d6babf94907e02d0b3a2f67e3cd96264b26e0c74
[linux-2.6-block.git] / fs / ext4 / balloc.c
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  */
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
52 /*
53  * Check whether the 'block' lives within the 'block_group'. Returns 1 if so
54  * and 0 otherwise.
55  */
56 static inline int ext4_block_in_group(struct super_block *sb,
57                                       ext4_fsblk_t block,
58                                       ext4_group_t block_group)
59 {
60         ext4_group_t actual_group;
61
62         if (test_opt2(sb, STD_GROUP_SIZE))
63                 actual_group =
64                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) +
65                         block) >>
66                         (EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3);
67         else
68                 ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
69         return (actual_group == block_group) ? 1 : 0;
70 }
71
72 /* Return the number of clusters used for file system metadata; this
73  * represents the overhead needed by the file system.
74  */
75 unsigned ext4_num_overhead_clusters(struct super_block *sb,
76                                     ext4_group_t block_group,
77                                     struct ext4_group_desc *gdp)
78 {
79         unsigned num_clusters;
80         int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
81         ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
82         ext4_fsblk_t itbl_blk;
83         struct ext4_sb_info *sbi = EXT4_SB(sb);
84
85         /* This is the number of clusters used by the superblock,
86          * block group descriptors, and reserved block group
87          * descriptor blocks */
88         num_clusters = ext4_num_base_meta_clusters(sb, block_group);
89
90         /*
91          * For the allocation bitmaps and inode table, we first need
92          * to check to see if the block is in the block group.  If it
93          * is, then check to see if the cluster is already accounted
94          * for in the clusters used for the base metadata cluster, or
95          * if we can increment the base metadata cluster to include
96          * that block.  Otherwise, we will have to track the cluster
97          * used for the allocation bitmap or inode table explicitly.
98          * Normally all of these blocks are contiguous, so the special
99          * case handling shouldn't be necessary except for *very*
100          * unusual file system layouts.
101          */
102         if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
103                 block_cluster = EXT4_B2C(sbi,
104                                          ext4_block_bitmap(sb, gdp) - start);
105                 if (block_cluster < num_clusters)
106                         block_cluster = -1;
107                 else if (block_cluster == num_clusters) {
108                         num_clusters++;
109                         block_cluster = -1;
110                 }
111         }
112
113         if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
114                 inode_cluster = EXT4_B2C(sbi,
115                                          ext4_inode_bitmap(sb, gdp) - start);
116                 if (inode_cluster < num_clusters)
117                         inode_cluster = -1;
118                 else if (inode_cluster == num_clusters) {
119                         num_clusters++;
120                         inode_cluster = -1;
121                 }
122         }
123
124         itbl_blk = ext4_inode_table(sb, gdp);
125         for (i = 0; i < sbi->s_itb_per_group; i++) {
126                 if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
127                         c = EXT4_B2C(sbi, itbl_blk + i - start);
128                         if ((c < num_clusters) || (c == inode_cluster) ||
129                             (c == block_cluster) || (c == itbl_cluster))
130                                 continue;
131                         if (c == num_clusters) {
132                                 num_clusters++;
133                                 continue;
134                         }
135                         num_clusters++;
136                         itbl_cluster = c;
137                 }
138         }
139
140         if (block_cluster != -1)
141                 num_clusters++;
142         if (inode_cluster != -1)
143                 num_clusters++;
144
145         return num_clusters;
146 }
147
148 static unsigned int num_clusters_in_group(struct super_block *sb,
149                                           ext4_group_t block_group)
150 {
151         unsigned int blocks;
152
153         if (block_group == ext4_get_groups_count(sb) - 1) {
154                 /*
155                  * Even though mke2fs always initializes the first and
156                  * last group, just in case some other tool was used,
157                  * we need to make sure we calculate the right free
158                  * blocks.
159                  */
160                 blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
161                         ext4_group_first_block_no(sb, block_group);
162         } else
163                 blocks = EXT4_BLOCKS_PER_GROUP(sb);
164         return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
165 }
166
167 /* Initializes an uninitialized block bitmap */
168 void ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
169                             ext4_group_t block_group,
170                             struct ext4_group_desc *gdp)
171 {
172         unsigned int bit, bit_max;
173         struct ext4_sb_info *sbi = EXT4_SB(sb);
174         ext4_fsblk_t start, tmp;
175         int flex_bg = 0;
176
177         J_ASSERT_BH(bh, buffer_locked(bh));
178
179         /* If checksum is bad mark all blocks used to prevent allocation
180          * essentially implementing a per-group read-only flag. */
181         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
182                 ext4_error(sb, "Checksum bad for group %u", block_group);
183                 ext4_free_group_clusters_set(sb, gdp, 0);
184                 ext4_free_inodes_set(sb, gdp, 0);
185                 ext4_itable_unused_set(sb, gdp, 0);
186                 memset(bh->b_data, 0xff, sb->s_blocksize);
187                 ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
188                 return;
189         }
190         memset(bh->b_data, 0, sb->s_blocksize);
191
192         bit_max = ext4_num_base_meta_clusters(sb, block_group);
193         for (bit = 0; bit < bit_max; bit++)
194                 ext4_set_bit(bit, bh->b_data);
195
196         start = ext4_group_first_block_no(sb, block_group);
197
198         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
199                 flex_bg = 1;
200
201         /* Set bits for block and inode bitmaps, and inode table */
202         tmp = ext4_block_bitmap(sb, gdp);
203         if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
204                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
205
206         tmp = ext4_inode_bitmap(sb, gdp);
207         if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
208                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
209
210         tmp = ext4_inode_table(sb, gdp);
211         for (; tmp < ext4_inode_table(sb, gdp) +
212                      sbi->s_itb_per_group; tmp++) {
213                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
214                         ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
215         }
216
217         /*
218          * Also if the number of blocks within the group is less than
219          * the blocksize * 8 ( which is the size of bitmap ), set rest
220          * of the block bitmap to 1
221          */
222         ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
223                              sb->s_blocksize * 8, bh->b_data);
224         ext4_block_bitmap_csum_set(sb, block_group, gdp, bh);
225         ext4_group_desc_csum_set(sb, block_group, gdp);
226 }
227
228 /* Return the number of free blocks in a block group.  It is used when
229  * the block bitmap is uninitialized, so we can't just count the bits
230  * in the bitmap. */
231 unsigned ext4_free_clusters_after_init(struct super_block *sb,
232                                        ext4_group_t block_group,
233                                        struct ext4_group_desc *gdp)
234 {
235         return num_clusters_in_group(sb, block_group) - 
236                 ext4_num_overhead_clusters(sb, block_group, gdp);
237 }
238
239 /*
240  * The free blocks are managed by bitmaps.  A file system contains several
241  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
242  * block for inodes, N blocks for the inode table and data blocks.
243  *
244  * The file system contains group descriptors which are located after the
245  * super block.  Each descriptor contains the number of the bitmap block and
246  * the free blocks count in the block.  The descriptors are loaded in memory
247  * when a file system is mounted (see ext4_fill_super).
248  */
249
250 /**
251  * ext4_get_group_desc() -- load group descriptor from disk
252  * @sb:                 super block
253  * @block_group:        given block group
254  * @bh:                 pointer to the buffer head to store the block
255  *                      group descriptor
256  */
257 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
258                                              ext4_group_t block_group,
259                                              struct buffer_head **bh)
260 {
261         unsigned int group_desc;
262         unsigned int offset;
263         ext4_group_t ngroups = ext4_get_groups_count(sb);
264         struct ext4_group_desc *desc;
265         struct ext4_sb_info *sbi = EXT4_SB(sb);
266
267         if (block_group >= ngroups) {
268                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
269                            " groups_count = %u", block_group, ngroups);
270
271                 return NULL;
272         }
273
274         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
275         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
276         if (!sbi->s_group_desc[group_desc]) {
277                 ext4_error(sb, "Group descriptor not loaded - "
278                            "block_group = %u, group_desc = %u, desc = %u",
279                            block_group, group_desc, offset);
280                 return NULL;
281         }
282
283         desc = (struct ext4_group_desc *)(
284                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
285                 offset * EXT4_DESC_SIZE(sb));
286         if (bh)
287                 *bh = sbi->s_group_desc[group_desc];
288         return desc;
289 }
290
291 /*
292  * Return the block number which was discovered to be invalid, or 0 if
293  * the block bitmap is valid.
294  */
295 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
296                                             struct ext4_group_desc *desc,
297                                             unsigned int block_group,
298                                             struct buffer_head *bh)
299 {
300         ext4_grpblk_t offset;
301         ext4_grpblk_t next_zero_bit;
302         ext4_fsblk_t blk;
303         ext4_fsblk_t group_first_block;
304
305         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
306                 /* with FLEX_BG, the inode/block bitmaps and itable
307                  * blocks may not be in the group at all
308                  * so the bitmap validation will be skipped for those groups
309                  * or it has to also read the block group where the bitmaps
310                  * are located to verify they are set.
311                  */
312                 return 0;
313         }
314         group_first_block = ext4_group_first_block_no(sb, block_group);
315
316         /* check whether block bitmap block number is set */
317         blk = ext4_block_bitmap(sb, desc);
318         offset = blk - group_first_block;
319         if (!ext4_test_bit(offset, bh->b_data))
320                 /* bad block bitmap */
321                 return blk;
322
323         /* check whether the inode bitmap block number is set */
324         blk = ext4_inode_bitmap(sb, desc);
325         offset = blk - group_first_block;
326         if (!ext4_test_bit(offset, bh->b_data))
327                 /* bad block bitmap */
328                 return blk;
329
330         /* check whether the inode table block number is set */
331         blk = ext4_inode_table(sb, desc);
332         offset = blk - group_first_block;
333         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
334                                 offset + EXT4_SB(sb)->s_itb_per_group,
335                                 offset);
336         if (next_zero_bit < offset + EXT4_SB(sb)->s_itb_per_group)
337                 /* bad bitmap for inode tables */
338                 return blk;
339         return 0;
340 }
341
342 void ext4_validate_block_bitmap(struct super_block *sb,
343                                struct ext4_group_desc *desc,
344                                unsigned int block_group,
345                                struct buffer_head *bh)
346 {
347         ext4_fsblk_t    blk;
348
349         if (buffer_verified(bh))
350                 return;
351
352         ext4_lock_group(sb, block_group);
353         blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
354         if (unlikely(blk != 0)) {
355                 ext4_unlock_group(sb, block_group);
356                 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
357                            block_group, blk);
358                 return;
359         }
360         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
361                         desc, bh))) {
362                 ext4_unlock_group(sb, block_group);
363                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
364                 return;
365         }
366         set_buffer_verified(bh);
367         ext4_unlock_group(sb, block_group);
368 }
369
370 /**
371  * ext4_read_block_bitmap_nowait()
372  * @sb:                 super block
373  * @block_group:        given block group
374  *
375  * Read the bitmap for a given block_group,and validate the
376  * bits for block/inode/inode tables are set in the bitmaps
377  *
378  * Return buffer_head on success or NULL in case of failure.
379  */
380 struct buffer_head *
381 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
382 {
383         struct ext4_group_desc *desc;
384         struct buffer_head *bh;
385         ext4_fsblk_t bitmap_blk;
386
387         desc = ext4_get_group_desc(sb, block_group, NULL);
388         if (!desc)
389                 return NULL;
390         bitmap_blk = ext4_block_bitmap(sb, desc);
391         bh = sb_getblk(sb, bitmap_blk);
392         if (unlikely(!bh)) {
393                 ext4_error(sb, "Cannot get buffer for block bitmap - "
394                            "block_group = %u, block_bitmap = %llu",
395                            block_group, bitmap_blk);
396                 return NULL;
397         }
398
399         if (bitmap_uptodate(bh))
400                 goto verify;
401
402         lock_buffer(bh);
403         if (bitmap_uptodate(bh)) {
404                 unlock_buffer(bh);
405                 goto verify;
406         }
407         ext4_lock_group(sb, block_group);
408         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
409                 ext4_init_block_bitmap(sb, bh, block_group, desc);
410                 set_bitmap_uptodate(bh);
411                 set_buffer_uptodate(bh);
412                 ext4_unlock_group(sb, block_group);
413                 unlock_buffer(bh);
414                 return bh;
415         }
416         ext4_unlock_group(sb, block_group);
417         if (buffer_uptodate(bh)) {
418                 /*
419                  * if not uninit if bh is uptodate,
420                  * bitmap is also uptodate
421                  */
422                 set_bitmap_uptodate(bh);
423                 unlock_buffer(bh);
424                 goto verify;
425         }
426         /*
427          * submit the buffer_head for reading
428          */
429         set_buffer_new(bh);
430         trace_ext4_read_block_bitmap_load(sb, block_group);
431         bh->b_end_io = ext4_end_bitmap_read;
432         get_bh(bh);
433         submit_bh(READ, bh);
434         return bh;
435 verify:
436         ext4_validate_block_bitmap(sb, desc, block_group, bh);
437         return bh;
438 }
439
440 /* Returns 0 on success, 1 on error */
441 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
442                            struct buffer_head *bh)
443 {
444         struct ext4_group_desc *desc;
445
446         if (!buffer_new(bh))
447                 return 0;
448         desc = ext4_get_group_desc(sb, block_group, NULL);
449         if (!desc)
450                 return 1;
451         wait_on_buffer(bh);
452         if (!buffer_uptodate(bh)) {
453                 ext4_error(sb, "Cannot read block bitmap - "
454                            "block_group = %u, block_bitmap = %llu",
455                            block_group, (unsigned long long) bh->b_blocknr);
456                 return 1;
457         }
458         clear_buffer_new(bh);
459         /* Panic or remount fs read-only if block bitmap is invalid */
460         ext4_validate_block_bitmap(sb, desc, block_group, bh);
461         return 0;
462 }
463
464 struct buffer_head *
465 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
466 {
467         struct buffer_head *bh;
468
469         bh = ext4_read_block_bitmap_nowait(sb, block_group);
470         if (!bh)
471                 return NULL;
472         if (ext4_wait_block_bitmap(sb, block_group, bh)) {
473                 put_bh(bh);
474                 return NULL;
475         }
476         return bh;
477 }
478
479 /**
480  * ext4_has_free_clusters()
481  * @sbi:        in-core super block structure.
482  * @nclusters:  number of needed blocks
483  * @flags:      flags from ext4_mb_new_blocks()
484  *
485  * Check if filesystem has nclusters free & available for allocation.
486  * On success return 1, return 0 on failure.
487  */
488 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
489                                   s64 nclusters, unsigned int flags)
490 {
491         s64 free_clusters, dirty_clusters, root_clusters;
492         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
493         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
494
495         free_clusters  = percpu_counter_read_positive(fcc);
496         dirty_clusters = percpu_counter_read_positive(dcc);
497
498         /*
499          * r_blocks_count should always be multiple of the cluster ratio so
500          * we are safe to do a plane bit shift only.
501          */
502         root_clusters = ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits;
503
504         if (free_clusters - (nclusters + root_clusters + dirty_clusters) <
505                                         EXT4_FREECLUSTERS_WATERMARK) {
506                 free_clusters  = percpu_counter_sum_positive(fcc);
507                 dirty_clusters = percpu_counter_sum_positive(dcc);
508         }
509         /* Check whether we have space after accounting for current
510          * dirty clusters & root reserved clusters.
511          */
512         if (free_clusters >= ((root_clusters + nclusters) + dirty_clusters))
513                 return 1;
514
515         /* Hm, nope.  Are (enough) root reserved clusters available? */
516         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
517             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
518             capable(CAP_SYS_RESOURCE) ||
519                 (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
520
521                 if (free_clusters >= (nclusters + dirty_clusters))
522                         return 1;
523         }
524
525         return 0;
526 }
527
528 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
529                              s64 nclusters, unsigned int flags)
530 {
531         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
532                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
533                 return 0;
534         } else
535                 return -ENOSPC;
536 }
537
538 /**
539  * ext4_should_retry_alloc()
540  * @sb:                 super block
541  * @retries             number of attemps has been made
542  *
543  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
544  * it is profitable to retry the operation, this function will wait
545  * for the current or committing transaction to complete, and then
546  * return TRUE.
547  *
548  * if the total number of retries exceed three times, return FALSE.
549  */
550 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
551 {
552         if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
553             (*retries)++ > 3 ||
554             !EXT4_SB(sb)->s_journal)
555                 return 0;
556
557         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
558
559         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
560 }
561
562 /*
563  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
564  *
565  * @handle:             handle to this transaction
566  * @inode:              file inode
567  * @goal:               given target block(filesystem wide)
568  * @count:              pointer to total number of clusters needed
569  * @errp:               error code
570  *
571  * Return 1st allocated block number on success, *count stores total account
572  * error stores in errp pointer
573  */
574 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
575                                   ext4_fsblk_t goal, unsigned int flags,
576                                   unsigned long *count, int *errp)
577 {
578         struct ext4_allocation_request ar;
579         ext4_fsblk_t ret;
580
581         memset(&ar, 0, sizeof(ar));
582         /* Fill with neighbour allocated blocks */
583         ar.inode = inode;
584         ar.goal = goal;
585         ar.len = count ? *count : 1;
586         ar.flags = flags;
587
588         ret = ext4_mb_new_blocks(handle, &ar, errp);
589         if (count)
590                 *count = ar.len;
591         /*
592          * Account for the allocated meta blocks.  We will never
593          * fail EDQUOT for metdata, but we do account for it.
594          */
595         if (!(*errp) &&
596             ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
597                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
598                 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
599                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
600                 dquot_alloc_block_nofail(inode,
601                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
602         }
603         return ret;
604 }
605
606 /**
607  * ext4_count_free_clusters() -- count filesystem free clusters
608  * @sb:         superblock
609  *
610  * Adds up the number of free clusters from each block group.
611  */
612 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
613 {
614         ext4_fsblk_t desc_count;
615         struct ext4_group_desc *gdp;
616         ext4_group_t i;
617         ext4_group_t ngroups = ext4_get_groups_count(sb);
618 #ifdef EXT4FS_DEBUG
619         struct ext4_super_block *es;
620         ext4_fsblk_t bitmap_count;
621         unsigned int x;
622         struct buffer_head *bitmap_bh = NULL;
623
624         es = EXT4_SB(sb)->s_es;
625         desc_count = 0;
626         bitmap_count = 0;
627         gdp = NULL;
628
629         for (i = 0; i < ngroups; i++) {
630                 gdp = ext4_get_group_desc(sb, i, NULL);
631                 if (!gdp)
632                         continue;
633                 desc_count += ext4_free_group_clusters(sb, gdp);
634                 brelse(bitmap_bh);
635                 bitmap_bh = ext4_read_block_bitmap(sb, i);
636                 if (bitmap_bh == NULL)
637                         continue;
638
639                 x = ext4_count_free(bitmap_bh->b_data,
640                                     EXT4_BLOCKS_PER_GROUP(sb) / 8);
641                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
642                         i, ext4_free_group_clusters(sb, gdp), x);
643                 bitmap_count += x;
644         }
645         brelse(bitmap_bh);
646         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
647                ", computed = %llu, %llu\n",
648                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
649                desc_count, bitmap_count);
650         return bitmap_count;
651 #else
652         desc_count = 0;
653         for (i = 0; i < ngroups; i++) {
654                 gdp = ext4_get_group_desc(sb, i, NULL);
655                 if (!gdp)
656                         continue;
657                 desc_count += ext4_free_group_clusters(sb, gdp);
658         }
659
660         return desc_count;
661 #endif
662 }
663
664 static inline int test_root(ext4_group_t a, int b)
665 {
666         int num = b;
667
668         while (a > num)
669                 num *= b;
670         return num == a;
671 }
672
673 static int ext4_group_sparse(ext4_group_t group)
674 {
675         if (group <= 1)
676                 return 1;
677         if (!(group & 1))
678                 return 0;
679         return (test_root(group, 7) || test_root(group, 5) ||
680                 test_root(group, 3));
681 }
682
683 /**
684  *      ext4_bg_has_super - number of blocks used by the superblock in group
685  *      @sb: superblock for filesystem
686  *      @group: group number to check
687  *
688  *      Return the number of blocks used by the superblock (primary or backup)
689  *      in this group.  Currently this will be only 0 or 1.
690  */
691 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
692 {
693         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
694                                 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
695                         !ext4_group_sparse(group))
696                 return 0;
697         return 1;
698 }
699
700 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
701                                         ext4_group_t group)
702 {
703         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
704         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
705         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
706
707         if (group == first || group == first + 1 || group == last)
708                 return 1;
709         return 0;
710 }
711
712 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
713                                         ext4_group_t group)
714 {
715         if (!ext4_bg_has_super(sb, group))
716                 return 0;
717
718         if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
719                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
720         else
721                 return EXT4_SB(sb)->s_gdb_count;
722 }
723
724 /**
725  *      ext4_bg_num_gdb - number of blocks used by the group table in group
726  *      @sb: superblock for filesystem
727  *      @group: group number to check
728  *
729  *      Return the number of blocks used by the group descriptor table
730  *      (primary or backup) in this group.  In the future there may be a
731  *      different number of descriptor blocks in each group.
732  */
733 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
734 {
735         unsigned long first_meta_bg =
736                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
737         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
738
739         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
740                         metagroup < first_meta_bg)
741                 return ext4_bg_num_gdb_nometa(sb, group);
742
743         return ext4_bg_num_gdb_meta(sb,group);
744
745 }
746
747 /*
748  * This function returns the number of file system metadata clusters at
749  * the beginning of a block group, including the reserved gdt blocks.
750  */
751 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
752                                      ext4_group_t block_group)
753 {
754         struct ext4_sb_info *sbi = EXT4_SB(sb);
755         unsigned num;
756
757         /* Check for superblock and gdt backups in this group */
758         num = ext4_bg_has_super(sb, block_group);
759
760         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
761             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
762                           sbi->s_desc_per_block) {
763                 if (num) {
764                         num += ext4_bg_num_gdb(sb, block_group);
765                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
766                 }
767         } else { /* For META_BG_BLOCK_GROUPS */
768                 num += ext4_bg_num_gdb(sb, block_group);
769         }
770         return EXT4_NUM_B2C(sbi, num);
771 }
772 /**
773  *      ext4_inode_to_goal_block - return a hint for block allocation
774  *      @inode: inode for block allocation
775  *
776  *      Return the ideal location to start allocating blocks for a
777  *      newly created inode.
778  */
779 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
780 {
781         struct ext4_inode_info *ei = EXT4_I(inode);
782         ext4_group_t block_group;
783         ext4_grpblk_t colour;
784         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
785         ext4_fsblk_t bg_start;
786         ext4_fsblk_t last_block;
787
788         block_group = ei->i_block_group;
789         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
790                 /*
791                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
792                  * block groups per flexgroup, reserve the first block
793                  * group for directories and special files.  Regular
794                  * files will start at the second block group.  This
795                  * tends to speed up directory access and improves
796                  * fsck times.
797                  */
798                 block_group &= ~(flex_size-1);
799                 if (S_ISREG(inode->i_mode))
800                         block_group++;
801         }
802         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
803         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
804
805         /*
806          * If we are doing delayed allocation, we don't need take
807          * colour into account.
808          */
809         if (test_opt(inode->i_sb, DELALLOC))
810                 return bg_start;
811
812         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
813                 colour = (current->pid % 16) *
814                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
815         else
816                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
817         return bg_start + colour;
818 }
819