]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/ext4/balloc.c
ext4: Fix race between read_block_bitmap() and mark_diskspace_used()
[net-next-2.6.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"
717d50e4 22#include "group.h"
e21675d4 23#include "mballoc.h"
3dcf5451 24
ac27a0ec
DK
25/*
26 * balloc.c contains the blocks allocation and deallocation routines
27 */
28
72b64b59
AM
29/*
30 * Calculate the block group number and offset, given a block number
31 */
32void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
fd2d4291 33 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
72b64b59 34{
8c55e204 35 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
72b64b59
AM
36 ext4_grpblk_t offset;
37
8c55e204 38 blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
f4e5bc24 39 offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
72b64b59
AM
40 if (offsetp)
41 *offsetp = offset;
42 if (blockgrpp)
8c55e204 43 *blockgrpp = blocknr;
72b64b59
AM
44
45}
46
0bf7e837
JS
47static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
48 ext4_group_t block_group)
49{
50 ext4_group_t actual_group;
7477827f 51 ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
0bf7e837
JS
52 if (actual_group == block_group)
53 return 1;
54 return 0;
55}
56
57static int ext4_group_used_meta_blocks(struct super_block *sb,
58 ext4_group_t block_group)
59{
60 ext4_fsblk_t tmp;
61 struct ext4_sb_info *sbi = EXT4_SB(sb);
62 /* block bitmap, inode bitmap, and inode table blocks */
63 int used_blocks = sbi->s_itb_per_group + 2;
64
65 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
66 struct ext4_group_desc *gdp;
67 struct buffer_head *bh;
68
69 gdp = ext4_get_group_desc(sb, block_group, &bh);
70 if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
71 block_group))
72 used_blocks--;
73
74 if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
75 block_group))
76 used_blocks--;
77
78 tmp = ext4_inode_table(sb, gdp);
79 for (; tmp < ext4_inode_table(sb, gdp) +
80 sbi->s_itb_per_group; tmp++) {
81 if (!ext4_block_in_group(sb, tmp, block_group))
82 used_blocks -= 1;
83 }
84 }
85 return used_blocks;
86}
c2ea3fde 87
717d50e4
AD
88/* Initializes an uninitialized block bitmap if given, and returns the
89 * number of blocks free in the group. */
90unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
fd2d4291 91 ext4_group_t block_group, struct ext4_group_desc *gdp)
717d50e4 92{
717d50e4
AD
93 int bit, bit_max;
94 unsigned free_blocks, group_blocks;
95 struct ext4_sb_info *sbi = EXT4_SB(sb);
96
97 if (bh) {
98 J_ASSERT_BH(bh, buffer_locked(bh));
99
100 /* If checksum is bad mark all blocks used to prevent allocation
101 * essentially implementing a per-group read-only flag. */
102 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
46e665e9 103 ext4_error(sb, __func__,
a9df9a49 104 "Checksum bad for group %u", block_group);
717d50e4
AD
105 gdp->bg_free_blocks_count = 0;
106 gdp->bg_free_inodes_count = 0;
107 gdp->bg_itable_unused = 0;
108 memset(bh->b_data, 0xff, sb->s_blocksize);
109 return 0;
110 }
111 memset(bh->b_data, 0, sb->s_blocksize);
112 }
113
114 /* Check for superblock and gdt backups in this group */
115 bit_max = ext4_bg_has_super(sb, block_group);
116
117 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
118 block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
119 sbi->s_desc_per_block) {
120 if (bit_max) {
121 bit_max += ext4_bg_num_gdb(sb, block_group);
122 bit_max +=
123 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
124 }
125 } else { /* For META_BG_BLOCK_GROUPS */
6afd6707 126 bit_max += ext4_bg_num_gdb(sb, block_group);
717d50e4
AD
127 }
128
129 if (block_group == sbi->s_groups_count - 1) {
130 /*
131 * Even though mke2fs always initialize first and last group
132 * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
133 * to make sure we calculate the right free blocks
134 */
135 group_blocks = ext4_blocks_count(sbi->s_es) -
136 le32_to_cpu(sbi->s_es->s_first_data_block) -
af5bc92d 137 (EXT4_BLOCKS_PER_GROUP(sb) * (sbi->s_groups_count - 1));
717d50e4
AD
138 } else {
139 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
140 }
141
142 free_blocks = group_blocks - bit_max;
143
144 if (bh) {
0bf7e837
JS
145 ext4_fsblk_t start, tmp;
146 int flex_bg = 0;
d00a6d7b 147
717d50e4
AD
148 for (bit = 0; bit < bit_max; bit++)
149 ext4_set_bit(bit, bh->b_data);
150
d00a6d7b 151 start = ext4_group_first_block_no(sb, block_group);
717d50e4 152
0bf7e837
JS
153 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
154 EXT4_FEATURE_INCOMPAT_FLEX_BG))
155 flex_bg = 1;
717d50e4 156
0bf7e837
JS
157 /* Set bits for block and inode bitmaps, and inode table */
158 tmp = ext4_block_bitmap(sb, gdp);
159 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
160 ext4_set_bit(tmp - start, bh->b_data);
161
162 tmp = ext4_inode_bitmap(sb, gdp);
163 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
164 ext4_set_bit(tmp - start, bh->b_data);
165
166 tmp = ext4_inode_table(sb, gdp);
167 for (; tmp < ext4_inode_table(sb, gdp) +
168 sbi->s_itb_per_group; tmp++) {
169 if (!flex_bg ||
170 ext4_block_in_group(sb, tmp, block_group))
171 ext4_set_bit(tmp - start, bh->b_data);
172 }
717d50e4
AD
173 /*
174 * Also if the number of blocks within the group is
175 * less than the blocksize * 8 ( which is the size
176 * of bitmap ), set rest of the block bitmap to 1
177 */
178 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
179 }
0bf7e837 180 return free_blocks - ext4_group_used_meta_blocks(sb, block_group);
717d50e4
AD
181}
182
183
ac27a0ec
DK
184/*
185 * The free blocks are managed by bitmaps. A file system contains several
186 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
187 * block for inodes, N blocks for the inode table and data blocks.
188 *
189 * The file system contains group descriptors which are located after the
190 * super block. Each descriptor contains the number of the bitmap block and
191 * the free blocks count in the block. The descriptors are loaded in memory
e627432c 192 * when a file system is mounted (see ext4_fill_super).
ac27a0ec
DK
193 */
194
195
196#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
197
198/**
617ba13b 199 * ext4_get_group_desc() -- load group descriptor from disk
ac27a0ec
DK
200 * @sb: super block
201 * @block_group: given block group
202 * @bh: pointer to the buffer head to store the block
203 * group descriptor
204 */
af5bc92d 205struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
fd2d4291 206 ext4_group_t block_group,
af5bc92d 207 struct buffer_head **bh)
ac27a0ec 208{
498e5f24
TT
209 unsigned int group_desc;
210 unsigned int offset;
af5bc92d 211 struct ext4_group_desc *desc;
617ba13b 212 struct ext4_sb_info *sbi = EXT4_SB(sb);
ac27a0ec
DK
213
214 if (block_group >= sbi->s_groups_count) {
af5bc92d
TT
215 ext4_error(sb, "ext4_get_group_desc",
216 "block_group >= groups_count - "
a9df9a49 217 "block_group = %u, groups_count = %u",
af5bc92d 218 block_group, sbi->s_groups_count);
ac27a0ec
DK
219
220 return NULL;
221 }
222 smp_rmb();
223
617ba13b
MC
224 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
225 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
ac27a0ec 226 if (!sbi->s_group_desc[group_desc]) {
af5bc92d
TT
227 ext4_error(sb, "ext4_get_group_desc",
228 "Group descriptor not loaded - "
498e5f24 229 "block_group = %u, group_desc = %u, desc = %u",
af5bc92d 230 block_group, group_desc, offset);
ac27a0ec
DK
231 return NULL;
232 }
233
0d1ee42f
AR
234 desc = (struct ext4_group_desc *)(
235 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
236 offset * EXT4_DESC_SIZE(sb));
ac27a0ec
DK
237 if (bh)
238 *bh = sbi->s_group_desc[group_desc];
0d1ee42f 239 return desc;
ac27a0ec
DK
240}
241
abcb2947
AK
242static int ext4_valid_block_bitmap(struct super_block *sb,
243 struct ext4_group_desc *desc,
244 unsigned int block_group,
245 struct buffer_head *bh)
246{
247 ext4_grpblk_t offset;
248 ext4_grpblk_t next_zero_bit;
249 ext4_fsblk_t bitmap_blk;
250 ext4_fsblk_t group_first_block;
251
252 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
253 /* with FLEX_BG, the inode/block bitmaps and itable
254 * blocks may not be in the group at all
255 * so the bitmap validation will be skipped for those groups
256 * or it has to also read the block group where the bitmaps
257 * are located to verify they are set.
258 */
259 return 1;
260 }
261 group_first_block = ext4_group_first_block_no(sb, block_group);
262
263 /* check whether block bitmap block number is set */
264 bitmap_blk = ext4_block_bitmap(sb, desc);
265 offset = bitmap_blk - group_first_block;
266 if (!ext4_test_bit(offset, bh->b_data))
267 /* bad block bitmap */
268 goto err_out;
269
270 /* check whether the inode bitmap block number is set */
271 bitmap_blk = ext4_inode_bitmap(sb, desc);
272 offset = bitmap_blk - group_first_block;
273 if (!ext4_test_bit(offset, bh->b_data))
274 /* bad block bitmap */
275 goto err_out;
276
277 /* check whether the inode table block number is set */
278 bitmap_blk = ext4_inode_table(sb, desc);
279 offset = bitmap_blk - group_first_block;
280 next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
281 offset + EXT4_SB(sb)->s_itb_per_group,
282 offset);
283 if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
284 /* good bitmap for inode tables */
285 return 1;
286
287err_out:
46e665e9 288 ext4_error(sb, __func__,
abcb2947
AK
289 "Invalid block bitmap - "
290 "block_group = %d, block = %llu",
291 block_group, bitmap_blk);
292 return 0;
293}
ac27a0ec 294/**
574ca174 295 * ext4_read_block_bitmap()
ac27a0ec
DK
296 * @sb: super block
297 * @block_group: given block group
298 *
abcb2947
AK
299 * Read the bitmap for a given block_group,and validate the
300 * bits for block/inode/inode tables are set in the bitmaps
ac27a0ec
DK
301 *
302 * Return buffer_head on success or NULL in case of failure.
303 */
717d50e4 304struct buffer_head *
574ca174 305ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
ac27a0ec 306{
af5bc92d
TT
307 struct ext4_group_desc *desc;
308 struct buffer_head *bh = NULL;
7c9e69fa 309 ext4_fsblk_t bitmap_blk;
ac27a0ec 310
717d50e4 311 desc = ext4_get_group_desc(sb, block_group, NULL);
ac27a0ec 312 if (!desc)
7c9e69fa
AK
313 return NULL;
314 bitmap_blk = ext4_block_bitmap(sb, desc);
abcb2947
AK
315 bh = sb_getblk(sb, bitmap_blk);
316 if (unlikely(!bh)) {
46e665e9 317 ext4_error(sb, __func__,
abcb2947 318 "Cannot read block bitmap - "
a9df9a49 319 "block_group = %u, block_bitmap = %llu",
e29d1cde 320 block_group, bitmap_blk);
abcb2947
AK
321 return NULL;
322 }
c806e68f
FB
323 if (buffer_uptodate(bh) &&
324 !(desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)))
abcb2947
AK
325 return bh;
326
c806e68f 327 lock_buffer(bh);
b5f10eed 328 spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
717d50e4 329 if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
abcb2947
AK
330 ext4_init_block_bitmap(sb, bh, block_group, desc);
331 set_buffer_uptodate(bh);
332 unlock_buffer(bh);
b5f10eed 333 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
abcb2947 334 return bh;
717d50e4 335 }
b5f10eed 336 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
abcb2947
AK
337 if (bh_submit_read(bh) < 0) {
338 put_bh(bh);
46e665e9 339 ext4_error(sb, __func__,
ac27a0ec 340 "Cannot read block bitmap - "
a9df9a49 341 "block_group = %u, block_bitmap = %llu",
e29d1cde 342 block_group, bitmap_blk);
abcb2947
AK
343 return NULL;
344 }
519deca0
AK
345 ext4_valid_block_bitmap(sb, desc, block_group, bh);
346 /*
347 * file system mounted not to panic on error,
348 * continue with corrupt bitmap
349 */
ac27a0ec
DK
350 return bh;
351}
ac27a0ec
DK
352
353/**
e21675d4 354 * ext4_add_groupblocks() -- Add given blocks to an existing group
ac27a0ec
DK
355 * @handle: handle to this transaction
356 * @sb: super block
e21675d4 357 * @block: start physcial block to add to the block group
ac27a0ec 358 * @count: number of blocks to free
c2ea3fde 359 *
e21675d4
AK
360 * This marks the blocks as free in the bitmap. We ask the
361 * mballoc to reload the buddy after this by setting group
362 * EXT4_GROUP_INFO_NEED_INIT_BIT flag
ac27a0ec 363 */
e21675d4
AK
364void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
365 ext4_fsblk_t block, unsigned long count)
ac27a0ec
DK
366{
367 struct buffer_head *bitmap_bh = NULL;
368 struct buffer_head *gd_bh;
fd2d4291 369 ext4_group_t block_group;
617ba13b 370 ext4_grpblk_t bit;
498e5f24 371 unsigned int i;
af5bc92d
TT
372 struct ext4_group_desc *desc;
373 struct ext4_super_block *es;
617ba13b 374 struct ext4_sb_info *sbi;
ac27a0ec 375 int err = 0, ret;
e21675d4
AK
376 ext4_grpblk_t blocks_freed;
377 struct ext4_group_info *grp;
ac27a0ec 378
617ba13b 379 sbi = EXT4_SB(sb);
ac27a0ec 380 es = sbi->s_es;
e21675d4 381 ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
ac27a0ec 382
3a5b2ecd 383 ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
920313a7 384 grp = ext4_get_group_info(sb, block_group);
ac27a0ec
DK
385 /*
386 * Check to see if we are freeing blocks across a group
387 * boundary.
388 */
617ba13b 389 if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
e21675d4 390 goto error_return;
ac27a0ec 391 }
574ca174 392 bitmap_bh = ext4_read_block_bitmap(sb, block_group);
ac27a0ec
DK
393 if (!bitmap_bh)
394 goto error_return;
af5bc92d 395 desc = ext4_get_group_desc(sb, block_group, &gd_bh);
ac27a0ec
DK
396 if (!desc)
397 goto error_return;
398
8fadc143
AR
399 if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
400 in_range(ext4_inode_bitmap(sb, desc), block, count) ||
401 in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
402 in_range(block + count - 1, ext4_inode_table(sb, desc),
cb47dce7 403 sbi->s_itb_per_group)) {
e21675d4
AK
404 ext4_error(sb, __func__,
405 "Adding blocks in system zones - "
af5bc92d
TT
406 "Block = %llu, count = %lu",
407 block, count);
cb47dce7
AK
408 goto error_return;
409 }
ac27a0ec
DK
410
411 /*
e21675d4 412 * We are about to add blocks to the bitmap,
ac27a0ec
DK
413 * so we need undo access.
414 */
ac27a0ec 415 BUFFER_TRACE(bitmap_bh, "getting undo access");
617ba13b 416 err = ext4_journal_get_undo_access(handle, bitmap_bh);
ac27a0ec
DK
417 if (err)
418 goto error_return;
419
420 /*
421 * We are about to modify some metadata. Call the journal APIs
422 * to unshare ->b_data if a currently-committing transaction is
423 * using it
424 */
425 BUFFER_TRACE(gd_bh, "get_write_access");
617ba13b 426 err = ext4_journal_get_write_access(handle, gd_bh);
ac27a0ec
DK
427 if (err)
428 goto error_return;
920313a7
AK
429 /*
430 * make sure we don't allow a parallel init on other groups in the
431 * same buddy cache
432 */
433 down_write(&grp->alloc_sem);
e21675d4 434 for (i = 0, blocks_freed = 0; i < count; i++) {
ac27a0ec 435 BUFFER_TRACE(bitmap_bh, "clear bit");
617ba13b 436 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
ac27a0ec 437 bit + i, bitmap_bh->b_data)) {
46e665e9 438 ext4_error(sb, __func__,
2ae02107 439 "bit already cleared for block %llu",
bd81d8ee 440 (ext4_fsblk_t)(block + i));
ac27a0ec
DK
441 BUFFER_TRACE(bitmap_bh, "bit already cleared");
442 } else {
e21675d4 443 blocks_freed++;
ac27a0ec
DK
444 }
445 }
ac27a0ec 446 spin_lock(sb_bgl_lock(sbi, block_group));
e21675d4 447 le16_add_cpu(&desc->bg_free_blocks_count, blocks_freed);
717d50e4 448 desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
ac27a0ec 449 spin_unlock(sb_bgl_lock(sbi, block_group));
e21675d4 450 percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
ac27a0ec 451
772cb7c8
JS
452 if (sbi->s_log_groups_per_flex) {
453 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
454 spin_lock(sb_bgl_lock(sbi, flex_group));
e21675d4 455 sbi->s_flex_groups[flex_group].free_blocks += blocks_freed;
772cb7c8
JS
456 spin_unlock(sb_bgl_lock(sbi, flex_group));
457 }
920313a7
AK
458 /*
459 * request to reload the buddy with the
460 * new bitmap information
461 */
462 set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
463 ext4_mb_update_group_info(grp, blocks_freed);
464 up_write(&grp->alloc_sem);
772cb7c8 465
ac27a0ec
DK
466 /* We dirtied the bitmap block */
467 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
0390131b 468 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
ac27a0ec
DK
469
470 /* And the group descriptor block */
471 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
0390131b 472 ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
e21675d4
AK
473 if (!err)
474 err = ret;
ac27a0ec 475 sb->s_dirt = 1;
e21675d4 476
ac27a0ec
DK
477error_return:
478 brelse(bitmap_bh);
617ba13b 479 ext4_std_error(sb, err);
ac27a0ec
DK
480 return;
481}
482
483/**
617ba13b 484 * ext4_free_blocks() -- Free given blocks and update quota
ac27a0ec
DK
485 * @handle: handle for this transaction
486 * @inode: inode
487 * @block: start physical block to free
488 * @count: number of blocks to count
c9de560d 489 * @metadata: Are these metadata blocks
ac27a0ec 490 */
617ba13b 491void ext4_free_blocks(handle_t *handle, struct inode *inode,
c9de560d
AT
492 ext4_fsblk_t block, unsigned long count,
493 int metadata)
ac27a0ec 494{
af5bc92d 495 struct super_block *sb;
ac27a0ec
DK
496 unsigned long dquot_freed_blocks;
497
c9de560d
AT
498 /* this isn't the right place to decide whether block is metadata
499 * inode.c/extents.c knows better, but for safety ... */
a1aebc1e
AK
500 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
501 metadata = 1;
502
503 /* We need to make sure we don't reuse
504 * block released untill the transaction commit.
505 * writeback mode have weak data consistency so
506 * don't force data as metadata when freeing block
507 * for writeback mode.
508 */
509 if (metadata == 0 && !ext4_should_writeback_data(inode))
c9de560d
AT
510 metadata = 1;
511
ac27a0ec 512 sb = inode->i_sb;
c9de560d 513
c2ea3fde
TT
514 ext4_mb_free_blocks(handle, inode, block, count,
515 metadata, &dquot_freed_blocks);
ac27a0ec
DK
516 if (dquot_freed_blocks)
517 DQUOT_FREE_BLOCK(inode, dquot_freed_blocks);
518 return;
519}
520
8c3bf8a0
ES
521/**
522 * ext4_has_free_blocks()
523 * @sbi: in-core super block structure.
524 * @nblocks: number of needed blocks
525 *
526 * Check if filesystem has nblocks free & available for allocation.
527 * On success return 1, return 0 on failure.
528 */
529int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
a30d542a 530{
a996031c 531 s64 free_blocks, dirty_blocks, root_blocks;
a30d542a 532 struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
6bc6e63f 533 struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
a30d542a 534
6bc6e63f
AK
535 free_blocks = percpu_counter_read_positive(fbc);
536 dirty_blocks = percpu_counter_read_positive(dbc);
a996031c 537 root_blocks = ext4_r_blocks_count(sbi->s_es);
a30d542a 538
6bc6e63f
AK
539 if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
540 EXT4_FREEBLOCKS_WATERMARK) {
02d21168
AM
541 free_blocks = percpu_counter_sum_positive(fbc);
542 dirty_blocks = percpu_counter_sum_positive(dbc);
6bc6e63f
AK
543 if (dirty_blocks < 0) {
544 printk(KERN_CRIT "Dirty block accounting "
545 "went wrong %lld\n",
8f72fbdf 546 (long long)dirty_blocks);
6bc6e63f
AK
547 }
548 }
549 /* Check whether we have space after
a996031c 550 * accounting for current dirty blocks & root reserved blocks.
6bc6e63f 551 */
a996031c
ES
552 if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
553 return 1;
a30d542a 554
a996031c 555 /* Hm, nope. Are (enough) root reserved blocks available? */
4c9c544e 556 if (sbi->s_resuid == current_fsuid() ||
a996031c
ES
557 ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
558 capable(CAP_SYS_RESOURCE)) {
559 if (free_blocks >= (nblocks + dirty_blocks))
560 return 1;
561 }
562
563 return 0;
a30d542a
AK
564}
565
8c3bf8a0 566int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
5c791616 567 s64 nblocks)
ac27a0ec 568{
8c3bf8a0
ES
569 if (ext4_has_free_blocks(sbi, nblocks)) {
570 percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
16eb7295 571 return 0;
8c3bf8a0
ES
572 } else
573 return -ENOSPC;
a30d542a 574}
07031431 575
ac27a0ec 576/**
617ba13b 577 * ext4_should_retry_alloc()
ac27a0ec
DK
578 * @sb: super block
579 * @retries number of attemps has been made
580 *
617ba13b 581 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
ac27a0ec
DK
582 * it is profitable to retry the operation, this function will wait
583 * for the current or commiting transaction to complete, and then
584 * return TRUE.
585 *
586 * if the total number of retries exceed three times, return FALSE.
587 */
617ba13b 588int ext4_should_retry_alloc(struct super_block *sb, int *retries)
ac27a0ec 589{
07031431 590 if (!ext4_has_free_blocks(EXT4_SB(sb), 1) || (*retries)++ > 3)
ac27a0ec
DK
591 return 0;
592
593 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
594
dab291af 595 return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
ac27a0ec
DK
596}
597
654b4908 598/*
d2a17637 599 * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
654b4908
AK
600 *
601 * @handle: handle to this transaction
602 * @inode: file inode
603 * @goal: given target block(filesystem wide)
97df5d15 604 * @count: pointer to total number of blocks needed
654b4908
AK
605 * @errp: error code
606 *
97df5d15 607 * Return 1st allocated block number on success, *count stores total account
d2a17637 608 * error stores in errp pointer
654b4908 609 */
d2a17637
MC
610ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
611 ext4_fsblk_t goal, unsigned long *count, int *errp)
654b4908 612{
97df5d15 613 struct ext4_allocation_request ar;
d2a17637 614 ext4_fsblk_t ret;
97df5d15
TT
615
616 memset(&ar, 0, sizeof(ar));
617 /* Fill with neighbour allocated blocks */
618 ar.inode = inode;
619 ar.goal = goal;
620 ar.len = count ? *count : 1;
621
622 ret = ext4_mb_new_blocks(handle, &ar, errp);
623 if (count)
624 *count = ar.len;
625
d2a17637
MC
626 /*
627 * Account for the allocated meta blocks
628 */
166348dd 629 if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
d2a17637 630 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
97df5d15 631 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
d2a17637
MC
632 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
633 }
634 return ret;
654b4908
AK
635}
636
ac27a0ec 637/**
617ba13b 638 * ext4_count_free_blocks() -- count filesystem free blocks
ac27a0ec
DK
639 * @sb: superblock
640 *
641 * Adds up the number of free blocks from each block group.
642 */
617ba13b 643ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
ac27a0ec 644{
617ba13b
MC
645 ext4_fsblk_t desc_count;
646 struct ext4_group_desc *gdp;
fd2d4291
AM
647 ext4_group_t i;
648 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
617ba13b
MC
649#ifdef EXT4FS_DEBUG
650 struct ext4_super_block *es;
651 ext4_fsblk_t bitmap_count;
498e5f24 652 unsigned int x;
ac27a0ec
DK
653 struct buffer_head *bitmap_bh = NULL;
654
617ba13b 655 es = EXT4_SB(sb)->s_es;
ac27a0ec
DK
656 desc_count = 0;
657 bitmap_count = 0;
658 gdp = NULL;
659
660 smp_rmb();
661 for (i = 0; i < ngroups; i++) {
617ba13b 662 gdp = ext4_get_group_desc(sb, i, NULL);
ac27a0ec
DK
663 if (!gdp)
664 continue;
665 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
666 brelse(bitmap_bh);
574ca174 667 bitmap_bh = ext4_read_block_bitmap(sb, i);
ac27a0ec
DK
668 if (bitmap_bh == NULL)
669 continue;
670
617ba13b 671 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
498e5f24 672 printk(KERN_DEBUG "group %lu: stored = %d, counted = %u\n",
ac27a0ec
DK
673 i, le16_to_cpu(gdp->bg_free_blocks_count), x);
674 bitmap_count += x;
675 }
676 brelse(bitmap_bh);
4776004f
TT
677 printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
678 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
679 desc_count, bitmap_count);
ac27a0ec
DK
680 return bitmap_count;
681#else
682 desc_count = 0;
683 smp_rmb();
684 for (i = 0; i < ngroups; i++) {
617ba13b 685 gdp = ext4_get_group_desc(sb, i, NULL);
ac27a0ec
DK
686 if (!gdp)
687 continue;
688 desc_count += le16_to_cpu(gdp->bg_free_blocks_count);
689 }
690
691 return desc_count;
692#endif
693}
694
fd2d4291 695static inline int test_root(ext4_group_t a, int b)
ac27a0ec
DK
696{
697 int num = b;
698
699 while (a > num)
700 num *= b;
701 return num == a;
702}
703
fd2d4291 704static int ext4_group_sparse(ext4_group_t group)
ac27a0ec
DK
705{
706 if (group <= 1)
707 return 1;
708 if (!(group & 1))
709 return 0;
710 return (test_root(group, 7) || test_root(group, 5) ||
711 test_root(group, 3));
712}
713
714/**
617ba13b 715 * ext4_bg_has_super - number of blocks used by the superblock in group
ac27a0ec
DK
716 * @sb: superblock for filesystem
717 * @group: group number to check
718 *
719 * Return the number of blocks used by the superblock (primary or backup)
720 * in this group. Currently this will be only 0 or 1.
721 */
fd2d4291 722int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
ac27a0ec 723{
617ba13b
MC
724 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
725 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
726 !ext4_group_sparse(group))
ac27a0ec
DK
727 return 0;
728 return 1;
729}
730
fd2d4291
AM
731static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
732 ext4_group_t group)
ac27a0ec 733{
617ba13b 734 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
fd2d4291
AM
735 ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
736 ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
ac27a0ec
DK
737
738 if (group == first || group == first + 1 || group == last)
739 return 1;
740 return 0;
741}
742
fd2d4291
AM
743static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
744 ext4_group_t group)
ac27a0ec 745{
859cb936 746 return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0;
ac27a0ec
DK
747}
748
749/**
617ba13b 750 * ext4_bg_num_gdb - number of blocks used by the group table in group
ac27a0ec
DK
751 * @sb: superblock for filesystem
752 * @group: group number to check
753 *
754 * Return the number of blocks used by the group descriptor table
755 * (primary or backup) in this group. In the future there may be a
756 * different number of descriptor blocks in each group.
757 */
fd2d4291 758unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
ac27a0ec
DK
759{
760 unsigned long first_meta_bg =
617ba13b
MC
761 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
762 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
ac27a0ec 763
617ba13b 764 if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
ac27a0ec 765 metagroup < first_meta_bg)
af5bc92d 766 return ext4_bg_num_gdb_nometa(sb, group);
ac27a0ec 767
617ba13b 768 return ext4_bg_num_gdb_meta(sb,group);
ac27a0ec
DK
769
770}
c2ea3fde 771