]>
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> |
5b825c3a IM |
24 | #include <linux/cred.h> |
25 | ||
ac27a0ec | 26 | #include <asm/byteorder.h> |
9bffad1e | 27 | |
3dcf5451 CH |
28 | #include "ext4.h" |
29 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
30 | #include "xattr.h" |
31 | #include "acl.h" | |
32 | ||
9bffad1e TT |
33 | #include <trace/events/ext4.h> |
34 | ||
ac27a0ec DK |
35 | /* |
36 | * ialloc.c contains the inodes allocation and deallocation routines | |
37 | */ | |
38 | ||
39 | /* | |
40 | * The free inodes are managed by bitmaps. A file system contains several | |
41 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
42 | * block for inodes, N blocks for the inode table and data blocks. | |
43 | * | |
44 | * The file system contains group descriptors which are located after the | |
45 | * super block. Each descriptor contains the number of the bitmap block and | |
46 | * the free blocks count in the block. | |
47 | */ | |
48 | ||
717d50e4 AD |
49 | /* |
50 | * To avoid calling the atomic setbit hundreds or thousands of times, we only | |
51 | * need to use it within a single byte (to ensure we get endianness right). | |
52 | * We can use memset for the rest of the bitmap as there are no other users. | |
53 | */ | |
61d08673 | 54 | void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap) |
717d50e4 AD |
55 | { |
56 | int i; | |
57 | ||
58 | if (start_bit >= end_bit) | |
59 | return; | |
60 | ||
61 | ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); | |
62 | for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) | |
63 | ext4_set_bit(i, bitmap); | |
64 | if (i < end_bit) | |
65 | memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); | |
66 | } | |
67 | ||
68 | /* Initializes an uninitialized inode bitmap */ | |
9008a58e | 69 | static int ext4_init_inode_bitmap(struct super_block *sb, |
1f109d5a TT |
70 | struct buffer_head *bh, |
71 | ext4_group_t block_group, | |
72 | struct ext4_group_desc *gdp) | |
717d50e4 | 73 | { |
bdfb6ff4 | 74 | struct ext4_group_info *grp; |
e43bb4e6 | 75 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
717d50e4 AD |
76 | J_ASSERT_BH(bh, buffer_locked(bh)); |
77 | ||
78 | /* If checksum is bad mark all blocks and inodes use to prevent | |
79 | * allocation, essentially implementing a per-group read-only flag. */ | |
feb0ab32 | 80 | if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) { |
bdfb6ff4 | 81 | grp = ext4_get_group_info(sb, block_group); |
e43bb4e6 NJ |
82 | if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) |
83 | percpu_counter_sub(&sbi->s_freeclusters_counter, | |
84 | grp->bb_free); | |
bdfb6ff4 | 85 | set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state); |
e43bb4e6 NJ |
86 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
87 | int count; | |
88 | count = ext4_free_inodes_count(sb, gdp); | |
89 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
90 | count); | |
91 | } | |
bdfb6ff4 | 92 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
9008a58e | 93 | return -EFSBADCRC; |
717d50e4 AD |
94 | } |
95 | ||
96 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
61d08673 | 97 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
717d50e4 | 98 | bh->b_data); |
41a246d1 DW |
99 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh, |
100 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 101 | ext4_group_desc_csum_set(sb, block_group, gdp); |
717d50e4 | 102 | |
9008a58e | 103 | return 0; |
717d50e4 | 104 | } |
ac27a0ec | 105 | |
813e5727 TT |
106 | void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate) |
107 | { | |
108 | if (uptodate) { | |
109 | set_buffer_uptodate(bh); | |
110 | set_bitmap_uptodate(bh); | |
111 | } | |
112 | unlock_buffer(bh); | |
113 | put_bh(bh); | |
114 | } | |
115 | ||
9008a58e DW |
116 | static int ext4_validate_inode_bitmap(struct super_block *sb, |
117 | struct ext4_group_desc *desc, | |
118 | ext4_group_t block_group, | |
119 | struct buffer_head *bh) | |
120 | { | |
121 | ext4_fsblk_t blk; | |
122 | struct ext4_group_info *grp = ext4_get_group_info(sb, block_group); | |
123 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
124 | ||
125 | if (buffer_verified(bh)) | |
126 | return 0; | |
127 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) | |
128 | return -EFSCORRUPTED; | |
129 | ||
130 | ext4_lock_group(sb, block_group); | |
131 | blk = ext4_inode_bitmap(sb, desc); | |
132 | if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh, | |
133 | EXT4_INODES_PER_GROUP(sb) / 8)) { | |
134 | ext4_unlock_group(sb, block_group); | |
135 | ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " | |
136 | "inode_bitmap = %llu", block_group, blk); | |
137 | grp = ext4_get_group_info(sb, block_group); | |
138 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { | |
139 | int count; | |
140 | count = ext4_free_inodes_count(sb, desc); | |
141 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
142 | count); | |
143 | } | |
144 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); | |
145 | return -EFSBADCRC; | |
146 | } | |
147 | set_buffer_verified(bh); | |
148 | ext4_unlock_group(sb, block_group); | |
149 | return 0; | |
150 | } | |
151 | ||
ac27a0ec DK |
152 | /* |
153 | * Read the inode allocation bitmap for a given block_group, reading | |
154 | * into the specified slot in the superblock's bitmap cache. | |
155 | * | |
156 | * Return buffer_head of bitmap on success or NULL. | |
157 | */ | |
158 | static struct buffer_head * | |
e29d1cde | 159 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 160 | { |
617ba13b | 161 | struct ext4_group_desc *desc; |
ac27a0ec | 162 | struct buffer_head *bh = NULL; |
e29d1cde | 163 | ext4_fsblk_t bitmap_blk; |
9008a58e | 164 | int err; |
ac27a0ec | 165 | |
617ba13b | 166 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 167 | if (!desc) |
9008a58e | 168 | return ERR_PTR(-EFSCORRUPTED); |
bfff6873 | 169 | |
e29d1cde ES |
170 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
171 | bh = sb_getblk(sb, bitmap_blk); | |
172 | if (unlikely(!bh)) { | |
12062ddd | 173 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 174 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde | 175 | block_group, bitmap_blk); |
9008a58e | 176 | return ERR_PTR(-EIO); |
e29d1cde | 177 | } |
2ccb5fb9 | 178 | if (bitmap_uptodate(bh)) |
41a246d1 | 179 | goto verify; |
e29d1cde | 180 | |
c806e68f | 181 | lock_buffer(bh); |
2ccb5fb9 AK |
182 | if (bitmap_uptodate(bh)) { |
183 | unlock_buffer(bh); | |
41a246d1 | 184 | goto verify; |
2ccb5fb9 | 185 | } |
bfff6873 | 186 | |
955ce5f5 | 187 | ext4_lock_group(sb, block_group); |
717d50e4 | 188 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
9008a58e | 189 | err = ext4_init_inode_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 190 | set_bitmap_uptodate(bh); |
e29d1cde | 191 | set_buffer_uptodate(bh); |
41a246d1 | 192 | set_buffer_verified(bh); |
955ce5f5 | 193 | ext4_unlock_group(sb, block_group); |
3300beda | 194 | unlock_buffer(bh); |
05145bd7 JK |
195 | if (err) { |
196 | ext4_error(sb, "Failed to init inode bitmap for group " | |
197 | "%u: %d", block_group, err); | |
9008a58e | 198 | goto out; |
05145bd7 | 199 | } |
e29d1cde | 200 | return bh; |
717d50e4 | 201 | } |
955ce5f5 | 202 | ext4_unlock_group(sb, block_group); |
bfff6873 | 203 | |
2ccb5fb9 AK |
204 | if (buffer_uptodate(bh)) { |
205 | /* | |
206 | * if not uninit if bh is uptodate, | |
207 | * bitmap is also uptodate | |
208 | */ | |
209 | set_bitmap_uptodate(bh); | |
210 | unlock_buffer(bh); | |
41a246d1 | 211 | goto verify; |
2ccb5fb9 AK |
212 | } |
213 | /* | |
813e5727 | 214 | * submit the buffer_head for reading |
2ccb5fb9 | 215 | */ |
0562e0ba | 216 | trace_ext4_load_inode_bitmap(sb, block_group); |
813e5727 TT |
217 | bh->b_end_io = ext4_end_bitmap_read; |
218 | get_bh(bh); | |
2a222ca9 | 219 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
813e5727 TT |
220 | wait_on_buffer(bh); |
221 | if (!buffer_uptodate(bh)) { | |
e29d1cde | 222 | put_bh(bh); |
12062ddd | 223 | ext4_error(sb, "Cannot read inode bitmap - " |
813e5727 TT |
224 | "block_group = %u, inode_bitmap = %llu", |
225 | block_group, bitmap_blk); | |
9008a58e | 226 | return ERR_PTR(-EIO); |
e29d1cde | 227 | } |
41a246d1 DW |
228 | |
229 | verify: | |
9008a58e DW |
230 | err = ext4_validate_inode_bitmap(sb, desc, block_group, bh); |
231 | if (err) | |
232 | goto out; | |
ac27a0ec | 233 | return bh; |
9008a58e DW |
234 | out: |
235 | put_bh(bh); | |
236 | return ERR_PTR(err); | |
ac27a0ec DK |
237 | } |
238 | ||
239 | /* | |
240 | * NOTE! When we get the inode, we're the only people | |
241 | * that have access to it, and as such there are no | |
242 | * race conditions we have to worry about. The inode | |
243 | * is not on the hash-lists, and it cannot be reached | |
244 | * through the filesystem because the directory entry | |
245 | * has been deleted earlier. | |
246 | * | |
247 | * HOWEVER: we must make sure that we get no aliases, | |
248 | * which means that we have to call "clear_inode()" | |
249 | * _before_ we mark the inode not in use in the inode | |
250 | * bitmaps. Otherwise a newly created file might use | |
251 | * the same inode number (not actually the same pointer | |
252 | * though), and then we'd have two inodes sharing the | |
253 | * same inode number and space on the harddisk. | |
254 | */ | |
af5bc92d | 255 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 256 | { |
af5bc92d | 257 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
258 | int is_directory; |
259 | unsigned long ino; | |
260 | struct buffer_head *bitmap_bh = NULL; | |
261 | struct buffer_head *bh2; | |
fd2d4291 | 262 | ext4_group_t block_group; |
ac27a0ec | 263 | unsigned long bit; |
af5bc92d TT |
264 | struct ext4_group_desc *gdp; |
265 | struct ext4_super_block *es; | |
617ba13b | 266 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 267 | int fatal = 0, err, count, cleared; |
87a39389 | 268 | struct ext4_group_info *grp; |
ac27a0ec | 269 | |
92b97816 TT |
270 | if (!sb) { |
271 | printk(KERN_ERR "EXT4-fs: %s:%d: inode on " | |
272 | "nonexistent device\n", __func__, __LINE__); | |
ac27a0ec DK |
273 | return; |
274 | } | |
92b97816 TT |
275 | if (atomic_read(&inode->i_count) > 1) { |
276 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", | |
277 | __func__, __LINE__, inode->i_ino, | |
278 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
279 | return; |
280 | } | |
92b97816 TT |
281 | if (inode->i_nlink) { |
282 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", | |
283 | __func__, __LINE__, inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
284 | return; |
285 | } | |
617ba13b | 286 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
287 | |
288 | ino = inode->i_ino; | |
af5bc92d | 289 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 290 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
291 | |
292 | /* | |
293 | * Note: we must free any quota before locking the superblock, | |
294 | * as writing the quota to disk may need the lock as well. | |
295 | */ | |
871a2931 | 296 | dquot_initialize(inode); |
617ba13b | 297 | ext4_xattr_delete_inode(handle, inode); |
63936dda | 298 | dquot_free_inode(inode); |
9f754758 | 299 | dquot_drop(inode); |
ac27a0ec DK |
300 | |
301 | is_directory = S_ISDIR(inode->i_mode); | |
302 | ||
303 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 304 | ext4_clear_inode(inode); |
ac27a0ec | 305 | |
617ba13b MC |
306 | es = EXT4_SB(sb)->s_es; |
307 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 308 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
309 | goto error_return; |
310 | } | |
617ba13b MC |
311 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
312 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 313 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
87a39389 DW |
314 | /* Don't bother if the inode bitmap is corrupt. */ |
315 | grp = ext4_get_group_info(sb, block_group); | |
9008a58e DW |
316 | if (IS_ERR(bitmap_bh)) { |
317 | fatal = PTR_ERR(bitmap_bh); | |
318 | bitmap_bh = NULL; | |
319 | goto error_return; | |
320 | } | |
321 | if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) { | |
322 | fatal = -EFSCORRUPTED; | |
ac27a0ec | 323 | goto error_return; |
9008a58e | 324 | } |
ac27a0ec DK |
325 | |
326 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 327 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
328 | if (fatal) |
329 | goto error_return; | |
330 | ||
d17413c0 DM |
331 | fatal = -ESRCH; |
332 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
333 | if (gdp) { | |
ac27a0ec | 334 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 335 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
336 | } |
337 | ext4_lock_group(sb, block_group); | |
597d508c | 338 | cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); |
d17413c0 DM |
339 | if (fatal || !cleared) { |
340 | ext4_unlock_group(sb, block_group); | |
341 | goto out; | |
342 | } | |
7d39db14 | 343 | |
d17413c0 DM |
344 | count = ext4_free_inodes_count(sb, gdp) + 1; |
345 | ext4_free_inodes_set(sb, gdp, count); | |
346 | if (is_directory) { | |
347 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
348 | ext4_used_dirs_set(sb, gdp, count); | |
349 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 350 | } |
41a246d1 DW |
351 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, |
352 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 353 | ext4_group_desc_csum_set(sb, block_group, gdp); |
d17413c0 | 354 | ext4_unlock_group(sb, block_group); |
ac27a0ec | 355 | |
d17413c0 DM |
356 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
357 | if (sbi->s_log_groups_per_flex) { | |
358 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 359 | |
d17413c0 DM |
360 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
361 | if (is_directory) | |
362 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 363 | } |
d17413c0 DM |
364 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
365 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
366 | out: | |
367 | if (cleared) { | |
368 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
369 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
370 | if (!fatal) | |
371 | fatal = err; | |
87a39389 | 372 | } else { |
d17413c0 | 373 | ext4_error(sb, "bit already cleared for inode %lu", ino); |
bf40c926 | 374 | if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
e43bb4e6 NJ |
375 | int count; |
376 | count = ext4_free_inodes_count(sb, gdp); | |
377 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
378 | count); | |
379 | } | |
87a39389 DW |
380 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
381 | } | |
d17413c0 | 382 | |
ac27a0ec DK |
383 | error_return: |
384 | brelse(bitmap_bh); | |
617ba13b | 385 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
386 | } |
387 | ||
a4912123 | 388 | struct orlov_stats { |
90ba983f | 389 | __u64 free_clusters; |
a4912123 | 390 | __u32 free_inodes; |
a4912123 TT |
391 | __u32 used_dirs; |
392 | }; | |
393 | ||
394 | /* | |
395 | * Helper function for Orlov's allocator; returns critical information | |
396 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
397 | * is a block group number; otherwise it is flex_bg number. | |
398 | */ | |
1f109d5a TT |
399 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
400 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
401 | { |
402 | struct ext4_group_desc *desc; | |
7d39db14 | 403 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 404 | |
7d39db14 TT |
405 | if (flex_size > 1) { |
406 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
90ba983f | 407 | stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); |
7d39db14 TT |
408 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); |
409 | return; | |
410 | } | |
a4912123 | 411 | |
7d39db14 TT |
412 | desc = ext4_get_group_desc(sb, g, NULL); |
413 | if (desc) { | |
414 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
021b65bb | 415 | stats->free_clusters = ext4_free_group_clusters(sb, desc); |
7d39db14 TT |
416 | stats->used_dirs = ext4_used_dirs_count(sb, desc); |
417 | } else { | |
418 | stats->free_inodes = 0; | |
24aaa8ef | 419 | stats->free_clusters = 0; |
7d39db14 | 420 | stats->used_dirs = 0; |
a4912123 TT |
421 | } |
422 | } | |
423 | ||
ac27a0ec DK |
424 | /* |
425 | * Orlov's allocator for directories. | |
426 | * | |
427 | * We always try to spread first-level directories. | |
428 | * | |
429 | * If there are blockgroups with both free inodes and free blocks counts | |
430 | * not worse than average we return one with smallest directory count. | |
431 | * Otherwise we simply return a random group. | |
432 | * | |
433 | * For the rest rules look so: | |
434 | * | |
435 | * It's OK to put directory into a group unless | |
436 | * it has too many directories already (max_dirs) or | |
437 | * it has too few free inodes left (min_inodes) or | |
438 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 439 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
440 | * conditions we search cyclically through the rest. If none |
441 | * of the groups look good we just look for a group with more | |
442 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
443 | */ |
444 | ||
2aa9fc4c | 445 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
dcca3fec | 446 | ext4_group_t *group, umode_t mode, |
f157a4aa | 447 | const struct qstr *qstr) |
ac27a0ec | 448 | { |
fd2d4291 | 449 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 450 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 451 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 452 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
14c83c9f | 453 | unsigned int freei, avefreei, grp_free; |
24aaa8ef | 454 | ext4_fsblk_t freeb, avefreec; |
ac27a0ec | 455 | unsigned int ndirs; |
a4912123 | 456 | int max_dirs, min_inodes; |
24aaa8ef | 457 | ext4_grpblk_t min_clusters; |
8df9675f | 458 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 459 | struct ext4_group_desc *desc; |
a4912123 TT |
460 | struct orlov_stats stats; |
461 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 462 | struct dx_hash_info hinfo; |
a4912123 | 463 | |
8df9675f | 464 | ngroups = real_ngroups; |
a4912123 | 465 | if (flex_size > 1) { |
8df9675f | 466 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
467 | sbi->s_log_groups_per_flex; |
468 | parent_group >>= sbi->s_log_groups_per_flex; | |
469 | } | |
ac27a0ec DK |
470 | |
471 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
472 | avefreei = freei / ngroups; | |
57042651 TT |
473 | freeb = EXT4_C2B(sbi, |
474 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
24aaa8ef TT |
475 | avefreec = freeb; |
476 | do_div(avefreec, ngroups); | |
ac27a0ec DK |
477 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
478 | ||
a4912123 | 479 | if (S_ISDIR(mode) && |
2b0143b5 | 480 | ((parent == d_inode(sb->s_root)) || |
12e9b892 | 481 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 482 | int best_ndir = inodes_per_group; |
2aa9fc4c | 483 | int ret = -1; |
ac27a0ec | 484 | |
f157a4aa TT |
485 | if (qstr) { |
486 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
487 | hinfo.seed = sbi->s_hash_seed; | |
488 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
489 | grp = hinfo.hash; | |
490 | } else | |
dd1f723b | 491 | grp = prandom_u32(); |
2aa9fc4c | 492 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 493 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
494 | g = (parent_group + i) % ngroups; |
495 | get_orlov_stats(sb, g, flex_size, &stats); | |
496 | if (!stats.free_inodes) | |
ac27a0ec | 497 | continue; |
a4912123 | 498 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 499 | continue; |
a4912123 | 500 | if (stats.free_inodes < avefreei) |
ac27a0ec | 501 | continue; |
24aaa8ef | 502 | if (stats.free_clusters < avefreec) |
ac27a0ec | 503 | continue; |
a4912123 | 504 | grp = g; |
2aa9fc4c | 505 | ret = 0; |
a4912123 TT |
506 | best_ndir = stats.used_dirs; |
507 | } | |
508 | if (ret) | |
509 | goto fallback; | |
510 | found_flex_bg: | |
511 | if (flex_size == 1) { | |
512 | *group = grp; | |
513 | return 0; | |
514 | } | |
515 | ||
516 | /* | |
517 | * We pack inodes at the beginning of the flexgroup's | |
518 | * inode tables. Block allocation decisions will do | |
519 | * something similar, although regular files will | |
520 | * start at 2nd block group of the flexgroup. See | |
521 | * ext4_ext_find_goal() and ext4_find_near(). | |
522 | */ | |
523 | grp *= flex_size; | |
524 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 525 | if (grp+i >= real_ngroups) |
a4912123 TT |
526 | break; |
527 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
528 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
529 | *group = grp+i; | |
530 | return 0; | |
531 | } | |
ac27a0ec | 532 | } |
ac27a0ec DK |
533 | goto fallback; |
534 | } | |
535 | ||
ac27a0ec | 536 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
537 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
538 | if (min_inodes < 1) | |
539 | min_inodes = 1; | |
24aaa8ef | 540 | min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; |
a4912123 TT |
541 | |
542 | /* | |
543 | * Start looking in the flex group where we last allocated an | |
544 | * inode for this parent directory | |
545 | */ | |
546 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
547 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
548 | if (flex_size > 1) | |
549 | parent_group >>= sbi->s_log_groups_per_flex; | |
550 | } | |
ac27a0ec DK |
551 | |
552 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
553 | grp = (parent_group + i) % ngroups; |
554 | get_orlov_stats(sb, grp, flex_size, &stats); | |
555 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 556 | continue; |
a4912123 | 557 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 558 | continue; |
24aaa8ef | 559 | if (stats.free_clusters < min_clusters) |
ac27a0ec | 560 | continue; |
a4912123 | 561 | goto found_flex_bg; |
ac27a0ec DK |
562 | } |
563 | ||
564 | fallback: | |
8df9675f | 565 | ngroups = real_ngroups; |
a4912123 | 566 | avefreei = freei / ngroups; |
b5451f7b | 567 | fallback_retry: |
a4912123 | 568 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 569 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
570 | grp = (parent_group + i) % ngroups; |
571 | desc = ext4_get_group_desc(sb, grp, NULL); | |
bb3d132a DC |
572 | if (desc) { |
573 | grp_free = ext4_free_inodes_count(sb, desc); | |
574 | if (grp_free && grp_free >= avefreei) { | |
575 | *group = grp; | |
576 | return 0; | |
577 | } | |
a4912123 | 578 | } |
ac27a0ec DK |
579 | } |
580 | ||
581 | if (avefreei) { | |
582 | /* | |
583 | * The free-inodes counter is approximate, and for really small | |
584 | * filesystems the above test can fail to find any blockgroups | |
585 | */ | |
586 | avefreei = 0; | |
b5451f7b | 587 | goto fallback_retry; |
ac27a0ec DK |
588 | } |
589 | ||
590 | return -1; | |
591 | } | |
592 | ||
2aa9fc4c | 593 | static int find_group_other(struct super_block *sb, struct inode *parent, |
dcca3fec | 594 | ext4_group_t *group, umode_t mode) |
ac27a0ec | 595 | { |
fd2d4291 | 596 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 597 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 598 | struct ext4_group_desc *desc; |
a4912123 TT |
599 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
600 | ||
601 | /* | |
602 | * Try to place the inode is the same flex group as its | |
603 | * parent. If we can't find space, use the Orlov algorithm to | |
604 | * find another flex group, and store that information in the | |
605 | * parent directory's inode information so that use that flex | |
606 | * group for future allocations. | |
607 | */ | |
608 | if (flex_size > 1) { | |
609 | int retry = 0; | |
610 | ||
611 | try_again: | |
612 | parent_group &= ~(flex_size-1); | |
613 | last = parent_group + flex_size; | |
614 | if (last > ngroups) | |
615 | last = ngroups; | |
616 | for (i = parent_group; i < last; i++) { | |
617 | desc = ext4_get_group_desc(sb, i, NULL); | |
618 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
619 | *group = i; | |
620 | return 0; | |
621 | } | |
622 | } | |
623 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
624 | retry = 1; | |
625 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
626 | goto try_again; | |
627 | } | |
628 | /* | |
629 | * If this didn't work, use the Orlov search algorithm | |
630 | * to find a new flex group; we pass in the mode to | |
631 | * avoid the topdir algorithms. | |
632 | */ | |
633 | *group = parent_group + flex_size; | |
634 | if (*group > ngroups) | |
635 | *group = 0; | |
7dc57615 | 636 | return find_group_orlov(sb, parent, group, mode, NULL); |
a4912123 | 637 | } |
ac27a0ec DK |
638 | |
639 | /* | |
640 | * Try to place the inode in its parent directory | |
641 | */ | |
2aa9fc4c AM |
642 | *group = parent_group; |
643 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 644 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 645 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 646 | return 0; |
ac27a0ec DK |
647 | |
648 | /* | |
649 | * We're going to place this inode in a different blockgroup from its | |
650 | * parent. We want to cause files in a common directory to all land in | |
651 | * the same blockgroup. But we want files which are in a different | |
652 | * directory which shares a blockgroup with our parent to land in a | |
653 | * different blockgroup. | |
654 | * | |
655 | * So add our directory's i_ino into the starting point for the hash. | |
656 | */ | |
2aa9fc4c | 657 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
658 | |
659 | /* | |
660 | * Use a quadratic hash to find a group with a free inode and some free | |
661 | * blocks. | |
662 | */ | |
663 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
664 | *group += i; |
665 | if (*group >= ngroups) | |
666 | *group -= ngroups; | |
667 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 668 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 669 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 670 | return 0; |
ac27a0ec DK |
671 | } |
672 | ||
673 | /* | |
674 | * That failed: try linear search for a free inode, even if that group | |
675 | * has no free blocks. | |
676 | */ | |
2aa9fc4c | 677 | *group = parent_group; |
ac27a0ec | 678 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
679 | if (++*group >= ngroups) |
680 | *group = 0; | |
681 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 682 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 683 | return 0; |
ac27a0ec DK |
684 | } |
685 | ||
686 | return -1; | |
687 | } | |
688 | ||
19883bd9 TT |
689 | /* |
690 | * In no journal mode, if an inode has recently been deleted, we want | |
691 | * to avoid reusing it until we're reasonably sure the inode table | |
692 | * block has been written back to disk. (Yes, these values are | |
693 | * somewhat arbitrary...) | |
694 | */ | |
695 | #define RECENTCY_MIN 5 | |
696 | #define RECENTCY_DIRTY 30 | |
697 | ||
698 | static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) | |
699 | { | |
700 | struct ext4_group_desc *gdp; | |
701 | struct ext4_inode *raw_inode; | |
702 | struct buffer_head *bh; | |
703 | unsigned long dtime, now; | |
704 | int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
705 | int offset, ret = 0, recentcy = RECENTCY_MIN; | |
706 | ||
707 | gdp = ext4_get_group_desc(sb, group, NULL); | |
708 | if (unlikely(!gdp)) | |
709 | return 0; | |
710 | ||
711 | bh = sb_getblk(sb, ext4_inode_table(sb, gdp) + | |
712 | (ino / inodes_per_block)); | |
713 | if (unlikely(!bh) || !buffer_uptodate(bh)) | |
714 | /* | |
715 | * If the block is not in the buffer cache, then it | |
716 | * must have been written out. | |
717 | */ | |
718 | goto out; | |
719 | ||
720 | offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
721 | raw_inode = (struct ext4_inode *) (bh->b_data + offset); | |
722 | dtime = le32_to_cpu(raw_inode->i_dtime); | |
723 | now = get_seconds(); | |
724 | if (buffer_dirty(bh)) | |
725 | recentcy += RECENTCY_DIRTY; | |
726 | ||
727 | if (dtime && (dtime < now) && (now < dtime + recentcy)) | |
728 | ret = 1; | |
729 | out: | |
730 | brelse(bh); | |
731 | return ret; | |
732 | } | |
733 | ||
ac27a0ec DK |
734 | /* |
735 | * There are two policies for allocating an inode. If the new inode is | |
736 | * a directory, then a forward search is made for a block group with both | |
737 | * free space and a low directory-to-inode ratio; if that fails, then of | |
738 | * the groups with above-average free space, that group with the fewest | |
739 | * directories already is chosen. | |
740 | * | |
741 | * For other inodes, search forward from the parent directory's block | |
742 | * group to find a free inode. | |
743 | */ | |
1139575a TT |
744 | struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, |
745 | umode_t mode, const struct qstr *qstr, | |
746 | __u32 goal, uid_t *owner, int handle_type, | |
747 | unsigned int line_no, int nblocks) | |
ac27a0ec DK |
748 | { |
749 | struct super_block *sb; | |
3300beda AK |
750 | struct buffer_head *inode_bitmap_bh = NULL; |
751 | struct buffer_head *group_desc_bh; | |
8df9675f | 752 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 753 | unsigned long ino = 0; |
af5bc92d TT |
754 | struct inode *inode; |
755 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
756 | struct ext4_inode_info *ei; |
757 | struct ext4_sb_info *sbi; | |
a7cdadee | 758 | int ret2, err; |
ac27a0ec | 759 | struct inode *ret; |
2aa9fc4c | 760 | ext4_group_t i; |
772cb7c8 | 761 | ext4_group_t flex_group; |
87a39389 | 762 | struct ext4_group_info *grp; |
e709e9df | 763 | int encrypt = 0; |
ac27a0ec DK |
764 | |
765 | /* Cannot create files in a deleted directory */ | |
766 | if (!dir || !dir->i_nlink) | |
767 | return ERR_PTR(-EPERM); | |
768 | ||
0db1ff22 TT |
769 | if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb)))) |
770 | return ERR_PTR(-EIO); | |
771 | ||
e709e9df TT |
772 | if ((ext4_encrypted_inode(dir) || |
773 | DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) && | |
774 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) { | |
a7550b30 | 775 | err = fscrypt_get_encryption_info(dir); |
e709e9df TT |
776 | if (err) |
777 | return ERR_PTR(err); | |
a7550b30 | 778 | if (!fscrypt_has_encryption_key(dir)) |
54475f53 | 779 | return ERR_PTR(-ENOKEY); |
e709e9df TT |
780 | if (!handle) |
781 | nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb); | |
782 | encrypt = 1; | |
783 | } | |
784 | ||
ac27a0ec | 785 | sb = dir->i_sb; |
8df9675f | 786 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 787 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
788 | inode = new_inode(sb); |
789 | if (!inode) | |
790 | return ERR_PTR(-ENOMEM); | |
617ba13b | 791 | ei = EXT4_I(inode); |
617ba13b | 792 | sbi = EXT4_SB(sb); |
772cb7c8 | 793 | |
eb9cc7e1 | 794 | /* |
b8a07463 | 795 | * Initialize owners and quota early so that we don't have to account |
eb9cc7e1 JK |
796 | * for quota initialization worst case in standard inode creating |
797 | * transaction | |
798 | */ | |
799 | if (owner) { | |
800 | inode->i_mode = mode; | |
801 | i_uid_write(inode, owner[0]); | |
802 | i_gid_write(inode, owner[1]); | |
803 | } else if (test_opt(sb, GRPID)) { | |
804 | inode->i_mode = mode; | |
805 | inode->i_uid = current_fsuid(); | |
806 | inode->i_gid = dir->i_gid; | |
807 | } else | |
808 | inode_init_owner(inode, dir, mode); | |
040cb378 | 809 | |
0b7b7779 | 810 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
811 | ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) |
812 | ei->i_projid = EXT4_I(dir)->i_projid; | |
813 | else | |
814 | ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID); | |
815 | ||
a7cdadee JK |
816 | err = dquot_initialize(inode); |
817 | if (err) | |
818 | goto out; | |
eb9cc7e1 | 819 | |
11013911 AD |
820 | if (!goal) |
821 | goal = sbi->s_inode_goal; | |
822 | ||
e6462869 | 823 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
824 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
825 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
826 | ret2 = 0; | |
827 | goto got_group; | |
828 | } | |
829 | ||
4113c4ca LC |
830 | if (S_ISDIR(mode)) |
831 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); | |
832 | else | |
a4912123 | 833 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 834 | |
772cb7c8 | 835 | got_group: |
a4912123 | 836 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 837 | err = -ENOSPC; |
2aa9fc4c | 838 | if (ret2 == -1) |
ac27a0ec DK |
839 | goto out; |
840 | ||
119c0d44 TT |
841 | /* |
842 | * Normally we will only go through one pass of this loop, | |
843 | * unless we get unlucky and it turns out the group we selected | |
844 | * had its last inode grabbed by someone else. | |
845 | */ | |
11013911 | 846 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
847 | err = -EIO; |
848 | ||
3300beda | 849 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec | 850 | if (!gdp) |
eb9cc7e1 | 851 | goto out; |
ac27a0ec | 852 | |
f2a09af6 YY |
853 | /* |
854 | * Check free inodes count before loading bitmap. | |
855 | */ | |
856 | if (ext4_free_inodes_count(sb, gdp) == 0) { | |
857 | if (++group == ngroups) | |
858 | group = 0; | |
859 | continue; | |
860 | } | |
861 | ||
87a39389 DW |
862 | grp = ext4_get_group_info(sb, group); |
863 | /* Skip groups with already-known suspicious inode tables */ | |
864 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { | |
865 | if (++group == ngroups) | |
866 | group = 0; | |
867 | continue; | |
868 | } | |
869 | ||
3300beda AK |
870 | brelse(inode_bitmap_bh); |
871 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
87a39389 | 872 | /* Skip groups with suspicious inode tables */ |
9008a58e DW |
873 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || |
874 | IS_ERR(inode_bitmap_bh)) { | |
875 | inode_bitmap_bh = NULL; | |
87a39389 DW |
876 | if (++group == ngroups) |
877 | group = 0; | |
878 | continue; | |
879 | } | |
ac27a0ec | 880 | |
ac27a0ec | 881 | repeat_in_this_group: |
617ba13b | 882 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
883 | inode_bitmap_bh->b_data, |
884 | EXT4_INODES_PER_GROUP(sb), ino); | |
a34eb503 TT |
885 | if (ino >= EXT4_INODES_PER_GROUP(sb)) |
886 | goto next_group; | |
119c0d44 TT |
887 | if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) { |
888 | ext4_error(sb, "reserved inode found cleared - " | |
889 | "inode=%lu", ino + 1); | |
890 | continue; | |
891 | } | |
19883bd9 TT |
892 | if ((EXT4_SB(sb)->s_journal == NULL) && |
893 | recently_deleted(sb, group, ino)) { | |
894 | ino++; | |
895 | goto next_inode; | |
896 | } | |
1139575a TT |
897 | if (!handle) { |
898 | BUG_ON(nblocks <= 0); | |
899 | handle = __ext4_journal_start_sb(dir->i_sb, line_no, | |
5fe2fe89 JK |
900 | handle_type, nblocks, |
901 | 0); | |
1139575a TT |
902 | if (IS_ERR(handle)) { |
903 | err = PTR_ERR(handle); | |
eb9cc7e1 JK |
904 | ext4_std_error(sb, err); |
905 | goto out; | |
1139575a TT |
906 | } |
907 | } | |
ffb5387e ES |
908 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
909 | err = ext4_journal_get_write_access(handle, inode_bitmap_bh); | |
eb9cc7e1 JK |
910 | if (err) { |
911 | ext4_std_error(sb, err); | |
912 | goto out; | |
913 | } | |
119c0d44 TT |
914 | ext4_lock_group(sb, group); |
915 | ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); | |
916 | ext4_unlock_group(sb, group); | |
917 | ino++; /* the inode bitmap is zero-based */ | |
918 | if (!ret2) | |
919 | goto got; /* we grabbed the inode! */ | |
19883bd9 | 920 | next_inode: |
119c0d44 TT |
921 | if (ino < EXT4_INODES_PER_GROUP(sb)) |
922 | goto repeat_in_this_group; | |
a34eb503 TT |
923 | next_group: |
924 | if (++group == ngroups) | |
925 | group = 0; | |
ac27a0ec DK |
926 | } |
927 | err = -ENOSPC; | |
928 | goto out; | |
929 | ||
930 | got: | |
ffb5387e ES |
931 | BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); |
932 | err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); | |
eb9cc7e1 JK |
933 | if (err) { |
934 | ext4_std_error(sb, err); | |
935 | goto out; | |
936 | } | |
ffb5387e | 937 | |
61c219f5 TT |
938 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
939 | err = ext4_journal_get_write_access(handle, group_desc_bh); | |
940 | if (err) { | |
941 | ext4_std_error(sb, err); | |
942 | goto out; | |
943 | } | |
944 | ||
717d50e4 | 945 | /* We may have to initialize the block bitmap if it isn't already */ |
feb0ab32 | 946 | if (ext4_has_group_desc_csum(sb) && |
717d50e4 | 947 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 948 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 949 | |
3300beda | 950 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
9008a58e DW |
951 | if (IS_ERR(block_bitmap_bh)) { |
952 | err = PTR_ERR(block_bitmap_bh); | |
599a9b77 JK |
953 | goto out; |
954 | } | |
3300beda AK |
955 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); |
956 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 957 | if (err) { |
3300beda | 958 | brelse(block_bitmap_bh); |
eb9cc7e1 JK |
959 | ext4_std_error(sb, err); |
960 | goto out; | |
717d50e4 AD |
961 | } |
962 | ||
fd034a84 TT |
963 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
964 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
fd034a84 | 965 | |
717d50e4 | 966 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 967 | ext4_lock_group(sb, group); |
717d50e4 | 968 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 969 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 970 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 971 | ext4_free_clusters_after_init(sb, group, gdp)); |
fa77dcfa | 972 | ext4_block_bitmap_csum_set(sb, group, gdp, |
79f1ba49 | 973 | block_bitmap_bh); |
feb0ab32 | 974 | ext4_group_desc_csum_set(sb, group, gdp); |
717d50e4 | 975 | } |
955ce5f5 | 976 | ext4_unlock_group(sb, group); |
aeb1e5d6 | 977 | brelse(block_bitmap_bh); |
717d50e4 | 978 | |
eb9cc7e1 JK |
979 | if (err) { |
980 | ext4_std_error(sb, err); | |
981 | goto out; | |
982 | } | |
717d50e4 | 983 | } |
119c0d44 | 984 | |
119c0d44 | 985 | /* Update the relevant bg descriptor fields */ |
41a246d1 | 986 | if (ext4_has_group_desc_csum(sb)) { |
119c0d44 TT |
987 | int free; |
988 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
989 | ||
990 | down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ | |
991 | ext4_lock_group(sb, group); /* while we modify the bg desc */ | |
992 | free = EXT4_INODES_PER_GROUP(sb) - | |
993 | ext4_itable_unused_count(sb, gdp); | |
994 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
995 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
996 | free = 0; | |
997 | } | |
998 | /* | |
999 | * Check the relative inode number against the last used | |
1000 | * relative inode number in this group. if it is greater | |
1001 | * we need to update the bg_itable_unused count | |
1002 | */ | |
1003 | if (ino > free) | |
1004 | ext4_itable_unused_set(sb, gdp, | |
1005 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
1006 | up_read(&grp->alloc_sem); | |
6f2e9f0e TM |
1007 | } else { |
1008 | ext4_lock_group(sb, group); | |
119c0d44 | 1009 | } |
6f2e9f0e | 1010 | |
119c0d44 TT |
1011 | ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); |
1012 | if (S_ISDIR(mode)) { | |
1013 | ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); | |
1014 | if (sbi->s_log_groups_per_flex) { | |
1015 | ext4_group_t f = ext4_flex_group(sbi, group); | |
1016 | ||
1017 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); | |
1018 | } | |
1019 | } | |
41a246d1 DW |
1020 | if (ext4_has_group_desc_csum(sb)) { |
1021 | ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, | |
1022 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 1023 | ext4_group_desc_csum_set(sb, group, gdp); |
119c0d44 | 1024 | } |
6f2e9f0e | 1025 | ext4_unlock_group(sb, group); |
119c0d44 | 1026 | |
3300beda AK |
1027 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
1028 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
eb9cc7e1 JK |
1029 | if (err) { |
1030 | ext4_std_error(sb, err); | |
1031 | goto out; | |
1032 | } | |
ac27a0ec DK |
1033 | |
1034 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
1035 | if (S_ISDIR(mode)) | |
1036 | percpu_counter_inc(&sbi->s_dirs_counter); | |
ac27a0ec | 1037 | |
772cb7c8 JS |
1038 | if (sbi->s_log_groups_per_flex) { |
1039 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 1040 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 | 1041 | } |
ac27a0ec | 1042 | |
717d50e4 | 1043 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
1044 | /* This is the optimal IO size (for stat), not the fs block size */ |
1045 | inode->i_blocks = 0; | |
ef7f3835 | 1046 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
eeca7ea1 | 1047 | current_time(inode); |
ac27a0ec DK |
1048 | |
1049 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
1050 | ei->i_dir_start_lookup = 0; | |
1051 | ei->i_disksize = 0; | |
1052 | ||
4af83508 | 1053 | /* Don't inherit extent flag from directory, amongst others. */ |
2dc6b0d4 DG |
1054 | ei->i_flags = |
1055 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 1056 | ei->i_file_acl = 0; |
ac27a0ec | 1057 | ei->i_dtime = 0; |
ac27a0ec | 1058 | ei->i_block_group = group; |
a4912123 | 1059 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1060 | |
617ba13b | 1061 | ext4_set_inode_flags(inode); |
ac27a0ec | 1062 | if (IS_DIRSYNC(inode)) |
0390131b | 1063 | ext4_handle_sync(handle); |
6b38e842 | 1064 | if (insert_inode_locked(inode) < 0) { |
acd6ad83 JK |
1065 | /* |
1066 | * Likely a bitmap corruption causing inode to be allocated | |
1067 | * twice. | |
1068 | */ | |
1069 | err = -EIO; | |
eb9cc7e1 JK |
1070 | ext4_error(sb, "failed to insert inode %lu: doubly allocated?", |
1071 | inode->i_ino); | |
1072 | goto out; | |
6b38e842 | 1073 | } |
ac27a0ec DK |
1074 | spin_lock(&sbi->s_next_gen_lock); |
1075 | inode->i_generation = sbi->s_next_generation++; | |
1076 | spin_unlock(&sbi->s_next_gen_lock); | |
1077 | ||
814525f4 | 1078 | /* Precompute checksum seed for inode metadata */ |
9aa5d32b | 1079 | if (ext4_has_metadata_csum(sb)) { |
814525f4 | 1080 | __u32 csum; |
814525f4 DW |
1081 | __le32 inum = cpu_to_le32(inode->i_ino); |
1082 | __le32 gen = cpu_to_le32(inode->i_generation); | |
1083 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
1084 | sizeof(inum)); | |
1085 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
1086 | sizeof(gen)); | |
1087 | } | |
1088 | ||
353eb83c | 1089 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 1090 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 KS |
1091 | |
1092 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
f08225d1 | 1093 | ei->i_inline_off = 0; |
e2b911c5 | 1094 | if (ext4_has_feature_inline_data(sb)) |
f08225d1 | 1095 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
ac27a0ec | 1096 | ret = inode; |
63936dda CH |
1097 | err = dquot_alloc_inode(inode); |
1098 | if (err) | |
ac27a0ec | 1099 | goto fail_drop; |
ac27a0ec | 1100 | |
617ba13b | 1101 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
1102 | if (err) |
1103 | goto fail_free_drop; | |
1104 | ||
2a7dba39 | 1105 | err = ext4_init_security(handle, inode, dir, qstr); |
ac27a0ec DK |
1106 | if (err) |
1107 | goto fail_free_drop; | |
1108 | ||
e2b911c5 | 1109 | if (ext4_has_feature_extents(sb)) { |
e4079a11 | 1110 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1111 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1112 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1113 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1114 | } |
a86c6181 | 1115 | } |
ac27a0ec | 1116 | |
688f869c TT |
1117 | if (ext4_handle_valid(handle)) { |
1118 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
1119 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
1120 | } | |
1121 | ||
e709e9df | 1122 | if (encrypt) { |
2f8f5e76 | 1123 | err = fscrypt_inherit_context(dir, inode, handle, true); |
e709e9df TT |
1124 | if (err) |
1125 | goto fail_free_drop; | |
1126 | } | |
1127 | ||
8753e88f AK |
1128 | err = ext4_mark_inode_dirty(handle, inode); |
1129 | if (err) { | |
1130 | ext4_std_error(sb, err); | |
1131 | goto fail_free_drop; | |
1132 | } | |
1133 | ||
617ba13b | 1134 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1135 | trace_ext4_allocate_inode(inode, dir, mode); |
3300beda | 1136 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1137 | return ret; |
1138 | ||
1139 | fail_free_drop: | |
63936dda | 1140 | dquot_free_inode(inode); |
ac27a0ec | 1141 | fail_drop: |
6d6b77f1 | 1142 | clear_nlink(inode); |
6b38e842 | 1143 | unlock_new_inode(inode); |
eb9cc7e1 JK |
1144 | out: |
1145 | dquot_drop(inode); | |
1146 | inode->i_flags |= S_NOQUOTA; | |
ac27a0ec | 1147 | iput(inode); |
3300beda | 1148 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1149 | return ERR_PTR(err); |
1150 | } | |
1151 | ||
1152 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1153 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1154 | { |
617ba13b | 1155 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1156 | ext4_group_t block_group; |
ac27a0ec | 1157 | int bit; |
7827a7f6 | 1158 | struct buffer_head *bitmap_bh = NULL; |
ac27a0ec | 1159 | struct inode *inode = NULL; |
7827a7f6 | 1160 | int err = -EFSCORRUPTED; |
ac27a0ec | 1161 | |
7827a7f6 TT |
1162 | if (ino < EXT4_FIRST_INO(sb) || ino > max_ino) |
1163 | goto bad_orphan; | |
ac27a0ec | 1164 | |
617ba13b MC |
1165 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1166 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1167 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
9008a58e | 1168 | if (IS_ERR(bitmap_bh)) { |
7827a7f6 TT |
1169 | ext4_error(sb, "inode bitmap error %ld for orphan %lu", |
1170 | ino, PTR_ERR(bitmap_bh)); | |
1171 | return (struct inode *) bitmap_bh; | |
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); | |
7827a7f6 TT |
1182 | if (IS_ERR(inode)) { |
1183 | err = PTR_ERR(inode); | |
1184 | ext4_error(sb, "couldn't read orphan inode %lu (err %d)", | |
1185 | ino, err); | |
1186 | return inode; | |
1187 | } | |
1d1fe1ee | 1188 | |
91ef4caf | 1189 | /* |
c9eb13a9 TT |
1190 | * If the orphans has i_nlinks > 0 then it should be able to |
1191 | * be truncated, otherwise it won't be removed from the orphan | |
1192 | * list during processing and an infinite loop will result. | |
1193 | * Similarly, it must not be a bad inode. | |
91ef4caf | 1194 | */ |
c9eb13a9 TT |
1195 | if ((inode->i_nlink && !ext4_can_truncate(inode)) || |
1196 | is_bad_inode(inode)) | |
91ef4caf DG |
1197 | goto bad_orphan; |
1198 | ||
1d1fe1ee DH |
1199 | if (NEXT_ORPHAN(inode) > max_ino) |
1200 | goto bad_orphan; | |
1201 | brelse(bitmap_bh); | |
1202 | return inode; | |
1203 | ||
1d1fe1ee | 1204 | bad_orphan: |
7827a7f6 TT |
1205 | ext4_error(sb, "bad orphan inode %lu", ino); |
1206 | if (bitmap_bh) | |
1207 | printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n", | |
1208 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
1209 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
1d1fe1ee | 1210 | if (inode) { |
7827a7f6 | 1211 | printk(KERN_ERR "is_bad_inode(inode)=%d\n", |
1d1fe1ee | 1212 | is_bad_inode(inode)); |
7827a7f6 | 1213 | printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n", |
1d1fe1ee | 1214 | NEXT_ORPHAN(inode)); |
7827a7f6 TT |
1215 | printk(KERN_ERR "max_ino=%lu\n", max_ino); |
1216 | printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink); | |
ac27a0ec | 1217 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1218 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1219 | inode->i_blocks = 0; |
1220 | iput(inode); | |
ac27a0ec | 1221 | } |
ac27a0ec | 1222 | brelse(bitmap_bh); |
1d1fe1ee | 1223 | return ERR_PTR(err); |
ac27a0ec DK |
1224 | } |
1225 | ||
af5bc92d | 1226 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1227 | { |
1228 | unsigned long desc_count; | |
617ba13b | 1229 | struct ext4_group_desc *gdp; |
8df9675f | 1230 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1231 | #ifdef EXT4FS_DEBUG |
1232 | struct ext4_super_block *es; | |
ac27a0ec DK |
1233 | unsigned long bitmap_count, x; |
1234 | struct buffer_head *bitmap_bh = NULL; | |
1235 | ||
617ba13b | 1236 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1237 | desc_count = 0; |
1238 | bitmap_count = 0; | |
1239 | gdp = NULL; | |
8df9675f | 1240 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1241 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1242 | if (!gdp) |
1243 | continue; | |
560671a0 | 1244 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1245 | brelse(bitmap_bh); |
e29d1cde | 1246 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
9008a58e DW |
1247 | if (IS_ERR(bitmap_bh)) { |
1248 | bitmap_bh = NULL; | |
ac27a0ec | 1249 | continue; |
9008a58e | 1250 | } |
ac27a0ec | 1251 | |
f6fb99ca TT |
1252 | x = ext4_count_free(bitmap_bh->b_data, |
1253 | EXT4_INODES_PER_GROUP(sb) / 8); | |
c549a95d | 1254 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1255 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1256 | bitmap_count += x; |
1257 | } | |
1258 | brelse(bitmap_bh); | |
4776004f TT |
1259 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1260 | "stored = %u, computed = %lu, %lu\n", | |
1261 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1262 | return desc_count; |
1263 | #else | |
1264 | desc_count = 0; | |
8df9675f | 1265 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1266 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1267 | if (!gdp) |
1268 | continue; | |
560671a0 | 1269 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1270 | cond_resched(); |
1271 | } | |
1272 | return desc_count; | |
1273 | #endif | |
1274 | } | |
1275 | ||
1276 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1277 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1278 | { |
1279 | unsigned long count = 0; | |
8df9675f | 1280 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1281 | |
8df9675f | 1282 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1283 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1284 | if (!gdp) |
1285 | continue; | |
560671a0 | 1286 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1287 | } |
1288 | return count; | |
1289 | } | |
bfff6873 LC |
1290 | |
1291 | /* | |
1292 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1293 | * inode table. Must be called without any spinlock held. The only place | |
1294 | * where it is called from on active part of filesystem is ext4lazyinit | |
1295 | * thread, so we do not need any special locks, however we have to prevent | |
1296 | * inode allocation from the current group, so we take alloc_sem lock, to | |
119c0d44 | 1297 | * block ext4_new_inode() until we are finished. |
bfff6873 | 1298 | */ |
e0cbee3e | 1299 | int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, |
bfff6873 LC |
1300 | int barrier) |
1301 | { | |
1302 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1303 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1304 | struct ext4_group_desc *gdp = NULL; | |
1305 | struct buffer_head *group_desc_bh; | |
1306 | handle_t *handle; | |
1307 | ext4_fsblk_t blk; | |
1308 | int num, ret = 0, used_blks = 0; | |
bfff6873 LC |
1309 | |
1310 | /* This should not happen, but just to be sure check this */ | |
1311 | if (sb->s_flags & MS_RDONLY) { | |
1312 | ret = 1; | |
1313 | goto out; | |
1314 | } | |
1315 | ||
1316 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1317 | if (!gdp) | |
1318 | goto out; | |
1319 | ||
1320 | /* | |
1321 | * We do not need to lock this, because we are the only one | |
1322 | * handling this flag. | |
1323 | */ | |
1324 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1325 | goto out; | |
1326 | ||
9924a92a | 1327 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
bfff6873 LC |
1328 | if (IS_ERR(handle)) { |
1329 | ret = PTR_ERR(handle); | |
1330 | goto out; | |
1331 | } | |
1332 | ||
1333 | down_write(&grp->alloc_sem); | |
1334 | /* | |
1335 | * If inode bitmap was already initialized there may be some | |
1336 | * used inodes so we need to skip blocks with used inodes in | |
1337 | * inode table. | |
1338 | */ | |
1339 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1340 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1341 | ext4_itable_unused_count(sb, gdp)), | |
1342 | sbi->s_inodes_per_block); | |
1343 | ||
857ac889 | 1344 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) { |
1084f252 TT |
1345 | ext4_error(sb, "Something is wrong with group %u: " |
1346 | "used itable blocks: %d; " | |
1347 | "itable unused count: %u", | |
857ac889 LC |
1348 | group, used_blks, |
1349 | ext4_itable_unused_count(sb, gdp)); | |
1350 | ret = 1; | |
33853a0d | 1351 | goto err_out; |
857ac889 LC |
1352 | } |
1353 | ||
bfff6873 LC |
1354 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1355 | num = sbi->s_itb_per_group - used_blks; | |
1356 | ||
1357 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1358 | ret = ext4_journal_get_write_access(handle, | |
1359 | group_desc_bh); | |
1360 | if (ret) | |
1361 | goto err_out; | |
1362 | ||
bfff6873 LC |
1363 | /* |
1364 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1365 | * flag anyway, because obviously, when it is full it does not need | |
1366 | * further zeroing. | |
1367 | */ | |
1368 | if (unlikely(num == 0)) | |
1369 | goto skip_zeroout; | |
1370 | ||
1371 | ext4_debug("going to zero out inode table in group %d\n", | |
1372 | group); | |
a107e5a3 | 1373 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1374 | if (ret < 0) |
1375 | goto err_out; | |
a107e5a3 TT |
1376 | if (barrier) |
1377 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); | |
bfff6873 LC |
1378 | |
1379 | skip_zeroout: | |
1380 | ext4_lock_group(sb, group); | |
1381 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
feb0ab32 | 1382 | ext4_group_desc_csum_set(sb, group, gdp); |
bfff6873 LC |
1383 | ext4_unlock_group(sb, group); |
1384 | ||
1385 | BUFFER_TRACE(group_desc_bh, | |
1386 | "call ext4_handle_dirty_metadata"); | |
1387 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1388 | group_desc_bh); | |
1389 | ||
1390 | err_out: | |
1391 | up_write(&grp->alloc_sem); | |
1392 | ext4_journal_stop(handle); | |
1393 | out: | |
1394 | return ret; | |
1395 | } |