]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/ext4/balloc.c
ext4: fix uninit block bitmap initialization when s_meta_first_bg is non-zero
[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"
e21675d4 22#include "mballoc.h"
3dcf5451 23
ac27a0ec
DK
24/*
25 * balloc.c contains the blocks allocation and deallocation routines
26 */
27
72b64b59
AM
28/*
29 * Calculate the block group number and offset, given a block number
30 */
31void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
fd2d4291 32 ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp)
72b64b59 33{
8c55e204 34 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
72b64b59
AM
35 ext4_grpblk_t offset;
36
8c55e204 37 blocknr = blocknr - le32_to_cpu(es->s_first_data_block);
f4e5bc24 38 offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb));
72b64b59
AM
39 if (offsetp)
40 *offsetp = offset;
41 if (blockgrpp)
8c55e204 42 *blockgrpp = blocknr;
72b64b59
AM
43
44}
45
0bf7e837
JS
46static int ext4_block_in_group(struct super_block *sb, ext4_fsblk_t block,
47 ext4_group_t block_group)
48{
49 ext4_group_t actual_group;
7477827f 50 ext4_get_group_no_and_offset(sb, block, &actual_group, NULL);
0bf7e837
JS
51 if (actual_group == block_group)
52 return 1;
53 return 0;
54}
55
56static int ext4_group_used_meta_blocks(struct super_block *sb,
e187c658
TT
57 ext4_group_t block_group,
58 struct ext4_group_desc *gdp)
0bf7e837
JS
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)) {
0bf7e837
JS
66 if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
67 block_group))
68 used_blocks--;
69
70 if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
71 block_group))
72 used_blocks--;
73
74 tmp = ext4_inode_table(sb, gdp);
75 for (; tmp < ext4_inode_table(sb, gdp) +
76 sbi->s_itb_per_group; tmp++) {
77 if (!ext4_block_in_group(sb, tmp, block_group))
78 used_blocks -= 1;
79 }
80 }
81 return used_blocks;
82}
c2ea3fde 83
717d50e4
AD
84/* Initializes an uninitialized block bitmap if given, and returns the
85 * number of blocks free in the group. */
86unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
fd2d4291 87 ext4_group_t block_group, struct ext4_group_desc *gdp)
717d50e4 88{
717d50e4 89 int bit, bit_max;
8df9675f 90 ext4_group_t ngroups = ext4_get_groups_count(sb);
717d50e4
AD
91 unsigned free_blocks, group_blocks;
92 struct ext4_sb_info *sbi = EXT4_SB(sb);
93
94 if (bh) {
95 J_ASSERT_BH(bh, buffer_locked(bh));
96
97 /* If checksum is bad mark all blocks used to prevent allocation
98 * essentially implementing a per-group read-only flag. */
99 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
46e665e9 100 ext4_error(sb, __func__,
a9df9a49 101 "Checksum bad for group %u", block_group);
560671a0
AK
102 ext4_free_blks_set(sb, gdp, 0);
103 ext4_free_inodes_set(sb, gdp, 0);
104 ext4_itable_unused_set(sb, gdp, 0);
717d50e4
AD
105 memset(bh->b_data, 0xff, sb->s_blocksize);
106 return 0;
107 }
108 memset(bh->b_data, 0, sb->s_blocksize);
109 }
110
111 /* Check for superblock and gdt backups in this group */
112 bit_max = ext4_bg_has_super(sb, block_group);
113
114 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
115 block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
116 sbi->s_desc_per_block) {
117 if (bit_max) {
118 bit_max += ext4_bg_num_gdb(sb, block_group);
119 bit_max +=
120 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
121 }
122 } else { /* For META_BG_BLOCK_GROUPS */
6afd6707 123 bit_max += ext4_bg_num_gdb(sb, block_group);
717d50e4
AD
124 }
125
8df9675f 126 if (block_group == ngroups - 1) {
717d50e4
AD
127 /*
128 * Even though mke2fs always initialize first and last group
129 * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
130 * to make sure we calculate the right free blocks
131 */
132 group_blocks = ext4_blocks_count(sbi->s_es) -
133 le32_to_cpu(sbi->s_es->s_first_data_block) -
8df9675f 134 (EXT4_BLOCKS_PER_GROUP(sb) * (ngroups - 1));
717d50e4
AD
135 } else {
136 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
137 }
138
139 free_blocks = group_blocks - bit_max;
140
141 if (bh) {
0bf7e837
JS
142 ext4_fsblk_t start, tmp;
143 int flex_bg = 0;
d00a6d7b 144
717d50e4
AD
145 for (bit = 0; bit < bit_max; bit++)
146 ext4_set_bit(bit, bh->b_data);
147
d00a6d7b 148 start = ext4_group_first_block_no(sb, block_group);
717d50e4 149
0bf7e837
JS
150 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
151 EXT4_FEATURE_INCOMPAT_FLEX_BG))
152 flex_bg = 1;
717d50e4 153
0bf7e837
JS
154 /* Set bits for block and inode bitmaps, and inode table */
155 tmp = ext4_block_bitmap(sb, gdp);
156 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
157 ext4_set_bit(tmp - start, bh->b_data);
158
159 tmp = ext4_inode_bitmap(sb, gdp);
160 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
161 ext4_set_bit(tmp - start, bh->b_data);
162
163 tmp = ext4_inode_table(sb, gdp);
164 for (; tmp < ext4_inode_table(sb, gdp) +
165 sbi->s_itb_per_group; tmp++) {
166 if (!flex_bg ||
167 ext4_block_in_group(sb, tmp, block_group))
168 ext4_set_bit(tmp - start, bh->b_data);
169 }
717d50e4
AD
170 /*
171 * Also if the number of blocks within the group is
172 * less than the blocksize * 8 ( which is the size
173 * of bitmap ), set rest of the block bitmap to 1
174 */
175 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
176 }
e187c658 177 return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp);
717d50e4
AD
178}
179
180
ac27a0ec
DK
181/*
182 * The free blocks are managed by bitmaps. A file system contains several
183 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
184 * block for inodes, N blocks for the inode table and data blocks.
185 *
186 * The file system contains group descriptors which are located after the
187 * super block. Each descriptor contains the number of the bitmap block and
188 * the free blocks count in the block. The descriptors are loaded in memory
e627432c 189 * when a file system is mounted (see ext4_fill_super).
ac27a0ec
DK
190 */
191
192
193#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
194
195/**
617ba13b 196 * ext4_get_group_desc() -- load group descriptor from disk
ac27a0ec
DK
197 * @sb: super block
198 * @block_group: given block group
199 * @bh: pointer to the buffer head to store the block
200 * group descriptor
201 */
af5bc92d 202struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
fd2d4291 203 ext4_group_t block_group,
af5bc92d 204 struct buffer_head **bh)
ac27a0ec 205{
498e5f24
TT
206 unsigned int group_desc;
207 unsigned int offset;
8df9675f 208 ext4_group_t ngroups = ext4_get_groups_count(sb);
af5bc92d 209 struct ext4_group_desc *desc;
617ba13b 210 struct ext4_sb_info *sbi = EXT4_SB(sb);
ac27a0ec 211
8df9675f 212 if (block_group >= ngroups) {
af5bc92d
TT
213 ext4_error(sb, "ext4_get_group_desc",
214 "block_group >= groups_count - "
a9df9a49 215 "block_group = %u, groups_count = %u",
8df9675f 216 block_group, ngroups);
ac27a0ec
DK
217
218 return NULL;
219 }
ac27a0ec 220
617ba13b
MC
221 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
222 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
ac27a0ec 223 if (!sbi->s_group_desc[group_desc]) {
af5bc92d
TT
224 ext4_error(sb, "ext4_get_group_desc",
225 "Group descriptor not loaded - "
498e5f24 226 "block_group = %u, group_desc = %u, desc = %u",
af5bc92d 227 block_group, group_desc, offset);
ac27a0ec
DK
228 return NULL;
229 }
230
0d1ee42f
AR
231 desc = (struct ext4_group_desc *)(
232 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
233 offset * EXT4_DESC_SIZE(sb));
ac27a0ec
DK
234 if (bh)
235 *bh = sbi->s_group_desc[group_desc];
0d1ee42f 236 return desc;
ac27a0ec
DK
237}
238
abcb2947
AK
239static int ext4_valid_block_bitmap(struct super_block *sb,
240 struct ext4_group_desc *desc,
241 unsigned int block_group,
242 struct buffer_head *bh)
243{
244 ext4_grpblk_t offset;
245 ext4_grpblk_t next_zero_bit;
246 ext4_fsblk_t bitmap_blk;
247 ext4_fsblk_t group_first_block;
248
249 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
250 /* with FLEX_BG, the inode/block bitmaps and itable
251 * blocks may not be in the group at all
252 * so the bitmap validation will be skipped for those groups
253 * or it has to also read the block group where the bitmaps
254 * are located to verify they are set.
255 */
256 return 1;
257 }
258 group_first_block = ext4_group_first_block_no(sb, block_group);
259
260 /* check whether block bitmap block number is set */
261 bitmap_blk = ext4_block_bitmap(sb, desc);
262 offset = bitmap_blk - group_first_block;
263 if (!ext4_test_bit(offset, bh->b_data))
264 /* bad block bitmap */
265 goto err_out;
266
267 /* check whether the inode bitmap block number is set */
268 bitmap_blk = ext4_inode_bitmap(sb, desc);
269 offset = bitmap_blk - group_first_block;
270 if (!ext4_test_bit(offset, bh->b_data))
271 /* bad block bitmap */
272 goto err_out;
273
274 /* check whether the inode table block number is set */
275 bitmap_blk = ext4_inode_table(sb, desc);
276 offset = bitmap_blk - group_first_block;
277 next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
278 offset + EXT4_SB(sb)->s_itb_per_group,
279 offset);
280 if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
281 /* good bitmap for inode tables */
282 return 1;
283
284err_out:
46e665e9 285 ext4_error(sb, __func__,
abcb2947
AK
286 "Invalid block bitmap - "
287 "block_group = %d, block = %llu",
288 block_group, bitmap_blk);
289 return 0;
290}
ac27a0ec 291/**
574ca174 292 * ext4_read_block_bitmap()
ac27a0ec
DK
293 * @sb: super block
294 * @block_group: given block group
295 *
abcb2947
AK
296 * Read the bitmap for a given block_group,and validate the
297 * bits for block/inode/inode tables are set in the bitmaps
ac27a0ec
DK
298 *
299 * Return buffer_head on success or NULL in case of failure.
300 */
717d50e4 301struct buffer_head *
574ca174 302ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
ac27a0ec 303{
af5bc92d
TT
304 struct ext4_group_desc *desc;
305 struct buffer_head *bh = NULL;
7c9e69fa 306 ext4_fsblk_t bitmap_blk;
ac27a0ec 307
717d50e4 308 desc = ext4_get_group_desc(sb, block_group, NULL);
ac27a0ec 309 if (!desc)
7c9e69fa
AK
310 return NULL;
311 bitmap_blk = ext4_block_bitmap(sb, desc);
abcb2947
AK
312 bh = sb_getblk(sb, bitmap_blk);
313 if (unlikely(!bh)) {
46e665e9 314 ext4_error(sb, __func__,
abcb2947 315 "Cannot read block bitmap - "
a9df9a49 316 "block_group = %u, block_bitmap = %llu",
e29d1cde 317 block_group, bitmap_blk);
abcb2947
AK
318 return NULL;
319 }
2ccb5fb9
AK
320
321 if (bitmap_uptodate(bh))
abcb2947
AK
322 return bh;
323
c806e68f 324 lock_buffer(bh);
2ccb5fb9
AK
325 if (bitmap_uptodate(bh)) {
326 unlock_buffer(bh);
327 return bh;
328 }
955ce5f5 329 ext4_lock_group(sb, block_group);
717d50e4 330 if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
abcb2947 331 ext4_init_block_bitmap(sb, bh, block_group, desc);
2ccb5fb9 332 set_bitmap_uptodate(bh);
abcb2947 333 set_buffer_uptodate(bh);
955ce5f5 334 ext4_unlock_group(sb, block_group);
3300beda 335 unlock_buffer(bh);
abcb2947 336 return bh;
717d50e4 337 }
955ce5f5 338 ext4_unlock_group(sb, block_group);
2ccb5fb9
AK
339 if (buffer_uptodate(bh)) {
340 /*
341 * if not uninit if bh is uptodate,
342 * bitmap is also uptodate
343 */
344 set_bitmap_uptodate(bh);
345 unlock_buffer(bh);
346 return bh;
347 }
348 /*
349 * submit the buffer_head for read. We can
350 * safely mark the bitmap as uptodate now.
351 * We do it here so the bitmap uptodate bit
352 * get set with buffer lock held.
353 */
354 set_bitmap_uptodate(bh);
abcb2947
AK
355 if (bh_submit_read(bh) < 0) {
356 put_bh(bh);
46e665e9 357 ext4_error(sb, __func__,
ac27a0ec 358 "Cannot read block bitmap - "
a9df9a49 359 "block_group = %u, block_bitmap = %llu",
e29d1cde 360 block_group, bitmap_blk);
abcb2947
AK
361 return NULL;
362 }
519deca0
AK
363 ext4_valid_block_bitmap(sb, desc, block_group, bh);
364 /*
365 * file system mounted not to panic on error,
366 * continue with corrupt bitmap
367 */
ac27a0ec
DK
368 return bh;
369}
ac27a0ec
DK
370
371/**
e21675d4 372 * ext4_add_groupblocks() -- Add given blocks to an existing group
ac27a0ec
DK
373 * @handle: handle to this transaction
374 * @sb: super block
e21675d4 375 * @block: start physcial block to add to the block group
ac27a0ec 376 * @count: number of blocks to free
c2ea3fde 377 *
e21675d4
AK
378 * This marks the blocks as free in the bitmap. We ask the
379 * mballoc to reload the buddy after this by setting group
380 * EXT4_GROUP_INFO_NEED_INIT_BIT flag
ac27a0ec 381 */
e21675d4
AK
382void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
383 ext4_fsblk_t block, unsigned long count)
ac27a0ec
DK
384{
385 struct buffer_head *bitmap_bh = NULL;
386 struct buffer_head *gd_bh;
fd2d4291 387 ext4_group_t block_group;
617ba13b 388 ext4_grpblk_t bit;
498e5f24 389 unsigned int i;
af5bc92d
TT
390 struct ext4_group_desc *desc;
391 struct ext4_super_block *es;
617ba13b 392 struct ext4_sb_info *sbi;
560671a0 393 int err = 0, ret, blk_free_count;
e21675d4
AK
394 ext4_grpblk_t blocks_freed;
395 struct ext4_group_info *grp;
ac27a0ec 396
617ba13b 397 sbi = EXT4_SB(sb);
ac27a0ec 398 es = sbi->s_es;
e21675d4 399 ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
ac27a0ec 400
3a5b2ecd 401 ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
920313a7 402 grp = ext4_get_group_info(sb, block_group);
ac27a0ec
DK
403 /*
404 * Check to see if we are freeing blocks across a group
405 * boundary.
406 */
617ba13b 407 if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
e21675d4 408 goto error_return;
ac27a0ec 409 }
574ca174 410 bitmap_bh = ext4_read_block_bitmap(sb, block_group);
ac27a0ec
DK
411 if (!bitmap_bh)
412 goto error_return;
af5bc92d 413 desc = ext4_get_group_desc(sb, block_group, &gd_bh);
ac27a0ec
DK
414 if (!desc)
415 goto error_return;
416
8fadc143
AR
417 if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
418 in_range(ext4_inode_bitmap(sb, desc), block, count) ||
419 in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
420 in_range(block + count - 1, ext4_inode_table(sb, desc),
cb47dce7 421 sbi->s_itb_per_group)) {
e21675d4
AK
422 ext4_error(sb, __func__,
423 "Adding blocks in system zones - "
af5bc92d
TT
424 "Block = %llu, count = %lu",
425 block, count);
cb47dce7
AK
426 goto error_return;
427 }
ac27a0ec
DK
428
429 /*
e21675d4 430 * We are about to add blocks to the bitmap,
ac27a0ec
DK
431 * so we need undo access.
432 */
ac27a0ec 433 BUFFER_TRACE(bitmap_bh, "getting undo access");
617ba13b 434 err = ext4_journal_get_undo_access(handle, bitmap_bh);
ac27a0ec
DK
435 if (err)
436 goto error_return;
437
438 /*
439 * We are about to modify some metadata. Call the journal APIs
440 * to unshare ->b_data if a currently-committing transaction is
441 * using it
442 */
443 BUFFER_TRACE(gd_bh, "get_write_access");
617ba13b 444 err = ext4_journal_get_write_access(handle, gd_bh);
ac27a0ec
DK
445 if (err)
446 goto error_return;
920313a7
AK
447 /*
448 * make sure we don't allow a parallel init on other groups in the
449 * same buddy cache
450 */
451 down_write(&grp->alloc_sem);
e21675d4 452 for (i = 0, blocks_freed = 0; i < count; i++) {
ac27a0ec 453 BUFFER_TRACE(bitmap_bh, "clear bit");
955ce5f5 454 if (!ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group),
ac27a0ec 455 bit + i, bitmap_bh->b_data)) {
46e665e9 456 ext4_error(sb, __func__,
2ae02107 457 "bit already cleared for block %llu",
bd81d8ee 458 (ext4_fsblk_t)(block + i));
ac27a0ec
DK
459 BUFFER_TRACE(bitmap_bh, "bit already cleared");
460 } else {
e21675d4 461 blocks_freed++;
ac27a0ec
DK
462 }
463 }
955ce5f5 464 ext4_lock_group(sb, block_group);
560671a0
AK
465 blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
466 ext4_free_blks_set(sb, desc, blk_free_count);
717d50e4 467 desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
955ce5f5 468 ext4_unlock_group(sb, block_group);
e21675d4 469 percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
ac27a0ec 470
772cb7c8
JS
471 if (sbi->s_log_groups_per_flex) {
472 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
9f24e420
TT
473 atomic_add(blocks_freed,
474 &sbi->s_flex_groups[flex_group].free_blocks);
772cb7c8 475 }
920313a7
AK
476 /*
477 * request to reload the buddy with the
478 * new bitmap information
479 */
480 set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
0373130d 481 grp->bb_free += blocks_freed;
920313a7 482 up_write(&grp->alloc_sem);
772cb7c8 483
ac27a0ec
DK
484 /* We dirtied the bitmap block */
485 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
0390131b 486 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
ac27a0ec
DK
487
488 /* And the group descriptor block */
489 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
0390131b 490 ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
e21675d4
AK
491 if (!err)
492 err = ret;
ac27a0ec 493 sb->s_dirt = 1;
e21675d4 494
ac27a0ec
DK
495error_return:
496 brelse(bitmap_bh);
617ba13b 497 ext4_std_error(sb, err);
ac27a0ec
DK
498 return;
499}
500
501/**
617ba13b 502 * ext4_free_blocks() -- Free given blocks and update quota
ac27a0ec
DK
503 * @handle: handle for this transaction
504 * @inode: inode
505 * @block: start physical block to free
506 * @count: number of blocks to count
c9de560d 507 * @metadata: Are these metadata blocks
ac27a0ec 508 */
617ba13b 509void ext4_free_blocks(handle_t *handle, struct inode *inode,
c9de560d
AT
510 ext4_fsblk_t block, unsigned long count,
511 int metadata)
ac27a0ec 512{
af5bc92d 513 struct super_block *sb;
ac27a0ec
DK
514 unsigned long dquot_freed_blocks;
515
c9de560d
AT
516 /* this isn't the right place to decide whether block is metadata
517 * inode.c/extents.c knows better, but for safety ... */
a1aebc1e
AK
518 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
519 metadata = 1;
520
521 /* We need to make sure we don't reuse
522 * block released untill the transaction commit.
523 * writeback mode have weak data consistency so
524 * don't force data as metadata when freeing block
525 * for writeback mode.
526 */
527 if (metadata == 0 && !ext4_should_writeback_data(inode))
c9de560d
AT
528 metadata = 1;
529
ac27a0ec 530 sb = inode->i_sb;
c9de560d 531
c2ea3fde
TT
532 ext4_mb_free_blocks(handle, inode, block, count,
533 metadata, &dquot_freed_blocks);
ac27a0ec 534 if (dquot_freed_blocks)
a269eb18 535 vfs_dq_free_block(inode, dquot_freed_blocks);
ac27a0ec
DK
536 return;
537}
538
8c3bf8a0
ES
539/**
540 * ext4_has_free_blocks()
541 * @sbi: in-core super block structure.
542 * @nblocks: number of needed blocks
543 *
544 * Check if filesystem has nblocks free & available for allocation.
545 * On success return 1, return 0 on failure.
546 */
547int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
a30d542a 548{
a996031c 549 s64 free_blocks, dirty_blocks, root_blocks;
a30d542a 550 struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
6bc6e63f 551 struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
a30d542a 552
6bc6e63f
AK
553 free_blocks = percpu_counter_read_positive(fbc);
554 dirty_blocks = percpu_counter_read_positive(dbc);
a996031c 555 root_blocks = ext4_r_blocks_count(sbi->s_es);
a30d542a 556
6bc6e63f
AK
557 if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
558 EXT4_FREEBLOCKS_WATERMARK) {
02d21168
AM
559 free_blocks = percpu_counter_sum_positive(fbc);
560 dirty_blocks = percpu_counter_sum_positive(dbc);
6bc6e63f
AK
561 if (dirty_blocks < 0) {
562 printk(KERN_CRIT "Dirty block accounting "
563 "went wrong %lld\n",
8f72fbdf 564 (long long)dirty_blocks);
6bc6e63f
AK
565 }
566 }
567 /* Check whether we have space after
a996031c 568 * accounting for current dirty blocks & root reserved blocks.
6bc6e63f 569 */
a996031c
ES
570 if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
571 return 1;
a30d542a 572
a996031c 573 /* Hm, nope. Are (enough) root reserved blocks available? */
4c9c544e 574 if (sbi->s_resuid == current_fsuid() ||
a996031c
ES
575 ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
576 capable(CAP_SYS_RESOURCE)) {
577 if (free_blocks >= (nblocks + dirty_blocks))
578 return 1;
579 }
580
581 return 0;
a30d542a
AK
582}
583
8c3bf8a0 584int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
5c791616 585 s64 nblocks)
ac27a0ec 586{
8c3bf8a0
ES
587 if (ext4_has_free_blocks(sbi, nblocks)) {
588 percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
16eb7295 589 return 0;
8c3bf8a0
ES
590 } else
591 return -ENOSPC;
a30d542a 592}
07031431 593
ac27a0ec 594/**
617ba13b 595 * ext4_should_retry_alloc()
ac27a0ec
DK
596 * @sb: super block
597 * @retries number of attemps has been made
598 *
617ba13b 599 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
ac27a0ec
DK
600 * it is profitable to retry the operation, this function will wait
601 * for the current or commiting transaction to complete, and then
602 * return TRUE.
603 *
604 * if the total number of retries exceed three times, return FALSE.
605 */
617ba13b 606int ext4_should_retry_alloc(struct super_block *sb, int *retries)
ac27a0ec 607{
8f64b32e
ES
608 if (!ext4_has_free_blocks(EXT4_SB(sb), 1) ||
609 (*retries)++ > 3 ||
610 !EXT4_SB(sb)->s_journal)
ac27a0ec
DK
611 return 0;
612
613 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
614
dab291af 615 return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
ac27a0ec
DK
616}
617
654b4908 618/*
d2a17637 619 * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
654b4908
AK
620 *
621 * @handle: handle to this transaction
622 * @inode: file inode
623 * @goal: given target block(filesystem wide)
97df5d15 624 * @count: pointer to total number of blocks needed
654b4908
AK
625 * @errp: error code
626 *
97df5d15 627 * Return 1st allocated block number on success, *count stores total account
d2a17637 628 * error stores in errp pointer
654b4908 629 */
d2a17637
MC
630ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
631 ext4_fsblk_t goal, unsigned long *count, int *errp)
654b4908 632{
97df5d15 633 struct ext4_allocation_request ar;
d2a17637 634 ext4_fsblk_t ret;
97df5d15
TT
635
636 memset(&ar, 0, sizeof(ar));
637 /* Fill with neighbour allocated blocks */
638 ar.inode = inode;
639 ar.goal = goal;
640 ar.len = count ? *count : 1;
641
642 ret = ext4_mb_new_blocks(handle, &ar, errp);
643 if (count)
644 *count = ar.len;
645
d2a17637
MC
646 /*
647 * Account for the allocated meta blocks
648 */
166348dd 649 if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
d2a17637 650 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
97df5d15 651 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
d2a17637
MC
652 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
653 }
654 return ret;
654b4908
AK
655}
656
ac27a0ec 657/**
617ba13b 658 * ext4_count_free_blocks() -- count filesystem free blocks
ac27a0ec
DK
659 * @sb: superblock
660 *
661 * Adds up the number of free blocks from each block group.
662 */
617ba13b 663ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
ac27a0ec 664{
617ba13b
MC
665 ext4_fsblk_t desc_count;
666 struct ext4_group_desc *gdp;
fd2d4291 667 ext4_group_t i;
8df9675f 668 ext4_group_t ngroups = ext4_get_groups_count(sb);
617ba13b
MC
669#ifdef EXT4FS_DEBUG
670 struct ext4_super_block *es;
671 ext4_fsblk_t bitmap_count;
498e5f24 672 unsigned int x;
ac27a0ec
DK
673 struct buffer_head *bitmap_bh = NULL;
674
617ba13b 675 es = EXT4_SB(sb)->s_es;
ac27a0ec
DK
676 desc_count = 0;
677 bitmap_count = 0;
678 gdp = NULL;
679
ac27a0ec 680 for (i = 0; i < ngroups; i++) {
617ba13b 681 gdp = ext4_get_group_desc(sb, i, NULL);
ac27a0ec
DK
682 if (!gdp)
683 continue;
9fd9784c 684 desc_count += ext4_free_blks_count(sb, gdp);
ac27a0ec 685 brelse(bitmap_bh);
574ca174 686 bitmap_bh = ext4_read_block_bitmap(sb, i);
ac27a0ec
DK
687 if (bitmap_bh == NULL)
688 continue;
689
617ba13b 690 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
9fd9784c
TLSC
691 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
692 i, ext4_free_blks_count(sb, gdp), x);
ac27a0ec
DK
693 bitmap_count += x;
694 }
695 brelse(bitmap_bh);
4776004f
TT
696 printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
697 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
698 desc_count, bitmap_count);
ac27a0ec
DK
699 return bitmap_count;
700#else
701 desc_count = 0;
ac27a0ec 702 for (i = 0; i < ngroups; i++) {
617ba13b 703 gdp = ext4_get_group_desc(sb, i, NULL);
ac27a0ec
DK
704 if (!gdp)
705 continue;
560671a0 706 desc_count += ext4_free_blks_count(sb, gdp);
ac27a0ec
DK
707 }
708
709 return desc_count;
710#endif
711}
712
fd2d4291 713static inline int test_root(ext4_group_t a, int b)
ac27a0ec
DK
714{
715 int num = b;
716
717 while (a > num)
718 num *= b;
719 return num == a;
720}
721
fd2d4291 722static int ext4_group_sparse(ext4_group_t group)
ac27a0ec
DK
723{
724 if (group <= 1)
725 return 1;
726 if (!(group & 1))
727 return 0;
728 return (test_root(group, 7) || test_root(group, 5) ||
729 test_root(group, 3));
730}
731
732/**
617ba13b 733 * ext4_bg_has_super - number of blocks used by the superblock in group
ac27a0ec
DK
734 * @sb: superblock for filesystem
735 * @group: group number to check
736 *
737 * Return the number of blocks used by the superblock (primary or backup)
738 * in this group. Currently this will be only 0 or 1.
739 */
fd2d4291 740int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
ac27a0ec 741{
617ba13b
MC
742 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
743 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
744 !ext4_group_sparse(group))
ac27a0ec
DK
745 return 0;
746 return 1;
747}
748
fd2d4291
AM
749static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
750 ext4_group_t group)
ac27a0ec 751{
617ba13b 752 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
fd2d4291
AM
753 ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
754 ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
ac27a0ec
DK
755
756 if (group == first || group == first + 1 || group == last)
757 return 1;
758 return 0;
759}
760
fd2d4291
AM
761static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
762 ext4_group_t group)
ac27a0ec 763{
8dadb198
TT
764 if (!ext4_bg_has_super(sb, group))
765 return 0;
766
767 if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG))
768 return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
769 else
770 return EXT4_SB(sb)->s_gdb_count;
ac27a0ec
DK
771}
772
773/**
617ba13b 774 * ext4_bg_num_gdb - number of blocks used by the group table in group
ac27a0ec
DK
775 * @sb: superblock for filesystem
776 * @group: group number to check
777 *
778 * Return the number of blocks used by the group descriptor table
779 * (primary or backup) in this group. In the future there may be a
780 * different number of descriptor blocks in each group.
781 */
fd2d4291 782unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
ac27a0ec
DK
783{
784 unsigned long first_meta_bg =
617ba13b
MC
785 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
786 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
ac27a0ec 787
617ba13b 788 if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
ac27a0ec 789 metagroup < first_meta_bg)
af5bc92d 790 return ext4_bg_num_gdb_nometa(sb, group);
ac27a0ec 791
617ba13b 792 return ext4_bg_num_gdb_meta(sb,group);
ac27a0ec
DK
793
794}
c2ea3fde 795