			return 0;
731 		a = a / b;
732 	}
733 }
734 
735 /**
736  *	ext4_bg_has_super - number of blocks used by the superblock in group
737  *	@sb: superblock for filesystem
738  *	@group: group number to check
739  *
740  *	Return the number of blocks used by the superblock (primary or backup)
741  *	in this group.  Currently this will be only 0 or 1.
742  */
ext4_bg_has_super(struct super_block * sb,ext4_group_t group)743 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
744 {
745 	struct ext4_super_block *es = EXT4_SB(sb)->s_es;
746 
747 	if (group == 0)
748 		return 1;
749 	if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_SPARSE_SUPER2)) {
750 		if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
751 		    group == le32_to_cpu(es->s_backup_bgs[1]))
752 			return 1;
753 		return 0;
754 	}
755 	if ((group <= 1) || !EXT4_HAS_RO_COMPAT_FEATURE(sb,
756 					EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER))
757 		return 1;
758 	if (!(group & 1))
759 		return 0;
760 	if (test_root(group, 3) || (test_root(group, 5)) ||
761 	    test_root(group, 7))
762 		return 1;
763 
764 	return 0;
765 }
766 
ext4_bg_num_gdb_meta(struct super_block * sb,ext4_group_t group)767 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
768 					ext4_group_t group)
769 {
770 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
771 	ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
772 	ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
773 
774 	if (group == first || group == first + 1 || group == last)
775 		return 1;
776 	return 0;
777 }
778 
ext4_bg_num_gdb_nometa(struct super_block * sb,ext4_group_t group)779 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
780 					ext4_group_t group)
781 {
782 	if (!ext4_bg_has_super(sb, group))
783 		return 0;
784 
785 	if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
786 		return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
787 	else
788 		return EXT4_SB(sb)->s_gdb_count;
789 }
790 
791 /**
792  *	ext4_bg_num_gdb - number of blocks used by the group table in group
793  *	@sb: superblock for filesystem
794  *	@group: group number to check
795  *
796  *	Return the number of blocks used by the group descriptor table
797  *	(primary or backup) in this group.  In the future there may be a
798  *	different number of descriptor blocks in each group.
799  */
ext4_bg_num_gdb(struct super_block * sb,ext4_group_t group)800 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
801 {
802 	unsigned long first_meta_bg =
803 			le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
804 	unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
805 
806 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
807 			metagroup < first_meta_bg)
808 		return ext4_bg_num_gdb_nometa(sb, group);
809 
810 	return ext4_bg_num_gdb_meta(sb,group);
811 
812 }
813 
814 /*
815  * This function returns the number of file system metadata clusters at
816  * the beginning of a block group, including the reserved gdt blocks.
817  */
ext4_num_base_meta_clusters(struct super_block * sb,ext4_group_t block_group)818 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
819 				     ext4_group_t block_group)
820 {
821 	struct ext4_sb_info *sbi = EXT4_SB(sb);
822 	unsigned num;
823 
824 	/* Check for superblock and gdt backups in this group */
825 	num = ext4_bg_has_super(sb, block_group);
826 
827 	if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
828 	    block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
829 			  sbi->s_desc_per_block) {
830 		if (num) {
831 			num += ext4_bg_num_gdb(sb, block_group);
832 			num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
833 		}
834 	} else { /* For META_BG_BLOCK_GROUPS */
835 		num += ext4_bg_num_gdb(sb, block_group);
836 	}
837 	return EXT4_NUM_B2C(sbi, num);
838 }
839 /**
840  *	ext4_inode_to_goal_block - return a hint for block allocation
841  *	@inode: inode for block allocation
842  *
843  *	Return the ideal location to start allocating blocks for a
844  *	newly created inode.
845  */
ext4_inode_to_goal_block(struct inode * inode)846 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
847 {
848 	struct ext4_inode_info *ei = EXT4_I(inode);
849 	ext4_group_t block_group;
850 	ext4_grpblk_t colour;
851 	int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
852 	ext4_fsblk_t bg_start;
853 	ext4_fsblk_t last_block;
854 
855 	block_group = ei->i_block_group;
856 	if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
857 		/*
858 		 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
859 		 * block groups per flexgroup, reserve the first block
860 		 * group for directories and special files.  Regular
861 		 * files will start at the second block group.  This
862 		 * tends to speed up directory access and improves
863 		 * fsck times.
864 		 */
865 		block_group &= ~(flex_size-1);
866 		if (S_ISREG(inode->i_mode))
867 			block_group++;
868 	}
869 	bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
870 	last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
871 
872 	/*
873 	 * If we are doing delayed allocation, we don't need take
874 	 * colour into account.
875 	 */
876 	if (test_opt(inode->i_sb, DELALLOC))
877 		return bg_start;
878 
879 	if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
880 		colour = (current->pid % 16) *
881 			(EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
882 	else
883 		colour = (current->pid % 16) * ((last_block - bg_start) / 16);
884 	return bg_start + colour;
885 }
886 
887