]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/ialloc.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 | * BSD ufs-inspired inode and directory allocation by | |
10 | * Stephen Tweedie (sct@redhat.com), 1993 | |
11 | * Big-endian to little-endian byte-swapping/bitmaps by | |
12 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
13 | */ | |
14 | ||
15 | #include <linux/time.h> | |
16 | #include <linux/fs.h> | |
ac27a0ec DK |
17 | #include <linux/stat.h> |
18 | #include <linux/string.h> | |
19 | #include <linux/quotaops.h> | |
20 | #include <linux/buffer_head.h> | |
21 | #include <linux/random.h> | |
22 | #include <linux/bitops.h> | |
3a5b2ecd | 23 | #include <linux/blkdev.h> |
ac27a0ec | 24 | #include <asm/byteorder.h> |
9bffad1e | 25 | |
3dcf5451 CH |
26 | #include "ext4.h" |
27 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
28 | #include "xattr.h" |
29 | #include "acl.h" | |
30 | ||
9bffad1e TT |
31 | #include <trace/events/ext4.h> |
32 | ||
ac27a0ec DK |
33 | /* |
34 | * ialloc.c contains the inodes allocation and deallocation routines | |
35 | */ | |
36 | ||
37 | /* | |
38 | * The free inodes are managed by bitmaps. A file system contains several | |
39 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
40 | * block for inodes, N blocks for the inode table and data blocks. | |
41 | * | |
42 | * The file system contains group descriptors which are located after the | |
43 | * super block. Each descriptor contains the number of the bitmap block and | |
44 | * the free blocks count in the block. | |
45 | */ | |
46 | ||
717d50e4 AD |
47 | /* |
48 | * To avoid calling the atomic setbit hundreds or thousands of times, we only | |
49 | * need to use it within a single byte (to ensure we get endianness right). | |
50 | * We can use memset for the rest of the bitmap as there are no other users. | |
51 | */ | |
61d08673 | 52 | void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap) |
717d50e4 AD |
53 | { |
54 | int i; | |
55 | ||
56 | if (start_bit >= end_bit) | |
57 | return; | |
58 | ||
59 | ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); | |
60 | for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) | |
61 | ext4_set_bit(i, bitmap); | |
62 | if (i < end_bit) | |
63 | memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); | |
64 | } | |
65 | ||
66 | /* Initializes an uninitialized inode bitmap */ | |
9008a58e | 67 | static int ext4_init_inode_bitmap(struct super_block *sb, |
1f109d5a TT |
68 | struct buffer_head *bh, |
69 | ext4_group_t block_group, | |
70 | struct ext4_group_desc *gdp) | |
717d50e4 | 71 | { |
bdfb6ff4 | 72 | struct ext4_group_info *grp; |
e43bb4e6 | 73 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
717d50e4 AD |
74 | J_ASSERT_BH(bh, buffer_locked(bh)); |
75 | ||
76 | /* If checksum is bad mark all blocks and inodes use to prevent | |
77 | * allocation, essentially implementing a per-group read-only flag. */ | |
feb0ab32 | 78 | if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) { |
bdfb6ff4 | 79 | grp = ext4_get_group_info(sb, block_group); |
e43bb4e6 NJ |
80 | if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) |
81 | percpu_counter_sub(&sbi->s_freeclusters_counter, | |
82 | grp->bb_free); | |
bdfb6ff4 | 83 | set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state); |
e43bb4e6 NJ |
84 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
85 | int count; | |
86 | count = ext4_free_inodes_count(sb, gdp); | |
87 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
88 | count); | |
89 | } | |
bdfb6ff4 | 90 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
9008a58e | 91 | return -EFSBADCRC; |
717d50e4 AD |
92 | } |
93 | ||
94 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
61d08673 | 95 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
717d50e4 | 96 | bh->b_data); |
41a246d1 DW |
97 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh, |
98 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 99 | ext4_group_desc_csum_set(sb, block_group, gdp); |
717d50e4 | 100 | |
9008a58e | 101 | return 0; |
717d50e4 | 102 | } |
ac27a0ec | 103 | |
813e5727 TT |
104 | void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate) |
105 | { | |
106 | if (uptodate) { | |
107 | set_buffer_uptodate(bh); | |
108 | set_bitmap_uptodate(bh); | |
109 | } | |
110 | unlock_buffer(bh); | |
111 | put_bh(bh); | |
112 | } | |
113 | ||
9008a58e DW |
114 | static int ext4_validate_inode_bitmap(struct super_block *sb, |
115 | struct ext4_group_desc *desc, | |
116 | ext4_group_t block_group, | |
117 | struct buffer_head *bh) | |
118 | { | |
119 | ext4_fsblk_t blk; | |
120 | struct ext4_group_info *grp = ext4_get_group_info(sb, block_group); | |
121 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
122 | ||
123 | if (buffer_verified(bh)) | |
124 | return 0; | |
125 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) | |
126 | return -EFSCORRUPTED; | |
127 | ||
128 | ext4_lock_group(sb, block_group); | |
129 | blk = ext4_inode_bitmap(sb, desc); | |
130 | if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh, | |
131 | EXT4_INODES_PER_GROUP(sb) / 8)) { | |
132 | ext4_unlock_group(sb, block_group); | |
133 | ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " | |
134 | "inode_bitmap = %llu", block_group, blk); | |
135 | grp = ext4_get_group_info(sb, block_group); | |
136 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { | |
137 | int count; | |
138 | count = ext4_free_inodes_count(sb, desc); | |
139 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
140 | count); | |
141 | } | |
142 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); | |
143 | return -EFSBADCRC; | |
144 | } | |
145 | set_buffer_verified(bh); | |
146 | ext4_unlock_group(sb, block_group); | |
147 | return 0; | |
148 | } | |
149 | ||
ac27a0ec DK |
150 | /* |
151 | * Read the inode allocation bitmap for a given block_group, reading | |
152 | * into the specified slot in the superblock's bitmap cache. | |
153 | * | |
154 | * Return buffer_head of bitmap on success or NULL. | |
155 | */ | |
156 | static struct buffer_head * | |
e29d1cde | 157 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 158 | { |
617ba13b | 159 | struct ext4_group_desc *desc; |
ac27a0ec | 160 | struct buffer_head *bh = NULL; |
e29d1cde | 161 | ext4_fsblk_t bitmap_blk; |
9008a58e | 162 | int err; |
ac27a0ec | 163 | |
617ba13b | 164 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 165 | if (!desc) |
9008a58e | 166 | return ERR_PTR(-EFSCORRUPTED); |
bfff6873 | 167 | |
e29d1cde ES |
168 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
169 | bh = sb_getblk(sb, bitmap_blk); | |
170 | if (unlikely(!bh)) { | |
12062ddd | 171 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 172 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde | 173 | block_group, bitmap_blk); |
9008a58e | 174 | return ERR_PTR(-EIO); |
e29d1cde | 175 | } |
2ccb5fb9 | 176 | if (bitmap_uptodate(bh)) |
41a246d1 | 177 | goto verify; |
e29d1cde | 178 | |
c806e68f | 179 | lock_buffer(bh); |
2ccb5fb9 AK |
180 | if (bitmap_uptodate(bh)) { |
181 | unlock_buffer(bh); | |
41a246d1 | 182 | goto verify; |
2ccb5fb9 | 183 | } |
bfff6873 | 184 | |
955ce5f5 | 185 | ext4_lock_group(sb, block_group); |
717d50e4 | 186 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
9008a58e | 187 | err = ext4_init_inode_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 188 | set_bitmap_uptodate(bh); |
e29d1cde | 189 | set_buffer_uptodate(bh); |
41a246d1 | 190 | set_buffer_verified(bh); |
955ce5f5 | 191 | ext4_unlock_group(sb, block_group); |
3300beda | 192 | unlock_buffer(bh); |
05145bd7 JK |
193 | if (err) { |
194 | ext4_error(sb, "Failed to init inode bitmap for group " | |
195 | "%u: %d", block_group, err); | |
9008a58e | 196 | goto out; |
05145bd7 | 197 | } |
e29d1cde | 198 | return bh; |
717d50e4 | 199 | } |
955ce5f5 | 200 | ext4_unlock_group(sb, block_group); |
bfff6873 | 201 | |
2ccb5fb9 AK |
202 | if (buffer_uptodate(bh)) { |
203 | /* | |
204 | * if not uninit if bh is uptodate, | |
205 | * bitmap is also uptodate | |
206 | */ | |
207 | set_bitmap_uptodate(bh); | |
208 | unlock_buffer(bh); | |
41a246d1 | 209 | goto verify; |
2ccb5fb9 AK |
210 | } |
211 | /* | |
813e5727 | 212 | * submit the buffer_head for reading |
2ccb5fb9 | 213 | */ |
0562e0ba | 214 | trace_ext4_load_inode_bitmap(sb, block_group); |
813e5727 TT |
215 | bh->b_end_io = ext4_end_bitmap_read; |
216 | get_bh(bh); | |
9f203507 | 217 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
813e5727 TT |
218 | wait_on_buffer(bh); |
219 | if (!buffer_uptodate(bh)) { | |
e29d1cde | 220 | put_bh(bh); |
12062ddd | 221 | ext4_error(sb, "Cannot read inode bitmap - " |
813e5727 TT |
222 | "block_group = %u, inode_bitmap = %llu", |
223 | block_group, bitmap_blk); | |
9008a58e | 224 | return ERR_PTR(-EIO); |
e29d1cde | 225 | } |
41a246d1 DW |
226 | |
227 | verify: | |
9008a58e DW |
228 | err = ext4_validate_inode_bitmap(sb, desc, block_group, bh); |
229 | if (err) | |
230 | goto out; | |
ac27a0ec | 231 | return bh; |
9008a58e DW |
232 | out: |
233 | put_bh(bh); | |
234 | return ERR_PTR(err); | |
ac27a0ec DK |
235 | } |
236 | ||
237 | /* | |
238 | * NOTE! When we get the inode, we're the only people | |
239 | * that have access to it, and as such there are no | |
240 | * race conditions we have to worry about. The inode | |
241 | * is not on the hash-lists, and it cannot be reached | |
242 | * through the filesystem because the directory entry | |
243 | * has been deleted earlier. | |
244 | * | |
245 | * HOWEVER: we must make sure that we get no aliases, | |
246 | * which means that we have to call "clear_inode()" | |
247 | * _before_ we mark the inode not in use in the inode | |
248 | * bitmaps. Otherwise a newly created file might use | |
249 | * the same inode number (not actually the same pointer | |
250 | * though), and then we'd have two inodes sharing the | |
251 | * same inode number and space on the harddisk. | |
252 | */ | |
af5bc92d | 253 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 254 | { |
af5bc92d | 255 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
256 | int is_directory; |
257 | unsigned long ino; | |
258 | struct buffer_head *bitmap_bh = NULL; | |
259 | struct buffer_head *bh2; | |
fd2d4291 | 260 | ext4_group_t block_group; |
ac27a0ec | 261 | unsigned long bit; |
af5bc92d TT |
262 | struct ext4_group_desc *gdp; |
263 | struct ext4_super_block *es; | |
617ba13b | 264 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 265 | int fatal = 0, err, count, cleared; |
87a39389 | 266 | struct ext4_group_info *grp; |
ac27a0ec | 267 | |
92b97816 TT |
268 | if (!sb) { |
269 | printk(KERN_ERR "EXT4-fs: %s:%d: inode on " | |
270 | "nonexistent device\n", __func__, __LINE__); | |
ac27a0ec DK |
271 | return; |
272 | } | |
92b97816 TT |
273 | if (atomic_read(&inode->i_count) > 1) { |
274 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", | |
275 | __func__, __LINE__, inode->i_ino, | |
276 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
277 | return; |
278 | } | |
92b97816 TT |
279 | if (inode->i_nlink) { |
280 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", | |
281 | __func__, __LINE__, inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
282 | return; |
283 | } | |
617ba13b | 284 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
285 | |
286 | ino = inode->i_ino; | |
af5bc92d | 287 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 288 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
289 | |
290 | /* | |
291 | * Note: we must free any quota before locking the superblock, | |
292 | * as writing the quota to disk may need the lock as well. | |
293 | */ | |
871a2931 | 294 | dquot_initialize(inode); |
617ba13b | 295 | ext4_xattr_delete_inode(handle, inode); |
63936dda | 296 | dquot_free_inode(inode); |
9f754758 | 297 | dquot_drop(inode); |
ac27a0ec DK |
298 | |
299 | is_directory = S_ISDIR(inode->i_mode); | |
300 | ||
301 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 302 | ext4_clear_inode(inode); |
ac27a0ec | 303 | |
617ba13b MC |
304 | es = EXT4_SB(sb)->s_es; |
305 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 306 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
307 | goto error_return; |
308 | } | |
617ba13b MC |
309 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
310 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 311 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
87a39389 DW |
312 | /* Don't bother if the inode bitmap is corrupt. */ |
313 | grp = ext4_get_group_info(sb, block_group); | |
9008a58e DW |
314 | if (IS_ERR(bitmap_bh)) { |
315 | fatal = PTR_ERR(bitmap_bh); | |
316 | bitmap_bh = NULL; | |
317 | goto error_return; | |
318 | } | |
319 | if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) { | |
320 | fatal = -EFSCORRUPTED; | |
ac27a0ec | 321 | goto error_return; |
9008a58e | 322 | } |
ac27a0ec DK |
323 | |
324 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 325 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
326 | if (fatal) |
327 | goto error_return; | |
328 | ||
d17413c0 DM |
329 | fatal = -ESRCH; |
330 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
331 | if (gdp) { | |
ac27a0ec | 332 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 333 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
334 | } |
335 | ext4_lock_group(sb, block_group); | |
597d508c | 336 | cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); |
d17413c0 DM |
337 | if (fatal || !cleared) { |
338 | ext4_unlock_group(sb, block_group); | |
339 | goto out; | |
340 | } | |
7d39db14 | 341 | |
d17413c0 DM |
342 | count = ext4_free_inodes_count(sb, gdp) + 1; |
343 | ext4_free_inodes_set(sb, gdp, count); | |
344 | if (is_directory) { | |
345 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
346 | ext4_used_dirs_set(sb, gdp, count); | |
347 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 348 | } |
41a246d1 DW |
349 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, |
350 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 351 | ext4_group_desc_csum_set(sb, block_group, gdp); |
d17413c0 | 352 | ext4_unlock_group(sb, block_group); |
ac27a0ec | 353 | |
d17413c0 DM |
354 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
355 | if (sbi->s_log_groups_per_flex) { | |
356 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 357 | |
d17413c0 DM |
358 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
359 | if (is_directory) | |
360 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 361 | } |
d17413c0 DM |
362 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
363 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
364 | out: | |
365 | if (cleared) { | |
366 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
367 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
368 | if (!fatal) | |
369 | fatal = err; | |
87a39389 | 370 | } else { |
d17413c0 | 371 | ext4_error(sb, "bit already cleared for inode %lu", ino); |
bf40c926 | 372 | if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
e43bb4e6 NJ |
373 | int count; |
374 | count = ext4_free_inodes_count(sb, gdp); | |
375 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
376 | count); | |
377 | } | |
87a39389 DW |
378 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
379 | } | |
d17413c0 | 380 | |
ac27a0ec DK |
381 | error_return: |
382 | brelse(bitmap_bh); | |
617ba13b | 383 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
384 | } |
385 | ||
a4912123 | 386 | struct orlov_stats { |
90ba983f | 387 | __u64 free_clusters; |
a4912123 | 388 | __u32 free_inodes; |
a4912123 TT |
389 | __u32 used_dirs; |
390 | }; | |
391 | ||
392 | /* | |
393 | * Helper function for Orlov's allocator; returns critical information | |
394 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
395 | * is a block group number; otherwise it is flex_bg number. | |
396 | */ | |
1f109d5a TT |
397 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
398 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
399 | { |
400 | struct ext4_group_desc *desc; | |
7d39db14 | 401 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 402 | |
7d39db14 TT |
403 | if (flex_size > 1) { |
404 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
90ba983f | 405 | stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); |
7d39db14 TT |
406 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); |
407 | return; | |
408 | } | |
a4912123 | 409 | |
7d39db14 TT |
410 | desc = ext4_get_group_desc(sb, g, NULL); |
411 | if (desc) { | |
412 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
021b65bb | 413 | stats->free_clusters = ext4_free_group_clusters(sb, desc); |
7d39db14 TT |
414 | stats->used_dirs = ext4_used_dirs_count(sb, desc); |
415 | } else { | |
416 | stats->free_inodes = 0; | |
24aaa8ef | 417 | stats->free_clusters = 0; |
7d39db14 | 418 | stats->used_dirs = 0; |
a4912123 TT |
419 | } |
420 | } | |
421 | ||
ac27a0ec DK |
422 | /* |
423 | * Orlov's allocator for directories. | |
424 | * | |
425 | * We always try to spread first-level directories. | |
426 | * | |
427 | * If there are blockgroups with both free inodes and free blocks counts | |
428 | * not worse than average we return one with smallest directory count. | |
429 | * Otherwise we simply return a random group. | |
430 | * | |
431 | * For the rest rules look so: | |
432 | * | |
433 | * It's OK to put directory into a group unless | |
434 | * it has too many directories already (max_dirs) or | |
435 | * it has too few free inodes left (min_inodes) or | |
436 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 437 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
438 | * conditions we search cyclically through the rest. If none |
439 | * of the groups look good we just look for a group with more | |
440 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
441 | */ |
442 | ||
2aa9fc4c | 443 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
dcca3fec | 444 | ext4_group_t *group, umode_t mode, |
f157a4aa | 445 | const struct qstr *qstr) |
ac27a0ec | 446 | { |
fd2d4291 | 447 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 448 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 449 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 450 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
14c83c9f | 451 | unsigned int freei, avefreei, grp_free; |
24aaa8ef | 452 | ext4_fsblk_t freeb, avefreec; |
ac27a0ec | 453 | unsigned int ndirs; |
a4912123 | 454 | int max_dirs, min_inodes; |
24aaa8ef | 455 | ext4_grpblk_t min_clusters; |
8df9675f | 456 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 457 | struct ext4_group_desc *desc; |
a4912123 TT |
458 | struct orlov_stats stats; |
459 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 460 | struct dx_hash_info hinfo; |
a4912123 | 461 | |
8df9675f | 462 | ngroups = real_ngroups; |
a4912123 | 463 | if (flex_size > 1) { |
8df9675f | 464 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
465 | sbi->s_log_groups_per_flex; |
466 | parent_group >>= sbi->s_log_groups_per_flex; | |
467 | } | |
ac27a0ec DK |
468 | |
469 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
470 | avefreei = freei / ngroups; | |
57042651 TT |
471 | freeb = EXT4_C2B(sbi, |
472 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
24aaa8ef TT |
473 | avefreec = freeb; |
474 | do_div(avefreec, ngroups); | |
ac27a0ec DK |
475 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
476 | ||
a4912123 | 477 | if (S_ISDIR(mode) && |
2b0143b5 | 478 | ((parent == d_inode(sb->s_root)) || |
12e9b892 | 479 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 480 | int best_ndir = inodes_per_group; |
2aa9fc4c | 481 | int ret = -1; |
ac27a0ec | 482 | |
f157a4aa TT |
483 | if (qstr) { |
484 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
485 | hinfo.seed = sbi->s_hash_seed; | |
486 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
487 | grp = hinfo.hash; | |
488 | } else | |
dd1f723b | 489 | grp = prandom_u32(); |
2aa9fc4c | 490 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 491 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
492 | g = (parent_group + i) % ngroups; |
493 | get_orlov_stats(sb, g, flex_size, &stats); | |
494 | if (!stats.free_inodes) | |
ac27a0ec | 495 | continue; |
a4912123 | 496 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 497 | continue; |
a4912123 | 498 | if (stats.free_inodes < avefreei) |
ac27a0ec | 499 | continue; |
24aaa8ef | 500 | if (stats.free_clusters < avefreec) |
ac27a0ec | 501 | continue; |
a4912123 | 502 | grp = g; |
2aa9fc4c | 503 | ret = 0; |
a4912123 TT |
504 | best_ndir = stats.used_dirs; |
505 | } | |
506 | if (ret) | |
507 | goto fallback; | |
508 | found_flex_bg: | |
509 | if (flex_size == 1) { | |
510 | *group = grp; | |
511 | return 0; | |
512 | } | |
513 | ||
514 | /* | |
515 | * We pack inodes at the beginning of the flexgroup's | |
516 | * inode tables. Block allocation decisions will do | |
517 | * something similar, although regular files will | |
518 | * start at 2nd block group of the flexgroup. See | |
519 | * ext4_ext_find_goal() and ext4_find_near(). | |
520 | */ | |
521 | grp *= flex_size; | |
522 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 523 | if (grp+i >= real_ngroups) |
a4912123 TT |
524 | break; |
525 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
526 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
527 | *group = grp+i; | |
528 | return 0; | |
529 | } | |
ac27a0ec | 530 | } |
ac27a0ec DK |
531 | goto fallback; |
532 | } | |
533 | ||
ac27a0ec | 534 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
535 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
536 | if (min_inodes < 1) | |
537 | min_inodes = 1; | |
24aaa8ef | 538 | min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; |
a4912123 TT |
539 | |
540 | /* | |
541 | * Start looking in the flex group where we last allocated an | |
542 | * inode for this parent directory | |
543 | */ | |
544 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
545 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
546 | if (flex_size > 1) | |
547 | parent_group >>= sbi->s_log_groups_per_flex; | |
548 | } | |
ac27a0ec DK |
549 | |
550 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
551 | grp = (parent_group + i) % ngroups; |
552 | get_orlov_stats(sb, grp, flex_size, &stats); | |
553 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 554 | continue; |
a4912123 | 555 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 556 | continue; |
24aaa8ef | 557 | if (stats.free_clusters < min_clusters) |
ac27a0ec | 558 | continue; |
a4912123 | 559 | goto found_flex_bg; |
ac27a0ec DK |
560 | } |
561 | ||
562 | fallback: | |
8df9675f | 563 | ngroups = real_ngroups; |
a4912123 | 564 | avefreei = freei / ngroups; |
b5451f7b | 565 | fallback_retry: |
a4912123 | 566 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 567 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
568 | grp = (parent_group + i) % ngroups; |
569 | desc = ext4_get_group_desc(sb, grp, NULL); | |
bb3d132a DC |
570 | if (desc) { |
571 | grp_free = ext4_free_inodes_count(sb, desc); | |
572 | if (grp_free && grp_free >= avefreei) { | |
573 | *group = grp; | |
574 | return 0; | |
575 | } | |
a4912123 | 576 | } |
ac27a0ec DK |
577 | } |
578 | ||
579 | if (avefreei) { | |
580 | /* | |
581 | * The free-inodes counter is approximate, and for really small | |
582 | * filesystems the above test can fail to find any blockgroups | |
583 | */ | |
584 | avefreei = 0; | |
b5451f7b | 585 | goto fallback_retry; |
ac27a0ec DK |
586 | } |
587 | ||
588 | return -1; | |
589 | } | |
590 | ||
2aa9fc4c | 591 | static int find_group_other(struct super_block *sb, struct inode *parent, |
dcca3fec | 592 | ext4_group_t *group, umode_t mode) |
ac27a0ec | 593 | { |
fd2d4291 | 594 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 595 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 596 | struct ext4_group_desc *desc; |
a4912123 TT |
597 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
598 | ||
599 | /* | |
600 | * Try to place the inode is the same flex group as its | |
601 | * parent. If we can't find space, use the Orlov algorithm to | |
602 | * find another flex group, and store that information in the | |
603 | * parent directory's inode information so that use that flex | |
604 | * group for future allocations. | |
605 | */ | |
606 | if (flex_size > 1) { | |
607 | int retry = 0; | |
608 | ||
609 | try_again: | |
610 | parent_group &= ~(flex_size-1); | |
611 | last = parent_group + flex_size; | |
612 | if (last > ngroups) | |
613 | last = ngroups; | |
614 | for (i = parent_group; i < last; i++) { | |
615 | desc = ext4_get_group_desc(sb, i, NULL); | |
616 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
617 | *group = i; | |
618 | return 0; | |
619 | } | |
620 | } | |
621 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
622 | retry = 1; | |
623 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
624 | goto try_again; | |
625 | } | |
626 | /* | |
627 | * If this didn't work, use the Orlov search algorithm | |
628 | * to find a new flex group; we pass in the mode to | |
629 | * avoid the topdir algorithms. | |
630 | */ | |
631 | *group = parent_group + flex_size; | |
632 | if (*group > ngroups) | |
633 | *group = 0; | |
7dc57615 | 634 | return find_group_orlov(sb, parent, group, mode, NULL); |
a4912123 | 635 | } |
ac27a0ec DK |
636 | |
637 | /* | |
638 | * Try to place the inode in its parent directory | |
639 | */ | |
2aa9fc4c AM |
640 | *group = parent_group; |
641 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 642 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 643 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 644 | return 0; |
ac27a0ec DK |
645 | |
646 | /* | |
647 | * We're going to place this inode in a different blockgroup from its | |
648 | * parent. We want to cause files in a common directory to all land in | |
649 | * the same blockgroup. But we want files which are in a different | |
650 | * directory which shares a blockgroup with our parent to land in a | |
651 | * different blockgroup. | |
652 | * | |
653 | * So add our directory's i_ino into the starting point for the hash. | |
654 | */ | |
2aa9fc4c | 655 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
656 | |
657 | /* | |
658 | * Use a quadratic hash to find a group with a free inode and some free | |
659 | * blocks. | |
660 | */ | |
661 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
662 | *group += i; |
663 | if (*group >= ngroups) | |
664 | *group -= ngroups; | |
665 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 666 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 667 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 668 | return 0; |
ac27a0ec DK |
669 | } |
670 | ||
671 | /* | |
672 | * That failed: try linear search for a free inode, even if that group | |
673 | * has no free blocks. | |
674 | */ | |
2aa9fc4c | 675 | *group = parent_group; |
ac27a0ec | 676 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
677 | if (++*group >= ngroups) |
678 | *group = 0; | |
679 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 680 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 681 | return 0; |
ac27a0ec DK |
682 | } |
683 | ||
684 | return -1; | |
685 | } | |
686 | ||
19883bd9 TT |
687 | /* |
688 | * In no journal mode, if an inode has recently been deleted, we want | |
689 | * to avoid reusing it until we're reasonably sure the inode table | |
690 | * block has been written back to disk. (Yes, these values are | |
691 | * somewhat arbitrary...) | |
692 | */ | |
693 | #define RECENTCY_MIN 5 | |
694 | #define RECENTCY_DIRTY 30 | |
695 | ||
696 | static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) | |
697 | { | |
698 | struct ext4_group_desc *gdp; | |
699 | struct ext4_inode *raw_inode; | |
700 | struct buffer_head *bh; | |
701 | unsigned long dtime, now; | |
702 | int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
703 | int offset, ret = 0, recentcy = RECENTCY_MIN; | |
704 | ||
705 | gdp = ext4_get_group_desc(sb, group, NULL); | |
706 | if (unlikely(!gdp)) | |
707 | return 0; | |
708 | ||
709 | bh = sb_getblk(sb, ext4_inode_table(sb, gdp) + | |
710 | (ino / inodes_per_block)); | |
711 | if (unlikely(!bh) || !buffer_uptodate(bh)) | |
712 | /* | |
713 | * If the block is not in the buffer cache, then it | |
714 | * must have been written out. | |
715 | */ | |
716 | goto out; | |
717 | ||
718 | offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
719 | raw_inode = (struct ext4_inode *) (bh->b_data + offset); | |
720 | dtime = le32_to_cpu(raw_inode->i_dtime); | |
721 | now = get_seconds(); | |
722 | if (buffer_dirty(bh)) | |
723 | recentcy += RECENTCY_DIRTY; | |
724 | ||
725 | if (dtime && (dtime < now) && (now < dtime + recentcy)) | |
726 | ret = 1; | |
727 | out: | |
728 | brelse(bh); | |
729 | return ret; | |
730 | } | |
731 | ||
ac27a0ec DK |
732 | /* |
733 | * There are two policies for allocating an inode. If the new inode is | |
734 | * a directory, then a forward search is made for a block group with both | |
735 | * free space and a low directory-to-inode ratio; if that fails, then of | |
736 | * the groups with above-average free space, that group with the fewest | |
737 | * directories already is chosen. | |
738 | * | |
739 | * For other inodes, search forward from the parent directory's block | |
740 | * group to find a free inode. | |
741 | */ | |
1139575a TT |
742 | struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, |
743 | umode_t mode, const struct qstr *qstr, | |
744 | __u32 goal, uid_t *owner, int handle_type, | |
745 | unsigned int line_no, int nblocks) | |
ac27a0ec DK |
746 | { |
747 | struct super_block *sb; | |
3300beda AK |
748 | struct buffer_head *inode_bitmap_bh = NULL; |
749 | struct buffer_head *group_desc_bh; | |
8df9675f | 750 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 751 | unsigned long ino = 0; |
af5bc92d TT |
752 | struct inode *inode; |
753 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
754 | struct ext4_inode_info *ei; |
755 | struct ext4_sb_info *sbi; | |
a7cdadee | 756 | int ret2, err; |
ac27a0ec | 757 | struct inode *ret; |
2aa9fc4c | 758 | ext4_group_t i; |
772cb7c8 | 759 | ext4_group_t flex_group; |
87a39389 | 760 | struct ext4_group_info *grp; |
e709e9df | 761 | int encrypt = 0; |
ac27a0ec DK |
762 | |
763 | /* Cannot create files in a deleted directory */ | |
764 | if (!dir || !dir->i_nlink) | |
765 | return ERR_PTR(-EPERM); | |
766 | ||
e709e9df TT |
767 | if ((ext4_encrypted_inode(dir) || |
768 | DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) && | |
769 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
770 | err = ext4_get_encryption_info(dir); | |
771 | if (err) | |
772 | return ERR_PTR(err); | |
773 | if (ext4_encryption_info(dir) == NULL) | |
774 | return ERR_PTR(-EPERM); | |
775 | if (!handle) | |
776 | nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb); | |
777 | encrypt = 1; | |
778 | } | |
779 | ||
ac27a0ec | 780 | sb = dir->i_sb; |
8df9675f | 781 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 782 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
783 | inode = new_inode(sb); |
784 | if (!inode) | |
785 | return ERR_PTR(-ENOMEM); | |
617ba13b | 786 | ei = EXT4_I(inode); |
617ba13b | 787 | sbi = EXT4_SB(sb); |
772cb7c8 | 788 | |
eb9cc7e1 | 789 | /* |
b8a07463 | 790 | * Initialize owners and quota early so that we don't have to account |
eb9cc7e1 JK |
791 | * for quota initialization worst case in standard inode creating |
792 | * transaction | |
793 | */ | |
794 | if (owner) { | |
795 | inode->i_mode = mode; | |
796 | i_uid_write(inode, owner[0]); | |
797 | i_gid_write(inode, owner[1]); | |
798 | } else if (test_opt(sb, GRPID)) { | |
799 | inode->i_mode = mode; | |
800 | inode->i_uid = current_fsuid(); | |
801 | inode->i_gid = dir->i_gid; | |
802 | } else | |
803 | inode_init_owner(inode, dir, mode); | |
040cb378 LX |
804 | |
805 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_PROJECT) && | |
806 | ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) | |
807 | ei->i_projid = EXT4_I(dir)->i_projid; | |
808 | else | |
809 | ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID); | |
810 | ||
a7cdadee JK |
811 | err = dquot_initialize(inode); |
812 | if (err) | |
813 | goto out; | |
eb9cc7e1 | 814 | |
11013911 AD |
815 | if (!goal) |
816 | goal = sbi->s_inode_goal; | |
817 | ||
e6462869 | 818 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
819 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
820 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
821 | ret2 = 0; | |
822 | goto got_group; | |
823 | } | |
824 | ||
4113c4ca LC |
825 | if (S_ISDIR(mode)) |
826 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); | |
827 | else | |
a4912123 | 828 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 829 | |
772cb7c8 | 830 | got_group: |
a4912123 | 831 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 832 | err = -ENOSPC; |
2aa9fc4c | 833 | if (ret2 == -1) |
ac27a0ec DK |
834 | goto out; |
835 | ||
119c0d44 TT |
836 | /* |
837 | * Normally we will only go through one pass of this loop, | |
838 | * unless we get unlucky and it turns out the group we selected | |
839 | * had its last inode grabbed by someone else. | |
840 | */ | |
11013911 | 841 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
842 | err = -EIO; |
843 | ||
3300beda | 844 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec | 845 | if (!gdp) |
eb9cc7e1 | 846 | goto out; |
ac27a0ec | 847 | |
f2a09af6 YY |
848 | /* |
849 | * Check free inodes count before loading bitmap. | |
850 | */ | |
851 | if (ext4_free_inodes_count(sb, gdp) == 0) { | |
852 | if (++group == ngroups) | |
853 | group = 0; | |
854 | continue; | |
855 | } | |
856 | ||
87a39389 DW |
857 | grp = ext4_get_group_info(sb, group); |
858 | /* Skip groups with already-known suspicious inode tables */ | |
859 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { | |
860 | if (++group == ngroups) | |
861 | group = 0; | |
862 | continue; | |
863 | } | |
864 | ||
3300beda AK |
865 | brelse(inode_bitmap_bh); |
866 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
87a39389 | 867 | /* Skip groups with suspicious inode tables */ |
9008a58e DW |
868 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || |
869 | IS_ERR(inode_bitmap_bh)) { | |
870 | inode_bitmap_bh = NULL; | |
87a39389 DW |
871 | if (++group == ngroups) |
872 | group = 0; | |
873 | continue; | |
874 | } | |
ac27a0ec | 875 | |
ac27a0ec | 876 | repeat_in_this_group: |
617ba13b | 877 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
878 | inode_bitmap_bh->b_data, |
879 | EXT4_INODES_PER_GROUP(sb), ino); | |
a34eb503 TT |
880 | if (ino >= EXT4_INODES_PER_GROUP(sb)) |
881 | goto next_group; | |
119c0d44 TT |
882 | if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) { |
883 | ext4_error(sb, "reserved inode found cleared - " | |
884 | "inode=%lu", ino + 1); | |
885 | continue; | |
886 | } | |
19883bd9 TT |
887 | if ((EXT4_SB(sb)->s_journal == NULL) && |
888 | recently_deleted(sb, group, ino)) { | |
889 | ino++; | |
890 | goto next_inode; | |
891 | } | |
1139575a TT |
892 | if (!handle) { |
893 | BUG_ON(nblocks <= 0); | |
894 | handle = __ext4_journal_start_sb(dir->i_sb, line_no, | |
5fe2fe89 JK |
895 | handle_type, nblocks, |
896 | 0); | |
1139575a TT |
897 | if (IS_ERR(handle)) { |
898 | err = PTR_ERR(handle); | |
eb9cc7e1 JK |
899 | ext4_std_error(sb, err); |
900 | goto out; | |
1139575a TT |
901 | } |
902 | } | |
ffb5387e ES |
903 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
904 | err = ext4_journal_get_write_access(handle, inode_bitmap_bh); | |
eb9cc7e1 JK |
905 | if (err) { |
906 | ext4_std_error(sb, err); | |
907 | goto out; | |
908 | } | |
119c0d44 TT |
909 | ext4_lock_group(sb, group); |
910 | ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); | |
911 | ext4_unlock_group(sb, group); | |
912 | ino++; /* the inode bitmap is zero-based */ | |
913 | if (!ret2) | |
914 | goto got; /* we grabbed the inode! */ | |
19883bd9 | 915 | next_inode: |
119c0d44 TT |
916 | if (ino < EXT4_INODES_PER_GROUP(sb)) |
917 | goto repeat_in_this_group; | |
a34eb503 TT |
918 | next_group: |
919 | if (++group == ngroups) | |
920 | group = 0; | |
ac27a0ec DK |
921 | } |
922 | err = -ENOSPC; | |
923 | goto out; | |
924 | ||
925 | got: | |
ffb5387e ES |
926 | BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); |
927 | err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); | |
eb9cc7e1 JK |
928 | if (err) { |
929 | ext4_std_error(sb, err); | |
930 | goto out; | |
931 | } | |
ffb5387e | 932 | |
61c219f5 TT |
933 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
934 | err = ext4_journal_get_write_access(handle, group_desc_bh); | |
935 | if (err) { | |
936 | ext4_std_error(sb, err); | |
937 | goto out; | |
938 | } | |
939 | ||
717d50e4 | 940 | /* We may have to initialize the block bitmap if it isn't already */ |
feb0ab32 | 941 | if (ext4_has_group_desc_csum(sb) && |
717d50e4 | 942 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 943 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 944 | |
3300beda | 945 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
9008a58e DW |
946 | if (IS_ERR(block_bitmap_bh)) { |
947 | err = PTR_ERR(block_bitmap_bh); | |
599a9b77 JK |
948 | goto out; |
949 | } | |
3300beda AK |
950 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); |
951 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 952 | if (err) { |
3300beda | 953 | brelse(block_bitmap_bh); |
eb9cc7e1 JK |
954 | ext4_std_error(sb, err); |
955 | goto out; | |
717d50e4 AD |
956 | } |
957 | ||
fd034a84 TT |
958 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
959 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
fd034a84 | 960 | |
717d50e4 | 961 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 962 | ext4_lock_group(sb, group); |
717d50e4 | 963 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 964 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 965 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 966 | ext4_free_clusters_after_init(sb, group, gdp)); |
fa77dcfa | 967 | ext4_block_bitmap_csum_set(sb, group, gdp, |
79f1ba49 | 968 | block_bitmap_bh); |
feb0ab32 | 969 | ext4_group_desc_csum_set(sb, group, gdp); |
717d50e4 | 970 | } |
955ce5f5 | 971 | ext4_unlock_group(sb, group); |
aeb1e5d6 | 972 | brelse(block_bitmap_bh); |
717d50e4 | 973 | |
eb9cc7e1 JK |
974 | if (err) { |
975 | ext4_std_error(sb, err); | |
976 | goto out; | |
977 | } | |
717d50e4 | 978 | } |
119c0d44 | 979 | |
119c0d44 | 980 | /* Update the relevant bg descriptor fields */ |
41a246d1 | 981 | if (ext4_has_group_desc_csum(sb)) { |
119c0d44 TT |
982 | int free; |
983 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
984 | ||
985 | down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ | |
986 | ext4_lock_group(sb, group); /* while we modify the bg desc */ | |
987 | free = EXT4_INODES_PER_GROUP(sb) - | |
988 | ext4_itable_unused_count(sb, gdp); | |
989 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
990 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
991 | free = 0; | |
992 | } | |
993 | /* | |
994 | * Check the relative inode number against the last used | |
995 | * relative inode number in this group. if it is greater | |
996 | * we need to update the bg_itable_unused count | |
997 | */ | |
998 | if (ino > free) | |
999 | ext4_itable_unused_set(sb, gdp, | |
1000 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
1001 | up_read(&grp->alloc_sem); | |
6f2e9f0e TM |
1002 | } else { |
1003 | ext4_lock_group(sb, group); | |
119c0d44 | 1004 | } |
6f2e9f0e | 1005 | |
119c0d44 TT |
1006 | ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); |
1007 | if (S_ISDIR(mode)) { | |
1008 | ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); | |
1009 | if (sbi->s_log_groups_per_flex) { | |
1010 | ext4_group_t f = ext4_flex_group(sbi, group); | |
1011 | ||
1012 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); | |
1013 | } | |
1014 | } | |
41a246d1 DW |
1015 | if (ext4_has_group_desc_csum(sb)) { |
1016 | ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, | |
1017 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 1018 | ext4_group_desc_csum_set(sb, group, gdp); |
119c0d44 | 1019 | } |
6f2e9f0e | 1020 | ext4_unlock_group(sb, group); |
119c0d44 | 1021 | |
3300beda AK |
1022 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
1023 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
eb9cc7e1 JK |
1024 | if (err) { |
1025 | ext4_std_error(sb, err); | |
1026 | goto out; | |
1027 | } | |
ac27a0ec DK |
1028 | |
1029 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
1030 | if (S_ISDIR(mode)) | |
1031 | percpu_counter_inc(&sbi->s_dirs_counter); | |
ac27a0ec | 1032 | |
772cb7c8 JS |
1033 | if (sbi->s_log_groups_per_flex) { |
1034 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 1035 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 | 1036 | } |
ac27a0ec | 1037 | |
717d50e4 | 1038 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
1039 | /* This is the optimal IO size (for stat), not the fs block size */ |
1040 | inode->i_blocks = 0; | |
ef7f3835 KS |
1041 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
1042 | ext4_current_time(inode); | |
ac27a0ec DK |
1043 | |
1044 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
1045 | ei->i_dir_start_lookup = 0; | |
1046 | ei->i_disksize = 0; | |
1047 | ||
4af83508 | 1048 | /* Don't inherit extent flag from directory, amongst others. */ |
2dc6b0d4 DG |
1049 | ei->i_flags = |
1050 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 1051 | ei->i_file_acl = 0; |
ac27a0ec | 1052 | ei->i_dtime = 0; |
ac27a0ec | 1053 | ei->i_block_group = group; |
a4912123 | 1054 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1055 | |
617ba13b | 1056 | ext4_set_inode_flags(inode); |
ac27a0ec | 1057 | if (IS_DIRSYNC(inode)) |
0390131b | 1058 | ext4_handle_sync(handle); |
6b38e842 | 1059 | if (insert_inode_locked(inode) < 0) { |
acd6ad83 JK |
1060 | /* |
1061 | * Likely a bitmap corruption causing inode to be allocated | |
1062 | * twice. | |
1063 | */ | |
1064 | err = -EIO; | |
eb9cc7e1 JK |
1065 | ext4_error(sb, "failed to insert inode %lu: doubly allocated?", |
1066 | inode->i_ino); | |
1067 | goto out; | |
6b38e842 | 1068 | } |
ac27a0ec DK |
1069 | spin_lock(&sbi->s_next_gen_lock); |
1070 | inode->i_generation = sbi->s_next_generation++; | |
1071 | spin_unlock(&sbi->s_next_gen_lock); | |
1072 | ||
814525f4 | 1073 | /* Precompute checksum seed for inode metadata */ |
9aa5d32b | 1074 | if (ext4_has_metadata_csum(sb)) { |
814525f4 | 1075 | __u32 csum; |
814525f4 DW |
1076 | __le32 inum = cpu_to_le32(inode->i_ino); |
1077 | __le32 gen = cpu_to_le32(inode->i_generation); | |
1078 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
1079 | sizeof(inum)); | |
1080 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
1081 | sizeof(gen)); | |
1082 | } | |
1083 | ||
353eb83c | 1084 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 1085 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 KS |
1086 | |
1087 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
f08225d1 | 1088 | ei->i_inline_off = 0; |
e2b911c5 | 1089 | if (ext4_has_feature_inline_data(sb)) |
f08225d1 | 1090 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
ac27a0ec | 1091 | ret = inode; |
63936dda CH |
1092 | err = dquot_alloc_inode(inode); |
1093 | if (err) | |
ac27a0ec | 1094 | goto fail_drop; |
ac27a0ec | 1095 | |
617ba13b | 1096 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
1097 | if (err) |
1098 | goto fail_free_drop; | |
1099 | ||
2a7dba39 | 1100 | err = ext4_init_security(handle, inode, dir, qstr); |
ac27a0ec DK |
1101 | if (err) |
1102 | goto fail_free_drop; | |
1103 | ||
e2b911c5 | 1104 | if (ext4_has_feature_extents(sb)) { |
e4079a11 | 1105 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1106 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1107 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1108 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1109 | } |
a86c6181 | 1110 | } |
ac27a0ec | 1111 | |
688f869c TT |
1112 | if (ext4_handle_valid(handle)) { |
1113 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
1114 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
1115 | } | |
1116 | ||
e709e9df TT |
1117 | if (encrypt) { |
1118 | err = ext4_inherit_context(dir, inode); | |
1119 | if (err) | |
1120 | goto fail_free_drop; | |
1121 | } | |
1122 | ||
8753e88f AK |
1123 | err = ext4_mark_inode_dirty(handle, inode); |
1124 | if (err) { | |
1125 | ext4_std_error(sb, err); | |
1126 | goto fail_free_drop; | |
1127 | } | |
1128 | ||
617ba13b | 1129 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1130 | trace_ext4_allocate_inode(inode, dir, mode); |
3300beda | 1131 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1132 | return ret; |
1133 | ||
1134 | fail_free_drop: | |
63936dda | 1135 | dquot_free_inode(inode); |
ac27a0ec | 1136 | fail_drop: |
6d6b77f1 | 1137 | clear_nlink(inode); |
6b38e842 | 1138 | unlock_new_inode(inode); |
eb9cc7e1 JK |
1139 | out: |
1140 | dquot_drop(inode); | |
1141 | inode->i_flags |= S_NOQUOTA; | |
ac27a0ec | 1142 | iput(inode); |
3300beda | 1143 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1144 | return ERR_PTR(err); |
1145 | } | |
1146 | ||
1147 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1148 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1149 | { |
617ba13b | 1150 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1151 | ext4_group_t block_group; |
ac27a0ec | 1152 | int bit; |
1d1fe1ee | 1153 | struct buffer_head *bitmap_bh; |
ac27a0ec | 1154 | struct inode *inode = NULL; |
1d1fe1ee | 1155 | long err = -EIO; |
ac27a0ec DK |
1156 | |
1157 | /* Error cases - e2fsck has already cleaned up for us */ | |
1158 | if (ino > max_ino) { | |
12062ddd | 1159 | ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino); |
6a797d27 | 1160 | err = -EFSCORRUPTED; |
1d1fe1ee | 1161 | goto error; |
ac27a0ec DK |
1162 | } |
1163 | ||
617ba13b MC |
1164 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1165 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1166 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
9008a58e DW |
1167 | if (IS_ERR(bitmap_bh)) { |
1168 | err = PTR_ERR(bitmap_bh); | |
1169 | ext4_warning(sb, "inode bitmap error %ld for orphan %lu", | |
1170 | ino, err); | |
1d1fe1ee | 1171 | goto error; |
ac27a0ec DK |
1172 | } |
1173 | ||
1174 | /* Having the inode bit set should be a 100% indicator that this | |
1175 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1176 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1177 | */ | |
1d1fe1ee DH |
1178 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1179 | goto bad_orphan; | |
1180 | ||
1181 | inode = ext4_iget(sb, ino); | |
1182 | if (IS_ERR(inode)) | |
1183 | goto iget_failed; | |
1184 | ||
91ef4caf DG |
1185 | /* |
1186 | * If the orphans has i_nlinks > 0 then it should be able to be | |
1187 | * truncated, otherwise it won't be removed from the orphan list | |
1188 | * during processing and an infinite loop will result. | |
1189 | */ | |
1190 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
1191 | goto bad_orphan; | |
1192 | ||
1d1fe1ee DH |
1193 | if (NEXT_ORPHAN(inode) > max_ino) |
1194 | goto bad_orphan; | |
1195 | brelse(bitmap_bh); | |
1196 | return inode; | |
1197 | ||
1198 | iget_failed: | |
1199 | err = PTR_ERR(inode); | |
1200 | inode = NULL; | |
1201 | bad_orphan: | |
12062ddd | 1202 | ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino); |
8de5c325 | 1203 | printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n", |
1d1fe1ee DH |
1204 | bit, (unsigned long long)bitmap_bh->b_blocknr, |
1205 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
8de5c325 | 1206 | printk(KERN_WARNING "inode=%p\n", inode); |
1d1fe1ee | 1207 | if (inode) { |
8de5c325 | 1208 | printk(KERN_WARNING "is_bad_inode(inode)=%d\n", |
1d1fe1ee | 1209 | is_bad_inode(inode)); |
8de5c325 | 1210 | printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n", |
1d1fe1ee | 1211 | NEXT_ORPHAN(inode)); |
8de5c325 TT |
1212 | printk(KERN_WARNING "max_ino=%lu\n", max_ino); |
1213 | printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink); | |
ac27a0ec | 1214 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1215 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1216 | inode->i_blocks = 0; |
1217 | iput(inode); | |
ac27a0ec | 1218 | } |
ac27a0ec | 1219 | brelse(bitmap_bh); |
1d1fe1ee DH |
1220 | error: |
1221 | return ERR_PTR(err); | |
ac27a0ec DK |
1222 | } |
1223 | ||
af5bc92d | 1224 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1225 | { |
1226 | unsigned long desc_count; | |
617ba13b | 1227 | struct ext4_group_desc *gdp; |
8df9675f | 1228 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1229 | #ifdef EXT4FS_DEBUG |
1230 | struct ext4_super_block *es; | |
ac27a0ec DK |
1231 | unsigned long bitmap_count, x; |
1232 | struct buffer_head *bitmap_bh = NULL; | |
1233 | ||
617ba13b | 1234 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1235 | desc_count = 0; |
1236 | bitmap_count = 0; | |
1237 | gdp = NULL; | |
8df9675f | 1238 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1239 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1240 | if (!gdp) |
1241 | continue; | |
560671a0 | 1242 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1243 | brelse(bitmap_bh); |
e29d1cde | 1244 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
9008a58e DW |
1245 | if (IS_ERR(bitmap_bh)) { |
1246 | bitmap_bh = NULL; | |
ac27a0ec | 1247 | continue; |
9008a58e | 1248 | } |
ac27a0ec | 1249 | |
f6fb99ca TT |
1250 | x = ext4_count_free(bitmap_bh->b_data, |
1251 | EXT4_INODES_PER_GROUP(sb) / 8); | |
c549a95d | 1252 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1253 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1254 | bitmap_count += x; |
1255 | } | |
1256 | brelse(bitmap_bh); | |
4776004f TT |
1257 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1258 | "stored = %u, computed = %lu, %lu\n", | |
1259 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1260 | return desc_count; |
1261 | #else | |
1262 | desc_count = 0; | |
8df9675f | 1263 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1264 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1265 | if (!gdp) |
1266 | continue; | |
560671a0 | 1267 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1268 | cond_resched(); |
1269 | } | |
1270 | return desc_count; | |
1271 | #endif | |
1272 | } | |
1273 | ||
1274 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1275 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1276 | { |
1277 | unsigned long count = 0; | |
8df9675f | 1278 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1279 | |
8df9675f | 1280 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1281 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1282 | if (!gdp) |
1283 | continue; | |
560671a0 | 1284 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1285 | } |
1286 | return count; | |
1287 | } | |
bfff6873 LC |
1288 | |
1289 | /* | |
1290 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1291 | * inode table. Must be called without any spinlock held. The only place | |
1292 | * where it is called from on active part of filesystem is ext4lazyinit | |
1293 | * thread, so we do not need any special locks, however we have to prevent | |
1294 | * inode allocation from the current group, so we take alloc_sem lock, to | |
119c0d44 | 1295 | * block ext4_new_inode() until we are finished. |
bfff6873 | 1296 | */ |
e0cbee3e | 1297 | int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, |
bfff6873 LC |
1298 | int barrier) |
1299 | { | |
1300 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1301 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1302 | struct ext4_group_desc *gdp = NULL; | |
1303 | struct buffer_head *group_desc_bh; | |
1304 | handle_t *handle; | |
1305 | ext4_fsblk_t blk; | |
1306 | int num, ret = 0, used_blks = 0; | |
bfff6873 LC |
1307 | |
1308 | /* This should not happen, but just to be sure check this */ | |
1309 | if (sb->s_flags & MS_RDONLY) { | |
1310 | ret = 1; | |
1311 | goto out; | |
1312 | } | |
1313 | ||
1314 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1315 | if (!gdp) | |
1316 | goto out; | |
1317 | ||
1318 | /* | |
1319 | * We do not need to lock this, because we are the only one | |
1320 | * handling this flag. | |
1321 | */ | |
1322 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1323 | goto out; | |
1324 | ||
9924a92a | 1325 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
bfff6873 LC |
1326 | if (IS_ERR(handle)) { |
1327 | ret = PTR_ERR(handle); | |
1328 | goto out; | |
1329 | } | |
1330 | ||
1331 | down_write(&grp->alloc_sem); | |
1332 | /* | |
1333 | * If inode bitmap was already initialized there may be some | |
1334 | * used inodes so we need to skip blocks with used inodes in | |
1335 | * inode table. | |
1336 | */ | |
1337 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1338 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1339 | ext4_itable_unused_count(sb, gdp)), | |
1340 | sbi->s_inodes_per_block); | |
1341 | ||
857ac889 | 1342 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) { |
1084f252 TT |
1343 | ext4_error(sb, "Something is wrong with group %u: " |
1344 | "used itable blocks: %d; " | |
1345 | "itable unused count: %u", | |
857ac889 LC |
1346 | group, used_blks, |
1347 | ext4_itable_unused_count(sb, gdp)); | |
1348 | ret = 1; | |
33853a0d | 1349 | goto err_out; |
857ac889 LC |
1350 | } |
1351 | ||
bfff6873 LC |
1352 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1353 | num = sbi->s_itb_per_group - used_blks; | |
1354 | ||
1355 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1356 | ret = ext4_journal_get_write_access(handle, | |
1357 | group_desc_bh); | |
1358 | if (ret) | |
1359 | goto err_out; | |
1360 | ||
bfff6873 LC |
1361 | /* |
1362 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1363 | * flag anyway, because obviously, when it is full it does not need | |
1364 | * further zeroing. | |
1365 | */ | |
1366 | if (unlikely(num == 0)) | |
1367 | goto skip_zeroout; | |
1368 | ||
1369 | ext4_debug("going to zero out inode table in group %d\n", | |
1370 | group); | |
a107e5a3 | 1371 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1372 | if (ret < 0) |
1373 | goto err_out; | |
a107e5a3 TT |
1374 | if (barrier) |
1375 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); | |
bfff6873 LC |
1376 | |
1377 | skip_zeroout: | |
1378 | ext4_lock_group(sb, group); | |
1379 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
feb0ab32 | 1380 | ext4_group_desc_csum_set(sb, group, gdp); |
bfff6873 LC |
1381 | ext4_unlock_group(sb, group); |
1382 | ||
1383 | BUFFER_TRACE(group_desc_bh, | |
1384 | "call ext4_handle_dirty_metadata"); | |
1385 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1386 | group_desc_bh); | |
1387 | ||
1388 | err_out: | |
1389 | up_write(&grp->alloc_sem); | |
1390 | ext4_journal_stop(handle); | |
1391 | out: | |
1392 | return ret; | |
1393 | } |