]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/ext4/balloc.c
ext4: Avoid races caused by on-line resizing and SMP memory reordering
[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"
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,
e187c658
TT
58 ext4_group_t block_group,
59 struct ext4_group_desc *gdp)
0bf7e837
JS
60{
61 ext4_fsblk_t tmp;
62 struct ext4_sb_info *sbi = EXT4_SB(sb);
63 /* block bitmap, inode bitmap, and inode table blocks */
64 int used_blocks = sbi->s_itb_per_group + 2;
65
66 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
0bf7e837
JS
67 if (!ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp),
68 block_group))
69 used_blocks--;
70
71 if (!ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp),
72 block_group))
73 used_blocks--;
74
75 tmp = ext4_inode_table(sb, gdp);
76 for (; tmp < ext4_inode_table(sb, gdp) +
77 sbi->s_itb_per_group; tmp++) {
78 if (!ext4_block_in_group(sb, tmp, block_group))
79 used_blocks -= 1;
80 }
81 }
82 return used_blocks;
83}
c2ea3fde 84
717d50e4
AD
85/* Initializes an uninitialized block bitmap if given, and returns the
86 * number of blocks free in the group. */
87unsigned ext4_init_block_bitmap(struct super_block *sb, struct buffer_head *bh,
fd2d4291 88 ext4_group_t block_group, struct ext4_group_desc *gdp)
717d50e4 89{
717d50e4 90 int bit, bit_max;
8df9675f 91 ext4_group_t ngroups = ext4_get_groups_count(sb);
717d50e4
AD
92 unsigned free_blocks, group_blocks;
93 struct ext4_sb_info *sbi = EXT4_SB(sb);
94
95 if (bh) {
96 J_ASSERT_BH(bh, buffer_locked(bh));
97
98 /* If checksum is bad mark all blocks used to prevent allocation
99 * essentially implementing a per-group read-only flag. */
100 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
46e665e9 101 ext4_error(sb, __func__,
a9df9a49 102 "Checksum bad for group %u", block_group);
560671a0
AK
103 ext4_free_blks_set(sb, gdp, 0);
104 ext4_free_inodes_set(sb, gdp, 0);
105 ext4_itable_unused_set(sb, gdp, 0);
717d50e4
AD
106 memset(bh->b_data, 0xff, sb->s_blocksize);
107 return 0;
108 }
109 memset(bh->b_data, 0, sb->s_blocksize);
110 }
111
112 /* Check for superblock and gdt backups in this group */
113 bit_max = ext4_bg_has_super(sb, block_group);
114
115 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
116 block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) *
117 sbi->s_desc_per_block) {
118 if (bit_max) {
119 bit_max += ext4_bg_num_gdb(sb, block_group);
120 bit_max +=
121 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
122 }
123 } else { /* For META_BG_BLOCK_GROUPS */
6afd6707 124 bit_max += ext4_bg_num_gdb(sb, block_group);
717d50e4
AD
125 }
126
8df9675f 127 if (block_group == ngroups - 1) {
717d50e4
AD
128 /*
129 * Even though mke2fs always initialize first and last group
130 * if some other tool enabled the EXT4_BG_BLOCK_UNINIT we need
131 * to make sure we calculate the right free blocks
132 */
133 group_blocks = ext4_blocks_count(sbi->s_es) -
134 le32_to_cpu(sbi->s_es->s_first_data_block) -
8df9675f 135 (EXT4_BLOCKS_PER_GROUP(sb) * (ngroups - 1));
717d50e4
AD
136 } else {
137 group_blocks = EXT4_BLOCKS_PER_GROUP(sb);
138 }
139
140 free_blocks = group_blocks - bit_max;
141
142 if (bh) {
0bf7e837
JS
143 ext4_fsblk_t start, tmp;
144 int flex_bg = 0;
d00a6d7b 145
717d50e4
AD
146 for (bit = 0; bit < bit_max; bit++)
147 ext4_set_bit(bit, bh->b_data);
148
d00a6d7b 149 start = ext4_group_first_block_no(sb, block_group);
717d50e4 150
0bf7e837
JS
151 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
152 EXT4_FEATURE_INCOMPAT_FLEX_BG))
153 flex_bg = 1;
717d50e4 154
0bf7e837
JS
155 /* Set bits for block and inode bitmaps, and inode table */
156 tmp = ext4_block_bitmap(sb, gdp);
157 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
158 ext4_set_bit(tmp - start, bh->b_data);
159
160 tmp = ext4_inode_bitmap(sb, gdp);
161 if (!flex_bg || ext4_block_in_group(sb, tmp, block_group))
162 ext4_set_bit(tmp - start, bh->b_data);
163
164 tmp = ext4_inode_table(sb, gdp);
165 for (; tmp < ext4_inode_table(sb, gdp) +
166 sbi->s_itb_per_group; tmp++) {
167 if (!flex_bg ||
168 ext4_block_in_group(sb, tmp, block_group))
169 ext4_set_bit(tmp - start, bh->b_data);
170 }
717d50e4
AD
171 /*
172 * Also if the number of blocks within the group is
173 * less than the blocksize * 8 ( which is the size
174 * of bitmap ), set rest of the block bitmap to 1
175 */
176 mark_bitmap_end(group_blocks, sb->s_blocksize * 8, bh->b_data);
177 }
e187c658 178 return free_blocks - ext4_group_used_meta_blocks(sb, block_group, gdp);
717d50e4
AD
179}
180
181
ac27a0ec
DK
182/*
183 * The free blocks are managed by bitmaps. A file system contains several
184 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
185 * block for inodes, N blocks for the inode table and data blocks.
186 *
187 * The file system contains group descriptors which are located after the
188 * super block. Each descriptor contains the number of the bitmap block and
189 * the free blocks count in the block. The descriptors are loaded in memory
e627432c 190 * when a file system is mounted (see ext4_fill_super).
ac27a0ec
DK
191 */
192
193
194#define in_range(b, first, len) ((b) >= (first) && (b) <= (first) + (len) - 1)
195
196/**
617ba13b 197 * ext4_get_group_desc() -- load group descriptor from disk
ac27a0ec
DK
198 * @sb: super block
199 * @block_group: given block group
200 * @bh: pointer to the buffer head to store the block
201 * group descriptor
202 */
af5bc92d 203struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb,
fd2d4291 204 ext4_group_t block_group,
af5bc92d 205 struct buffer_head **bh)
ac27a0ec 206{
498e5f24
TT
207 unsigned int group_desc;
208 unsigned int offset;
8df9675f 209 ext4_group_t ngroups = ext4_get_groups_count(sb);
af5bc92d 210 struct ext4_group_desc *desc;
617ba13b 211 struct ext4_sb_info *sbi = EXT4_SB(sb);
ac27a0ec 212
8df9675f 213 if (block_group >= ngroups) {
af5bc92d
TT
214 ext4_error(sb, "ext4_get_group_desc",
215 "block_group >= groups_count - "
a9df9a49 216 "block_group = %u, groups_count = %u",
8df9675f 217 block_group, ngroups);
ac27a0ec
DK
218
219 return NULL;
220 }
ac27a0ec 221
617ba13b
MC
222 group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb);
223 offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1);
ac27a0ec 224 if (!sbi->s_group_desc[group_desc]) {
af5bc92d
TT
225 ext4_error(sb, "ext4_get_group_desc",
226 "Group descriptor not loaded - "
498e5f24 227 "block_group = %u, group_desc = %u, desc = %u",
af5bc92d 228 block_group, group_desc, offset);
ac27a0ec
DK
229 return NULL;
230 }
231
0d1ee42f
AR
232 desc = (struct ext4_group_desc *)(
233 (__u8 *)sbi->s_group_desc[group_desc]->b_data +
234 offset * EXT4_DESC_SIZE(sb));
ac27a0ec
DK
235 if (bh)
236 *bh = sbi->s_group_desc[group_desc];
0d1ee42f 237 return desc;
ac27a0ec
DK
238}
239
abcb2947
AK
240static int ext4_valid_block_bitmap(struct super_block *sb,
241 struct ext4_group_desc *desc,
242 unsigned int block_group,
243 struct buffer_head *bh)
244{
245 ext4_grpblk_t offset;
246 ext4_grpblk_t next_zero_bit;
247 ext4_fsblk_t bitmap_blk;
248 ext4_fsblk_t group_first_block;
249
250 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
251 /* with FLEX_BG, the inode/block bitmaps and itable
252 * blocks may not be in the group at all
253 * so the bitmap validation will be skipped for those groups
254 * or it has to also read the block group where the bitmaps
255 * are located to verify they are set.
256 */
257 return 1;
258 }
259 group_first_block = ext4_group_first_block_no(sb, block_group);
260
261 /* check whether block bitmap block number is set */
262 bitmap_blk = ext4_block_bitmap(sb, desc);
263 offset = bitmap_blk - group_first_block;
264 if (!ext4_test_bit(offset, bh->b_data))
265 /* bad block bitmap */
266 goto err_out;
267
268 /* check whether the inode bitmap block number is set */
269 bitmap_blk = ext4_inode_bitmap(sb, desc);
270 offset = bitmap_blk - group_first_block;
271 if (!ext4_test_bit(offset, bh->b_data))
272 /* bad block bitmap */
273 goto err_out;
274
275 /* check whether the inode table block number is set */
276 bitmap_blk = ext4_inode_table(sb, desc);
277 offset = bitmap_blk - group_first_block;
278 next_zero_bit = ext4_find_next_zero_bit(bh->b_data,
279 offset + EXT4_SB(sb)->s_itb_per_group,
280 offset);
281 if (next_zero_bit >= offset + EXT4_SB(sb)->s_itb_per_group)
282 /* good bitmap for inode tables */
283 return 1;
284
285err_out:
46e665e9 286 ext4_error(sb, __func__,
abcb2947
AK
287 "Invalid block bitmap - "
288 "block_group = %d, block = %llu",
289 block_group, bitmap_blk);
290 return 0;
291}
ac27a0ec 292/**
574ca174 293 * ext4_read_block_bitmap()
ac27a0ec
DK
294 * @sb: super block
295 * @block_group: given block group
296 *
abcb2947
AK
297 * Read the bitmap for a given block_group,and validate the
298 * bits for block/inode/inode tables are set in the bitmaps
ac27a0ec
DK
299 *
300 * Return buffer_head on success or NULL in case of failure.
301 */
717d50e4 302struct buffer_head *
574ca174 303ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group)
ac27a0ec 304{
af5bc92d
TT
305 struct ext4_group_desc *desc;
306 struct buffer_head *bh = NULL;
7c9e69fa 307 ext4_fsblk_t bitmap_blk;
ac27a0ec 308
717d50e4 309 desc = ext4_get_group_desc(sb, block_group, NULL);
ac27a0ec 310 if (!desc)
7c9e69fa
AK
311 return NULL;
312 bitmap_blk = ext4_block_bitmap(sb, desc);
abcb2947
AK
313 bh = sb_getblk(sb, bitmap_blk);
314 if (unlikely(!bh)) {
46e665e9 315 ext4_error(sb, __func__,
abcb2947 316 "Cannot read block bitmap - "
a9df9a49 317 "block_group = %u, block_bitmap = %llu",
e29d1cde 318 block_group, bitmap_blk);
abcb2947
AK
319 return NULL;
320 }
2ccb5fb9
AK
321
322 if (bitmap_uptodate(bh))
abcb2947
AK
323 return bh;
324
c806e68f 325 lock_buffer(bh);
2ccb5fb9
AK
326 if (bitmap_uptodate(bh)) {
327 unlock_buffer(bh);
328 return bh;
329 }
b5f10eed 330 spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
717d50e4 331 if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
abcb2947 332 ext4_init_block_bitmap(sb, bh, block_group, desc);
2ccb5fb9 333 set_bitmap_uptodate(bh);
abcb2947 334 set_buffer_uptodate(bh);
b5f10eed 335 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
3300beda 336 unlock_buffer(bh);
abcb2947 337 return bh;
717d50e4 338 }
b5f10eed 339 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
2ccb5fb9
AK
340 if (buffer_uptodate(bh)) {
341 /*
342 * if not uninit if bh is uptodate,
343 * bitmap is also uptodate
344 */
345 set_bitmap_uptodate(bh);
346 unlock_buffer(bh);
347 return bh;
348 }
349 /*
350 * submit the buffer_head for read. We can
351 * safely mark the bitmap as uptodate now.
352 * We do it here so the bitmap uptodate bit
353 * get set with buffer lock held.
354 */
355 set_bitmap_uptodate(bh);
abcb2947
AK
356 if (bh_submit_read(bh) < 0) {
357 put_bh(bh);
46e665e9 358 ext4_error(sb, __func__,
ac27a0ec 359 "Cannot read block bitmap - "
a9df9a49 360 "block_group = %u, block_bitmap = %llu",
e29d1cde 361 block_group, bitmap_blk);
abcb2947
AK
362 return NULL;
363 }
519deca0
AK
364 ext4_valid_block_bitmap(sb, desc, block_group, bh);
365 /*
366 * file system mounted not to panic on error,
367 * continue with corrupt bitmap
368 */
ac27a0ec
DK
369 return bh;
370}
ac27a0ec
DK
371
372/**
e21675d4 373 * ext4_add_groupblocks() -- Add given blocks to an existing group
ac27a0ec
DK
374 * @handle: handle to this transaction
375 * @sb: super block
e21675d4 376 * @block: start physcial block to add to the block group
ac27a0ec 377 * @count: number of blocks to free
c2ea3fde 378 *
e21675d4
AK
379 * This marks the blocks as free in the bitmap. We ask the
380 * mballoc to reload the buddy after this by setting group
381 * EXT4_GROUP_INFO_NEED_INIT_BIT flag
ac27a0ec 382 */
e21675d4
AK
383void ext4_add_groupblocks(handle_t *handle, struct super_block *sb,
384 ext4_fsblk_t block, unsigned long count)
ac27a0ec
DK
385{
386 struct buffer_head *bitmap_bh = NULL;
387 struct buffer_head *gd_bh;
fd2d4291 388 ext4_group_t block_group;
617ba13b 389 ext4_grpblk_t bit;
498e5f24 390 unsigned int i;
af5bc92d
TT
391 struct ext4_group_desc *desc;
392 struct ext4_super_block *es;
617ba13b 393 struct ext4_sb_info *sbi;
560671a0 394 int err = 0, ret, blk_free_count;
e21675d4
AK
395 ext4_grpblk_t blocks_freed;
396 struct ext4_group_info *grp;
ac27a0ec 397
617ba13b 398 sbi = EXT4_SB(sb);
ac27a0ec 399 es = sbi->s_es;
e21675d4 400 ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
ac27a0ec 401
3a5b2ecd 402 ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
920313a7 403 grp = ext4_get_group_info(sb, block_group);
ac27a0ec
DK
404 /*
405 * Check to see if we are freeing blocks across a group
406 * boundary.
407 */
617ba13b 408 if (bit + count > EXT4_BLOCKS_PER_GROUP(sb)) {
e21675d4 409 goto error_return;
ac27a0ec 410 }
574ca174 411 bitmap_bh = ext4_read_block_bitmap(sb, block_group);
ac27a0ec
DK
412 if (!bitmap_bh)
413 goto error_return;
af5bc92d 414 desc = ext4_get_group_desc(sb, block_group, &gd_bh);
ac27a0ec
DK
415 if (!desc)
416 goto error_return;
417
8fadc143
AR
418 if (in_range(ext4_block_bitmap(sb, desc), block, count) ||
419 in_range(ext4_inode_bitmap(sb, desc), block, count) ||
420 in_range(block, ext4_inode_table(sb, desc), sbi->s_itb_per_group) ||
421 in_range(block + count - 1, ext4_inode_table(sb, desc),
cb47dce7 422 sbi->s_itb_per_group)) {
e21675d4
AK
423 ext4_error(sb, __func__,
424 "Adding blocks in system zones - "
af5bc92d
TT
425 "Block = %llu, count = %lu",
426 block, count);
cb47dce7
AK
427 goto error_return;
428 }
ac27a0ec
DK
429
430 /*
e21675d4 431 * We are about to add blocks to the bitmap,
ac27a0ec
DK
432 * so we need undo access.
433 */
ac27a0ec 434 BUFFER_TRACE(bitmap_bh, "getting undo access");
617ba13b 435 err = ext4_journal_get_undo_access(handle, bitmap_bh);
ac27a0ec
DK
436 if (err)
437 goto error_return;
438
439 /*
440 * We are about to modify some metadata. Call the journal APIs
441 * to unshare ->b_data if a currently-committing transaction is
442 * using it
443 */
444 BUFFER_TRACE(gd_bh, "get_write_access");
617ba13b 445 err = ext4_journal_get_write_access(handle, gd_bh);
ac27a0ec
DK
446 if (err)
447 goto error_return;
920313a7
AK
448 /*
449 * make sure we don't allow a parallel init on other groups in the
450 * same buddy cache
451 */
452 down_write(&grp->alloc_sem);
e21675d4 453 for (i = 0, blocks_freed = 0; i < count; i++) {
ac27a0ec 454 BUFFER_TRACE(bitmap_bh, "clear bit");
617ba13b 455 if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
ac27a0ec 456 bit + i, bitmap_bh->b_data)) {
46e665e9 457 ext4_error(sb, __func__,
2ae02107 458 "bit already cleared for block %llu",
bd81d8ee 459 (ext4_fsblk_t)(block + i));
ac27a0ec
DK
460 BUFFER_TRACE(bitmap_bh, "bit already cleared");
461 } else {
e21675d4 462 blocks_freed++;
ac27a0ec
DK
463 }
464 }
ac27a0ec 465 spin_lock(sb_bgl_lock(sbi, block_group));
560671a0
AK
466 blk_free_count = blocks_freed + ext4_free_blks_count(sb, desc);
467 ext4_free_blks_set(sb, desc, blk_free_count);
717d50e4 468 desc->bg_checksum = ext4_group_desc_csum(sbi, block_group, desc);
ac27a0ec 469 spin_unlock(sb_bgl_lock(sbi, block_group));
e21675d4 470 percpu_counter_add(&sbi->s_freeblocks_counter, blocks_freed);
ac27a0ec 471
772cb7c8
JS
472 if (sbi->s_log_groups_per_flex) {
473 ext4_group_t flex_group = ext4_flex_group(sbi, block_group);
9f24e420
TT
474 atomic_add(blocks_freed,
475 &sbi->s_flex_groups[flex_group].free_blocks);
772cb7c8 476 }
920313a7
AK
477 /*
478 * request to reload the buddy with the
479 * new bitmap information
480 */
481 set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
482 ext4_mb_update_group_info(grp, blocks_freed);
483 up_write(&grp->alloc_sem);
772cb7c8 484
ac27a0ec
DK
485 /* We dirtied the bitmap block */
486 BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
0390131b 487 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
ac27a0ec
DK
488
489 /* And the group descriptor block */
490 BUFFER_TRACE(gd_bh, "dirtied group descriptor block");
0390131b 491 ret = ext4_handle_dirty_metadata(handle, NULL, gd_bh);
e21675d4
AK
492 if (!err)
493 err = ret;
ac27a0ec 494 sb->s_dirt = 1;
e21675d4 495
ac27a0ec
DK
496error_return:
497 brelse(bitmap_bh);
617ba13b 498 ext4_std_error(sb, err);
ac27a0ec
DK
499 return;
500}
501
502/**
617ba13b 503 * ext4_free_blocks() -- Free given blocks and update quota
ac27a0ec
DK
504 * @handle: handle for this transaction
505 * @inode: inode
506 * @block: start physical block to free
507 * @count: number of blocks to count
c9de560d 508 * @metadata: Are these metadata blocks
ac27a0ec 509 */
617ba13b 510void ext4_free_blocks(handle_t *handle, struct inode *inode,
c9de560d
AT
511 ext4_fsblk_t block, unsigned long count,
512 int metadata)
ac27a0ec 513{
af5bc92d 514 struct super_block *sb;
ac27a0ec
DK
515 unsigned long dquot_freed_blocks;
516
c9de560d
AT
517 /* this isn't the right place to decide whether block is metadata
518 * inode.c/extents.c knows better, but for safety ... */
a1aebc1e
AK
519 if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))
520 metadata = 1;
521
522 /* We need to make sure we don't reuse
523 * block released untill the transaction commit.
524 * writeback mode have weak data consistency so
525 * don't force data as metadata when freeing block
526 * for writeback mode.
527 */
528 if (metadata == 0 && !ext4_should_writeback_data(inode))
c9de560d
AT
529 metadata = 1;
530
ac27a0ec 531 sb = inode->i_sb;
c9de560d 532
c2ea3fde
TT
533 ext4_mb_free_blocks(handle, inode, block, count,
534 metadata, &dquot_freed_blocks);
ac27a0ec 535 if (dquot_freed_blocks)
a269eb18 536 vfs_dq_free_block(inode, dquot_freed_blocks);
ac27a0ec
DK
537 return;
538}
539
8c3bf8a0
ES
540/**
541 * ext4_has_free_blocks()
542 * @sbi: in-core super block structure.
543 * @nblocks: number of needed blocks
544 *
545 * Check if filesystem has nblocks free & available for allocation.
546 * On success return 1, return 0 on failure.
547 */
548int ext4_has_free_blocks(struct ext4_sb_info *sbi, s64 nblocks)
a30d542a 549{
a996031c 550 s64 free_blocks, dirty_blocks, root_blocks;
a30d542a 551 struct percpu_counter *fbc = &sbi->s_freeblocks_counter;
6bc6e63f 552 struct percpu_counter *dbc = &sbi->s_dirtyblocks_counter;
a30d542a 553
6bc6e63f
AK
554 free_blocks = percpu_counter_read_positive(fbc);
555 dirty_blocks = percpu_counter_read_positive(dbc);
a996031c 556 root_blocks = ext4_r_blocks_count(sbi->s_es);
a30d542a 557
6bc6e63f
AK
558 if (free_blocks - (nblocks + root_blocks + dirty_blocks) <
559 EXT4_FREEBLOCKS_WATERMARK) {
02d21168
AM
560 free_blocks = percpu_counter_sum_positive(fbc);
561 dirty_blocks = percpu_counter_sum_positive(dbc);
6bc6e63f
AK
562 if (dirty_blocks < 0) {
563 printk(KERN_CRIT "Dirty block accounting "
564 "went wrong %lld\n",
8f72fbdf 565 (long long)dirty_blocks);
6bc6e63f
AK
566 }
567 }
568 /* Check whether we have space after
a996031c 569 * accounting for current dirty blocks & root reserved blocks.
6bc6e63f 570 */
a996031c
ES
571 if (free_blocks >= ((root_blocks + nblocks) + dirty_blocks))
572 return 1;
a30d542a 573
a996031c 574 /* Hm, nope. Are (enough) root reserved blocks available? */
4c9c544e 575 if (sbi->s_resuid == current_fsuid() ||
a996031c
ES
576 ((sbi->s_resgid != 0) && in_group_p(sbi->s_resgid)) ||
577 capable(CAP_SYS_RESOURCE)) {
578 if (free_blocks >= (nblocks + dirty_blocks))
579 return 1;
580 }
581
582 return 0;
a30d542a
AK
583}
584
8c3bf8a0 585int ext4_claim_free_blocks(struct ext4_sb_info *sbi,
5c791616 586 s64 nblocks)
ac27a0ec 587{
8c3bf8a0
ES
588 if (ext4_has_free_blocks(sbi, nblocks)) {
589 percpu_counter_add(&sbi->s_dirtyblocks_counter, nblocks);
16eb7295 590 return 0;
8c3bf8a0
ES
591 } else
592 return -ENOSPC;
a30d542a 593}
07031431 594
ac27a0ec 595/**
617ba13b 596 * ext4_should_retry_alloc()
ac27a0ec
DK
597 * @sb: super block
598 * @retries number of attemps has been made
599 *
617ba13b 600 * ext4_should_retry_alloc() is called when ENOSPC is returned, and if
ac27a0ec
DK
601 * it is profitable to retry the operation, this function will wait
602 * for the current or commiting transaction to complete, and then
603 * return TRUE.
604 *
605 * if the total number of retries exceed three times, return FALSE.
606 */
617ba13b 607int ext4_should_retry_alloc(struct super_block *sb, int *retries)
ac27a0ec 608{
8f64b32e
ES
609 if (!ext4_has_free_blocks(EXT4_SB(sb), 1) ||
610 (*retries)++ > 3 ||
611 !EXT4_SB(sb)->s_journal)
ac27a0ec
DK
612 return 0;
613
614 jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id);
615
dab291af 616 return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal);
ac27a0ec
DK
617}
618
654b4908 619/*
d2a17637 620 * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks
654b4908
AK
621 *
622 * @handle: handle to this transaction
623 * @inode: file inode
624 * @goal: given target block(filesystem wide)
97df5d15 625 * @count: pointer to total number of blocks needed
654b4908
AK
626 * @errp: error code
627 *
97df5d15 628 * Return 1st allocated block number on success, *count stores total account
d2a17637 629 * error stores in errp pointer
654b4908 630 */
d2a17637
MC
631ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode,
632 ext4_fsblk_t goal, unsigned long *count, int *errp)
654b4908 633{
97df5d15 634 struct ext4_allocation_request ar;
d2a17637 635 ext4_fsblk_t ret;
97df5d15
TT
636
637 memset(&ar, 0, sizeof(ar));
638 /* Fill with neighbour allocated blocks */
639 ar.inode = inode;
640 ar.goal = goal;
641 ar.len = count ? *count : 1;
642
643 ret = ext4_mb_new_blocks(handle, &ar, errp);
644 if (count)
645 *count = ar.len;
646
d2a17637
MC
647 /*
648 * Account for the allocated meta blocks
649 */
166348dd 650 if (!(*errp) && EXT4_I(inode)->i_delalloc_reserved_flag) {
d2a17637 651 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
97df5d15 652 EXT4_I(inode)->i_allocated_meta_blocks += ar.len;
d2a17637
MC
653 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
654 }
655 return ret;
654b4908
AK
656}
657
ac27a0ec 658/**
617ba13b 659 * ext4_count_free_blocks() -- count filesystem free blocks
ac27a0ec
DK
660 * @sb: superblock
661 *
662 * Adds up the number of free blocks from each block group.
663 */
617ba13b 664ext4_fsblk_t ext4_count_free_blocks(struct super_block *sb)
ac27a0ec 665{
617ba13b
MC
666 ext4_fsblk_t desc_count;
667 struct ext4_group_desc *gdp;
fd2d4291 668 ext4_group_t i;
8df9675f 669 ext4_group_t ngroups = ext4_get_groups_count(sb);
617ba13b
MC
670#ifdef EXT4FS_DEBUG
671 struct ext4_super_block *es;
672 ext4_fsblk_t bitmap_count;
498e5f24 673 unsigned int x;
ac27a0ec
DK
674 struct buffer_head *bitmap_bh = NULL;
675
617ba13b 676 es = EXT4_SB(sb)->s_es;
ac27a0ec
DK
677 desc_count = 0;
678 bitmap_count = 0;
679 gdp = NULL;
680
ac27a0ec 681 for (i = 0; i < ngroups; i++) {
617ba13b 682 gdp = ext4_get_group_desc(sb, i, NULL);
ac27a0ec
DK
683 if (!gdp)
684 continue;
9fd9784c 685 desc_count += ext4_free_blks_count(sb, gdp);
ac27a0ec 686 brelse(bitmap_bh);
574ca174 687 bitmap_bh = ext4_read_block_bitmap(sb, i);
ac27a0ec
DK
688 if (bitmap_bh == NULL)
689 continue;
690
617ba13b 691 x = ext4_count_free(bitmap_bh, sb->s_blocksize);
9fd9784c
TLSC
692 printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n",
693 i, ext4_free_blks_count(sb, gdp), x);
ac27a0ec
DK
694 bitmap_count += x;
695 }
696 brelse(bitmap_bh);
4776004f
TT
697 printk(KERN_DEBUG "ext4_count_free_blocks: stored = %llu"
698 ", computed = %llu, %llu\n", ext4_free_blocks_count(es),
699 desc_count, bitmap_count);
ac27a0ec
DK
700 return bitmap_count;
701#else
702 desc_count = 0;
ac27a0ec 703 for (i = 0; i < ngroups; i++) {
617ba13b 704 gdp = ext4_get_group_desc(sb, i, NULL);
ac27a0ec
DK
705 if (!gdp)
706 continue;
560671a0 707 desc_count += ext4_free_blks_count(sb, gdp);
ac27a0ec
DK
708 }
709
710 return desc_count;
711#endif
712}
713
fd2d4291 714static inline int test_root(ext4_group_t a, int b)
ac27a0ec
DK
715{
716 int num = b;
717
718 while (a > num)
719 num *= b;
720 return num == a;
721}
722
fd2d4291 723static int ext4_group_sparse(ext4_group_t group)
ac27a0ec
DK
724{
725 if (group <= 1)
726 return 1;
727 if (!(group & 1))
728 return 0;
729 return (test_root(group, 7) || test_root(group, 5) ||
730 test_root(group, 3));
731}
732
733/**
617ba13b 734 * ext4_bg_has_super - number of blocks used by the superblock in group
ac27a0ec
DK
735 * @sb: superblock for filesystem
736 * @group: group number to check
737 *
738 * Return the number of blocks used by the superblock (primary or backup)
739 * in this group. Currently this will be only 0 or 1.
740 */
fd2d4291 741int ext4_bg_has_super(struct super_block *sb, ext4_group_t group)
ac27a0ec 742{
617ba13b
MC
743 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
744 EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) &&
745 !ext4_group_sparse(group))
ac27a0ec
DK
746 return 0;
747 return 1;
748}
749
fd2d4291
AM
750static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb,
751 ext4_group_t group)
ac27a0ec 752{
617ba13b 753 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
fd2d4291
AM
754 ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb);
755 ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1;
ac27a0ec
DK
756
757 if (group == first || group == first + 1 || group == last)
758 return 1;
759 return 0;
760}
761
fd2d4291
AM
762static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb,
763 ext4_group_t group)
ac27a0ec 764{
859cb936 765 return ext4_bg_has_super(sb, group) ? EXT4_SB(sb)->s_gdb_count : 0;
ac27a0ec
DK
766}
767
768/**
617ba13b 769 * ext4_bg_num_gdb - number of blocks used by the group table in group
ac27a0ec
DK
770 * @sb: superblock for filesystem
771 * @group: group number to check
772 *
773 * Return the number of blocks used by the group descriptor table
774 * (primary or backup) in this group. In the future there may be a
775 * different number of descriptor blocks in each group.
776 */
fd2d4291 777unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group)
ac27a0ec
DK
778{
779 unsigned long first_meta_bg =
617ba13b
MC
780 le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg);
781 unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb);
ac27a0ec 782
617ba13b 783 if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) ||
ac27a0ec 784 metagroup < first_meta_bg)
af5bc92d 785 return ext4_bg_num_gdb_nometa(sb, group);
ac27a0ec 786
617ba13b 787 return ext4_bg_num_gdb_meta(sb,group);
ac27a0ec
DK
788
789}
c2ea3fde 790