]>
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); |
63936dda | 297 | dquot_free_inode(inode); |
9f754758 | 298 | dquot_drop(inode); |
ac27a0ec DK |
299 | |
300 | is_directory = S_ISDIR(inode->i_mode); | |
301 | ||
302 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 303 | ext4_clear_inode(inode); |
ac27a0ec | 304 | |
617ba13b MC |
305 | es = EXT4_SB(sb)->s_es; |
306 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 307 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
308 | goto error_return; |
309 | } | |
617ba13b MC |
310 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
311 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 312 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
87a39389 DW |
313 | /* Don't bother if the inode bitmap is corrupt. */ |
314 | grp = ext4_get_group_info(sb, block_group); | |
9008a58e DW |
315 | if (IS_ERR(bitmap_bh)) { |
316 | fatal = PTR_ERR(bitmap_bh); | |
317 | bitmap_bh = NULL; | |
318 | goto error_return; | |
319 | } | |
320 | if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) { | |
321 | fatal = -EFSCORRUPTED; | |
ac27a0ec | 322 | goto error_return; |
9008a58e | 323 | } |
ac27a0ec DK |
324 | |
325 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 326 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
327 | if (fatal) |
328 | goto error_return; | |
329 | ||
d17413c0 DM |
330 | fatal = -ESRCH; |
331 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
332 | if (gdp) { | |
ac27a0ec | 333 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 334 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
335 | } |
336 | ext4_lock_group(sb, block_group); | |
597d508c | 337 | cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); |
d17413c0 DM |
338 | if (fatal || !cleared) { |
339 | ext4_unlock_group(sb, block_group); | |
340 | goto out; | |
341 | } | |
7d39db14 | 342 | |
d17413c0 DM |
343 | count = ext4_free_inodes_count(sb, gdp) + 1; |
344 | ext4_free_inodes_set(sb, gdp, count); | |
345 | if (is_directory) { | |
346 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
347 | ext4_used_dirs_set(sb, gdp, count); | |
348 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 349 | } |
41a246d1 DW |
350 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, |
351 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 352 | ext4_group_desc_csum_set(sb, block_group, gdp); |
d17413c0 | 353 | ext4_unlock_group(sb, block_group); |
ac27a0ec | 354 | |
d17413c0 DM |
355 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
356 | if (sbi->s_log_groups_per_flex) { | |
357 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 358 | |
d17413c0 DM |
359 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
360 | if (is_directory) | |
361 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 362 | } |
d17413c0 DM |
363 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
364 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
365 | out: | |
366 | if (cleared) { | |
367 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
368 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
369 | if (!fatal) | |
370 | fatal = err; | |
87a39389 | 371 | } else { |
d17413c0 | 372 | ext4_error(sb, "bit already cleared for inode %lu", ino); |
bf40c926 | 373 | if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
e43bb4e6 NJ |
374 | int count; |
375 | count = ext4_free_inodes_count(sb, gdp); | |
376 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
377 | count); | |
378 | } | |
87a39389 DW |
379 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
380 | } | |
d17413c0 | 381 | |
ac27a0ec DK |
382 | error_return: |
383 | brelse(bitmap_bh); | |
617ba13b | 384 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
385 | } |
386 | ||
a4912123 | 387 | struct orlov_stats { |
90ba983f | 388 | __u64 free_clusters; |
a4912123 | 389 | __u32 free_inodes; |
a4912123 TT |
390 | __u32 used_dirs; |
391 | }; | |
392 | ||
393 | /* | |
394 | * Helper function for Orlov's allocator; returns critical information | |
395 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
396 | * is a block group number; otherwise it is flex_bg number. | |
397 | */ | |
1f109d5a TT |
398 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
399 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
400 | { |
401 | struct ext4_group_desc *desc; | |
7d39db14 | 402 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 403 | |
7d39db14 TT |
404 | if (flex_size > 1) { |
405 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
90ba983f | 406 | stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); |
7d39db14 TT |
407 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); |
408 | return; | |
409 | } | |
a4912123 | 410 | |
7d39db14 TT |
411 | desc = ext4_get_group_desc(sb, g, NULL); |
412 | if (desc) { | |
413 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
021b65bb | 414 | stats->free_clusters = ext4_free_group_clusters(sb, desc); |
7d39db14 TT |
415 | stats->used_dirs = ext4_used_dirs_count(sb, desc); |
416 | } else { | |
417 | stats->free_inodes = 0; | |
24aaa8ef | 418 | stats->free_clusters = 0; |
7d39db14 | 419 | stats->used_dirs = 0; |
a4912123 TT |
420 | } |
421 | } | |
422 | ||
ac27a0ec DK |
423 | /* |
424 | * Orlov's allocator for directories. | |
425 | * | |
426 | * We always try to spread first-level directories. | |
427 | * | |
428 | * If there are blockgroups with both free inodes and free blocks counts | |
429 | * not worse than average we return one with smallest directory count. | |
430 | * Otherwise we simply return a random group. | |
431 | * | |
432 | * For the rest rules look so: | |
433 | * | |
434 | * It's OK to put directory into a group unless | |
435 | * it has too many directories already (max_dirs) or | |
436 | * it has too few free inodes left (min_inodes) or | |
437 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 438 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
439 | * conditions we search cyclically through the rest. If none |
440 | * of the groups look good we just look for a group with more | |
441 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
442 | */ |
443 | ||
2aa9fc4c | 444 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
dcca3fec | 445 | ext4_group_t *group, umode_t mode, |
f157a4aa | 446 | const struct qstr *qstr) |
ac27a0ec | 447 | { |
fd2d4291 | 448 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 449 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 450 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 451 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
14c83c9f | 452 | unsigned int freei, avefreei, grp_free; |
24aaa8ef | 453 | ext4_fsblk_t freeb, avefreec; |
ac27a0ec | 454 | unsigned int ndirs; |
a4912123 | 455 | int max_dirs, min_inodes; |
24aaa8ef | 456 | ext4_grpblk_t min_clusters; |
8df9675f | 457 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 458 | struct ext4_group_desc *desc; |
a4912123 TT |
459 | struct orlov_stats stats; |
460 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 461 | struct dx_hash_info hinfo; |
a4912123 | 462 | |
8df9675f | 463 | ngroups = real_ngroups; |
a4912123 | 464 | if (flex_size > 1) { |
8df9675f | 465 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
466 | sbi->s_log_groups_per_flex; |
467 | parent_group >>= sbi->s_log_groups_per_flex; | |
468 | } | |
ac27a0ec DK |
469 | |
470 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
471 | avefreei = freei / ngroups; | |
57042651 TT |
472 | freeb = EXT4_C2B(sbi, |
473 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
24aaa8ef TT |
474 | avefreec = freeb; |
475 | do_div(avefreec, ngroups); | |
ac27a0ec DK |
476 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
477 | ||
a4912123 | 478 | if (S_ISDIR(mode) && |
2b0143b5 | 479 | ((parent == d_inode(sb->s_root)) || |
12e9b892 | 480 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 481 | int best_ndir = inodes_per_group; |
2aa9fc4c | 482 | int ret = -1; |
ac27a0ec | 483 | |
f157a4aa TT |
484 | if (qstr) { |
485 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
486 | hinfo.seed = sbi->s_hash_seed; | |
487 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
488 | grp = hinfo.hash; | |
489 | } else | |
dd1f723b | 490 | grp = prandom_u32(); |
2aa9fc4c | 491 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 492 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
493 | g = (parent_group + i) % ngroups; |
494 | get_orlov_stats(sb, g, flex_size, &stats); | |
495 | if (!stats.free_inodes) | |
ac27a0ec | 496 | continue; |
a4912123 | 497 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 498 | continue; |
a4912123 | 499 | if (stats.free_inodes < avefreei) |
ac27a0ec | 500 | continue; |
24aaa8ef | 501 | if (stats.free_clusters < avefreec) |
ac27a0ec | 502 | continue; |
a4912123 | 503 | grp = g; |
2aa9fc4c | 504 | ret = 0; |
a4912123 TT |
505 | best_ndir = stats.used_dirs; |
506 | } | |
507 | if (ret) | |
508 | goto fallback; | |
509 | found_flex_bg: | |
510 | if (flex_size == 1) { | |
511 | *group = grp; | |
512 | return 0; | |
513 | } | |
514 | ||
515 | /* | |
516 | * We pack inodes at the beginning of the flexgroup's | |
517 | * inode tables. Block allocation decisions will do | |
518 | * something similar, although regular files will | |
519 | * start at 2nd block group of the flexgroup. See | |
520 | * ext4_ext_find_goal() and ext4_find_near(). | |
521 | */ | |
522 | grp *= flex_size; | |
523 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 524 | if (grp+i >= real_ngroups) |
a4912123 TT |
525 | break; |
526 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
527 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
528 | *group = grp+i; | |
529 | return 0; | |
530 | } | |
ac27a0ec | 531 | } |
ac27a0ec DK |
532 | goto fallback; |
533 | } | |
534 | ||
ac27a0ec | 535 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
536 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
537 | if (min_inodes < 1) | |
538 | min_inodes = 1; | |
24aaa8ef | 539 | min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; |
a4912123 TT |
540 | |
541 | /* | |
542 | * Start looking in the flex group where we last allocated an | |
543 | * inode for this parent directory | |
544 | */ | |
545 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
546 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
547 | if (flex_size > 1) | |
548 | parent_group >>= sbi->s_log_groups_per_flex; | |
549 | } | |
ac27a0ec DK |
550 | |
551 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
552 | grp = (parent_group + i) % ngroups; |
553 | get_orlov_stats(sb, grp, flex_size, &stats); | |
554 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 555 | continue; |
a4912123 | 556 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 557 | continue; |
24aaa8ef | 558 | if (stats.free_clusters < min_clusters) |
ac27a0ec | 559 | continue; |
a4912123 | 560 | goto found_flex_bg; |
ac27a0ec DK |
561 | } |
562 | ||
563 | fallback: | |
8df9675f | 564 | ngroups = real_ngroups; |
a4912123 | 565 | avefreei = freei / ngroups; |
b5451f7b | 566 | fallback_retry: |
a4912123 | 567 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 568 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
569 | grp = (parent_group + i) % ngroups; |
570 | desc = ext4_get_group_desc(sb, grp, NULL); | |
bb3d132a DC |
571 | if (desc) { |
572 | grp_free = ext4_free_inodes_count(sb, desc); | |
573 | if (grp_free && grp_free >= avefreei) { | |
574 | *group = grp; | |
575 | return 0; | |
576 | } | |
a4912123 | 577 | } |
ac27a0ec DK |
578 | } |
579 | ||
580 | if (avefreei) { | |
581 | /* | |
582 | * The free-inodes counter is approximate, and for really small | |
583 | * filesystems the above test can fail to find any blockgroups | |
584 | */ | |
585 | avefreei = 0; | |
b5451f7b | 586 | goto fallback_retry; |
ac27a0ec DK |
587 | } |
588 | ||
589 | return -1; | |
590 | } | |
591 | ||
2aa9fc4c | 592 | static int find_group_other(struct super_block *sb, struct inode *parent, |
dcca3fec | 593 | ext4_group_t *group, umode_t mode) |
ac27a0ec | 594 | { |
fd2d4291 | 595 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 596 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 597 | struct ext4_group_desc *desc; |
a4912123 TT |
598 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
599 | ||
600 | /* | |
601 | * Try to place the inode is the same flex group as its | |
602 | * parent. If we can't find space, use the Orlov algorithm to | |
603 | * find another flex group, and store that information in the | |
604 | * parent directory's inode information so that use that flex | |
605 | * group for future allocations. | |
606 | */ | |
607 | if (flex_size > 1) { | |
608 | int retry = 0; | |
609 | ||
610 | try_again: | |
611 | parent_group &= ~(flex_size-1); | |
612 | last = parent_group + flex_size; | |
613 | if (last > ngroups) | |
614 | last = ngroups; | |
615 | for (i = parent_group; i < last; i++) { | |
616 | desc = ext4_get_group_desc(sb, i, NULL); | |
617 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
618 | *group = i; | |
619 | return 0; | |
620 | } | |
621 | } | |
622 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
623 | retry = 1; | |
624 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
625 | goto try_again; | |
626 | } | |
627 | /* | |
628 | * If this didn't work, use the Orlov search algorithm | |
629 | * to find a new flex group; we pass in the mode to | |
630 | * avoid the topdir algorithms. | |
631 | */ | |
632 | *group = parent_group + flex_size; | |
633 | if (*group > ngroups) | |
634 | *group = 0; | |
7dc57615 | 635 | return find_group_orlov(sb, parent, group, mode, NULL); |
a4912123 | 636 | } |
ac27a0ec DK |
637 | |
638 | /* | |
639 | * Try to place the inode in its parent directory | |
640 | */ | |
2aa9fc4c AM |
641 | *group = parent_group; |
642 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 643 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 644 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 645 | return 0; |
ac27a0ec DK |
646 | |
647 | /* | |
648 | * We're going to place this inode in a different blockgroup from its | |
649 | * parent. We want to cause files in a common directory to all land in | |
650 | * the same blockgroup. But we want files which are in a different | |
651 | * directory which shares a blockgroup with our parent to land in a | |
652 | * different blockgroup. | |
653 | * | |
654 | * So add our directory's i_ino into the starting point for the hash. | |
655 | */ | |
2aa9fc4c | 656 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
657 | |
658 | /* | |
659 | * Use a quadratic hash to find a group with a free inode and some free | |
660 | * blocks. | |
661 | */ | |
662 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
663 | *group += i; |
664 | if (*group >= ngroups) | |
665 | *group -= ngroups; | |
666 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 667 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 668 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 669 | return 0; |
ac27a0ec DK |
670 | } |
671 | ||
672 | /* | |
673 | * That failed: try linear search for a free inode, even if that group | |
674 | * has no free blocks. | |
675 | */ | |
2aa9fc4c | 676 | *group = parent_group; |
ac27a0ec | 677 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
678 | if (++*group >= ngroups) |
679 | *group = 0; | |
680 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 681 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 682 | return 0; |
ac27a0ec DK |
683 | } |
684 | ||
685 | return -1; | |
686 | } | |
687 | ||
19883bd9 TT |
688 | /* |
689 | * In no journal mode, if an inode has recently been deleted, we want | |
690 | * to avoid reusing it until we're reasonably sure the inode table | |
691 | * block has been written back to disk. (Yes, these values are | |
692 | * somewhat arbitrary...) | |
693 | */ | |
694 | #define RECENTCY_MIN 5 | |
695 | #define RECENTCY_DIRTY 30 | |
696 | ||
697 | static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) | |
698 | { | |
699 | struct ext4_group_desc *gdp; | |
700 | struct ext4_inode *raw_inode; | |
701 | struct buffer_head *bh; | |
702 | unsigned long dtime, now; | |
703 | int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
704 | int offset, ret = 0, recentcy = RECENTCY_MIN; | |
705 | ||
706 | gdp = ext4_get_group_desc(sb, group, NULL); | |
707 | if (unlikely(!gdp)) | |
708 | return 0; | |
709 | ||
710 | bh = sb_getblk(sb, ext4_inode_table(sb, gdp) + | |
711 | (ino / inodes_per_block)); | |
712 | if (unlikely(!bh) || !buffer_uptodate(bh)) | |
713 | /* | |
714 | * If the block is not in the buffer cache, then it | |
715 | * must have been written out. | |
716 | */ | |
717 | goto out; | |
718 | ||
719 | offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
720 | raw_inode = (struct ext4_inode *) (bh->b_data + offset); | |
721 | dtime = le32_to_cpu(raw_inode->i_dtime); | |
722 | now = get_seconds(); | |
723 | if (buffer_dirty(bh)) | |
724 | recentcy += RECENTCY_DIRTY; | |
725 | ||
726 | if (dtime && (dtime < now) && (now < dtime + recentcy)) | |
727 | ret = 1; | |
728 | out: | |
729 | brelse(bh); | |
730 | return ret; | |
731 | } | |
732 | ||
ac27a0ec DK |
733 | /* |
734 | * There are two policies for allocating an inode. If the new inode is | |
735 | * a directory, then a forward search is made for a block group with both | |
736 | * free space and a low directory-to-inode ratio; if that fails, then of | |
737 | * the groups with above-average free space, that group with the fewest | |
738 | * directories already is chosen. | |
739 | * | |
740 | * For other inodes, search forward from the parent directory's block | |
741 | * group to find a free inode. | |
742 | */ | |
1139575a TT |
743 | struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, |
744 | umode_t mode, const struct qstr *qstr, | |
1b917ed8 TE |
745 | __u32 goal, uid_t *owner, __u32 i_flags, |
746 | int handle_type, unsigned int line_no, | |
747 | 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 | ||
af65207c TE |
769 | sb = dir->i_sb; |
770 | sbi = EXT4_SB(sb); | |
771 | ||
772 | if (unlikely(ext4_forced_shutdown(sbi))) | |
0db1ff22 TT |
773 | return ERR_PTR(-EIO); |
774 | ||
afdbd4a6 SF |
775 | /* Supplied owner must be valid */ |
776 | if (owner && (owner[0] == (uid_t)-1 || owner[1] == (uid_t)-1)) | |
777 | return ERR_PTR(-EOVERFLOW); | |
778 | ||
af65207c | 779 | if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) && |
ad47f953 TE |
780 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) && |
781 | !(i_flags & EXT4_EA_INODE_FL)) { | |
a7550b30 | 782 | err = fscrypt_get_encryption_info(dir); |
e709e9df TT |
783 | if (err) |
784 | return ERR_PTR(err); | |
a7550b30 | 785 | if (!fscrypt_has_encryption_key(dir)) |
54475f53 | 786 | return ERR_PTR(-ENOKEY); |
e709e9df TT |
787 | encrypt = 1; |
788 | } | |
789 | ||
af65207c TE |
790 | if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) { |
791 | #ifdef CONFIG_EXT4_FS_POSIX_ACL | |
792 | struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT); | |
793 | ||
794 | if (p) { | |
795 | int acl_size = p->a_count * sizeof(ext4_acl_entry); | |
796 | ||
797 | nblocks += (S_ISDIR(mode) ? 2 : 1) * | |
798 | __ext4_xattr_set_credits(sb, NULL /* inode */, | |
799 | NULL /* block_bh */, acl_size, | |
800 | true /* is_create */); | |
801 | posix_acl_release(p); | |
802 | } | |
803 | #endif | |
804 | ||
805 | #ifdef CONFIG_SECURITY | |
806 | { | |
807 | int num_security_xattrs = 1; | |
808 | ||
809 | #ifdef CONFIG_INTEGRITY | |
810 | num_security_xattrs++; | |
811 | #endif | |
812 | /* | |
813 | * We assume that security xattrs are never | |
814 | * more than 1k. In practice they are under | |
815 | * 128 bytes. | |
816 | */ | |
817 | nblocks += num_security_xattrs * | |
818 | __ext4_xattr_set_credits(sb, NULL /* inode */, | |
819 | NULL /* block_bh */, 1024, | |
820 | true /* is_create */); | |
821 | } | |
822 | #endif | |
823 | if (encrypt) | |
824 | nblocks += __ext4_xattr_set_credits(sb, | |
825 | NULL /* inode */, NULL /* block_bh */, | |
826 | FSCRYPT_SET_CONTEXT_MAX_SIZE, | |
827 | true /* is_create */); | |
828 | } | |
829 | ||
8df9675f | 830 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 831 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
832 | inode = new_inode(sb); |
833 | if (!inode) | |
834 | return ERR_PTR(-ENOMEM); | |
617ba13b | 835 | ei = EXT4_I(inode); |
772cb7c8 | 836 | |
eb9cc7e1 | 837 | /* |
b8a07463 | 838 | * Initialize owners and quota early so that we don't have to account |
eb9cc7e1 JK |
839 | * for quota initialization worst case in standard inode creating |
840 | * transaction | |
841 | */ | |
842 | if (owner) { | |
843 | inode->i_mode = mode; | |
844 | i_uid_write(inode, owner[0]); | |
845 | i_gid_write(inode, owner[1]); | |
846 | } else if (test_opt(sb, GRPID)) { | |
847 | inode->i_mode = mode; | |
848 | inode->i_uid = current_fsuid(); | |
849 | inode->i_gid = dir->i_gid; | |
850 | } else | |
851 | inode_init_owner(inode, dir, mode); | |
040cb378 | 852 | |
0b7b7779 | 853 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
854 | ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) |
855 | ei->i_projid = EXT4_I(dir)->i_projid; | |
856 | else | |
afdbd4a6 | 857 | ei->i_projid = make_kprojid(sb->s_user_ns, EXT4_DEF_PROJID); |
040cb378 | 858 | |
a7cdadee JK |
859 | err = dquot_initialize(inode); |
860 | if (err) | |
861 | goto out; | |
eb9cc7e1 | 862 | |
11013911 AD |
863 | if (!goal) |
864 | goal = sbi->s_inode_goal; | |
865 | ||
e6462869 | 866 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
867 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
868 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
869 | ret2 = 0; | |
870 | goto got_group; | |
871 | } | |
872 | ||
4113c4ca LC |
873 | if (S_ISDIR(mode)) |
874 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); | |
875 | else | |
a4912123 | 876 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 877 | |
772cb7c8 | 878 | got_group: |
a4912123 | 879 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 880 | err = -ENOSPC; |
2aa9fc4c | 881 | if (ret2 == -1) |
ac27a0ec DK |
882 | goto out; |
883 | ||
119c0d44 TT |
884 | /* |
885 | * Normally we will only go through one pass of this loop, | |
886 | * unless we get unlucky and it turns out the group we selected | |
887 | * had its last inode grabbed by someone else. | |
888 | */ | |
11013911 | 889 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
890 | err = -EIO; |
891 | ||
3300beda | 892 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec | 893 | if (!gdp) |
eb9cc7e1 | 894 | goto out; |
ac27a0ec | 895 | |
f2a09af6 YY |
896 | /* |
897 | * Check free inodes count before loading bitmap. | |
898 | */ | |
899 | if (ext4_free_inodes_count(sb, gdp) == 0) { | |
900 | if (++group == ngroups) | |
901 | group = 0; | |
902 | continue; | |
903 | } | |
904 | ||
87a39389 DW |
905 | grp = ext4_get_group_info(sb, group); |
906 | /* Skip groups with already-known suspicious inode tables */ | |
907 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { | |
908 | if (++group == ngroups) | |
909 | group = 0; | |
910 | continue; | |
911 | } | |
912 | ||
3300beda AK |
913 | brelse(inode_bitmap_bh); |
914 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
87a39389 | 915 | /* Skip groups with suspicious inode tables */ |
9008a58e DW |
916 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || |
917 | IS_ERR(inode_bitmap_bh)) { | |
918 | inode_bitmap_bh = NULL; | |
87a39389 DW |
919 | if (++group == ngroups) |
920 | group = 0; | |
921 | continue; | |
922 | } | |
ac27a0ec | 923 | |
ac27a0ec | 924 | repeat_in_this_group: |
617ba13b | 925 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
926 | inode_bitmap_bh->b_data, |
927 | EXT4_INODES_PER_GROUP(sb), ino); | |
a34eb503 TT |
928 | if (ino >= EXT4_INODES_PER_GROUP(sb)) |
929 | goto next_group; | |
119c0d44 TT |
930 | if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) { |
931 | ext4_error(sb, "reserved inode found cleared - " | |
932 | "inode=%lu", ino + 1); | |
933 | continue; | |
934 | } | |
19883bd9 TT |
935 | if ((EXT4_SB(sb)->s_journal == NULL) && |
936 | recently_deleted(sb, group, ino)) { | |
937 | ino++; | |
938 | goto next_inode; | |
939 | } | |
1139575a TT |
940 | if (!handle) { |
941 | BUG_ON(nblocks <= 0); | |
942 | handle = __ext4_journal_start_sb(dir->i_sb, line_no, | |
5fe2fe89 JK |
943 | handle_type, nblocks, |
944 | 0); | |
1139575a TT |
945 | if (IS_ERR(handle)) { |
946 | err = PTR_ERR(handle); | |
eb9cc7e1 JK |
947 | ext4_std_error(sb, err); |
948 | goto out; | |
1139575a TT |
949 | } |
950 | } | |
ffb5387e ES |
951 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
952 | err = ext4_journal_get_write_access(handle, inode_bitmap_bh); | |
eb9cc7e1 JK |
953 | if (err) { |
954 | ext4_std_error(sb, err); | |
955 | goto out; | |
956 | } | |
119c0d44 TT |
957 | ext4_lock_group(sb, group); |
958 | ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); | |
959 | ext4_unlock_group(sb, group); | |
960 | ino++; /* the inode bitmap is zero-based */ | |
961 | if (!ret2) | |
962 | goto got; /* we grabbed the inode! */ | |
19883bd9 | 963 | next_inode: |
119c0d44 TT |
964 | if (ino < EXT4_INODES_PER_GROUP(sb)) |
965 | goto repeat_in_this_group; | |
a34eb503 TT |
966 | next_group: |
967 | if (++group == ngroups) | |
968 | group = 0; | |
ac27a0ec DK |
969 | } |
970 | err = -ENOSPC; | |
971 | goto out; | |
972 | ||
973 | got: | |
ffb5387e ES |
974 | BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); |
975 | err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); | |
eb9cc7e1 JK |
976 | if (err) { |
977 | ext4_std_error(sb, err); | |
978 | goto out; | |
979 | } | |
ffb5387e | 980 | |
61c219f5 TT |
981 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
982 | err = ext4_journal_get_write_access(handle, group_desc_bh); | |
983 | if (err) { | |
984 | ext4_std_error(sb, err); | |
985 | goto out; | |
986 | } | |
987 | ||
717d50e4 | 988 | /* We may have to initialize the block bitmap if it isn't already */ |
feb0ab32 | 989 | if (ext4_has_group_desc_csum(sb) && |
717d50e4 | 990 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 991 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 992 | |
3300beda | 993 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
9008a58e DW |
994 | if (IS_ERR(block_bitmap_bh)) { |
995 | err = PTR_ERR(block_bitmap_bh); | |
599a9b77 JK |
996 | goto out; |
997 | } | |
3300beda AK |
998 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); |
999 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 1000 | if (err) { |
3300beda | 1001 | brelse(block_bitmap_bh); |
eb9cc7e1 JK |
1002 | ext4_std_error(sb, err); |
1003 | goto out; | |
717d50e4 AD |
1004 | } |
1005 | ||
fd034a84 TT |
1006 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
1007 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
fd034a84 | 1008 | |
717d50e4 | 1009 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 1010 | ext4_lock_group(sb, group); |
717d50e4 | 1011 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 1012 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 1013 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 1014 | ext4_free_clusters_after_init(sb, group, gdp)); |
fa77dcfa | 1015 | ext4_block_bitmap_csum_set(sb, group, gdp, |
79f1ba49 | 1016 | block_bitmap_bh); |
feb0ab32 | 1017 | ext4_group_desc_csum_set(sb, group, gdp); |
717d50e4 | 1018 | } |
955ce5f5 | 1019 | ext4_unlock_group(sb, group); |
aeb1e5d6 | 1020 | brelse(block_bitmap_bh); |
717d50e4 | 1021 | |
eb9cc7e1 JK |
1022 | if (err) { |
1023 | ext4_std_error(sb, err); | |
1024 | goto out; | |
1025 | } | |
717d50e4 | 1026 | } |
119c0d44 | 1027 | |
119c0d44 | 1028 | /* Update the relevant bg descriptor fields */ |
41a246d1 | 1029 | if (ext4_has_group_desc_csum(sb)) { |
119c0d44 TT |
1030 | int free; |
1031 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1032 | ||
1033 | down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ | |
1034 | ext4_lock_group(sb, group); /* while we modify the bg desc */ | |
1035 | free = EXT4_INODES_PER_GROUP(sb) - | |
1036 | ext4_itable_unused_count(sb, gdp); | |
1037 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
1038 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
1039 | free = 0; | |
1040 | } | |
1041 | /* | |
1042 | * Check the relative inode number against the last used | |
1043 | * relative inode number in this group. if it is greater | |
1044 | * we need to update the bg_itable_unused count | |
1045 | */ | |
1046 | if (ino > free) | |
1047 | ext4_itable_unused_set(sb, gdp, | |
1048 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
1049 | up_read(&grp->alloc_sem); | |
6f2e9f0e TM |
1050 | } else { |
1051 | ext4_lock_group(sb, group); | |
119c0d44 | 1052 | } |
6f2e9f0e | 1053 | |
119c0d44 TT |
1054 | ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); |
1055 | if (S_ISDIR(mode)) { | |
1056 | ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); | |
1057 | if (sbi->s_log_groups_per_flex) { | |
1058 | ext4_group_t f = ext4_flex_group(sbi, group); | |
1059 | ||
1060 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); | |
1061 | } | |
1062 | } | |
41a246d1 DW |
1063 | if (ext4_has_group_desc_csum(sb)) { |
1064 | ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, | |
1065 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 1066 | ext4_group_desc_csum_set(sb, group, gdp); |
119c0d44 | 1067 | } |
6f2e9f0e | 1068 | ext4_unlock_group(sb, group); |
119c0d44 | 1069 | |
3300beda AK |
1070 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
1071 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
eb9cc7e1 JK |
1072 | if (err) { |
1073 | ext4_std_error(sb, err); | |
1074 | goto out; | |
1075 | } | |
ac27a0ec DK |
1076 | |
1077 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
1078 | if (S_ISDIR(mode)) | |
1079 | percpu_counter_inc(&sbi->s_dirs_counter); | |
ac27a0ec | 1080 | |
772cb7c8 JS |
1081 | if (sbi->s_log_groups_per_flex) { |
1082 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 1083 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 | 1084 | } |
ac27a0ec | 1085 | |
717d50e4 | 1086 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
1087 | /* This is the optimal IO size (for stat), not the fs block size */ |
1088 | inode->i_blocks = 0; | |
ef7f3835 | 1089 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
eeca7ea1 | 1090 | current_time(inode); |
ac27a0ec DK |
1091 | |
1092 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
1093 | ei->i_dir_start_lookup = 0; | |
1094 | ei->i_disksize = 0; | |
1095 | ||
4af83508 | 1096 | /* Don't inherit extent flag from directory, amongst others. */ |
2dc6b0d4 DG |
1097 | ei->i_flags = |
1098 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
1b917ed8 | 1099 | ei->i_flags |= i_flags; |
ac27a0ec | 1100 | ei->i_file_acl = 0; |
ac27a0ec | 1101 | ei->i_dtime = 0; |
ac27a0ec | 1102 | ei->i_block_group = group; |
a4912123 | 1103 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1104 | |
617ba13b | 1105 | ext4_set_inode_flags(inode); |
ac27a0ec | 1106 | if (IS_DIRSYNC(inode)) |
0390131b | 1107 | ext4_handle_sync(handle); |
6b38e842 | 1108 | if (insert_inode_locked(inode) < 0) { |
acd6ad83 JK |
1109 | /* |
1110 | * Likely a bitmap corruption causing inode to be allocated | |
1111 | * twice. | |
1112 | */ | |
1113 | err = -EIO; | |
eb9cc7e1 JK |
1114 | ext4_error(sb, "failed to insert inode %lu: doubly allocated?", |
1115 | inode->i_ino); | |
1116 | goto out; | |
6b38e842 | 1117 | } |
ac27a0ec DK |
1118 | spin_lock(&sbi->s_next_gen_lock); |
1119 | inode->i_generation = sbi->s_next_generation++; | |
1120 | spin_unlock(&sbi->s_next_gen_lock); | |
1121 | ||
814525f4 | 1122 | /* Precompute checksum seed for inode metadata */ |
9aa5d32b | 1123 | if (ext4_has_metadata_csum(sb)) { |
814525f4 | 1124 | __u32 csum; |
814525f4 DW |
1125 | __le32 inum = cpu_to_le32(inode->i_ino); |
1126 | __le32 gen = cpu_to_le32(inode->i_generation); | |
1127 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
1128 | sizeof(inum)); | |
1129 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
1130 | sizeof(gen)); | |
1131 | } | |
1132 | ||
353eb83c | 1133 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 1134 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 KS |
1135 | |
1136 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
f08225d1 | 1137 | ei->i_inline_off = 0; |
e2b911c5 | 1138 | if (ext4_has_feature_inline_data(sb)) |
f08225d1 | 1139 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
ac27a0ec | 1140 | ret = inode; |
63936dda CH |
1141 | err = dquot_alloc_inode(inode); |
1142 | if (err) | |
ac27a0ec | 1143 | goto fail_drop; |
ac27a0ec | 1144 | |
aa1dca3b EB |
1145 | /* |
1146 | * Since the encryption xattr will always be unique, create it first so | |
1147 | * that it's less likely to end up in an external xattr block and | |
1148 | * prevent its deduplication. | |
1149 | */ | |
1150 | if (encrypt) { | |
1151 | err = fscrypt_inherit_context(dir, inode, handle, true); | |
1152 | if (err) | |
1153 | goto fail_free_drop; | |
1154 | } | |
1155 | ||
1b917ed8 TE |
1156 | if (!(ei->i_flags & EXT4_EA_INODE_FL)) { |
1157 | err = ext4_init_acl(handle, inode, dir); | |
1158 | if (err) | |
1159 | goto fail_free_drop; | |
ac27a0ec | 1160 | |
ad47f953 TE |
1161 | err = ext4_init_security(handle, inode, dir, qstr); |
1162 | if (err) | |
1163 | goto fail_free_drop; | |
1164 | } | |
ac27a0ec | 1165 | |
e2b911c5 | 1166 | if (ext4_has_feature_extents(sb)) { |
e4079a11 | 1167 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1168 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1169 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1170 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1171 | } |
a86c6181 | 1172 | } |
ac27a0ec | 1173 | |
688f869c TT |
1174 | if (ext4_handle_valid(handle)) { |
1175 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
1176 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
1177 | } | |
1178 | ||
8753e88f AK |
1179 | err = ext4_mark_inode_dirty(handle, inode); |
1180 | if (err) { | |
1181 | ext4_std_error(sb, err); | |
1182 | goto fail_free_drop; | |
1183 | } | |
1184 | ||
617ba13b | 1185 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1186 | trace_ext4_allocate_inode(inode, dir, mode); |
3300beda | 1187 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1188 | return ret; |
1189 | ||
1190 | fail_free_drop: | |
63936dda | 1191 | dquot_free_inode(inode); |
ac27a0ec | 1192 | fail_drop: |
6d6b77f1 | 1193 | clear_nlink(inode); |
6b38e842 | 1194 | unlock_new_inode(inode); |
eb9cc7e1 JK |
1195 | out: |
1196 | dquot_drop(inode); | |
1197 | inode->i_flags |= S_NOQUOTA; | |
ac27a0ec | 1198 | iput(inode); |
3300beda | 1199 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1200 | return ERR_PTR(err); |
1201 | } | |
1202 | ||
1203 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1204 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1205 | { |
617ba13b | 1206 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1207 | ext4_group_t block_group; |
ac27a0ec | 1208 | int bit; |
7827a7f6 | 1209 | struct buffer_head *bitmap_bh = NULL; |
ac27a0ec | 1210 | struct inode *inode = NULL; |
7827a7f6 | 1211 | int err = -EFSCORRUPTED; |
ac27a0ec | 1212 | |
7827a7f6 TT |
1213 | if (ino < EXT4_FIRST_INO(sb) || ino > max_ino) |
1214 | goto bad_orphan; | |
ac27a0ec | 1215 | |
617ba13b MC |
1216 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1217 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1218 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
9008a58e | 1219 | if (IS_ERR(bitmap_bh)) { |
7827a7f6 TT |
1220 | ext4_error(sb, "inode bitmap error %ld for orphan %lu", |
1221 | ino, PTR_ERR(bitmap_bh)); | |
1222 | return (struct inode *) bitmap_bh; | |
ac27a0ec DK |
1223 | } |
1224 | ||
1225 | /* Having the inode bit set should be a 100% indicator that this | |
1226 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1227 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1228 | */ | |
1d1fe1ee DH |
1229 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1230 | goto bad_orphan; | |
1231 | ||
1232 | inode = ext4_iget(sb, ino); | |
7827a7f6 TT |
1233 | if (IS_ERR(inode)) { |
1234 | err = PTR_ERR(inode); | |
1235 | ext4_error(sb, "couldn't read orphan inode %lu (err %d)", | |
1236 | ino, err); | |
1237 | return inode; | |
1238 | } | |
1d1fe1ee | 1239 | |
91ef4caf | 1240 | /* |
c9eb13a9 TT |
1241 | * If the orphans has i_nlinks > 0 then it should be able to |
1242 | * be truncated, otherwise it won't be removed from the orphan | |
1243 | * list during processing and an infinite loop will result. | |
1244 | * Similarly, it must not be a bad inode. | |
91ef4caf | 1245 | */ |
c9eb13a9 TT |
1246 | if ((inode->i_nlink && !ext4_can_truncate(inode)) || |
1247 | is_bad_inode(inode)) | |
91ef4caf DG |
1248 | goto bad_orphan; |
1249 | ||
1d1fe1ee DH |
1250 | if (NEXT_ORPHAN(inode) > max_ino) |
1251 | goto bad_orphan; | |
1252 | brelse(bitmap_bh); | |
1253 | return inode; | |
1254 | ||
1d1fe1ee | 1255 | bad_orphan: |
7827a7f6 TT |
1256 | ext4_error(sb, "bad orphan inode %lu", ino); |
1257 | if (bitmap_bh) | |
1258 | printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n", | |
1259 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
1260 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
1d1fe1ee | 1261 | if (inode) { |
7827a7f6 | 1262 | printk(KERN_ERR "is_bad_inode(inode)=%d\n", |
1d1fe1ee | 1263 | is_bad_inode(inode)); |
7827a7f6 | 1264 | printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n", |
1d1fe1ee | 1265 | NEXT_ORPHAN(inode)); |
7827a7f6 TT |
1266 | printk(KERN_ERR "max_ino=%lu\n", max_ino); |
1267 | printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink); | |
ac27a0ec | 1268 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1269 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1270 | inode->i_blocks = 0; |
1271 | iput(inode); | |
ac27a0ec | 1272 | } |
ac27a0ec | 1273 | brelse(bitmap_bh); |
1d1fe1ee | 1274 | return ERR_PTR(err); |
ac27a0ec DK |
1275 | } |
1276 | ||
af5bc92d | 1277 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1278 | { |
1279 | unsigned long desc_count; | |
617ba13b | 1280 | struct ext4_group_desc *gdp; |
8df9675f | 1281 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1282 | #ifdef EXT4FS_DEBUG |
1283 | struct ext4_super_block *es; | |
ac27a0ec DK |
1284 | unsigned long bitmap_count, x; |
1285 | struct buffer_head *bitmap_bh = NULL; | |
1286 | ||
617ba13b | 1287 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1288 | desc_count = 0; |
1289 | bitmap_count = 0; | |
1290 | gdp = NULL; | |
8df9675f | 1291 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1292 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1293 | if (!gdp) |
1294 | continue; | |
560671a0 | 1295 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1296 | brelse(bitmap_bh); |
e29d1cde | 1297 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
9008a58e DW |
1298 | if (IS_ERR(bitmap_bh)) { |
1299 | bitmap_bh = NULL; | |
ac27a0ec | 1300 | continue; |
9008a58e | 1301 | } |
ac27a0ec | 1302 | |
f6fb99ca TT |
1303 | x = ext4_count_free(bitmap_bh->b_data, |
1304 | EXT4_INODES_PER_GROUP(sb) / 8); | |
c549a95d | 1305 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1306 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1307 | bitmap_count += x; |
1308 | } | |
1309 | brelse(bitmap_bh); | |
4776004f TT |
1310 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1311 | "stored = %u, computed = %lu, %lu\n", | |
1312 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1313 | return desc_count; |
1314 | #else | |
1315 | desc_count = 0; | |
8df9675f | 1316 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1317 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1318 | if (!gdp) |
1319 | continue; | |
560671a0 | 1320 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1321 | cond_resched(); |
1322 | } | |
1323 | return desc_count; | |
1324 | #endif | |
1325 | } | |
1326 | ||
1327 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1328 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1329 | { |
1330 | unsigned long count = 0; | |
8df9675f | 1331 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1332 | |
8df9675f | 1333 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1334 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1335 | if (!gdp) |
1336 | continue; | |
560671a0 | 1337 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1338 | } |
1339 | return count; | |
1340 | } | |
bfff6873 LC |
1341 | |
1342 | /* | |
1343 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1344 | * inode table. Must be called without any spinlock held. The only place | |
1345 | * where it is called from on active part of filesystem is ext4lazyinit | |
1346 | * thread, so we do not need any special locks, however we have to prevent | |
1347 | * inode allocation from the current group, so we take alloc_sem lock, to | |
119c0d44 | 1348 | * block ext4_new_inode() until we are finished. |
bfff6873 | 1349 | */ |
e0cbee3e | 1350 | int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, |
bfff6873 LC |
1351 | int barrier) |
1352 | { | |
1353 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1354 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1355 | struct ext4_group_desc *gdp = NULL; | |
1356 | struct buffer_head *group_desc_bh; | |
1357 | handle_t *handle; | |
1358 | ext4_fsblk_t blk; | |
1359 | int num, ret = 0, used_blks = 0; | |
bfff6873 LC |
1360 | |
1361 | /* This should not happen, but just to be sure check this */ | |
1362 | if (sb->s_flags & MS_RDONLY) { | |
1363 | ret = 1; | |
1364 | goto out; | |
1365 | } | |
1366 | ||
1367 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1368 | if (!gdp) | |
1369 | goto out; | |
1370 | ||
1371 | /* | |
1372 | * We do not need to lock this, because we are the only one | |
1373 | * handling this flag. | |
1374 | */ | |
1375 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1376 | goto out; | |
1377 | ||
9924a92a | 1378 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
bfff6873 LC |
1379 | if (IS_ERR(handle)) { |
1380 | ret = PTR_ERR(handle); | |
1381 | goto out; | |
1382 | } | |
1383 | ||
1384 | down_write(&grp->alloc_sem); | |
1385 | /* | |
1386 | * If inode bitmap was already initialized there may be some | |
1387 | * used inodes so we need to skip blocks with used inodes in | |
1388 | * inode table. | |
1389 | */ | |
1390 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1391 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1392 | ext4_itable_unused_count(sb, gdp)), | |
1393 | sbi->s_inodes_per_block); | |
1394 | ||
857ac889 | 1395 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) { |
1084f252 TT |
1396 | ext4_error(sb, "Something is wrong with group %u: " |
1397 | "used itable blocks: %d; " | |
1398 | "itable unused count: %u", | |
857ac889 LC |
1399 | group, used_blks, |
1400 | ext4_itable_unused_count(sb, gdp)); | |
1401 | ret = 1; | |
33853a0d | 1402 | goto err_out; |
857ac889 LC |
1403 | } |
1404 | ||
bfff6873 LC |
1405 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1406 | num = sbi->s_itb_per_group - used_blks; | |
1407 | ||
1408 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1409 | ret = ext4_journal_get_write_access(handle, | |
1410 | group_desc_bh); | |
1411 | if (ret) | |
1412 | goto err_out; | |
1413 | ||
bfff6873 LC |
1414 | /* |
1415 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1416 | * flag anyway, because obviously, when it is full it does not need | |
1417 | * further zeroing. | |
1418 | */ | |
1419 | if (unlikely(num == 0)) | |
1420 | goto skip_zeroout; | |
1421 | ||
1422 | ext4_debug("going to zero out inode table in group %d\n", | |
1423 | group); | |
a107e5a3 | 1424 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1425 | if (ret < 0) |
1426 | goto err_out; | |
a107e5a3 TT |
1427 | if (barrier) |
1428 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); | |
bfff6873 LC |
1429 | |
1430 | skip_zeroout: | |
1431 | ext4_lock_group(sb, group); | |
1432 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
feb0ab32 | 1433 | ext4_group_desc_csum_set(sb, group, gdp); |
bfff6873 LC |
1434 | ext4_unlock_group(sb, group); |
1435 | ||
1436 | BUFFER_TRACE(group_desc_bh, | |
1437 | "call ext4_handle_dirty_metadata"); | |
1438 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1439 | group_desc_bh); | |
1440 | ||
1441 | err_out: | |
1442 | up_write(&grp->alloc_sem); | |
1443 | ext4_journal_stop(handle); | |
1444 | out: | |
1445 | return ret; | |
1446 | } |