]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/ext4/balloc.c
ext4: fix race when setting bitmap_uptodate flag
[mirror_ubuntu-bionic-kernel.git] / fs / ext4 / balloc.c
CommitLineData
ac27a0ec 1/*
617ba13b 2 * linux/fs/ext4/balloc.c
ac27a0ec
DK
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>
dab291af 17#include <linux/jbd2.h>
ac27a0ec
DK
18#include <linux/quotaops.h>
19#include <linux/buffer_head.h>
3dcf5451
CH
20#include "ext4.h"
21#include "ext4_jbd2.h"
e21675d4 22#include "mballoc.h"
3dcf5451 23
0562e0ba
JZ
24#include <trace/events/ext4.h>
25
5f163cc7
ES
26static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
27 ext4_group_t block_group);
ac27a0ec
DK
28/*
29 * balloc.c contains the blocks allocation and deallocation routines
30 */
31
72b64b59 32/*
3212a80a
TT
33 * Calculate the block group number and offset into the block/cluster
34 * allocation bitmap, given a block number
72b64b59
AM
35 */
36void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
fd2d4291 37 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
72b64b59 38{
8c55e204 39 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
72b64b59
AM
40 ext4_grpblk_t offset;
41
8c55e204 42 blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
3212a80a
TT
43 offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >>
44 EXT4_SB(sb)->s_cluster_bits;
72b64b59
AM
45 if (offsetp)
46 *offsetp = offset;
47 if (blockgrpp)
8c55e204 48 *blockgrpp = blocknr;
72b64b59
AM
49
50}
51
0bf7e837
JS
52static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
53 ext4_group_t block_group)
54{
55 ext4_group_t actual_group;
7477827f 56 ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
0bf7e837
JS
57 if (actual_group == block_group)
58 return 1;
59 return 0;
60}
61
d5b8f310
TT
62/* Return the number of clusters used for file system metadata; this
63 * represents the overhead needed by the file system.
64 */
65unsigned ext4_num_overhead_clusters(struct super_block *sb,
66 ext4_group_t block_group,
67 struct ext4_group_desc *gdp)
0bf7e837 68{
d5b8f310
TT
69 unsigned num_clusters;
70 int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c;
71 ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group);
72 ext4_fsblk_t itbl_blk;
0bf7e837 73 struct ext4_sb_info *sbi = EXT4_SB(sb);
0bf7e837 74
d5b8f310
TT
75 /* This is the number of clusters used by the superblock,
76 * block group descriptors, and reserved block group
77 * descriptor blocks */
78 num_clusters = ext4_num_base_meta_clusters(sb, block_group);
79
80 /*
81 * For the allocation bitmaps and inode table, we first need
82 * to check to see if the block is in the block group. If it
83 * is, then check to see if the cluster is already accounted
84 * for in the clusters used for the base metadata cluster, or
85 * if we can increment the base metadata cluster to include
86 * that block. Otherwise, we will have to track the cluster
87 * used for the allocation bitmap or inode table explicitly.
88 * Normally all of these blocks are contiguous, so the special
89 * case handling shouldn't be necessary except for *very*
90 * unusual file system layouts.
91 */
92 if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) {
93 block_cluster = EXT4_B2C(sbi, (start -
94 ext4_block_bitmap(sb, gdp)));
95 if (block_cluster < num_clusters)
96 block_cluster = -1;
97 else if (block_cluster == num_clusters) {
98 num_clusters++;
99 block_cluster = -1;
100 }
101 }
102
103 if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) {
104 inode_cluster = EXT4_B2C(sbi,
105 start - ext4_inode_bitmap(sb, gdp));
106 if (inode_cluster < num_clusters)
107 inode_cluster = -1;
108 else if (inode_cluster == num_clusters) {
109 num_clusters++;
110 inode_cluster = -1;
111 }
112 }
113
114 itbl_blk = ext4_inode_table(sb, gdp);
115 for (i = 0; i < sbi->s_itb_per_group; i++) {
116 if (ext4_block_in_group(sb, itbl_blk + i, block_group)) {
117 c = EXT4_B2C(sbi, start - itbl_blk + i);
118 if ((c < num_clusters) || (c == inode_cluster) ||
119 (c == block_cluster) || (c == itbl_cluster))
120 continue;
121 if (c == num_clusters) {
122 num_clusters++;
123 continue;
124 }
125 num_clusters++;
126 itbl_cluster = c;
0bf7e837
JS
127 }
128 }
d5b8f310
TT
129
130 if (block_cluster != -1)
131 num_clusters++;
132 if (inode_cluster != -1)
133 num_clusters++;
134
135 return num_clusters;
0bf7e837 136}
c2ea3fde 137
d5b8f310
TT
138static unsigned int num_clusters_in_group(struct super_block *sb,
139 ext4_group_t block_group)
49f7f9af 140{
d5b8f310
TT
141 unsigned int blocks;
142
49f7f9af
TT
143 if (block_group == ext4_get_groups_count(sb) - 1) {
144 /*
145 * Even though mke2fs always initializes the first and
146 * last group, just in case some other tool was used,
147 * we need to make sure we calculate the right free
148 * blocks.
149 */
d5b8f310 150 blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) -
49f7f9af
TT
151 ext4_group_first_block_no(sb, block_group);
152 } else
d5b8f310
TT
153 blocks = EXT4_BLOCKS_PER_GROUP(sb);
154 return EXT4_NUM_B2C(EXT4_SB(sb), blocks);
49f7f9af
TT
155}
156
fd034a84
TT
157/* Initializes an uninitialized block bitmap */
158void ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
159 ext4_group_t block_group,
160 struct ext4_group_desc *gdp)
717d50e4 161{
d5b8f310 162 unsigned int bit, bit_max;
717d50e4 163 struct ext4_sb_info *sbi = EXT4_SB(sb);
fd034a84
TT
164 ext4_fsblk_t start, tmp;
165 int flex_bg = 0;
166
167 J_ASSERT_BH(bh, buffer_locked(bh));
168
169 /* If checksum is bad mark all blocks used to prevent allocation
170 * essentially implementing a per-group read-only flag. */
171 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
172 ext4_error(sb, "Checksum bad for group %u", block_group);
021b65bb 173 ext4_free_group_clusters_set(sb, gdp, 0);
fd034a84
TT
174 ext4_free_inodes_set(sb, gdp, 0);
175 ext4_itable_unused_set(sb, gdp, 0);
176 memset(bh->b_data, 0xff, sb->s_blocksize);
177 return;
717d50e4 178 }
fd034a84 179 memset(bh->b_data, 0, sb->s_blocksize);
717d50e4 180
d5b8f310 181 bit_max = ext4_num_base_meta_clusters(sb, block_group);
fd034a84
TT
182 for (bit = 0; bit < bit_max; bit++)
183 ext4_set_bit(bit, bh->b_data);
d00a6d7b 184
fd034a84 185 start = ext4_group_first_block_no(sb, block_group);
717d50e4 186
fd034a84
TT
187 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
188 flex_bg = 1;
717d50e4 189
fd034a84
TT
190 /* Set bits for block and inode bitmaps, and inode table */
191 tmp = ext4_block_bitmap(sb, gdp);
192 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
d5b8f310 193 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
717d50e4 194
fd034a84
TT
195 tmp = ext4_inode_bitmap(sb, gdp);
196 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
d5b8f310 197 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
0bf7e837 198
fd034a84
TT
199 tmp = ext4_inode_table(sb, gdp);
200 for (; tmp < ext4_inode_table(sb, gdp) +
201 sbi->s_itb_per_group; tmp++) {
0bf7e837 202 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
d5b8f310 203 ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data);
717d50e4 204 }
d5b8f310 205
fd034a84
TT
206 /*
207 * Also if the number of blocks within the group is less than
208 * the blocksize * 8 ( which is the size of bitmap ), set rest
209 * of the block bitmap to 1
210 */
d5b8f310 211 ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group),
fd034a84 212 sb->s_blocksize * 8, bh->b_data);
717d50e4
AD
213}
214
fd034a84
TT
215/* Return the number of free blocks in a block group. It is used when
216 * the block bitmap is uninitialized, so we can't just count the bits
217 * in the bitmap. */
cff1dfd7
TT
218unsigned ext4_free_clusters_after_init(struct super_block *sb,
219 ext4_group_t block_group,
220 struct ext4_group_desc *gdp)
fd034a84 221{
d5b8f310
TT
222 return num_clusters_in_group(sb, block_group) -
223 ext4_num_overhead_clusters(sb, block_group, gdp);
fd034a84 224}
717d50e4 225
ac27a0ec
DK
226/*
227 * The free blocks are managed by bitmaps. A file system contains several
228 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
229 * block for inodes, N blocks for the inode table and data blocks.
230 *
231 * The file system contains group descriptors which are located after the
232 * super block. Each descriptor contains the number of the bitmap block and
233 * the free blocks count in the block. The descriptors are loaded in memory
e627432c 234 * when a file system is mounted (see ext4_fill_super).
ac27a0ec
DK
235 */
236
ac27a0ec 237/**
617ba13b 238 * ext4_get_group_desc() -- load group descriptor from disk
ac27a0ec
DK
239 * @sb: super block
240 * @block_group: given block group
241 * @bh: pointer to the buffer head to store the block
242 * group descriptor
243 */
af5bc92d 244struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
fd2d4291 245 ext4_group_t block_group,
af5bc92d 246 struct buffer_head **bh)
ac27a0ec 247{
498e5f24
TT
248 unsigned int group_desc;
249 unsigned int offset;
8df9675f 250 ext4_group_t ngroups = ext4_get_groups_count(sb);
af5bc92d 251 struct ext4_group_desc *desc;
617ba13b 252 struct ext4_sb_info *sbi = EXT4_SB(sb);
ac27a0ec 253
8df9675f 254 if (block_group >= ngroups) {
12062ddd
ES
255 ext4_error(sb, "block_group >= groups_count - block_group = %u,"
256 " groups_count = %u", block_group, ngroups);
ac27a0ec
DK
257
258 return NULL;
259 }
ac27a0ec 260
617ba13b
MC
261 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
262 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
ac27a0ec 263 if (!sbi->s_group_desc[group_desc]) {
12062ddd 264 ext4_error(sb, "Group descriptor not loaded - "
498e5f24 265 "block_group = %u, group_desc = %u, desc = %u",
af5bc92d 266 block_group, group_desc, offset);
ac27a0ec
DK
267 return NULL;
268 }
269
0d1ee42f
AR
270 desc = (struct ext4_group_desc *)(
271 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
272 offset * EXT4_DESC_SIZE(sb));
ac27a0ec
DK
273 if (bh)
274 *bh = sbi->s_group_desc[group_desc];
0d1ee42f 275 return desc;
ac27a0ec
DK
276}
277
abcb2947
AK
278static int ext4_valid_block_bitmap(struct super_block *sb,
279 struct ext4_group_desc *desc,
280 unsigned int block_group,
281 struct buffer_head *bh)
282{
283 ext4_grpblk_t offset;
284 ext4_grpblk_t next_zero_bit;
285 ext4_fsblk_t bitmap_blk;
286 ext4_fsblk_t group_first_block;
287
288 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
289 /* with FLEX_BG, the inode/block bitmaps and itable
290 * blocks may not be in the group at all
291 * so the bitmap validation will be skipped for those groups
292 * or it has to also read the block group where the bitmaps
293 * are located to verify they are set.
294 */
295 return 1;
296 }
297 group_first_block = ext4_group_first_block_no(sb, block_group);
298
299 /* check whether block bitmap block number is set */
300 bitmap_blk = ext4_block_bitmap(sb, desc);
301 offset = bitmap_blk - group_first_block;
302 if (!ext4_test_bit(offset, bh->b_data))
303 /* bad block bitmap */
304 goto err_out;
305
306 /* check whether the inode bitmap block number is set */
307 bitmap_blk = ext4_inode_bitmap(sb, desc);
308 offset = bitmap_blk - group_first_block;
309 if (!ext4_test_bit(offset, bh->b_data))
310 /* bad block bitmap */
311 goto err_out;
312
313 /* check whether the inode table block number is set */
314 bitmap_blk = ext4_inode_table(sb, desc);
315 offset = bitmap_blk - group_first_block;
316 next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
317 offset + EXT4_SB(sb)->s_itb_per_group,
318 offset);
319 if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
320 /* good bitmap for inode tables */
321 return 1;
322
323err_out:
12062ddd 324 ext4_error(sb, "Invalid block bitmap - block_group = %d, block = %llu",
abcb2947
AK
325 block_group, bitmap_blk);
326 return 0;
327}
ac27a0ec 328/**
574ca174 329 * ext4_read_block_bitmap()
ac27a0ec
DK
330 * @sb: super block
331 * @block_group: given block group
332 *
abcb2947
AK
333 * Read the bitmap for a given block_group,and validate the
334 * bits for block/inode/inode tables are set in the bitmaps
ac27a0ec
DK
335 *
336 * Return buffer_head on success or NULL in case of failure.
337 */
717d50e4 338struct buffer_head *
813e5727 339ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group)
ac27a0ec 340{
af5bc92d 341 struct ext4_group_desc *desc;
813e5727 342 struct buffer_head *bh;
7c9e69fa 343 ext4_fsblk_t bitmap_blk;
ac27a0ec 344
717d50e4 345 desc = ext4_get_group_desc(sb, block_group, NULL);
ac27a0ec 346 if (!desc)
7c9e69fa
AK
347 return NULL;
348 bitmap_blk = ext4_block_bitmap(sb, desc);
abcb2947
AK
349 bh = sb_getblk(sb, bitmap_blk);
350 if (unlikely(!bh)) {
813e5727
TT
351 ext4_error(sb, "Cannot get buffer for block bitmap - "
352 "block_group = %u, block_bitmap = %llu",
353 block_group, bitmap_blk);
abcb2947
AK
354 return NULL;
355 }
2ccb5fb9
AK
356
357 if (bitmap_uptodate(bh))
abcb2947
AK
358 return bh;
359
c806e68f 360 lock_buffer(bh);
2ccb5fb9
AK
361 if (bitmap_uptodate(bh)) {
362 unlock_buffer(bh);
363 return bh;
364 }
955ce5f5 365 ext4_lock_group(sb, block_group);
717d50e4 366 if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
abcb2947 367 ext4_init_block_bitmap(sb, bh, block_group, desc);
2ccb5fb9 368 set_bitmap_uptodate(bh);
abcb2947 369 set_buffer_uptodate(bh);
955ce5f5 370 ext4_unlock_group(sb, block_group);
3300beda 371 unlock_buffer(bh);
abcb2947 372 return bh;
717d50e4 373 }
955ce5f5 374 ext4_unlock_group(sb, block_group);
2ccb5fb9
AK
375 if (buffer_uptodate(bh)) {
376 /*
377 * if not uninit if bh is uptodate,
378 * bitmap is also uptodate
379 */
380 set_bitmap_uptodate(bh);
381 unlock_buffer(bh);
382 return bh;
383 }
384 /*
813e5727 385 * submit the buffer_head for reading
2ccb5fb9 386 */
813e5727 387 set_buffer_new(bh);
0562e0ba 388 trace_ext4_read_block_bitmap_load(sb, block_group);
813e5727
TT
389 bh->b_end_io = ext4_end_bitmap_read;
390 get_bh(bh);
391 submit_bh(READ, bh);
392 return bh;
393}
394
395/* Returns 0 on success, 1 on error */
396int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group,
397 struct buffer_head *bh)
398{
399 struct ext4_group_desc *desc;
400
401 if (!buffer_new(bh))
402 return 0;
403 desc = ext4_get_group_desc(sb, block_group, NULL);
404 if (!desc)
405 return 1;
406 wait_on_buffer(bh);
407 if (!buffer_uptodate(bh)) {
12062ddd 408 ext4_error(sb, "Cannot read block bitmap - "
813e5727
TT
409 "block_group = %u, block_bitmap = %llu",
410 block_group, bh->b_blocknr);
411 return 1;
abcb2947 412 }
813e5727
TT
413 clear_buffer_new(bh);
414 /* Panic or remount fs read-only if block bitmap is invalid */
519deca0 415 ext4_valid_block_bitmap(sb, desc, block_group, bh);
813e5727
TT
416 return 0;
417}
418
419struct buffer_head *
420ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
421{
422 struct buffer_head *bh;
423
424 bh = ext4_read_block_bitmap_nowait(sb, block_group);
425 if (ext4_wait_block_bitmap(sb, block_group, bh)) {
426 put_bh(bh);
427 return NULL;
428 }
ac27a0ec
DK
429 return bh;
430}
ac27a0ec 431
8c3bf8a0 432/**
df55c99d 433 * ext4_has_free_clusters()
8c3bf8a0 434 * @sbi: in-core super block structure.
df55c99d
TT
435 * @nclusters: number of needed blocks
436 * @flags: flags from ext4_mb_new_blocks()
8c3bf8a0 437 *
df55c99d 438 * Check if filesystem has nclusters free & available for allocation.
8c3bf8a0
ES
439 * On success return 1, return 0 on failure.
440 */
df55c99d
TT
441static int ext4_has_free_clusters(struct ext4_sb_info *sbi,
442 s64 nclusters, unsigned int flags)
a30d542a 443{
df55c99d 444 s64 free_clusters, dirty_clusters, root_clusters;
57042651 445 struct percpu_counter *fcc = &sbi->s_freeclusters_counter;
df55c99d 446 struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter;
a30d542a 447
df55c99d
TT
448 free_clusters = percpu_counter_read_positive(fcc);
449 dirty_clusters = percpu_counter_read_positive(dcc);
450 root_clusters = EXT4_B2C(sbi, ext4_r_blocks_count(sbi->s_es));
a30d542a 451
df55c99d
TT
452 if (free_clusters - (nclusters + root_clusters + dirty_clusters) <
453 EXT4_FREECLUSTERS_WATERMARK) {
454 free_clusters = EXT4_C2B(sbi, percpu_counter_sum_positive(fcc));
455 dirty_clusters = percpu_counter_sum_positive(dcc);
6bc6e63f 456 }
df55c99d
TT
457 /* Check whether we have space after accounting for current
458 * dirty clusters & root reserved clusters.
6bc6e63f 459 */
df55c99d 460 if (free_clusters >= ((root_clusters + nclusters) + dirty_clusters))
a996031c 461 return 1;
a30d542a 462
df55c99d 463 /* Hm, nope. Are (enough) root reserved clusters available? */
4c9c544e 464 if (sbi->s_resuid == current_fsuid() ||
a996031c 465 ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
55f020db
AH
466 capable(CAP_SYS_RESOURCE) ||
467 (flags & EXT4_MB_USE_ROOT_BLOCKS)) {
468
df55c99d 469 if (free_clusters >= (nclusters + dirty_clusters))
a996031c
ES
470 return 1;
471 }
472
473 return 0;
a30d542a
AK
474}
475
e7d5f315
TT
476int ext4_claim_free_clusters(struct ext4_sb_info *sbi,
477 s64 nclusters, unsigned int flags)
ac27a0ec 478{
df55c99d 479 if (ext4_has_free_clusters(sbi, nclusters, flags)) {
e7d5f315 480 percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters);
16eb7295 481 return 0;
8c3bf8a0
ES
482 } else
483 return -ENOSPC;
a30d542a 484}
07031431 485
ac27a0ec 486/**
617ba13b 487 * ext4_should_retry_alloc()
ac27a0ec
DK
488 * @sb: super block
489 * @retries number of attemps has been made
490 *
617ba13b 491 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
ac27a0ec 492 * it is profitable to retry the operation, this function will wait
25985edc 493 * for the current or committing transaction to complete, and then
ac27a0ec
DK
494 * return TRUE.
495 *
496 * if the total number of retries exceed three times, return FALSE.
497 */
617ba13b 498int ext4_should_retry_alloc(struct super_block *sb, int *retries)
ac27a0ec 499{
df55c99d 500 if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) ||
8f64b32e
ES
501 (*retries)++ > 3 ||
502 !EXT4_SB(sb)->s_journal)
ac27a0ec
DK
503 return 0;
504
505 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
506
dab291af 507 return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
ac27a0ec
DK
508}
509
654b4908 510/*
d2a17637 511 * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
654b4908
AK
512 *
513 * @handle: handle to this transaction
514 * @inode: file inode
515 * @goal: given target block(filesystem wide)
7b415bf6 516 * @count: pointer to total number of clusters needed
654b4908
AK
517 * @errp: error code
518 *
97df5d15 519 * Return 1st allocated block number on success, *count stores total account
d2a17637 520 * error stores in errp pointer
654b4908 521 */
d2a17637 522ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
55f020db
AH
523 ext4_fsblk_t goal, unsigned int flags,
524 unsigned long *count, int *errp)
654b4908 525{
97df5d15 526 struct ext4_allocation_request ar;
d2a17637 527 ext4_fsblk_t ret;
97df5d15
TT
528
529 memset(&ar, 0, sizeof(ar));
530 /* Fill with neighbour allocated blocks */
531 ar.inode = inode;
532 ar.goal = goal;
533 ar.len = count ? *count : 1;
55f020db 534 ar.flags = flags;
97df5d15
TT
535
536 ret = ext4_mb_new_blocks(handle, &ar, errp);
537 if (count)
538 *count = ar.len;
d2a17637 539 /*
72b8ab9d
ES
540 * Account for the allocated meta blocks. We will never
541 * fail EDQUOT for metdata, but we do account for it.
d2a17637 542 */
f2321097
TT
543 if (!(*errp) &&
544 ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) {
d2a17637 545 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
97df5d15 546 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
d2a17637 547 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
7b415bf6
AK
548 dquot_alloc_block_nofail(inode,
549 EXT4_C2B(EXT4_SB(inode->i_sb), ar.len));
d2a17637
MC
550 }
551 return ret;
654b4908
AK
552}
553
ac27a0ec 554/**
5dee5437 555 * ext4_count_free_clusters() -- count filesystem free clusters
ac27a0ec
DK
556 * @sb: superblock
557 *
5dee5437 558 * Adds up the number of free clusters from each block group.
ac27a0ec 559 */
5dee5437 560ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb)
ac27a0ec 561{
617ba13b
MC
562 ext4_fsblk_t desc_count;
563 struct ext4_group_desc *gdp;
fd2d4291 564 ext4_group_t i;
8df9675f 565 ext4_group_t ngroups = ext4_get_groups_count(sb);
617ba13b
MC
566#ifdef EXT4FS_DEBUG
567 struct ext4_super_block *es;
568 ext4_fsblk_t bitmap_count;
498e5f24 569 unsigned int x;
ac27a0ec
DK
570 struct buffer_head *bitmap_bh = NULL;
571
617ba13b 572 es = EXT4_SB(sb)->s_es;
ac27a0ec
DK
573 desc_count = 0;
574 bitmap_count = 0;
575 gdp = NULL;
576
ac27a0ec 577 for (i = 0; i < ngroups; i++) {
617ba13b 578 gdp = ext4_get_group_desc(sb, i, NULL);
ac27a0ec
DK
579 if (!gdp)
580 continue;
021b65bb 581 desc_count += ext4_free_group_clusters(sb, gdp);
ac27a0ec 582 brelse(bitmap_bh);
574ca174 583 bitmap_bh = ext4_read_block_bitmap(sb, i);
ac27a0ec
DK
584 if (bitmap_bh == NULL)
585 continue;
586
617ba13b 587 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
9fd9784c 588 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
021b65bb 589 i, ext4_free_group_clusters(sb, gdp), x);
ac27a0ec
DK
590 bitmap_count += x;
591 }
592 brelse(bitmap_bh);
5dee5437
TT
593 printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu"
594 ", computed = %llu, %llu\n",
6e58ad69 595 EXT4_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)),
4776004f 596 desc_count, bitmap_count);
ac27a0ec
DK
597 return bitmap_count;
598#else
599 desc_count = 0;
ac27a0ec 600 for (i = 0; i < ngroups; i++) {
617ba13b 601 gdp = ext4_get_group_desc(sb, i, NULL);
ac27a0ec
DK
602 if (!gdp)
603 continue;
021b65bb 604 desc_count += ext4_free_group_clusters(sb, gdp);
ac27a0ec
DK
605 }
606
607 return desc_count;
608#endif
609}
610
fd2d4291 611static inline int test_root(ext4_group_t a, int b)
ac27a0ec
DK
612{
613 int num = b;
614
615 while (a > num)
616 num *= b;
617 return num == a;
618}
619
fd2d4291 620static int ext4_group_sparse(ext4_group_t group)
ac27a0ec
DK
621{
622 if (group <= 1)
623 return 1;
624 if (!(group & 1))
625 return 0;
626 return (test_root(group, 7) || test_root(group, 5) ||
627 test_root(group, 3));
628}
629
630/**
617ba13b 631 * ext4_bg_has_super - number of blocks used by the superblock in group
ac27a0ec
DK
632 * @sb: superblock for filesystem
633 * @group: group number to check
634 *
635 * Return the number of blocks used by the superblock (primary or backup)
636 * in this group. Currently this will be only 0 or 1.
637 */
fd2d4291 638int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
ac27a0ec 639{
617ba13b
MC
640 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
641 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
642 !ext4_group_sparse(group))
ac27a0ec
DK
643 return 0;
644 return 1;
645}
646
fd2d4291
AM
647static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
648 ext4_group_t group)
ac27a0ec 649{
617ba13b 650 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
fd2d4291
AM
651 ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
652 ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
ac27a0ec
DK
653
654 if (group == first || group == first + 1 || group == last)
655 return 1;
656 return 0;
657}
658
fd2d4291
AM
659static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
660 ext4_group_t group)
ac27a0ec 661{
8dadb198
TT
662 if (!ext4_bg_has_super(sb, group))
663 return 0;
664
665 if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
666 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
667 else
668 return EXT4_SB(sb)->s_gdb_count;
ac27a0ec
DK
669}
670
671/**
617ba13b 672 * ext4_bg_num_gdb - number of blocks used by the group table in group
ac27a0ec
DK
673 * @sb: superblock for filesystem
674 * @group: group number to check
675 *
676 * Return the number of blocks used by the group descriptor table
677 * (primary or backup) in this group. In the future there may be a
678 * different number of descriptor blocks in each group.
679 */
fd2d4291 680unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
ac27a0ec
DK
681{
682 unsigned long first_meta_bg =
617ba13b
MC
683 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
684 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
ac27a0ec 685
617ba13b 686 if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
ac27a0ec 687 metagroup < first_meta_bg)
af5bc92d 688 return ext4_bg_num_gdb_nometa(sb, group);
ac27a0ec 689
617ba13b 690 return ext4_bg_num_gdb_meta(sb,group);
ac27a0ec
DK
691
692}
c2ea3fde 693
49f7f9af 694/*
d5b8f310 695 * This function returns the number of file system metadata clusters at
49f7f9af
TT
696 * the beginning of a block group, including the reserved gdt blocks.
697 */
5f163cc7 698static unsigned ext4_num_base_meta_clusters(struct super_block *sb,
d5b8f310 699 ext4_group_t block_group)
49f7f9af
TT
700{
701 struct ext4_sb_info *sbi = EXT4_SB(sb);
d5b8f310 702 unsigned num;
49f7f9af
TT
703
704 /* Check for superblock and gdt backups in this group */
705 num = ext4_bg_has_super(sb, block_group);
706
707 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
708 block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
709 sbi->s_desc_per_block) {
710 if (num) {
711 num += ext4_bg_num_gdb(sb, block_group);
712 num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
713 }
714 } else { /* For META_BG_BLOCK_GROUPS */
715 num += ext4_bg_num_gdb(sb, block_group);
716 }
d5b8f310 717 return EXT4_NUM_B2C(sbi, num);
49f7f9af 718}
f86186b4
ES
719/**
720 * ext4_inode_to_goal_block - return a hint for block allocation
721 * @inode: inode for block allocation
722 *
723 * Return the ideal location to start allocating blocks for a
724 * newly created inode.
725 */
726ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode)
727{
728 struct ext4_inode_info *ei = EXT4_I(inode);
729 ext4_group_t block_group;
730 ext4_grpblk_t colour;
731 int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb));
732 ext4_fsblk_t bg_start;
733 ext4_fsblk_t last_block;
734
735 block_group = ei->i_block_group;
736 if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) {
737 /*
738 * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME
739 * block groups per flexgroup, reserve the first block
740 * group for directories and special files. Regular
741 * files will start at the second block group. This
742 * tends to speed up directory access and improves
743 * fsck times.
744 */
745 block_group &= ~(flex_size-1);
746 if (S_ISREG(inode->i_mode))
747 block_group++;
748 }
749 bg_start = ext4_group_first_block_no(inode->i_sb, block_group);
750 last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1;
751
752 /*
753 * If we are doing delayed allocation, we don't need take
754 * colour into account.
755 */
756 if (test_opt(inode->i_sb, DELALLOC))
757 return bg_start;
758
759 if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block)
760 colour = (current->pid % 16) *
761 (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16);
762 else
763 colour = (current->pid % 16) * ((last_block - bg_start) / 16);
764 return bg_start + colour;
765}
766