Commit | Line | Data |
---|---|---|
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 |
26 | static 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 | ||
bd86298e LC |
32 | /* |
33 | * Calculate block group number for a given block number | |
34 | */ | |
35 | ext4_group_t ext4_get_group_number(struct super_block *sb, | |
36 | ext4_fsblk_t block) | |
37 | { | |
38 | ext4_group_t group; | |
39 | ||
40 | if (test_opt2(sb, STD_GROUP_SIZE)) | |
960fd856 TT |
41 | group = (block - |
42 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) >> | |
bd86298e LC |
43 | (EXT4_BLOCK_SIZE_BITS(sb) + EXT4_CLUSTER_BITS(sb) + 3); |
44 | else | |
45 | ext4_get_group_no_and_offset(sb, block, &group, NULL); | |
46 | return group; | |
47 | } | |
48 | ||
72b64b59 | 49 | /* |
3212a80a TT |
50 | * Calculate the block group number and offset into the block/cluster |
51 | * allocation bitmap, given a block number | |
72b64b59 AM |
52 | */ |
53 | void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr, | |
fd2d4291 | 54 | ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp) |
72b64b59 | 55 | { |
8c55e204 | 56 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
72b64b59 AM |
57 | ext4_grpblk_t offset; |
58 | ||
8c55e204 | 59 | blocknr = blocknr - le32_to_cpu(es->s_first_data_block); |
3212a80a TT |
60 | offset = do_div(blocknr, EXT4_BLOCKS_PER_GROUP(sb)) >> |
61 | EXT4_SB(sb)->s_cluster_bits; | |
72b64b59 AM |
62 | if (offsetp) |
63 | *offsetp = offset; | |
64 | if (blockgrpp) | |
8c55e204 | 65 | *blockgrpp = blocknr; |
72b64b59 AM |
66 | |
67 | } | |
68 | ||
68911009 LC |
69 | /* |
70 | * Check whether the 'block' lives within the 'block_group'. Returns 1 if so | |
71 | * and 0 otherwise. | |
72 | */ | |
73 | static inline int ext4_block_in_group(struct super_block *sb, | |
74 | ext4_fsblk_t block, | |
75 | ext4_group_t block_group) | |
0bf7e837 JS |
76 | { |
77 | ext4_group_t actual_group; | |
68911009 | 78 | |
bd86298e | 79 | actual_group = ext4_get_group_number(sb, block); |
68911009 | 80 | return (actual_group == block_group) ? 1 : 0; |
0bf7e837 JS |
81 | } |
82 | ||
d5b8f310 TT |
83 | /* Return the number of clusters used for file system metadata; this |
84 | * represents the overhead needed by the file system. | |
85 | */ | |
c197855e SH |
86 | static unsigned ext4_num_overhead_clusters(struct super_block *sb, |
87 | ext4_group_t block_group, | |
88 | struct ext4_group_desc *gdp) | |
0bf7e837 | 89 | { |
d5b8f310 TT |
90 | unsigned num_clusters; |
91 | int block_cluster = -1, inode_cluster = -1, itbl_cluster = -1, i, c; | |
92 | ext4_fsblk_t start = ext4_group_first_block_no(sb, block_group); | |
93 | ext4_fsblk_t itbl_blk; | |
0bf7e837 | 94 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
0bf7e837 | 95 | |
d5b8f310 TT |
96 | /* This is the number of clusters used by the superblock, |
97 | * block group descriptors, and reserved block group | |
98 | * descriptor blocks */ | |
99 | num_clusters = ext4_num_base_meta_clusters(sb, block_group); | |
100 | ||
101 | /* | |
102 | * For the allocation bitmaps and inode table, we first need | |
103 | * to check to see if the block is in the block group. If it | |
104 | * is, then check to see if the cluster is already accounted | |
105 | * for in the clusters used for the base metadata cluster, or | |
106 | * if we can increment the base metadata cluster to include | |
107 | * that block. Otherwise, we will have to track the cluster | |
108 | * used for the allocation bitmap or inode table explicitly. | |
109 | * Normally all of these blocks are contiguous, so the special | |
110 | * case handling shouldn't be necessary except for *very* | |
111 | * unusual file system layouts. | |
112 | */ | |
113 | if (ext4_block_in_group(sb, ext4_block_bitmap(sb, gdp), block_group)) { | |
b0dd6b70 TT |
114 | block_cluster = EXT4_B2C(sbi, |
115 | ext4_block_bitmap(sb, gdp) - start); | |
d5b8f310 TT |
116 | if (block_cluster < num_clusters) |
117 | block_cluster = -1; | |
118 | else if (block_cluster == num_clusters) { | |
119 | num_clusters++; | |
120 | block_cluster = -1; | |
121 | } | |
122 | } | |
123 | ||
124 | if (ext4_block_in_group(sb, ext4_inode_bitmap(sb, gdp), block_group)) { | |
125 | inode_cluster = EXT4_B2C(sbi, | |
b0dd6b70 | 126 | ext4_inode_bitmap(sb, gdp) - start); |
d5b8f310 TT |
127 | if (inode_cluster < num_clusters) |
128 | inode_cluster = -1; | |
129 | else if (inode_cluster == num_clusters) { | |
130 | num_clusters++; | |
131 | inode_cluster = -1; | |
132 | } | |
133 | } | |
134 | ||
135 | itbl_blk = ext4_inode_table(sb, gdp); | |
136 | for (i = 0; i < sbi->s_itb_per_group; i++) { | |
137 | if (ext4_block_in_group(sb, itbl_blk + i, block_group)) { | |
b0dd6b70 | 138 | c = EXT4_B2C(sbi, itbl_blk + i - start); |
d5b8f310 TT |
139 | if ((c < num_clusters) || (c == inode_cluster) || |
140 | (c == block_cluster) || (c == itbl_cluster)) | |
141 | continue; | |
142 | if (c == num_clusters) { | |
143 | num_clusters++; | |
144 | continue; | |
145 | } | |
146 | num_clusters++; | |
147 | itbl_cluster = c; | |
0bf7e837 JS |
148 | } |
149 | } | |
d5b8f310 TT |
150 | |
151 | if (block_cluster != -1) | |
152 | num_clusters++; | |
153 | if (inode_cluster != -1) | |
154 | num_clusters++; | |
155 | ||
156 | return num_clusters; | |
0bf7e837 | 157 | } |
c2ea3fde | 158 | |
d5b8f310 TT |
159 | static unsigned int num_clusters_in_group(struct super_block *sb, |
160 | ext4_group_t block_group) | |
49f7f9af | 161 | { |
d5b8f310 TT |
162 | unsigned int blocks; |
163 | ||
49f7f9af TT |
164 | if (block_group == ext4_get_groups_count(sb) - 1) { |
165 | /* | |
166 | * Even though mke2fs always initializes the first and | |
167 | * last group, just in case some other tool was used, | |
168 | * we need to make sure we calculate the right free | |
169 | * blocks. | |
170 | */ | |
d5b8f310 | 171 | blocks = ext4_blocks_count(EXT4_SB(sb)->s_es) - |
49f7f9af TT |
172 | ext4_group_first_block_no(sb, block_group); |
173 | } else | |
d5b8f310 TT |
174 | blocks = EXT4_BLOCKS_PER_GROUP(sb); |
175 | return EXT4_NUM_B2C(EXT4_SB(sb), blocks); | |
49f7f9af TT |
176 | } |
177 | ||
fd034a84 | 178 | /* Initializes an uninitialized block bitmap */ |
c197855e SH |
179 | static void ext4_init_block_bitmap(struct super_block *sb, |
180 | struct buffer_head *bh, | |
181 | ext4_group_t block_group, | |
182 | struct ext4_group_desc *gdp) | |
717d50e4 | 183 | { |
d5b8f310 | 184 | unsigned int bit, bit_max; |
717d50e4 | 185 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
fd034a84 TT |
186 | ext4_fsblk_t start, tmp; |
187 | int flex_bg = 0; | |
bdfb6ff4 | 188 | struct ext4_group_info *grp; |
fd034a84 TT |
189 | |
190 | J_ASSERT_BH(bh, buffer_locked(bh)); | |
191 | ||
192 | /* If checksum is bad mark all blocks used to prevent allocation | |
193 | * essentially implementing a per-group read-only flag. */ | |
feb0ab32 | 194 | if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) { |
fd034a84 | 195 | ext4_error(sb, "Checksum bad for group %u", block_group); |
bdfb6ff4 | 196 | grp = ext4_get_group_info(sb, block_group); |
e43bb4e6 NJ |
197 | if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) |
198 | percpu_counter_sub(&sbi->s_freeclusters_counter, | |
199 | grp->bb_free); | |
bdfb6ff4 | 200 | set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state); |
e43bb4e6 NJ |
201 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
202 | int count; | |
203 | count = ext4_free_inodes_count(sb, gdp); | |
204 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
205 | count); | |
206 | } | |
bdfb6ff4 | 207 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
fd034a84 | 208 | return; |
717d50e4 | 209 | } |
fd034a84 | 210 | memset(bh->b_data, 0, sb->s_blocksize); |
717d50e4 | 211 | |
d5b8f310 | 212 | bit_max = ext4_num_base_meta_clusters(sb, block_group); |
fd034a84 TT |
213 | for (bit = 0; bit < bit_max; bit++) |
214 | ext4_set_bit(bit, bh->b_data); | |
d00a6d7b | 215 | |
fd034a84 | 216 | start = ext4_group_first_block_no(sb, block_group); |
717d50e4 | 217 | |
fd034a84 TT |
218 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) |
219 | flex_bg = 1; | |
717d50e4 | 220 | |
fd034a84 TT |
221 | /* Set bits for block and inode bitmaps, and inode table */ |
222 | tmp = ext4_block_bitmap(sb, gdp); | |
223 | if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) | |
d5b8f310 | 224 | ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); |
717d50e4 | 225 | |
fd034a84 TT |
226 | tmp = ext4_inode_bitmap(sb, gdp); |
227 | if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) | |
d5b8f310 | 228 | ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); |
0bf7e837 | 229 | |
fd034a84 TT |
230 | tmp = ext4_inode_table(sb, gdp); |
231 | for (; tmp < ext4_inode_table(sb, gdp) + | |
232 | sbi->s_itb_per_group; tmp++) { | |
0bf7e837 | 233 | if (!flex_bg || ext4_block_in_group(sb, tmp, block_group)) |
d5b8f310 | 234 | ext4_set_bit(EXT4_B2C(sbi, tmp - start), bh->b_data); |
717d50e4 | 235 | } |
d5b8f310 | 236 | |
fd034a84 TT |
237 | /* |
238 | * Also if the number of blocks within the group is less than | |
239 | * the blocksize * 8 ( which is the size of bitmap ), set rest | |
240 | * of the block bitmap to 1 | |
241 | */ | |
d5b8f310 | 242 | ext4_mark_bitmap_end(num_clusters_in_group(sb, block_group), |
fd034a84 | 243 | sb->s_blocksize * 8, bh->b_data); |
79f1ba49 | 244 | ext4_block_bitmap_csum_set(sb, block_group, gdp, bh); |
feb0ab32 | 245 | ext4_group_desc_csum_set(sb, block_group, gdp); |
717d50e4 AD |
246 | } |
247 | ||
fd034a84 TT |
248 | /* Return the number of free blocks in a block group. It is used when |
249 | * the block bitmap is uninitialized, so we can't just count the bits | |
250 | * in the bitmap. */ | |
cff1dfd7 TT |
251 | unsigned ext4_free_clusters_after_init(struct super_block *sb, |
252 | ext4_group_t block_group, | |
253 | struct ext4_group_desc *gdp) | |
fd034a84 | 254 | { |
d5b8f310 TT |
255 | return num_clusters_in_group(sb, block_group) - |
256 | ext4_num_overhead_clusters(sb, block_group, gdp); | |
fd034a84 | 257 | } |
717d50e4 | 258 | |
ac27a0ec DK |
259 | /* |
260 | * The free blocks are managed by bitmaps. A file system contains several | |
261 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
262 | * block for inodes, N blocks for the inode table and data blocks. | |
263 | * | |
264 | * The file system contains group descriptors which are located after the | |
265 | * super block. Each descriptor contains the number of the bitmap block and | |
266 | * the free blocks count in the block. The descriptors are loaded in memory | |
e627432c | 267 | * when a file system is mounted (see ext4_fill_super). |
ac27a0ec DK |
268 | */ |
269 | ||
ac27a0ec | 270 | /** |
617ba13b | 271 | * ext4_get_group_desc() -- load group descriptor from disk |
ac27a0ec DK |
272 | * @sb: super block |
273 | * @block_group: given block group | |
274 | * @bh: pointer to the buffer head to store the block | |
275 | * group descriptor | |
276 | */ | |
af5bc92d | 277 | struct ext4_group_desc * ext4_get_group_desc(struct super_block *sb, |
fd2d4291 | 278 | ext4_group_t block_group, |
af5bc92d | 279 | struct buffer_head **bh) |
ac27a0ec | 280 | { |
498e5f24 TT |
281 | unsigned int group_desc; |
282 | unsigned int offset; | |
8df9675f | 283 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
af5bc92d | 284 | struct ext4_group_desc *desc; |
617ba13b | 285 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
ac27a0ec | 286 | |
8df9675f | 287 | if (block_group >= ngroups) { |
12062ddd ES |
288 | ext4_error(sb, "block_group >= groups_count - block_group = %u," |
289 | " groups_count = %u", block_group, ngroups); | |
ac27a0ec DK |
290 | |
291 | return NULL; | |
292 | } | |
ac27a0ec | 293 | |
617ba13b MC |
294 | group_desc = block_group >> EXT4_DESC_PER_BLOCK_BITS(sb); |
295 | offset = block_group & (EXT4_DESC_PER_BLOCK(sb) - 1); | |
ac27a0ec | 296 | if (!sbi->s_group_desc[group_desc]) { |
12062ddd | 297 | ext4_error(sb, "Group descriptor not loaded - " |
498e5f24 | 298 | "block_group = %u, group_desc = %u, desc = %u", |
af5bc92d | 299 | block_group, group_desc, offset); |
ac27a0ec DK |
300 | return NULL; |
301 | } | |
302 | ||
0d1ee42f AR |
303 | desc = (struct ext4_group_desc *)( |
304 | (__u8 *)sbi->s_group_desc[group_desc]->b_data + | |
305 | offset * EXT4_DESC_SIZE(sb)); | |
ac27a0ec DK |
306 | if (bh) |
307 | *bh = sbi->s_group_desc[group_desc]; | |
0d1ee42f | 308 | return desc; |
ac27a0ec DK |
309 | } |
310 | ||
7a4c5de2 TT |
311 | /* |
312 | * Return the block number which was discovered to be invalid, or 0 if | |
313 | * the block bitmap is valid. | |
314 | */ | |
315 | static ext4_fsblk_t ext4_valid_block_bitmap(struct super_block *sb, | |
316 | struct ext4_group_desc *desc, | |
dbde0abe | 317 | ext4_group_t block_group, |
7a4c5de2 | 318 | struct buffer_head *bh) |
abcb2947 | 319 | { |
e674e5cb | 320 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
abcb2947 AK |
321 | ext4_grpblk_t offset; |
322 | ext4_grpblk_t next_zero_bit; | |
7a4c5de2 | 323 | ext4_fsblk_t blk; |
abcb2947 AK |
324 | ext4_fsblk_t group_first_block; |
325 | ||
326 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) { | |
327 | /* with FLEX_BG, the inode/block bitmaps and itable | |
328 | * blocks may not be in the group at all | |
329 | * so the bitmap validation will be skipped for those groups | |
330 | * or it has to also read the block group where the bitmaps | |
331 | * are located to verify they are set. | |
332 | */ | |
7a4c5de2 | 333 | return 0; |
abcb2947 AK |
334 | } |
335 | group_first_block = ext4_group_first_block_no(sb, block_group); | |
336 | ||
337 | /* check whether block bitmap block number is set */ | |
7a4c5de2 TT |
338 | blk = ext4_block_bitmap(sb, desc); |
339 | offset = blk - group_first_block; | |
e674e5cb | 340 | if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data)) |
abcb2947 | 341 | /* bad block bitmap */ |
7a4c5de2 | 342 | return blk; |
abcb2947 AK |
343 | |
344 | /* check whether the inode bitmap block number is set */ | |
7a4c5de2 TT |
345 | blk = ext4_inode_bitmap(sb, desc); |
346 | offset = blk - group_first_block; | |
e674e5cb | 347 | if (!ext4_test_bit(EXT4_B2C(sbi, offset), bh->b_data)) |
abcb2947 | 348 | /* bad block bitmap */ |
7a4c5de2 | 349 | return blk; |
abcb2947 AK |
350 | |
351 | /* check whether the inode table block number is set */ | |
7a4c5de2 TT |
352 | blk = ext4_inode_table(sb, desc); |
353 | offset = blk - group_first_block; | |
abcb2947 | 354 | next_zero_bit = ext4_find_next_zero_bit(bh->b_data, |
e674e5cb DW |
355 | EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group), |
356 | EXT4_B2C(sbi, offset)); | |
357 | if (next_zero_bit < | |
358 | EXT4_B2C(sbi, offset + EXT4_SB(sb)->s_itb_per_group)) | |
7a4c5de2 TT |
359 | /* bad bitmap for inode tables */ |
360 | return blk; | |
abcb2947 AK |
361 | return 0; |
362 | } | |
fa77dcfa | 363 | |
c197855e SH |
364 | static void ext4_validate_block_bitmap(struct super_block *sb, |
365 | struct ext4_group_desc *desc, | |
366 | ext4_group_t block_group, | |
367 | struct buffer_head *bh) | |
fa77dcfa | 368 | { |
7a4c5de2 | 369 | ext4_fsblk_t blk; |
163a203d | 370 | struct ext4_group_info *grp = ext4_get_group_info(sb, block_group); |
e43bb4e6 | 371 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
7a4c5de2 | 372 | |
fa77dcfa DW |
373 | if (buffer_verified(bh)) |
374 | return; | |
375 | ||
376 | ext4_lock_group(sb, block_group); | |
7a4c5de2 TT |
377 | blk = ext4_valid_block_bitmap(sb, desc, block_group, bh); |
378 | if (unlikely(blk != 0)) { | |
379 | ext4_unlock_group(sb, block_group); | |
380 | ext4_error(sb, "bg %u: block %llu: invalid block bitmap", | |
381 | block_group, blk); | |
e43bb4e6 NJ |
382 | if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) |
383 | percpu_counter_sub(&sbi->s_freeclusters_counter, | |
384 | grp->bb_free); | |
163a203d | 385 | set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state); |
7a4c5de2 TT |
386 | return; |
387 | } | |
388 | if (unlikely(!ext4_block_bitmap_csum_verify(sb, block_group, | |
79f1ba49 | 389 | desc, bh))) { |
7a4c5de2 TT |
390 | ext4_unlock_group(sb, block_group); |
391 | ext4_error(sb, "bg %u: bad block bitmap checksum", block_group); | |
e43bb4e6 NJ |
392 | if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) |
393 | percpu_counter_sub(&sbi->s_freeclusters_counter, | |
394 | grp->bb_free); | |
163a203d | 395 | set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state); |
7a4c5de2 TT |
396 | return; |
397 | } | |
398 | set_buffer_verified(bh); | |
fa77dcfa DW |
399 | ext4_unlock_group(sb, block_group); |
400 | } | |
401 | ||
ac27a0ec | 402 | /** |
15b49132 | 403 | * ext4_read_block_bitmap_nowait() |
ac27a0ec DK |
404 | * @sb: super block |
405 | * @block_group: given block group | |
406 | * | |
abcb2947 AK |
407 | * Read the bitmap for a given block_group,and validate the |
408 | * bits for block/inode/inode tables are set in the bitmaps | |
ac27a0ec DK |
409 | * |
410 | * Return buffer_head on success or NULL in case of failure. | |
411 | */ | |
717d50e4 | 412 | struct buffer_head * |
813e5727 | 413 | ext4_read_block_bitmap_nowait(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 414 | { |
af5bc92d | 415 | struct ext4_group_desc *desc; |
813e5727 | 416 | struct buffer_head *bh; |
7c9e69fa | 417 | ext4_fsblk_t bitmap_blk; |
ac27a0ec | 418 | |
717d50e4 | 419 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 420 | if (!desc) |
7c9e69fa AK |
421 | return NULL; |
422 | bitmap_blk = ext4_block_bitmap(sb, desc); | |
abcb2947 AK |
423 | bh = sb_getblk(sb, bitmap_blk); |
424 | if (unlikely(!bh)) { | |
813e5727 TT |
425 | ext4_error(sb, "Cannot get buffer for block bitmap - " |
426 | "block_group = %u, block_bitmap = %llu", | |
427 | block_group, bitmap_blk); | |
abcb2947 AK |
428 | return NULL; |
429 | } | |
2ccb5fb9 AK |
430 | |
431 | if (bitmap_uptodate(bh)) | |
fa77dcfa | 432 | goto verify; |
abcb2947 | 433 | |
c806e68f | 434 | lock_buffer(bh); |
2ccb5fb9 AK |
435 | if (bitmap_uptodate(bh)) { |
436 | unlock_buffer(bh); | |
fa77dcfa | 437 | goto verify; |
2ccb5fb9 | 438 | } |
955ce5f5 | 439 | ext4_lock_group(sb, block_group); |
717d50e4 | 440 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
abcb2947 | 441 | ext4_init_block_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 442 | set_bitmap_uptodate(bh); |
abcb2947 | 443 | set_buffer_uptodate(bh); |
955ce5f5 | 444 | ext4_unlock_group(sb, block_group); |
3300beda | 445 | unlock_buffer(bh); |
abcb2947 | 446 | return bh; |
717d50e4 | 447 | } |
955ce5f5 | 448 | ext4_unlock_group(sb, block_group); |
2ccb5fb9 AK |
449 | if (buffer_uptodate(bh)) { |
450 | /* | |
451 | * if not uninit if bh is uptodate, | |
452 | * bitmap is also uptodate | |
453 | */ | |
454 | set_bitmap_uptodate(bh); | |
455 | unlock_buffer(bh); | |
fa77dcfa | 456 | goto verify; |
2ccb5fb9 AK |
457 | } |
458 | /* | |
813e5727 | 459 | * submit the buffer_head for reading |
2ccb5fb9 | 460 | */ |
813e5727 | 461 | set_buffer_new(bh); |
0562e0ba | 462 | trace_ext4_read_block_bitmap_load(sb, block_group); |
813e5727 TT |
463 | bh->b_end_io = ext4_end_bitmap_read; |
464 | get_bh(bh); | |
9f203507 | 465 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
813e5727 | 466 | return bh; |
fa77dcfa DW |
467 | verify: |
468 | ext4_validate_block_bitmap(sb, desc, block_group, bh); | |
48d9eb97 DW |
469 | if (buffer_verified(bh)) |
470 | return bh; | |
471 | put_bh(bh); | |
472 | return NULL; | |
813e5727 TT |
473 | } |
474 | ||
475 | /* Returns 0 on success, 1 on error */ | |
476 | int ext4_wait_block_bitmap(struct super_block *sb, ext4_group_t block_group, | |
477 | struct buffer_head *bh) | |
478 | { | |
479 | struct ext4_group_desc *desc; | |
480 | ||
481 | if (!buffer_new(bh)) | |
482 | return 0; | |
483 | desc = ext4_get_group_desc(sb, block_group, NULL); | |
484 | if (!desc) | |
485 | return 1; | |
486 | wait_on_buffer(bh); | |
487 | if (!buffer_uptodate(bh)) { | |
12062ddd | 488 | ext4_error(sb, "Cannot read block bitmap - " |
813e5727 | 489 | "block_group = %u, block_bitmap = %llu", |
d4dc462f | 490 | block_group, (unsigned long long) bh->b_blocknr); |
813e5727 | 491 | return 1; |
abcb2947 | 492 | } |
813e5727 TT |
493 | clear_buffer_new(bh); |
494 | /* Panic or remount fs read-only if block bitmap is invalid */ | |
fa77dcfa | 495 | ext4_validate_block_bitmap(sb, desc, block_group, bh); |
48d9eb97 DW |
496 | /* ...but check for error just in case errors=continue. */ |
497 | return !buffer_verified(bh); | |
813e5727 TT |
498 | } |
499 | ||
500 | struct buffer_head * | |
501 | ext4_read_block_bitmap(struct super_block *sb, ext4_group_t block_group) | |
502 | { | |
503 | struct buffer_head *bh; | |
504 | ||
505 | bh = ext4_read_block_bitmap_nowait(sb, block_group); | |
15b49132 EG |
506 | if (!bh) |
507 | return NULL; | |
813e5727 TT |
508 | if (ext4_wait_block_bitmap(sb, block_group, bh)) { |
509 | put_bh(bh); | |
510 | return NULL; | |
511 | } | |
ac27a0ec DK |
512 | return bh; |
513 | } | |
ac27a0ec | 514 | |
8c3bf8a0 | 515 | /** |
df55c99d | 516 | * ext4_has_free_clusters() |
8c3bf8a0 | 517 | * @sbi: in-core super block structure. |
df55c99d TT |
518 | * @nclusters: number of needed blocks |
519 | * @flags: flags from ext4_mb_new_blocks() | |
8c3bf8a0 | 520 | * |
df55c99d | 521 | * Check if filesystem has nclusters free & available for allocation. |
8c3bf8a0 ES |
522 | * On success return 1, return 0 on failure. |
523 | */ | |
df55c99d TT |
524 | static int ext4_has_free_clusters(struct ext4_sb_info *sbi, |
525 | s64 nclusters, unsigned int flags) | |
a30d542a | 526 | { |
27dd4385 | 527 | s64 free_clusters, dirty_clusters, rsv, resv_clusters; |
57042651 | 528 | struct percpu_counter *fcc = &sbi->s_freeclusters_counter; |
df55c99d | 529 | struct percpu_counter *dcc = &sbi->s_dirtyclusters_counter; |
a30d542a | 530 | |
df55c99d TT |
531 | free_clusters = percpu_counter_read_positive(fcc); |
532 | dirty_clusters = percpu_counter_read_positive(dcc); | |
27dd4385 | 533 | resv_clusters = atomic64_read(&sbi->s_resv_clusters); |
304e220f LC |
534 | |
535 | /* | |
536 | * r_blocks_count should always be multiple of the cluster ratio so | |
537 | * we are safe to do a plane bit shift only. | |
538 | */ | |
27dd4385 LC |
539 | rsv = (ext4_r_blocks_count(sbi->s_es) >> sbi->s_cluster_bits) + |
540 | resv_clusters; | |
a30d542a | 541 | |
27dd4385 | 542 | if (free_clusters - (nclusters + rsv + dirty_clusters) < |
df55c99d | 543 | EXT4_FREECLUSTERS_WATERMARK) { |
304e220f | 544 | free_clusters = percpu_counter_sum_positive(fcc); |
df55c99d | 545 | dirty_clusters = percpu_counter_sum_positive(dcc); |
6bc6e63f | 546 | } |
df55c99d TT |
547 | /* Check whether we have space after accounting for current |
548 | * dirty clusters & root reserved clusters. | |
6bc6e63f | 549 | */ |
27dd4385 | 550 | if (free_clusters >= (rsv + nclusters + dirty_clusters)) |
a996031c | 551 | return 1; |
a30d542a | 552 | |
df55c99d | 553 | /* Hm, nope. Are (enough) root reserved clusters available? */ |
08cefc7a EB |
554 | if (uid_eq(sbi->s_resuid, current_fsuid()) || |
555 | (!gid_eq(sbi->s_resgid, GLOBAL_ROOT_GID) && in_group_p(sbi->s_resgid)) || | |
55f020db | 556 | capable(CAP_SYS_RESOURCE) || |
27dd4385 | 557 | (flags & EXT4_MB_USE_ROOT_BLOCKS)) { |
55f020db | 558 | |
27dd4385 LC |
559 | if (free_clusters >= (nclusters + dirty_clusters + |
560 | resv_clusters)) | |
561 | return 1; | |
562 | } | |
563 | /* No free blocks. Let's see if we can dip into reserved pool */ | |
564 | if (flags & EXT4_MB_USE_RESERVED) { | |
df55c99d | 565 | if (free_clusters >= (nclusters + dirty_clusters)) |
a996031c ES |
566 | return 1; |
567 | } | |
568 | ||
569 | return 0; | |
a30d542a AK |
570 | } |
571 | ||
e7d5f315 TT |
572 | int ext4_claim_free_clusters(struct ext4_sb_info *sbi, |
573 | s64 nclusters, unsigned int flags) | |
ac27a0ec | 574 | { |
df55c99d | 575 | if (ext4_has_free_clusters(sbi, nclusters, flags)) { |
e7d5f315 | 576 | percpu_counter_add(&sbi->s_dirtyclusters_counter, nclusters); |
16eb7295 | 577 | return 0; |
8c3bf8a0 ES |
578 | } else |
579 | return -ENOSPC; | |
a30d542a | 580 | } |
07031431 | 581 | |
ac27a0ec | 582 | /** |
617ba13b | 583 | * ext4_should_retry_alloc() |
ac27a0ec DK |
584 | * @sb: super block |
585 | * @retries number of attemps has been made | |
586 | * | |
617ba13b | 587 | * ext4_should_retry_alloc() is called when ENOSPC is returned, and if |
ac27a0ec | 588 | * it is profitable to retry the operation, this function will wait |
25985edc | 589 | * for the current or committing transaction to complete, and then |
ac27a0ec DK |
590 | * return TRUE. |
591 | * | |
592 | * if the total number of retries exceed three times, return FALSE. | |
593 | */ | |
617ba13b | 594 | int ext4_should_retry_alloc(struct super_block *sb, int *retries) |
ac27a0ec | 595 | { |
df55c99d | 596 | if (!ext4_has_free_clusters(EXT4_SB(sb), 1, 0) || |
8f64b32e ES |
597 | (*retries)++ > 3 || |
598 | !EXT4_SB(sb)->s_journal) | |
ac27a0ec DK |
599 | return 0; |
600 | ||
601 | jbd_debug(1, "%s: retrying operation after ENOSPC\n", sb->s_id); | |
602 | ||
dab291af | 603 | return jbd2_journal_force_commit_nested(EXT4_SB(sb)->s_journal); |
ac27a0ec DK |
604 | } |
605 | ||
654b4908 | 606 | /* |
d2a17637 | 607 | * ext4_new_meta_blocks() -- allocate block for meta data (indexing) blocks |
654b4908 AK |
608 | * |
609 | * @handle: handle to this transaction | |
610 | * @inode: file inode | |
611 | * @goal: given target block(filesystem wide) | |
7b415bf6 | 612 | * @count: pointer to total number of clusters needed |
654b4908 AK |
613 | * @errp: error code |
614 | * | |
97df5d15 | 615 | * Return 1st allocated block number on success, *count stores total account |
d2a17637 | 616 | * error stores in errp pointer |
654b4908 | 617 | */ |
d2a17637 | 618 | ext4_fsblk_t ext4_new_meta_blocks(handle_t *handle, struct inode *inode, |
55f020db AH |
619 | ext4_fsblk_t goal, unsigned int flags, |
620 | unsigned long *count, int *errp) | |
654b4908 | 621 | { |
97df5d15 | 622 | struct ext4_allocation_request ar; |
d2a17637 | 623 | ext4_fsblk_t ret; |
97df5d15 TT |
624 | |
625 | memset(&ar, 0, sizeof(ar)); | |
626 | /* Fill with neighbour allocated blocks */ | |
627 | ar.inode = inode; | |
628 | ar.goal = goal; | |
629 | ar.len = count ? *count : 1; | |
55f020db | 630 | ar.flags = flags; |
97df5d15 TT |
631 | |
632 | ret = ext4_mb_new_blocks(handle, &ar, errp); | |
633 | if (count) | |
634 | *count = ar.len; | |
d2a17637 | 635 | /* |
72b8ab9d ES |
636 | * Account for the allocated meta blocks. We will never |
637 | * fail EDQUOT for metdata, but we do account for it. | |
d2a17637 | 638 | */ |
f2321097 TT |
639 | if (!(*errp) && |
640 | ext4_test_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED)) { | |
d2a17637 | 641 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
97df5d15 | 642 | EXT4_I(inode)->i_allocated_meta_blocks += ar.len; |
d2a17637 | 643 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
7b415bf6 AK |
644 | dquot_alloc_block_nofail(inode, |
645 | EXT4_C2B(EXT4_SB(inode->i_sb), ar.len)); | |
d2a17637 MC |
646 | } |
647 | return ret; | |
654b4908 AK |
648 | } |
649 | ||
ac27a0ec | 650 | /** |
5dee5437 | 651 | * ext4_count_free_clusters() -- count filesystem free clusters |
ac27a0ec DK |
652 | * @sb: superblock |
653 | * | |
5dee5437 | 654 | * Adds up the number of free clusters from each block group. |
ac27a0ec | 655 | */ |
5dee5437 | 656 | ext4_fsblk_t ext4_count_free_clusters(struct super_block *sb) |
ac27a0ec | 657 | { |
617ba13b MC |
658 | ext4_fsblk_t desc_count; |
659 | struct ext4_group_desc *gdp; | |
fd2d4291 | 660 | ext4_group_t i; |
8df9675f | 661 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
2746f7a1 | 662 | struct ext4_group_info *grp; |
617ba13b MC |
663 | #ifdef EXT4FS_DEBUG |
664 | struct ext4_super_block *es; | |
665 | ext4_fsblk_t bitmap_count; | |
498e5f24 | 666 | unsigned int x; |
ac27a0ec DK |
667 | struct buffer_head *bitmap_bh = NULL; |
668 | ||
617ba13b | 669 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
670 | desc_count = 0; |
671 | bitmap_count = 0; | |
672 | gdp = NULL; | |
673 | ||
ac27a0ec | 674 | for (i = 0; i < ngroups; i++) { |
617ba13b | 675 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
676 | if (!gdp) |
677 | continue; | |
2746f7a1 DW |
678 | grp = NULL; |
679 | if (EXT4_SB(sb)->s_group_info) | |
680 | grp = ext4_get_group_info(sb, i); | |
681 | if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) | |
682 | desc_count += ext4_free_group_clusters(sb, gdp); | |
ac27a0ec | 683 | brelse(bitmap_bh); |
574ca174 | 684 | bitmap_bh = ext4_read_block_bitmap(sb, i); |
ac27a0ec DK |
685 | if (bitmap_bh == NULL) |
686 | continue; | |
687 | ||
f6fb99ca | 688 | x = ext4_count_free(bitmap_bh->b_data, |
036acea2 | 689 | EXT4_CLUSTERS_PER_GROUP(sb) / 8); |
9fd9784c | 690 | printk(KERN_DEBUG "group %u: stored = %d, counted = %u\n", |
021b65bb | 691 | i, ext4_free_group_clusters(sb, gdp), x); |
ac27a0ec DK |
692 | bitmap_count += x; |
693 | } | |
694 | brelse(bitmap_bh); | |
5dee5437 TT |
695 | printk(KERN_DEBUG "ext4_count_free_clusters: stored = %llu" |
696 | ", computed = %llu, %llu\n", | |
810da240 | 697 | EXT4_NUM_B2C(EXT4_SB(sb), ext4_free_blocks_count(es)), |
4776004f | 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; | |
2746f7a1 DW |
706 | grp = NULL; |
707 | if (EXT4_SB(sb)->s_group_info) | |
708 | grp = ext4_get_group_info(sb, i); | |
709 | if (!grp || !EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) | |
710 | desc_count += ext4_free_group_clusters(sb, gdp); | |
ac27a0ec DK |
711 | } |
712 | ||
713 | return desc_count; | |
714 | #endif | |
715 | } | |
716 | ||
fd2d4291 | 717 | static inline int test_root(ext4_group_t a, int b) |
ac27a0ec | 718 | { |
f4afb4f4 TT |
719 | while (1) { |
720 | if (a < b) | |
721 | return 0; | |
722 | if (a == b) | |
723 | return 1; | |
724 | if ((a % b) != 0) | |
725 | return 0; | |
726 | a = a / b; | |
727 | } | |
ac27a0ec DK |
728 | } |
729 | ||
ac27a0ec | 730 | /** |
617ba13b | 731 | * ext4_bg_has_super - number of blocks used by the superblock in group |
ac27a0ec DK |
732 | * @sb: superblock for filesystem |
733 | * @group: group number to check | |
734 | * | |
735 | * Return the number of blocks used by the superblock (primary or backup) | |
736 | * in this group. Currently this will be only 0 or 1. | |
737 | */ | |
fd2d4291 | 738 | int ext4_bg_has_super(struct super_block *sb, ext4_group_t group) |
ac27a0ec | 739 | { |
1beeef1b DW |
740 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
741 | ||
742 | if (group == 0) | |
743 | return 1; | |
744 | if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_SPARSE_SUPER2)) { | |
745 | if (group == le32_to_cpu(es->s_backup_bgs[0]) || | |
746 | group == le32_to_cpu(es->s_backup_bgs[1])) | |
747 | return 1; | |
748 | return 0; | |
749 | } | |
750 | if ((group <= 1) || !EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
751 | EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) | |
752 | return 1; | |
753 | if (!(group & 1)) | |
ac27a0ec | 754 | return 0; |
1beeef1b DW |
755 | if (test_root(group, 3) || (test_root(group, 5)) || |
756 | test_root(group, 7)) | |
757 | return 1; | |
758 | ||
759 | return 0; | |
ac27a0ec DK |
760 | } |
761 | ||
fd2d4291 AM |
762 | static unsigned long ext4_bg_num_gdb_meta(struct super_block *sb, |
763 | ext4_group_t group) | |
ac27a0ec | 764 | { |
617ba13b | 765 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); |
fd2d4291 AM |
766 | ext4_group_t first = metagroup * EXT4_DESC_PER_BLOCK(sb); |
767 | ext4_group_t last = first + EXT4_DESC_PER_BLOCK(sb) - 1; | |
ac27a0ec DK |
768 | |
769 | if (group == first || group == first + 1 || group == last) | |
770 | return 1; | |
771 | return 0; | |
772 | } | |
773 | ||
fd2d4291 AM |
774 | static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, |
775 | ext4_group_t group) | |
ac27a0ec | 776 | { |
8dadb198 TT |
777 | if (!ext4_bg_has_super(sb, group)) |
778 | return 0; | |
779 | ||
780 | if (EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG)) | |
781 | return le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); | |
782 | else | |
783 | return EXT4_SB(sb)->s_gdb_count; | |
ac27a0ec DK |
784 | } |
785 | ||
786 | /** | |
617ba13b | 787 | * ext4_bg_num_gdb - number of blocks used by the group table in group |
ac27a0ec DK |
788 | * @sb: superblock for filesystem |
789 | * @group: group number to check | |
790 | * | |
791 | * Return the number of blocks used by the group descriptor table | |
792 | * (primary or backup) in this group. In the future there may be a | |
793 | * different number of descriptor blocks in each group. | |
794 | */ | |
fd2d4291 | 795 | unsigned long ext4_bg_num_gdb(struct super_block *sb, ext4_group_t group) |
ac27a0ec DK |
796 | { |
797 | unsigned long first_meta_bg = | |
617ba13b MC |
798 | le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); |
799 | unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); | |
ac27a0ec | 800 | |
617ba13b | 801 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) || |
ac27a0ec | 802 | metagroup < first_meta_bg) |
af5bc92d | 803 | return ext4_bg_num_gdb_nometa(sb, group); |
ac27a0ec | 804 | |
617ba13b | 805 | return ext4_bg_num_gdb_meta(sb,group); |
ac27a0ec DK |
806 | |
807 | } | |
c2ea3fde | 808 | |
49f7f9af | 809 | /* |
d5b8f310 | 810 | * This function returns the number of file system metadata clusters at |
49f7f9af TT |
811 | * the beginning of a block group, including the reserved gdt blocks. |
812 | */ | |
5f163cc7 | 813 | static unsigned ext4_num_base_meta_clusters(struct super_block *sb, |
d5b8f310 | 814 | ext4_group_t block_group) |
49f7f9af TT |
815 | { |
816 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
d5b8f310 | 817 | unsigned num; |
49f7f9af TT |
818 | |
819 | /* Check for superblock and gdt backups in this group */ | |
820 | num = ext4_bg_has_super(sb, block_group); | |
821 | ||
822 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) || | |
823 | block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) * | |
824 | sbi->s_desc_per_block) { | |
825 | if (num) { | |
826 | num += ext4_bg_num_gdb(sb, block_group); | |
827 | num += le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); | |
828 | } | |
829 | } else { /* For META_BG_BLOCK_GROUPS */ | |
830 | num += ext4_bg_num_gdb(sb, block_group); | |
831 | } | |
d5b8f310 | 832 | return EXT4_NUM_B2C(sbi, num); |
49f7f9af | 833 | } |
f86186b4 ES |
834 | /** |
835 | * ext4_inode_to_goal_block - return a hint for block allocation | |
836 | * @inode: inode for block allocation | |
837 | * | |
838 | * Return the ideal location to start allocating blocks for a | |
839 | * newly created inode. | |
840 | */ | |
841 | ext4_fsblk_t ext4_inode_to_goal_block(struct inode *inode) | |
842 | { | |
843 | struct ext4_inode_info *ei = EXT4_I(inode); | |
844 | ext4_group_t block_group; | |
845 | ext4_grpblk_t colour; | |
846 | int flex_size = ext4_flex_bg_size(EXT4_SB(inode->i_sb)); | |
847 | ext4_fsblk_t bg_start; | |
848 | ext4_fsblk_t last_block; | |
849 | ||
850 | block_group = ei->i_block_group; | |
851 | if (flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) { | |
852 | /* | |
853 | * If there are at least EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME | |
854 | * block groups per flexgroup, reserve the first block | |
855 | * group for directories and special files. Regular | |
856 | * files will start at the second block group. This | |
857 | * tends to speed up directory access and improves | |
858 | * fsck times. | |
859 | */ | |
860 | block_group &= ~(flex_size-1); | |
861 | if (S_ISREG(inode->i_mode)) | |
862 | block_group++; | |
863 | } | |
864 | bg_start = ext4_group_first_block_no(inode->i_sb, block_group); | |
865 | last_block = ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es) - 1; | |
866 | ||
867 | /* | |
868 | * If we are doing delayed allocation, we don't need take | |
869 | * colour into account. | |
870 | */ | |
871 | if (test_opt(inode->i_sb, DELALLOC)) | |
872 | return bg_start; | |
873 | ||
874 | if (bg_start + EXT4_BLOCKS_PER_GROUP(inode->i_sb) <= last_block) | |
875 | colour = (current->pid % 16) * | |
876 | (EXT4_BLOCKS_PER_GROUP(inode->i_sb) / 16); | |
877 | else | |
878 | colour = (current->pid % 16) * ((last_block - bg_start) / 16); | |
879 | return bg_start + colour; | |
880 | } | |
881 |