GNU Linux-libre 4.19.211-gnu1
[releases.git] / fs / ext4 / balloc.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/balloc.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  Enhanced block allocation by Stephen Tweedie (sct@redhat.com), 1993
11  *  Big-endian to little-endian byte-swapping/bitmaps by
12  *        David S. Miller (davem@caip.rutgers.edu), 1995
13  */
14
15 #include <linux/time.h>
16 #include <linux/capability.h>
17 #include <linux/fs.h>
18 #include <linux/quotaops.h>
19 #include <linux/buffer_head.h>
20 #include "ext4.h"
21 #include "ext4_jbd2.h"
22 #include "mballoc.h"
23
24 #include <trace/events/ext4.h>
25
26 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
27                                             ext4_group_t block_group);
28 /*
29  * balloc.c contains the blocks allocation and deallocation routines
30  */
31
32 /*
33  * Calculate block group number for a given block number
34  */
35 ext4_group_t ext4_get_group_number(struct super_block *sb,
36                                    ext4_fsblk_t block)
37 {
38         ext4_group_t group;
39
40         if (test_opt2(sb, STD_GROUP_SIZE))
41                 group = (block -
42                          le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >>
43                         (EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3);
44         else
45                 ext4_get_group_no_and_offset(sb, block, &group, NULL);
46         return group;
47 }
48
49 /*
50  * Calculate the block group number and offset into the block/cluster
51  * allocation bitmap, given a block number
52  */
53 void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
54                 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
55 {
56         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
57         ext4_grpblk_t offset;
58
59         blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
60         offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
61                 EXT4_SB(sb)->s_cluster_bits;
62         if (offsetp)
63                 *offsetp = offset;
64         if (blockgrpp)
65                 *blockgrpp = blocknr;
66
67 }
68
69 /*
70  * Check whether the 'block' lives within the 'block_group'. Returns 1 if so
71  * and 0 otherwise.
72  */
73 static inline int ext4_block_in_group(struct super_block *sb,
74                                       ext4_fsblk_t block,
75                                       ext4_group_t block_group)
76 {
77         ext4_group_t actual_group;
78
79         actual_group = ext4_get_group_number(sb, block);
80         return (actual_group == block_group) ? 1 : 0;
81 }
82
83 /* Return the number of clusters used for file system metadata; this
84  * represents the overhead needed by the file system.
85  */
86 static unsigned ext4_num_overhead_clusters(struct super_block *sb,
87                                            ext4_group_t block_group,
88                                            struct ext4_group_desc *gdp)
89 {
90         unsigned num_clusters;
91         int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
92         ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
93         ext4_fsblk_t itbl_blk;
94         struct ext4_sb_info *sbi = EXT4_SB(sb);
95
96         /* This is the number of clusters used by the superblock,
97          * block group descriptors, and reserved block group
98          * descriptor blocks */
99         num_clusters = ext4_num_base_meta_clusters(sb, block_group);
100
101         /*
102          * For the allocation bitmaps and inode table, we first need
103          * to check to see if the block is in the block group.  If it
104          * is, then check to see if the cluster is already accounted
105          * for in the clusters used for the base metadata cluster, or
106          * if we can increment the base metadata cluster to include
107          * that block.  Otherwise, we will have to track the cluster
108          * used for the allocation bitmap or inode table explicitly.
109          * Normally all of these blocks are contiguous, so the special
110          * case handling shouldn't be necessary except for *very*
111          * unusual file system layouts.
112          */
113         if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
114                 block_cluster = EXT4_B2C(sbi,
115                                          ext4_block_bitmap(sb, gdp) - start);
116                 if (block_cluster < num_clusters)
117                         block_cluster = -1;
118                 else if (block_cluster == num_clusters) {
119                         num_clusters++;
120                         block_cluster = -1;
121                 }
122         }
123
124         if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
125                 inode_cluster = EXT4_B2C(sbi,
126                                          ext4_inode_bitmap(sb, gdp) - start);
127                 if (inode_cluster < num_clusters)
128                         inode_cluster = -1;
129                 else if (inode_cluster == num_clusters) {
130                         num_clusters++;
131                         inode_cluster = -1;
132                 }
133         }
134
135         itbl_blk = ext4_inode_table(sb, gdp);
136         for (i = 0; i < sbi->s_itb_per_group; i++) {
137                 if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
138                         c = EXT4_B2C(sbi, itbl_blk + i - start);
139                         if ((c < num_clusters) || (c == inode_cluster) ||
140                             (c == block_cluster) || (c == itbl_cluster))
141                                 continue;
142                         if (c == num_clusters) {
143                                 num_clusters++;
144                                 continue;
145                         }
146                         num_clusters++;
147                         itbl_cluster = c;
148                 }
149         }
150
151         if (block_cluster != -1)
152                 num_clusters++;
153         if (inode_cluster != -1)
154                 num_clusters++;
155
156         return num_clusters;
157 }
158
159 static unsigned int num_clusters_in_group(struct super_block *sb,
160                                           ext4_group_t block_group)
161 {
162         unsigned int blocks;
163
164         if (block_group == ext4_get_groups_count(sb) - 1) {
165                 /*
166                  * Even though mke2fs always initializes the first and
167                  * last group, just in case some other tool was used,
168                  * we need to make sure we calculate the right free
169                  * blocks.
170                  */
171                 blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
172                         ext4_group_first_block_no(sb, block_group);
173         } else
174                 blocks = EXT4_BLOCKS_PER_GROUP(sb);
175         return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
176 }
177
178 /* Initializes an uninitialized block bitmap */
179 static int ext4_init_block_bitmap(struct super_block *sb,
180                                    struct buffer_head *bh,
181                                    ext4_group_t block_group,
182                                    struct ext4_group_desc *gdp)
183 {
184         unsigned int bit, bit_max;
185         struct ext4_sb_info *sbi = EXT4_SB(sb);
186         ext4_fsblk_t start, tmp;
187
188         J_ASSERT_BH(bh, buffer_locked(bh));
189
190         /* If checksum is bad mark all blocks used to prevent allocation
191          * essentially implementing a per-group read-only flag. */
192         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
193                 ext4_mark_group_bitmap_corrupted(sb, block_group,
194                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT |
195                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
196                 return -EFSBADCRC;
197         }
198         memset(bh->b_data, 0, sb->s_blocksize);
199
200         bit_max = ext4_num_base_meta_clusters(sb, block_group);
201         if ((bit_max >> 3) >= bh->b_size)
202                 return -EFSCORRUPTED;
203
204         for (bit = 0; bit < bit_max; bit++)
205                 ext4_set_bit(bit, bh->b_data);
206
207         start = ext4_group_first_block_no(sb, block_group);
208
209         /* Set bits for block and inode bitmaps, and inode table */
210         tmp = ext4_block_bitmap(sb, gdp);
211         if (ext4_block_in_group(sb, tmp, block_group))
212                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
213
214         tmp = ext4_inode_bitmap(sb, gdp);
215         if (ext4_block_in_group(sb, tmp, block_group))
216                 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
217
218         tmp = ext4_inode_table(sb, gdp);
219         for (; tmp < ext4_inode_table(sb, gdp) +
220                      sbi->s_itb_per_group; tmp++) {
221                 if (ext4_block_in_group(sb, tmp, block_group))
222                         ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
223         }
224
225         /*
226          * Also if the number of blocks within the group is less than
227          * the blocksize * 8 ( which is the size of bitmap ), set rest
228          * of the block bitmap to 1
229          */
230         ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
231                              sb->s_blocksize * 8, bh->b_data);
232         return 0;
233 }
234
235 /* Return the number of free blocks in a block group.  It is used when
236  * the block bitmap is uninitialized, so we can't just count the bits
237  * in the bitmap. */
238 unsigned ext4_free_clusters_after_init(struct super_block *sb,
239                                        ext4_group_t block_group,
240                                        struct ext4_group_desc *gdp)
241 {
242         return num_clusters_in_group(sb, block_group) - 
243                 ext4_num_overhead_clusters(sb, block_group, gdp);
244 }
245
246 /*
247  * The free blocks are managed by bitmaps.  A file system contains several
248  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
249  * block for inodes, N blocks for the inode table and data blocks.
250  *
251  * The file system contains group descriptors which are located after the
252  * super block.  Each descriptor contains the number of the bitmap block and
253  * the free blocks count in the block.  The descriptors are loaded in memory
254  * when a file system is mounted (see ext4_fill_super).
255  */
256
257 /**
258  * ext4_get_group_desc() -- load group descriptor from disk
259  * @sb:                 super block
260  * @block_group:        given block group
261  * @bh:                 pointer to the buffer head to store the block
262  *                      group descriptor
263  */
264 struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
265                                              ext4_group_t block_group,
266                                              struct buffer_head **bh)
267 {
268         unsigned int group_desc;
269         unsigned int offset;
270         ext4_group_t ngroups = ext4_get_groups_count(sb);
271         struct ext4_group_desc *desc;
272         struct ext4_sb_info *sbi = EXT4_SB(sb);
273         struct buffer_head *bh_p;
274
275         if (block_group >= ngroups) {
276                 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
277                            " groups_count = %u", block_group, ngroups);
278
279                 return NULL;
280         }
281
282         group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
283         offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
284         bh_p = sbi_array_rcu_deref(sbi, s_group_desc, group_desc);
285         /*
286          * sbi_array_rcu_deref returns with rcu unlocked, this is ok since
287          * the pointer being dereferenced won't be dereferenced again. By
288          * looking at the usage in add_new_gdb() the value isn't modified,
289          * just the pointer, and so it remains valid.
290          */
291         if (!bh_p) {
292                 ext4_error(sb, "Group descriptor not loaded - "
293                            "block_group = %u, group_desc = %u, desc = %u",
294                            block_group, group_desc, offset);
295                 return NULL;
296         }
297
298         desc = (struct ext4_group_desc *)(
299                 (__u8 *)bh_p->b_data +
300                 offset * EXT4_DESC_SIZE(sb));
301         if (bh)
302                 *bh = bh_p;
303         return desc;
304 }
305
306 /*
307  * Return the block number which was discovered to be invalid, or 0 if
308  * the block bitmap is valid.
309  */
310 static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb,
311                                             struct ext4_group_desc *desc,
312                                             ext4_group_t block_group,
313                                             struct buffer_head *bh)
314 {
315         struct ext4_sb_info *sbi = EXT4_SB(sb);
316         ext4_grpblk_t offset;
317         ext4_grpblk_t next_zero_bit;
318         ext4_grpblk_t max_bit = EXT4_CLUSTERS_PER_GROUP(sb);
319         ext4_fsblk_t blk;
320         ext4_fsblk_t group_first_block;
321
322         if (ext4_has_feature_flex_bg(sb)) {
323                 /* with FLEX_BG, the inode/block bitmaps and itable
324                  * blocks may not be in the group at all
325                  * so the bitmap validation will be skipped for those groups
326                  * or it has to also read the block group where the bitmaps
327                  * are located to verify they are set.
328                  */
329                 return 0;
330         }
331         group_first_block = ext4_group_first_block_no(sb, block_group);
332
333         /* check whether block bitmap block number is set */
334         blk = ext4_block_bitmap(sb, desc);
335         offset = blk - group_first_block;
336         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
337             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
338                 /* bad block bitmap */
339                 return blk;
340
341         /* check whether the inode bitmap block number is set */
342         blk = ext4_inode_bitmap(sb, desc);
343         offset = blk - group_first_block;
344         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
345             !ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data))
346                 /* bad block bitmap */
347                 return blk;
348
349         /* check whether the inode table block number is set */
350         blk = ext4_inode_table(sb, desc);
351         offset = blk - group_first_block;
352         if (offset < 0 || EXT4_B2C(sbi, offset) >= max_bit ||
353             EXT4_B2C(sbi, offset + sbi->s_itb_per_group) >= max_bit)
354                 return blk;
355         next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
356                         EXT4_B2C(sbi, offset + sbi->s_itb_per_group),
357                         EXT4_B2C(sbi, offset));
358         if (next_zero_bit <
359             EXT4_B2C(sbi, offset + sbi->s_itb_per_group))
360                 /* bad bitmap for inode tables */
361                 return blk;
362         return 0;
363 }
364
365 static int ext4_validate_block_bitmap(struct super_block *sb,
366                                       struct ext4_group_desc *desc,
367                                       ext4_group_t block_group,
368                                       struct buffer_head *bh)
369 {
370         ext4_fsblk_t    blk;
371         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
372
373         if (buffer_verified(bh))
374                 return 0;
375         if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
376                 return -EFSCORRUPTED;
377
378         ext4_lock_group(sb, block_group);
379         if (buffer_verified(bh))
380                 goto verified;
381         if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group,
382                         desc, bh))) {
383                 ext4_unlock_group(sb, block_group);
384                 ext4_error(sb, "bg %u: bad block bitmap checksum", block_group);
385                 ext4_mark_group_bitmap_corrupted(sb, block_group,
386                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
387                 return -EFSBADCRC;
388         }
389         blk = ext4_valid_block_bitmap(sb, desc, block_group, bh);
390         if (unlikely(blk != 0)) {
391                 ext4_unlock_group(sb, block_group);
392                 ext4_error(sb, "bg %u: block %llu: invalid block bitmap",
393                            block_group, blk);
394                 ext4_mark_group_bitmap_corrupted(sb, block_group,
395                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
396                 return -EFSCORRUPTED;
397         }
398         set_buffer_verified(bh);
399 verified:
400         ext4_unlock_group(sb, block_group);
401         return 0;
402 }
403
404 /**
405  * ext4_read_block_bitmap_nowait()
406  * @sb:                 super block
407  * @block_group:        given block group
408  *
409  * Read the bitmap for a given block_group,and validate the
410  * bits for block/inode/inode tables are set in the bitmaps
411  *
412  * Return buffer_head on success or NULL in case of failure.
413  */
414 struct buffer_head *
415 ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
416 {
417         struct ext4_group_desc *desc;
418         struct ext4_sb_info *sbi = EXT4_SB(sb);
419         struct buffer_head *bh;
420         ext4_fsblk_t bitmap_blk;
421         int err;
422
423         desc = ext4_get_group_desc(sb, block_group, NULL);
424         if (!desc)
425                 return ERR_PTR(-EFSCORRUPTED);
426         bitmap_blk = ext4_block_bitmap(sb, desc);
427         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
428             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
429                 ext4_error(sb, "Invalid block bitmap block %llu in "
430                            "block_group %u", bitmap_blk, block_group);
431                 ext4_mark_group_bitmap_corrupted(sb, block_group,
432                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
433                 return ERR_PTR(-EFSCORRUPTED);
434         }
435         bh = sb_getblk(sb, bitmap_blk);
436         if (unlikely(!bh)) {
437                 ext4_warning(sb, "Cannot get buffer for block bitmap - "
438                              "block_group = %u, block_bitmap = %llu",
439                              block_group, bitmap_blk);
440                 return ERR_PTR(-ENOMEM);
441         }
442
443         if (bitmap_uptodate(bh))
444                 goto verify;
445
446         lock_buffer(bh);
447         if (bitmap_uptodate(bh)) {
448                 unlock_buffer(bh);
449                 goto verify;
450         }
451         ext4_lock_group(sb, block_group);
452         if (ext4_has_group_desc_csum(sb) &&
453             (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
454                 if (block_group == 0) {
455                         ext4_unlock_group(sb, block_group);
456                         unlock_buffer(bh);
457                         ext4_error(sb, "Block bitmap for bg 0 marked "
458                                    "uninitialized");
459                         err = -EFSCORRUPTED;
460                         goto out;
461                 }
462                 err = ext4_init_block_bitmap(sb, bh, block_group, desc);
463                 set_bitmap_uptodate(bh);
464                 set_buffer_uptodate(bh);
465                 set_buffer_verified(bh);
466                 ext4_unlock_group(sb, block_group);
467                 unlock_buffer(bh);
468                 if (err) {
469                         ext4_error(sb, "Failed to init block bitmap for group "
470                                    "%u: %d", block_group, err);
471                         goto out;
472                 }
473                 goto verify;
474         }
475         ext4_unlock_group(sb, block_group);
476         if (buffer_uptodate(bh)) {
477                 /*
478                  * if not uninit if bh is uptodate,
479                  * bitmap is also uptodate
480                  */
481                 set_bitmap_uptodate(bh);
482                 unlock_buffer(bh);
483                 goto verify;
484         }
485         /*
486          * submit the buffer_head for reading
487          */
488         set_buffer_new(bh);
489         trace_ext4_read_block_bitmap_load(sb, block_group);
490         bh->b_end_io = ext4_end_bitmap_read;
491         get_bh(bh);
492         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
493         return bh;
494 verify:
495         err = ext4_validate_block_bitmap(sb, desc, block_group, bh);
496         if (err)
497                 goto out;
498         return bh;
499 out:
500         put_bh(bh);
501         return ERR_PTR(err);
502 }
503
504 /* Returns 0 on success, 1 on error */
505 int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
506                            struct buffer_head *bh)
507 {
508         struct ext4_group_desc *desc;
509
510         if (!buffer_new(bh))
511                 return 0;
512         desc = ext4_get_group_desc(sb, block_group, NULL);
513         if (!desc)
514                 return -EFSCORRUPTED;
515         wait_on_buffer(bh);
516         if (!buffer_uptodate(bh)) {
517                 ext4_error(sb, "Cannot read block bitmap - "
518                            "block_group = %u, block_bitmap = %llu",
519                            block_group, (unsigned long long) bh->b_blocknr);
520                 ext4_mark_group_bitmap_corrupted(sb, block_group,
521                                         EXT4_GROUP_INFO_BBITMAP_CORRUPT);
522                 return -EIO;
523         }
524         clear_buffer_new(bh);
525         /* Panic or remount fs read-only if block bitmap is invalid */
526         return ext4_validate_block_bitmap(sb, desc, block_group, bh);
527 }
528
529 struct buffer_head *
530 ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
531 {
532         struct buffer_head *bh;
533         int err;
534
535         bh = ext4_read_block_bitmap_nowait(sb, block_group);
536         if (IS_ERR(bh))
537                 return bh;
538         err = ext4_wait_block_bitmap(sb, block_group, bh);
539         if (err) {
540                 put_bh(bh);
541                 return ERR_PTR(err);
542         }
543         return bh;
544 }
545
546 /**
547  * ext4_has_free_clusters()
548  * @sbi:        in-core super block structure.
549  * @nclusters:  number of needed blocks
550  * @flags:      flags from ext4_mb_new_blocks()
551  *
552  * Check if filesystem has nclusters free & available for allocation.
553  * On success return 1, return 0 on failure.
554  */
555 static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
556                                   s64 nclusters, unsigned int flags)
557 {
558         s64 free_clusters, dirty_clusters, rsv, resv_clusters;
559         struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
560         struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
561
562         free_clusters  = percpu_counter_read_positive(fcc);
563         dirty_clusters = percpu_counter_read_positive(dcc);
564         resv_clusters = atomic64_read(&sbi->s_resv_clusters);
565
566         /*
567          * r_blocks_count should always be multiple of the cluster ratio so
568          * we are safe to do a plane bit shift only.
569          */
570         rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) +
571               resv_clusters;
572
573         if (free_clusters - (nclusters + rsv + dirty_clusters) <
574                                         EXT4_FREECLUSTERS_WATERMARK) {
575                 free_clusters  = percpu_counter_sum_positive(fcc);
576                 dirty_clusters = percpu_counter_sum_positive(dcc);
577         }
578         /* Check whether we have space after accounting for current
579          * dirty clusters & root reserved clusters.
580          */
581         if (free_clusters >= (rsv + nclusters + dirty_clusters))
582                 return 1;
583
584         /* Hm, nope.  Are (enough) root reserved clusters available? */
585         if (uid_eq(sbi->s_resuid, current_fsuid()) ||
586             (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) ||
587             capable(CAP_SYS_RESOURCE) ||
588             (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
589
590                 if (free_clusters >= (nclusters + dirty_clusters +
591                                       resv_clusters))
592                         return 1;
593         }
594         /* No free blocks. Let's see if we can dip into reserved pool */
595         if (flags & EXT4_MB_USE_RESERVED) {
596                 if (free_clusters >= (nclusters + dirty_clusters))
597                         return 1;
598         }
599
600         return 0;
601 }
602
603 int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
604                              s64 nclusters, unsigned int flags)
605 {
606         if (ext4_has_free_clusters(sbi, nclusters, flags)) {
607                 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
608                 return 0;
609         } else
610                 return -ENOSPC;
611 }
612
613 /**
614  * ext4_should_retry_alloc()
615  * @sb:                 super block
616  * @retries             number of attemps has been made
617  *
618  * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
619  * it is profitable to retry the operation, this function will wait
620  * for the current or committing transaction to complete, and then
621  * return TRUE.  We will only retry once.
622  */
623 int ext4_should_retry_alloc(struct super_block *sb, int *retries)
624 {
625         if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
626             (*retries)++ > 1 ||
627             !EXT4_SB(sb)->s_journal)
628                 return 0;
629
630         smp_mb();
631         if (EXT4_SB(sb)->s_mb_free_pending == 0)
632                 return 0;
633
634         jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
635         jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
636         return 1;
637 }
638
639 /*
640  * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
641  *
642  * @handle:             handle to this transaction
643  * @inode:              file inode
644  * @goal:               given target block(filesystem wide)
645  * @count:              pointer to total number of clusters needed
646  * @errp:               error code
647  *
648  * Return 1st allocated block number on success, *count stores total account
649  * error stores in errp pointer
650  */
651 ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
652                                   ext4_fsblk_t goal, unsigned int flags,
653                                   unsigned long *count, int *errp)
654 {
655         struct ext4_allocation_request ar;
656         ext4_fsblk_t ret;
657
658         memset(&ar, 0, sizeof(ar));
659         /* Fill with neighbour allocated blocks */
660         ar.inode = inode;
661         ar.goal = goal;
662         ar.len = count ? *count : 1;
663         ar.flags = flags;
664
665         ret = ext4_mb_new_blocks(handle, &ar, errp);
666         if (count)
667                 *count = ar.len;
668         /*
669          * Account for the allocated meta blocks.  We will never
670          * fail EDQUOT for metdata, but we do account for it.
671          */
672         if (!(*errp) && (flags & EXT4_MB_DELALLOC_RESERVED)) {
673                 dquot_alloc_block_nofail(inode,
674                                 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
675         }
676         return ret;
677 }
678
679 /**
680  * ext4_count_free_clusters() -- count filesystem free clusters
681  * @sb:         superblock
682  *
683  * Adds up the number of free clusters from each block group.
684  */
685 ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
686 {
687         ext4_fsblk_t desc_count;
688         struct ext4_group_desc *gdp;
689         ext4_group_t i;
690         ext4_group_t ngroups = ext4_get_groups_count(sb);
691         struct ext4_group_info *grp;
692 #ifdef EXT4FS_DEBUG
693         struct ext4_super_block *es;
694         ext4_fsblk_t bitmap_count;
695         unsigned int x;
696         struct buffer_head *bitmap_bh = NULL;
697
698         es = EXT4_SB(sb)->s_es;
699         desc_count = 0;
700         bitmap_count = 0;
701         gdp = NULL;
702
703         for (i = 0; i < ngroups; i++) {
704                 gdp = ext4_get_group_desc(sb, i, NULL);
705                 if (!gdp)
706                         continue;
707                 grp = NULL;
708                 if (EXT4_SB(sb)->s_group_info)
709                         grp = ext4_get_group_info(sb, i);
710                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
711                         desc_count += ext4_free_group_clusters(sb, gdp);
712                 brelse(bitmap_bh);
713                 bitmap_bh = ext4_read_block_bitmap(sb, i);
714                 if (IS_ERR(bitmap_bh)) {
715                         bitmap_bh = NULL;
716                         continue;
717                 }
718
719                 x = ext4_count_free(bitmap_bh->b_data,
720                                     EXT4_CLUSTERS_PER_GROUP(sb) / 8);
721                 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
722                         i, ext4_free_group_clusters(sb, gdp), x);
723                 bitmap_count += x;
724         }
725         brelse(bitmap_bh);
726         printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
727                ", computed = %llu, %llu\n",
728                EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
729                desc_count, bitmap_count);
730         return bitmap_count;
731 #else
732         desc_count = 0;
733         for (i = 0; i < ngroups; i++) {
734                 gdp = ext4_get_group_desc(sb, i, NULL);
735                 if (!gdp)
736                         continue;
737                 grp = NULL;
738                 if (EXT4_SB(sb)->s_group_info)
739                         grp = ext4_get_group_info(sb, i);
740                 if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
741                         desc_count += ext4_free_group_clusters(sb, gdp);
742         }
743
744         return desc_count;
745 #endif
746 }
747
748 static inline int test_root(ext4_group_t a, int b)
749 {
750         while (1) {
751                 if (a < b)
752                         return 0;
753                 if (a == b)
754                         return 1;
755                 if ((a % b) != 0)
756                         return 0;
757                 a = a / b;
758         }
759 }
760
761 /**
762  *      ext4_bg_has_super - number of blocks used by the superblock in group
763  *      @sb: superblock for filesystem
764  *      @group: group number to check
765  *
766  *      Return the number of blocks used by the superblock (primary or backup)
767  *      in this group.  Currently this will be only 0 or 1.
768  */
769 int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
770 {
771         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
772
773         if (group == 0)
774                 return 1;
775         if (ext4_has_feature_sparse_super2(sb)) {
776                 if (group == le32_to_cpu(es->s_backup_bgs[0]) ||
777                     group == le32_to_cpu(es->s_backup_bgs[1]))
778                         return 1;
779                 return 0;
780         }
781         if ((group <= 1) || !ext4_has_feature_sparse_super(sb))
782                 return 1;
783         if (!(group & 1))
784                 return 0;
785         if (test_root(group, 3) || (test_root(group, 5)) ||
786             test_root(group, 7))
787                 return 1;
788
789         return 0;
790 }
791
792 static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
793                                         ext4_group_t group)
794 {
795         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
796         ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
797         ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
798
799         if (group == first || group == first + 1 || group == last)
800                 return 1;
801         return 0;
802 }
803
804 static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
805                                         ext4_group_t group)
806 {
807         if (!ext4_bg_has_super(sb, group))
808                 return 0;
809
810         if (ext4_has_feature_meta_bg(sb))
811                 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
812         else
813                 return EXT4_SB(sb)->s_gdb_count;
814 }
815
816 /**
817  *      ext4_bg_num_gdb - number of blocks used by the group table in group
818  *      @sb: superblock for filesystem
819  *      @group: group number to check
820  *
821  *      Return the number of blocks used by the group descriptor table
822  *      (primary or backup) in this group.  In the future there may be a
823  *      different number of descriptor blocks in each group.
824  */
825 unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
826 {
827         unsigned long first_meta_bg =
828                         le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
829         unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
830
831         if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg)
832                 return ext4_bg_num_gdb_nometa(sb, group);
833
834         return ext4_bg_num_gdb_meta(sb,group);
835
836 }
837
838 /*
839  * This function returns the number of file system metadata clusters at
840  * the beginning of a block group, including the reserved gdt blocks.
841  */
842 static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
843                                      ext4_group_t block_group)
844 {
845         struct ext4_sb_info *sbi = EXT4_SB(sb);
846         unsigned num;
847
848         /* Check for superblock and gdt backups in this group */
849         num = ext4_bg_has_super(sb, block_group);
850
851         if (!ext4_has_feature_meta_bg(sb) ||
852             block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
853                           sbi->s_desc_per_block) {
854                 if (num) {
855                         num += ext4_bg_num_gdb(sb, block_group);
856                         num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
857                 }
858         } else { /* For META_BG_BLOCK_GROUPS */
859                 num += ext4_bg_num_gdb(sb, block_group);
860         }
861         return EXT4_NUM_B2C(sbi, num);
862 }
863 /**
864  *      ext4_inode_to_goal_block - return a hint for block allocation
865  *      @inode: inode for block allocation
866  *
867  *      Return the ideal location to start allocating blocks for a
868  *      newly created inode.
869  */
870 ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
871 {
872         struct ext4_inode_info *ei = EXT4_I(inode);
873         ext4_group_t block_group;
874         ext4_grpblk_t colour;
875         int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
876         ext4_fsblk_t bg_start;
877         ext4_fsblk_t last_block;
878
879         block_group = ei->i_block_group;
880         if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
881                 /*
882                  * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
883                  * block groups per flexgroup, reserve the first block
884                  * group for directories and special files.  Regular
885                  * files will start at the second block group.  This
886                  * tends to speed up directory access and improves
887                  * fsck times.
888                  */
889                 block_group &= ~(flex_size-1);
890                 if (S_ISREG(inode->i_mode))
891                         block_group++;
892         }
893         bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
894         last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
895
896         /*
897          * If we are doing delayed allocation, we don't need take
898          * colour into account.
899          */
900         if (test_opt(inode->i_sb, DELALLOC))
901                 return bg_start;
902
903         if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
904                 colour = (current->pid % 16) *
905                         (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
906         else
907                 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
908         return bg_start + colour;
909 }
910