Merge branches 'core-fixes-for-linus' and 'irq-fixes-for-linus' of git://git.kernel...
[pandora-kernel.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 /*
27  * balloc.c contains the blocks allocation and deallocation routines
28  */
29
30 /*
31  * Calculate the block group number and offset, given a block number
32  */
33 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
34                 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
35 {
36         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
37         ext4_grpblk_t offset;
38
39         blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
40         offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
41         if (offsetp)
42                 *offsetp = offset;
43         if (blockgrpp)
44                 *blockgrpp = blocknr;
45
46 }
47
48 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
49                         ext4_group_t block_group)
50 {
51         ext4_group_t actual_group;
52         ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
53         if (actual_group == block_group)
54                 return 1;
55         return 0;
56 }
57
58 static int ext4_group_used_meta_blocks(struct super_block *sb,
59                                        ext4_group_t block_group,
60                                        struct ext4_group_desc *gdp)
61 {
62         ext4_fsblk_t tmp;
63         struct ext4_sb_info *sbi = EXT4_SB(sb);
64         /* block bitmap, inode bitmap, and inode table blocks */
65         int used_blocks = sbi->s_itb_per_group + 2;
66
67         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
68                 if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
69                                         block_group))
70                         used_blocks--;
71
72                 if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
73                                         block_group))
74                         used_blocks--;
75
76                 tmp = ext4_inode_table(sb, gdp);
77                 for (; tmp < ext4_inode_table(sb, gdp) +
78                                 sbi->s_itb_per_group; tmp++) {
79                         if (!ext4_block_in_group(sb, tmp, block_group))
80                                 used_blocks -= 1;
81                 }
82         }
83         return used_blocks;
84 }
85
86 /* Initializes an uninitialized block bitmap if given, and returns the
87  * number of blocks free in the group. */
88 unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
89                  ext4_group_t block_group, struct ext4_group_desc *gdp)
90 {
91         int bit, bit_max;
92         ext4_group_t ngroups = ext4_get_groups_count(sb);
93         unsigned free_blocks, group_blocks;
94         struct ext4_sb_info *sbi = EXT4_SB(sb);
95
96         if (bh) {
97                 J_ASSERT_BH(bh, buffer_locked(bh));
98
99                 /* If checksum is bad mark all blocks used to prevent allocation
100                  * essentially implementing a per-group read-only flag. */
101                 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
102                         ext4_error(sb, "Checksum bad for group %u",
103                                         block_group);
104                         ext4_free_blks_set(sb, gdp, 0);
105                         ext4_free_inodes_set(sb, gdp, 0);
106                         ext4_itable_unused_set(sb, gdp, 0);
107                         memset(bh->b_data, 0xff, sb->s_blocksize);
108                         return 0;
109                 }
110                 memset(bh->b_data, 0, sb->s_blocksize);
111         }
112
113         /* Check for superblock and gdt backups in this group */
114         bit_max = ext4_bg_has_super(sb, block_group);
115
116         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
117             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
118                           sbi->s_desc_per_block) {
119                 if (bit_max) {
120                         bit_max += ext4_bg_num_gdb(sb, block_group);
121                         bit_max +=
122                                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
123                 }
124         } else { /* For META_BG_BLOCK_GROUPS */
125                 bit_max += ext4_bg_num_gdb(sb, block_group);
126         }
127
128         if (block_group == ngroups - 1) {
129                 /*
130                  * Even though mke2fs always initialize first and last group
131                  * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
132                  * to make sure we calculate the right free blocks
133                  */
134                 group_blocks = ext4_blocks_count(sbi->s_es) -
135                         ext4_group_first_block_no(sb, ngroups - 1);
136         } else {
137                 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
138         }
139
140         free_blocks = group_blocks - bit_max;
141
142         if (bh) {
143                 ext4_fsblk_t start, tmp;
144                 int flex_bg = 0;
145
146                 for (bit = 0; bit < bit_max; bit++)
147                         ext4_set_bit(bit, bh->b_data);
148
149                 start = ext4_group_first_block_no(sb, block_group);
150
151                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
152                                               EXT4_FEATURE_INCOMPAT_FLEX_BG))
153                         flex_bg = 1;
154
155                 /* Set bits for block and inode bitmaps, and inode table */
156                 tmp = ext4_block_bitmap(sb, gdp);
157                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
158                         ext4_set_bit(tmp - start, bh->b_data);
159
160                 tmp = ext4_inode_bitmap(sb, gdp);
161                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
162                         ext4_set_bit(tmp - start, bh->b_data);
163
164                 tmp = ext4_inode_table(sb, gdp);
165                 for (; tmp < ext4_inode_table(sb, gdp) +
166                                 sbi->s_itb_per_group; tmp++) {
167                         if (!flex_bg ||
168                                 ext4_block_in_group(sb, tmp, block_group))
169                                 ext4_set_bit(tmp - start, bh->b_data);
170                 }
171                 /*
172                  * Also if the number of blocks within the group is
173                  * less than the blocksize * 8 ( which is the size
174                  * of bitmap ), set rest of the block bitmap to 1
175                  */
176                 ext4_mark_bitmap_end(group_blocks, sb->s_blocksize * 8,
177                                      bh->b_data);
178         }
179         return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp);
180 }
181
182
183 /*
184  * The free blocks are managed by bitmaps.  A file system contains several
185  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
186  * block for inodes, N blocks for the inode table and data blocks.
187  *
188  * The file system contains group descriptors which are located after the
189  * super block.  Each descriptor contains the number of the bitmap block and
190  * the free blocks count in the block.  The descriptors are loaded in memory
191  * when a file system is mounted (see ext4_fill_super).
192  */
193
194 /**
195  * ext4_get_group_desc() -- load group descriptor from disk
196  * @sb:                 super block
197  * @block_group:        given block group
198  * @bh:                 pointer to the buffer head to store the block
199  *                      group descriptor
200  */
201 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
202                                              ext4_group_t block_group,
203                                              struct buffer_head **bh)
204 {
205         unsigned int group_desc;
206         unsigned int offset;
207         ext4_group_t ngroups = ext4_get_groups_count(sb);
208         struct ext4_group_desc *desc;
209         struct ext4_sb_info *sbi = EXT4_SB(sb);
210
211         if (block_group >= ngroups) {
212                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
213                            " groups_count = %u", block_group, ngroups);
214
215                 return NULL;
216         }
217
218         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
219         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
220         if (!sbi->s_group_desc[group_desc]) {
221                 ext4_error(sb, "Group descriptor not loaded - "
222                            "block_group = %u, group_desc = %u, desc = %u",
223                            block_group, group_desc, offset);
224                 return NULL;
225         }
226
227         desc = (struct ext4_group_desc *)(
228                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
229                 offset * EXT4_DESC_SIZE(sb));
230         if (bh)
231                 *bh = sbi->s_group_desc[group_desc];
232         return desc;
233 }
234
235 static int ext4_valid_block_bitmap(struct super_block *sb,
236                                         struct ext4_group_desc *desc,
237                                         unsigned int block_group,
238                                         struct buffer_head *bh)
239 {
240         ext4_grpblk_t offset;
241         ext4_grpblk_t next_zero_bit;
242         ext4_fsblk_t bitmap_blk;
243         ext4_fsblk_t group_first_block;
244
245         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
246                 /* with FLEX_BG, the inode/block bitmaps and itable
247                  * blocks may not be in the group at all
248                  * so the bitmap validation will be skipped for those groups
249                  * or it has to also read the block group where the bitmaps
250                  * are located to verify they are set.
251                  */
252                 return 1;
253         }
254         group_first_block = ext4_group_first_block_no(sb, block_group);
255
256         /* check whether block bitmap block number is set */
257         bitmap_blk = ext4_block_bitmap(sb, desc);
258         offset = bitmap_blk - group_first_block;
259         if (!ext4_test_bit(offset, bh->b_data))
260                 /* bad block bitmap */
261                 goto err_out;
262
263         /* check whether the inode bitmap block number is set */
264         bitmap_blk = ext4_inode_bitmap(sb, desc);
265         offset = bitmap_blk - group_first_block;
266         if (!ext4_test_bit(offset, bh->b_data))
267                 /* bad block bitmap */
268                 goto err_out;
269
270         /* check whether the inode table block number is set */
271         bitmap_blk = ext4_inode_table(sb, desc);
272         offset = bitmap_blk - group_first_block;
273         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
274                                 offset + EXT4_SB(sb)->s_itb_per_group,
275                                 offset);
276         if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
277                 /* good bitmap for inode tables */
278                 return 1;
279
280 err_out:
281         ext4_error(sb, "Invalid block bitmap - block_group = %d, block = %llu",
282                         block_group, bitmap_blk);
283         return 0;
284 }
285 /**
286  * ext4_read_block_bitmap()
287  * @sb:                 super block
288  * @block_group:        given block group
289  *
290  * Read the bitmap for a given block_group,and validate the
291  * bits for block/inode/inode tables are set in the bitmaps
292  *
293  * Return buffer_head on success or NULL in case of failure.
294  */
295 struct buffer_head *
296 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
297 {
298         struct ext4_group_desc *desc;
299         struct buffer_head *bh = NULL;
300         ext4_fsblk_t bitmap_blk;
301
302         desc = ext4_get_group_desc(sb, block_group, NULL);
303         if (!desc)
304                 return NULL;
305         bitmap_blk = ext4_block_bitmap(sb, desc);
306         bh = sb_getblk(sb, bitmap_blk);
307         if (unlikely(!bh)) {
308                 ext4_error(sb, "Cannot read block bitmap - "
309                             "block_group = %u, block_bitmap = %llu",
310                             block_group, bitmap_blk);
311                 return NULL;
312         }
313
314         if (bitmap_uptodate(bh))
315                 return bh;
316
317         lock_buffer(bh);
318         if (bitmap_uptodate(bh)) {
319                 unlock_buffer(bh);
320                 return bh;
321         }
322         ext4_lock_group(sb, block_group);
323         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
324                 ext4_init_block_bitmap(sb, bh, block_group, desc);
325                 set_bitmap_uptodate(bh);
326                 set_buffer_uptodate(bh);
327                 ext4_unlock_group(sb, block_group);
328                 unlock_buffer(bh);
329                 return bh;
330         }
331         ext4_unlock_group(sb, block_group);
332         if (buffer_uptodate(bh)) {
333                 /*
334                  * if not uninit if bh is uptodate,
335                  * bitmap is also uptodate
336                  */
337                 set_bitmap_uptodate(bh);
338                 unlock_buffer(bh);
339                 return bh;
340         }
341         /*
342          * submit the buffer_head for read. We can
343          * safely mark the bitmap as uptodate now.
344          * We do it here so the bitmap uptodate bit
345          * get set with buffer lock held.
346          */
347         trace_ext4_read_block_bitmap_load(sb, block_group);
348         set_bitmap_uptodate(bh);
349         if (bh_submit_read(bh) < 0) {
350                 put_bh(bh);
351                 ext4_error(sb, "Cannot read block bitmap - "
352                             "block_group = %u, block_bitmap = %llu",
353                             block_group, bitmap_blk);
354                 return NULL;
355         }
356         ext4_valid_block_bitmap(sb, desc, block_group, bh);
357         /*
358          * file system mounted not to panic on error,
359          * continue with corrupt bitmap
360          */
361         return bh;
362 }
363
364 /**
365  * ext4_has_free_blocks()
366  * @sbi:        in-core super block structure.
367  * @nblocks:    number of needed blocks
368  *
369  * Check if filesystem has nblocks free & available for allocation.
370  * On success return 1, return 0 on failure.
371  */
372 static int ext4_has_free_blocks(struct ext4_sb_info *sbi,
373                                 s64 nblocks, unsigned int flags)
374 {
375         s64 free_blocks, dirty_blocks, root_blocks;
376         struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
377         struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
378
379         free_blocks  = percpu_counter_read_positive(fbc);
380         dirty_blocks = percpu_counter_read_positive(dbc);
381         root_blocks = ext4_r_blocks_count(sbi->s_es);
382
383         if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
384                                                 EXT4_FREEBLOCKS_WATERMARK) {
385                 free_blocks  = percpu_counter_sum_positive(fbc);
386                 dirty_blocks = percpu_counter_sum_positive(dbc);
387         }
388         /* Check whether we have space after
389          * accounting for current dirty blocks & root reserved blocks.
390          */
391         if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
392                 return 1;
393
394         /* Hm, nope.  Are (enough) root reserved blocks available? */
395         if (sbi->s_resuid == current_fsuid() ||
396             ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
397             capable(CAP_SYS_RESOURCE) ||
398                 (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
399
400                 if (free_blocks >= (nblocks + dirty_blocks))
401                         return 1;
402         }
403
404         return 0;
405 }
406
407 int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
408                            s64 nblocks, unsigned int flags)
409 {
410         if (ext4_has_free_blocks(sbi, nblocks, flags)) {
411                 percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
412                 return 0;
413         } else
414                 return -ENOSPC;
415 }
416
417 /**
418  * ext4_should_retry_alloc()
419  * @sb:                 super block
420  * @retries             number of attemps has been made
421  *
422  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
423  * it is profitable to retry the operation, this function will wait
424  * for the current or committing transaction to complete, and then
425  * return TRUE.
426  *
427  * if the total number of retries exceed three times, return FALSE.
428  */
429 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
430 {
431         if (!ext4_has_free_blocks(EXT4_SB(sb), 1, 0) ||
432             (*retries)++ > 3 ||
433             !EXT4_SB(sb)->s_journal)
434                 return 0;
435
436         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
437
438         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
439 }
440
441 /*
442  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
443  *
444  * @handle:             handle to this transaction
445  * @inode:              file inode
446  * @goal:               given target block(filesystem wide)
447  * @count:              pointer to total number of blocks needed
448  * @errp:               error code
449  *
450  * Return 1st allocated block number on success, *count stores total account
451  * error stores in errp pointer
452  */
453 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
454                                   ext4_fsblk_t goal, unsigned int flags,
455                                   unsigned long *count, int *errp)
456 {
457         struct ext4_allocation_request ar;
458         ext4_fsblk_t ret;
459
460         memset(&ar, 0, sizeof(ar));
461         /* Fill with neighbour allocated blocks */
462         ar.inode = inode;
463         ar.goal = goal;
464         ar.len = count ? *count : 1;
465         ar.flags = flags;
466
467         ret = ext4_mb_new_blocks(handle, &ar, errp);
468         if (count)
469                 *count = ar.len;
470         /*
471          * Account for the allocated meta blocks.  We will never
472          * fail EDQUOT for metdata, but we do account for it.
473          */
474         if (!(*errp) &&
475             ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
476                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
477                 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
478                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
479                 dquot_alloc_block_nofail(inode, ar.len);
480         }
481         return ret;
482 }
483
484 /**
485  * ext4_count_free_blocks() -- count filesystem free blocks
486  * @sb:         superblock
487  *
488  * Adds up the number of free blocks from each block group.
489  */
490 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
491 {
492         ext4_fsblk_t desc_count;
493         struct ext4_group_desc *gdp;
494         ext4_group_t i;
495         ext4_group_t ngroups = ext4_get_groups_count(sb);
496 #ifdef EXT4FS_DEBUG
497         struct ext4_super_block *es;
498         ext4_fsblk_t bitmap_count;
499         unsigned int x;
500         struct buffer_head *bitmap_bh = NULL;
501
502         es = EXT4_SB(sb)->s_es;
503         desc_count = 0;
504         bitmap_count = 0;
505         gdp = NULL;
506
507         for (i = 0; i < ngroups; i++) {
508                 gdp = ext4_get_group_desc(sb, i, NULL);
509                 if (!gdp)
510                         continue;
511                 desc_count += ext4_free_blks_count(sb, gdp);
512                 brelse(bitmap_bh);
513                 bitmap_bh = ext4_read_block_bitmap(sb, i);
514                 if (bitmap_bh == NULL)
515                         continue;
516
517                 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
518                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
519                         i, ext4_free_blks_count(sb, gdp), x);
520                 bitmap_count += x;
521         }
522         brelse(bitmap_bh);
523         printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
524                 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
525                desc_count, bitmap_count);
526         return bitmap_count;
527 #else
528         desc_count = 0;
529         for (i = 0; i < ngroups; i++) {
530                 gdp = ext4_get_group_desc(sb, i, NULL);
531                 if (!gdp)
532                         continue;
533                 desc_count += ext4_free_blks_count(sb, gdp);
534         }
535
536         return desc_count;
537 #endif
538 }
539
540 static inline int test_root(ext4_group_t a, int b)
541 {
542         int num = b;
543
544         while (a > num)
545                 num *= b;
546         return num == a;
547 }
548
549 static int ext4_group_sparse(ext4_group_t group)
550 {
551         if (group <= 1)
552                 return 1;
553         if (!(group & 1))
554                 return 0;
555         return (test_root(group, 7) || test_root(group, 5) ||
556                 test_root(group, 3));
557 }
558
559 /**
560  *      ext4_bg_has_super - number of blocks used by the superblock in group
561  *      @sb: superblock for filesystem
562  *      @group: group number to check
563  *
564  *      Return the number of blocks used by the superblock (primary or backup)
565  *      in this group.  Currently this will be only 0 or 1.
566  */
567 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
568 {
569         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
570                                 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
571                         !ext4_group_sparse(group))
572                 return 0;
573         return 1;
574 }
575
576 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
577                                         ext4_group_t group)
578 {
579         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
580         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
581         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
582
583         if (group == first || group == first + 1 || group == last)
584                 return 1;
585         return 0;
586 }
587
588 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
589                                         ext4_group_t group)
590 {
591         if (!ext4_bg_has_super(sb, group))
592                 return 0;
593
594         if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
595                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
596         else
597                 return EXT4_SB(sb)->s_gdb_count;
598 }
599
600 /**
601  *      ext4_bg_num_gdb - number of blocks used by the group table in group
602  *      @sb: superblock for filesystem
603  *      @group: group number to check
604  *
605  *      Return the number of blocks used by the group descriptor table
606  *      (primary or backup) in this group.  In the future there may be a
607  *      different number of descriptor blocks in each group.
608  */
609 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
610 {
611         unsigned long first_meta_bg =
612                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
613         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
614
615         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
616                         metagroup < first_meta_bg)
617                 return ext4_bg_num_gdb_nometa(sb, group);
618
619         return ext4_bg_num_gdb_meta(sb,group);
620
621 }
622