Merge branch 'for-2.6.28' of git://linux-nfs.org/~bfields/linux
[linux-2.6] / 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 "group.h"
23
24 /*
25  * balloc.c contains the blocks allocation and deallocation routines
26  */
27
28 /*
29  * Calculate the block group number and offset, given a block number
30  */
31 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
32                 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
33 {
34         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
35         ext4_grpblk_t offset;
36
37         blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
38         offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
39         if (offsetp)
40                 *offsetp = offset;
41         if (blockgrpp)
42                 *blockgrpp = blocknr;
43
44 }
45
46 static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
47                         ext4_group_t block_group)
48 {
49         ext4_group_t actual_group;
50         ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
51         if (actual_group == block_group)
52                 return 1;
53         return 0;
54 }
55
56 static int ext4_group_used_meta_blocks(struct super_block *sb,
57                                 ext4_group_t block_group)
58 {
59         ext4_fsblk_t tmp;
60         struct ext4_sb_info *sbi = EXT4_SB(sb);
61         /* block bitmap, inode bitmap, and inode table blocks */
62         int used_blocks = sbi->s_itb_per_group + 2;
63
64         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
65                 struct ext4_group_desc *gdp;
66                 struct buffer_head *bh;
67
68                 gdp = ext4_get_group_desc(sb, block_group, &bh);
69                 if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
70                                         block_group))
71                         used_blocks--;
72
73                 if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
74                                         block_group))
75                         used_blocks--;
76
77                 tmp = ext4_inode_table(sb, gdp);
78                 for (; tmp < ext4_inode_table(sb, gdp) +
79                                 sbi->s_itb_per_group; tmp++) {
80                         if (!ext4_block_in_group(sb, tmp, block_group))
81                                 used_blocks -= 1;
82                 }
83         }
84         return used_blocks;
85 }
86
87 /* Initializes an uninitialized block bitmap if given, and returns the
88  * number of blocks free in the group. */
89 unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
90                  ext4_group_t block_group, struct ext4_group_desc *gdp)
91 {
92         int bit, bit_max;
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, __func__,
103                                   "Checksum bad for group %lu\n", block_group);
104                         gdp->bg_free_blocks_count = 0;
105                         gdp->bg_free_inodes_count = 0;
106                         gdp->bg_itable_unused = 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 == sbi->s_groups_count - 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                         le32_to_cpu(sbi->s_es->s_first_data_block) -
136                         (EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count - 1));
137         } else {
138                 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
139         }
140
141         free_blocks = group_blocks - bit_max;
142
143         if (bh) {
144                 ext4_fsblk_t start, tmp;
145                 int flex_bg = 0;
146
147                 for (bit = 0; bit < bit_max; bit++)
148                         ext4_set_bit(bit, bh->b_data);
149
150                 start = ext4_group_first_block_no(sb, block_group);
151
152                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
153                                               EXT4_FEATURE_INCOMPAT_FLEX_BG))
154                         flex_bg = 1;
155
156                 /* Set bits for block and inode bitmaps, and inode table */
157                 tmp = ext4_block_bitmap(sb, gdp);
158                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
159                         ext4_set_bit(tmp - start, bh->b_data);
160
161                 tmp = ext4_inode_bitmap(sb, gdp);
162                 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
163                         ext4_set_bit(tmp - start, bh->b_data);
164
165                 tmp = ext4_inode_table(sb, gdp);
166                 for (; tmp < ext4_inode_table(sb, gdp) +
167                                 sbi->s_itb_per_group; tmp++) {
168                         if (!flex_bg ||
169                                 ext4_block_in_group(sb, tmp, block_group))
170                                 ext4_set_bit(tmp - start, bh->b_data);
171                 }
172                 /*
173                  * Also if the number of blocks within the group is
174                  * less than the blocksize * 8 ( which is the size
175                  * of bitmap ), set rest of the block bitmap to 1
176                  */
177                 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
178         }
179         return free_blocks - ext4_group_used_meta_blocks(sb, block_group);
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 #define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
196
197 /**
198  * ext4_get_group_desc() -- load group descriptor from disk
199  * @sb:                 super block
200  * @block_group:        given block group
201  * @bh:                 pointer to the buffer head to store the block
202  *                      group descriptor
203  */
204 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
205                                              ext4_group_t block_group,
206                                              struct buffer_head **bh)
207 {
208         unsigned long group_desc;
209         unsigned long offset;
210         struct ext4_group_desc *desc;
211         struct ext4_sb_info *sbi = EXT4_SB(sb);
212
213         if (block_group >= sbi->s_groups_count) {
214                 ext4_error(sb, "ext4_get_group_desc",
215                            "block_group >= groups_count - "
216                            "block_group = %lu, groups_count = %lu",
217                            block_group, sbi->s_groups_count);
218
219                 return NULL;
220         }
221         smp_rmb();
222
223         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
224         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
225         if (!sbi->s_group_desc[group_desc]) {
226                 ext4_error(sb, "ext4_get_group_desc",
227                            "Group descriptor not loaded - "
228                            "block_group = %lu, group_desc = %lu, desc = %lu",
229                            block_group, group_desc, offset);
230                 return NULL;
231         }
232
233         desc = (struct ext4_group_desc *)(
234                 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
235                 offset * EXT4_DESC_SIZE(sb));
236         if (bh)
237                 *bh = sbi->s_group_desc[group_desc];
238         return desc;
239 }
240
241 static int ext4_valid_block_bitmap(struct super_block *sb,
242                                         struct ext4_group_desc *desc,
243                                         unsigned int block_group,
244                                         struct buffer_head *bh)
245 {
246         ext4_grpblk_t offset;
247         ext4_grpblk_t next_zero_bit;
248         ext4_fsblk_t bitmap_blk;
249         ext4_fsblk_t group_first_block;
250
251         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
252                 /* with FLEX_BG, the inode/block bitmaps and itable
253                  * blocks may not be in the group at all
254                  * so the bitmap validation will be skipped for those groups
255                  * or it has to also read the block group where the bitmaps
256                  * are located to verify they are set.
257                  */
258                 return 1;
259         }
260         group_first_block = ext4_group_first_block_no(sb, block_group);
261
262         /* check whether block bitmap block number is set */
263         bitmap_blk = ext4_block_bitmap(sb, desc);
264         offset = bitmap_blk - group_first_block;
265         if (!ext4_test_bit(offset, bh->b_data))
266                 /* bad block bitmap */
267                 goto err_out;
268
269         /* check whether the inode bitmap block number is set */
270         bitmap_blk = ext4_inode_bitmap(sb, desc);
271         offset = bitmap_blk - group_first_block;
272         if (!ext4_test_bit(offset, bh->b_data))
273                 /* bad block bitmap */
274                 goto err_out;
275
276         /* check whether the inode table block number is set */
277         bitmap_blk = ext4_inode_table(sb, desc);
278         offset = bitmap_blk - group_first_block;
279         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
280                                 offset + EXT4_SB(sb)->s_itb_per_group,
281                                 offset);
282         if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
283                 /* good bitmap for inode tables */
284                 return 1;
285
286 err_out:
287         ext4_error(sb, __func__,
288                         "Invalid block bitmap - "
289                         "block_group = %d, block = %llu",
290                         block_group, bitmap_blk);
291         return 0;
292 }
293 /**
294  * ext4_read_block_bitmap()
295  * @sb:                 super block
296  * @block_group:        given block group
297  *
298  * Read the bitmap for a given block_group,and validate the
299  * bits for block/inode/inode tables are set in the bitmaps
300  *
301  * Return buffer_head on success or NULL in case of failure.
302  */
303 struct buffer_head *
304 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
305 {
306         struct ext4_group_desc *desc;
307         struct buffer_head *bh = NULL;
308         ext4_fsblk_t bitmap_blk;
309
310         desc = ext4_get_group_desc(sb, block_group, NULL);
311         if (!desc)
312                 return NULL;
313         bitmap_blk = ext4_block_bitmap(sb, desc);
314         bh = sb_getblk(sb, bitmap_blk);
315         if (unlikely(!bh)) {
316                 ext4_error(sb, __func__,
317                             "Cannot read block bitmap - "
318                             "block_group = %lu, block_bitmap = %llu",
319                             block_group, bitmap_blk);
320                 return NULL;
321         }
322         if (buffer_uptodate(bh) &&
323             !(desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)))
324                 return bh;
325
326         lock_buffer(bh);
327         spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
328         if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
329                 ext4_init_block_bitmap(sb, bh, block_group, desc);
330                 set_buffer_uptodate(bh);
331                 unlock_buffer(bh);
332                 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
333                 return bh;
334         }
335         spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
336         if (bh_submit_read(bh) < 0) {
337                 put_bh(bh);
338                 ext4_error(sb, __func__,
339                             "Cannot read block bitmap - "
340                             "block_group = %lu, block_bitmap = %llu",
341                             block_group, bitmap_blk);
342                 return NULL;
343         }
344         ext4_valid_block_bitmap(sb, desc, block_group, bh);
345         /*
346          * file system mounted not to panic on error,
347          * continue with corrupt bitmap
348          */
349         return bh;
350 }
351
352 /**
353  * ext4_free_blocks_sb() -- Free given blocks and update quota
354  * @handle:                     handle to this transaction
355  * @sb:                         super block
356  * @block:                      start physcial block to free
357  * @count:                      number of blocks to free
358  * @pdquot_freed_blocks:        pointer to quota
359  *
360  * XXX This function is only used by the on-line resizing code, which
361  * should probably be fixed up to call the mballoc variant.  There
362  * this needs to be cleaned up later; in fact, I'm not convinced this
363  * is 100% correct in the face of the mballoc code.  The online resizing
364  * code needs to be fixed up to more tightly (and correctly) interlock
365  * with the mballoc code.
366  */
367 void ext4_free_blocks_sb(handle_t *handle, struct super_block *sb,
368                          ext4_fsblk_t block, unsigned long count,
369                          unsigned long *pdquot_freed_blocks)
370 {
371         struct buffer_head *bitmap_bh = NULL;
372         struct buffer_head *gd_bh;
373         ext4_group_t block_group;
374         ext4_grpblk_t bit;
375         unsigned long i;
376         unsigned long overflow;
377         struct ext4_group_desc *desc;
378         struct ext4_super_block *es;
379         struct ext4_sb_info *sbi;
380         int err = 0, ret;
381         ext4_grpblk_t group_freed;
382
383         *pdquot_freed_blocks = 0;
384         sbi = EXT4_SB(sb);
385         es = sbi->s_es;
386         if (block < le32_to_cpu(es->s_first_data_block) ||
387             block + count < block ||
388             block + count > ext4_blocks_count(es)) {
389                 ext4_error(sb, "ext4_free_blocks",
390                            "Freeing blocks not in datazone - "
391                            "block = %llu, count = %lu", block, count);
392                 goto error_return;
393         }
394
395         ext4_debug("freeing block(s) %llu-%llu\n", block, block + count - 1);
396
397 do_more:
398         overflow = 0;
399         ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
400         /*
401          * Check to see if we are freeing blocks across a group
402          * boundary.
403          */
404         if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
405                 overflow = bit + count - EXT4_BLOCKS_PER_GROUP(sb);
406                 count -= overflow;
407         }
408         brelse(bitmap_bh);
409         bitmap_bh = ext4_read_block_bitmap(sb, block_group);
410         if (!bitmap_bh)
411                 goto error_return;
412         desc = ext4_get_group_desc(sb, block_group, &gd_bh);
413         if (!desc)
414                 goto error_return;
415
416         if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
417             in_range(ext4_inode_bitmap(sb, desc), block, count) ||
418             in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
419             in_range(block + count - 1, ext4_inode_table(sb, desc),
420                      sbi->s_itb_per_group)) {
421                 ext4_error(sb, "ext4_free_blocks",
422                            "Freeing blocks in system zones - "
423                            "Block = %llu, count = %lu",
424                            block, count);
425                 goto error_return;
426         }
427
428         /*
429          * We are about to start releasing blocks in the bitmap,
430          * so we need undo access.
431          */
432         /* @@@ check errors */
433         BUFFER_TRACE(bitmap_bh, "getting undo access");
434         err = ext4_journal_get_undo_access(handle, bitmap_bh);
435         if (err)
436                 goto error_return;
437
438         /*
439          * We are about to modify some metadata.  Call the journal APIs
440          * to unshare ->b_data if a currently-committing transaction is
441          * using it
442          */
443         BUFFER_TRACE(gd_bh, "get_write_access");
444         err = ext4_journal_get_write_access(handle, gd_bh);
445         if (err)
446                 goto error_return;
447
448         jbd_lock_bh_state(bitmap_bh);
449
450         for (i = 0, group_freed = 0; i < count; i++) {
451                 /*
452                  * An HJ special.  This is expensive...
453                  */
454 #ifdef CONFIG_JBD2_DEBUG
455                 jbd_unlock_bh_state(bitmap_bh);
456                 {
457                         struct buffer_head *debug_bh;
458                         debug_bh = sb_find_get_block(sb, block + i);
459                         if (debug_bh) {
460                                 BUFFER_TRACE(debug_bh, "Deleted!");
461                                 if (!bh2jh(bitmap_bh)->b_committed_data)
462                                         BUFFER_TRACE(debug_bh,
463                                                 "No commited data in bitmap");
464                                 BUFFER_TRACE2(debug_bh, bitmap_bh, "bitmap");
465                                 __brelse(debug_bh);
466                         }
467                 }
468                 jbd_lock_bh_state(bitmap_bh);
469 #endif
470                 if (need_resched()) {
471                         jbd_unlock_bh_state(bitmap_bh);
472                         cond_resched();
473                         jbd_lock_bh_state(bitmap_bh);
474                 }
475                 /* @@@ This prevents newly-allocated data from being
476                  * freed and then reallocated within the same
477                  * transaction.
478                  *
479                  * Ideally we would want to allow that to happen, but to
480                  * do so requires making jbd2_journal_forget() capable of
481                  * revoking the queued write of a data block, which
482                  * implies blocking on the journal lock.  *forget()
483                  * cannot block due to truncate races.
484                  *
485                  * Eventually we can fix this by making jbd2_journal_forget()
486                  * return a status indicating whether or not it was able
487                  * to revoke the buffer.  On successful revoke, it is
488                  * safe not to set the allocation bit in the committed
489                  * bitmap, because we know that there is no outstanding
490                  * activity on the buffer any more and so it is safe to
491                  * reallocate it.
492                  */
493                 BUFFER_TRACE(bitmap_bh, "set in b_committed_data");
494                 J_ASSERT_BH(bitmap_bh,
495                                 bh2jh(bitmap_bh)->b_committed_data != NULL);
496                 ext4_set_bit_atomic(sb_bgl_lock(sbi, block_group), bit + i,
497                                 bh2jh(bitmap_bh)->b_committed_data);
498
499                 /*
500                  * We clear the bit in the bitmap after setting the committed
501                  * data bit, because this is the reverse order to that which
502                  * the allocator uses.
503                  */
504                 BUFFER_TRACE(bitmap_bh, "clear bit");
505                 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
506                                                 bit + i, bitmap_bh->b_data)) {
507                         jbd_unlock_bh_state(bitmap_bh);
508                         ext4_error(sb, __func__,
509                                    "bit already cleared for block %llu",
510                                    (ext4_fsblk_t)(block + i));
511                         jbd_lock_bh_state(bitmap_bh);
512                         BUFFER_TRACE(bitmap_bh, "bit already cleared");
513                 } else {
514                         group_freed++;
515                 }
516         }
517         jbd_unlock_bh_state(bitmap_bh);
518
519         spin_lock(sb_bgl_lock(sbi, block_group));
520         le16_add_cpu(&desc->bg_free_blocks_count, group_freed);
521         desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
522         spin_unlock(sb_bgl_lock(sbi, block_group));
523         percpu_counter_add(&sbi->s_freeblocks_counter, count);
524
525         if (sbi->s_log_groups_per_flex) {
526                 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
527                 spin_lock(sb_bgl_lock(sbi, flex_group));
528                 sbi->s_flex_groups[flex_group].free_blocks += count;
529                 spin_unlock(sb_bgl_lock(sbi, flex_group));
530         }
531
532         /* We dirtied the bitmap block */
533         BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
534         err = ext4_journal_dirty_metadata(handle, bitmap_bh);
535
536         /* And the group descriptor block */
537         BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
538         ret = ext4_journal_dirty_metadata(handle, gd_bh);
539         if (!err) err = ret;
540         *pdquot_freed_blocks += group_freed;
541
542         if (overflow && !err) {
543                 block += count;
544                 count = overflow;
545                 goto do_more;
546         }
547         sb->s_dirt = 1;
548 error_return:
549         brelse(bitmap_bh);
550         ext4_std_error(sb, err);
551         return;
552 }
553
554 /**
555  * ext4_free_blocks() -- Free given blocks and update quota
556  * @handle:             handle for this transaction
557  * @inode:              inode
558  * @block:              start physical block to free
559  * @count:              number of blocks to count
560  * @metadata:           Are these metadata blocks
561  */
562 void ext4_free_blocks(handle_t *handle, struct inode *inode,
563                         ext4_fsblk_t block, unsigned long count,
564                         int metadata)
565 {
566         struct super_block *sb;
567         unsigned long dquot_freed_blocks;
568
569         /* this isn't the right place to decide whether block is metadata
570          * inode.c/extents.c knows better, but for safety ... */
571         if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
572                         ext4_should_journal_data(inode))
573                 metadata = 1;
574
575         sb = inode->i_sb;
576
577         ext4_mb_free_blocks(handle, inode, block, count,
578                             metadata, &dquot_freed_blocks);
579         if (dquot_freed_blocks)
580                 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
581         return;
582 }
583
584 int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
585                                                 s64 nblocks)
586 {
587         s64 free_blocks, dirty_blocks;
588         s64 root_blocks = 0;
589         struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
590         struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
591
592         free_blocks  = percpu_counter_read_positive(fbc);
593         dirty_blocks = percpu_counter_read_positive(dbc);
594
595         if (!capable(CAP_SYS_RESOURCE) &&
596                 sbi->s_resuid != current->fsuid &&
597                 (sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid)))
598                 root_blocks = ext4_r_blocks_count(sbi->s_es);
599
600         if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
601                                                 EXT4_FREEBLOCKS_WATERMARK) {
602                 free_blocks  = percpu_counter_sum(fbc);
603                 dirty_blocks = percpu_counter_sum(dbc);
604                 if (dirty_blocks < 0) {
605                         printk(KERN_CRIT "Dirty block accounting "
606                                         "went wrong %lld\n",
607                                         dirty_blocks);
608                 }
609         }
610         /* Check whether we have space after
611          * accounting for current dirty blocks
612          */
613         if (free_blocks < ((root_blocks + nblocks) + dirty_blocks))
614                 /* we don't have free space */
615                 return -ENOSPC;
616
617         /* Add the blocks to nblocks */
618         percpu_counter_add(dbc, nblocks);
619         return 0;
620 }
621
622 /**
623  * ext4_has_free_blocks()
624  * @sbi:        in-core super block structure.
625  * @nblocks:    number of neeed blocks
626  *
627  * Check if filesystem has free blocks available for allocation.
628  * Return the number of blocks avaible for allocation for this request
629  * On success, return nblocks
630  */
631 ext4_fsblk_t ext4_has_free_blocks(struct ext4_sb_info *sbi,
632                                                 s64 nblocks)
633 {
634         s64 free_blocks, dirty_blocks;
635         s64 root_blocks = 0;
636         struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
637         struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
638
639         free_blocks  = percpu_counter_read_positive(fbc);
640         dirty_blocks = percpu_counter_read_positive(dbc);
641
642         if (!capable(CAP_SYS_RESOURCE) &&
643                 sbi->s_resuid != current->fsuid &&
644                 (sbi->s_resgid == 0 || !in_group_p(sbi->s_resgid)))
645                 root_blocks = ext4_r_blocks_count(sbi->s_es);
646
647         if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
648                                                 EXT4_FREEBLOCKS_WATERMARK) {
649                 free_blocks  = percpu_counter_sum(fbc);
650                 dirty_blocks = percpu_counter_sum(dbc);
651         }
652         if (free_blocks <= (root_blocks + dirty_blocks))
653                 /* we don't have free space */
654                 return 0;
655
656         if (free_blocks - (root_blocks + dirty_blocks) < nblocks)
657                 return free_blocks - (root_blocks + dirty_blocks);
658         return nblocks;
659 }
660
661
662 /**
663  * ext4_should_retry_alloc()
664  * @sb:                 super block
665  * @retries             number of attemps has been made
666  *
667  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
668  * it is profitable to retry the operation, this function will wait
669  * for the current or commiting transaction to complete, and then
670  * return TRUE.
671  *
672  * if the total number of retries exceed three times, return FALSE.
673  */
674 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
675 {
676         if (!ext4_has_free_blocks(EXT4_SB(sb), 1) || (*retries)++ > 3)
677                 return 0;
678
679         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
680
681         return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
682 }
683
684 #define EXT4_META_BLOCK 0x1
685
686 static ext4_fsblk_t do_blk_alloc(handle_t *handle, struct inode *inode,
687                                 ext4_lblk_t iblock, ext4_fsblk_t goal,
688                                 unsigned long *count, int *errp, int flags)
689 {
690         struct ext4_allocation_request ar;
691         ext4_fsblk_t ret;
692
693         memset(&ar, 0, sizeof(ar));
694         /* Fill with neighbour allocated blocks */
695
696         ar.inode = inode;
697         ar.goal = goal;
698         ar.len = *count;
699         ar.logical = iblock;
700
701         if (S_ISREG(inode->i_mode) && !(flags & EXT4_META_BLOCK))
702                 /* enable in-core preallocation for data block allocation */
703                 ar.flags = EXT4_MB_HINT_DATA;
704         else
705                 /* disable in-core preallocation for non-regular files */
706                 ar.flags = 0;
707
708         ret = ext4_mb_new_blocks(handle, &ar, errp);
709         *count = ar.len;
710         return ret;
711 }
712
713 /*
714  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
715  *
716  * @handle:             handle to this transaction
717  * @inode:              file inode
718  * @goal:               given target block(filesystem wide)
719  * @count:              total number of blocks need
720  * @errp:               error code
721  *
722  * Return 1st allocated block numberon success, *count stores total account
723  * error stores in errp pointer
724  */
725 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
726                 ext4_fsblk_t goal, unsigned long *count, int *errp)
727 {
728         ext4_fsblk_t ret;
729         ret = do_blk_alloc(handle, inode, 0, goal,
730                                 count, errp, EXT4_META_BLOCK);
731         /*
732          * Account for the allocated meta blocks
733          */
734         if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
735                 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
736                 EXT4_I(inode)->i_allocated_meta_blocks += *count;
737                 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
738         }
739         return ret;
740 }
741
742 /*
743  * ext4_new_meta_block() -- allocate block for meta data (indexing) blocks
744  *
745  * @handle:             handle to this transaction
746  * @inode:              file inode
747  * @goal:               given target block(filesystem wide)
748  * @errp:               error code
749  *
750  * Return allocated block number on success
751  */
752 ext4_fsblk_t ext4_new_meta_block(handle_t *handle, struct inode *inode,
753                 ext4_fsblk_t goal, int *errp)
754 {
755         unsigned long count = 1;
756         return ext4_new_meta_blocks(handle, inode, goal, &count, errp);
757 }
758
759 /*
760  * ext4_new_blocks() -- allocate data blocks
761  *
762  * @handle:             handle to this transaction
763  * @inode:              file inode
764  * @goal:               given target block(filesystem wide)
765  * @count:              total number of blocks need
766  * @errp:               error code
767  *
768  * Return 1st allocated block numberon success, *count stores total account
769  * error stores in errp pointer
770  */
771
772 ext4_fsblk_t ext4_new_blocks(handle_t *handle, struct inode *inode,
773                                 ext4_lblk_t iblock, ext4_fsblk_t goal,
774                                 unsigned long *count, int *errp)
775 {
776         return do_blk_alloc(handle, inode, iblock, goal, count, errp, 0);
777 }
778
779 /**
780  * ext4_count_free_blocks() -- count filesystem free blocks
781  * @sb:         superblock
782  *
783  * Adds up the number of free blocks from each block group.
784  */
785 ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
786 {
787         ext4_fsblk_t desc_count;
788         struct ext4_group_desc *gdp;
789         ext4_group_t i;
790         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
791 #ifdef EXT4FS_DEBUG
792         struct ext4_super_block *es;
793         ext4_fsblk_t bitmap_count;
794         unsigned long x;
795         struct buffer_head *bitmap_bh = NULL;
796
797         es = EXT4_SB(sb)->s_es;
798         desc_count = 0;
799         bitmap_count = 0;
800         gdp = NULL;
801
802         smp_rmb();
803         for (i = 0; i < ngroups; i++) {
804                 gdp = ext4_get_group_desc(sb, i, NULL);
805                 if (!gdp)
806                         continue;
807                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
808                 brelse(bitmap_bh);
809                 bitmap_bh = ext4_read_block_bitmap(sb, i);
810                 if (bitmap_bh == NULL)
811                         continue;
812
813                 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
814                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
815                         i, le16_to_cpu(gdp->bg_free_blocks_count), x);
816                 bitmap_count += x;
817         }
818         brelse(bitmap_bh);
819         printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
820                 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
821                desc_count, bitmap_count);
822         return bitmap_count;
823 #else
824         desc_count = 0;
825         smp_rmb();
826         for (i = 0; i < ngroups; i++) {
827                 gdp = ext4_get_group_desc(sb, i, NULL);
828                 if (!gdp)
829                         continue;
830                 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
831         }
832
833         return desc_count;
834 #endif
835 }
836
837 static inline int test_root(ext4_group_t a, int b)
838 {
839         int num = b;
840
841         while (a > num)
842                 num *= b;
843         return num == a;
844 }
845
846 static int ext4_group_sparse(ext4_group_t group)
847 {
848         if (group <= 1)
849                 return 1;
850         if (!(group & 1))
851                 return 0;
852         return (test_root(group, 7) || test_root(group, 5) ||
853                 test_root(group, 3));
854 }
855
856 /**
857  *      ext4_bg_has_super - number of blocks used by the superblock in group
858  *      @sb: superblock for filesystem
859  *      @group: group number to check
860  *
861  *      Return the number of blocks used by the superblock (primary or backup)
862  *      in this group.  Currently this will be only 0 or 1.
863  */
864 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
865 {
866         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
867                                 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
868                         !ext4_group_sparse(group))
869                 return 0;
870         return 1;
871 }
872
873 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
874                                         ext4_group_t group)
875 {
876         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
877         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
878         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
879
880         if (group == first || group == first + 1 || group == last)
881                 return 1;
882         return 0;
883 }
884
885 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
886                                         ext4_group_t group)
887 {
888         return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0;
889 }
890
891 /**
892  *      ext4_bg_num_gdb - number of blocks used by the group table in group
893  *      @sb: superblock for filesystem
894  *      @group: group number to check
895  *
896  *      Return the number of blocks used by the group descriptor table
897  *      (primary or backup) in this group.  In the future there may be a
898  *      different number of descriptor blocks in each group.
899  */
900 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
901 {
902         unsigned long first_meta_bg =
903                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
904         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
905
906         if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
907                         metagroup < first_meta_bg)
908                 return ext4_bg_num_gdb_nometa(sb, group);
909
910         return ext4_bg_num_gdb_meta(sb,group);
911
912 }
913