]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
ac27a0ec | 2 | /* |
617ba13b | 3 | * linux/fs/ext4/ialloc.c |
ac27a0ec DK |
4 | * |
5 | * Copyright (C) 1992, 1993, 1994, 1995 | |
6 | * Remy Card (card@masi.ibp.fr) | |
7 | * Laboratoire MASI - Institut Blaise Pascal | |
8 | * Universite Pierre et Marie Curie (Paris VI) | |
9 | * | |
10 | * BSD ufs-inspired inode and directory allocation by | |
11 | * Stephen Tweedie (sct@redhat.com), 1993 | |
12 | * Big-endian to little-endian byte-swapping/bitmaps by | |
13 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
14 | */ | |
15 | ||
16 | #include <linux/time.h> | |
17 | #include <linux/fs.h> | |
ac27a0ec DK |
18 | #include <linux/stat.h> |
19 | #include <linux/string.h> | |
20 | #include <linux/quotaops.h> | |
21 | #include <linux/buffer_head.h> | |
22 | #include <linux/random.h> | |
23 | #include <linux/bitops.h> | |
3a5b2ecd | 24 | #include <linux/blkdev.h> |
5b825c3a IM |
25 | #include <linux/cred.h> |
26 | ||
ac27a0ec | 27 | #include <asm/byteorder.h> |
9bffad1e | 28 | |
3dcf5451 CH |
29 | #include "ext4.h" |
30 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
31 | #include "xattr.h" |
32 | #include "acl.h" | |
33 | ||
9bffad1e TT |
34 | #include <trace/events/ext4.h> |
35 | ||
ac27a0ec DK |
36 | /* |
37 | * ialloc.c contains the inodes allocation and deallocation routines | |
38 | */ | |
39 | ||
40 | /* | |
41 | * The free inodes are managed by bitmaps. A file system contains several | |
42 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
43 | * block for inodes, N blocks for the inode table and data blocks. | |
44 | * | |
45 | * The file system contains group descriptors which are located after the | |
46 | * super block. Each descriptor contains the number of the bitmap block and | |
47 | * the free blocks count in the block. | |
48 | */ | |
49 | ||
717d50e4 AD |
50 | /* |
51 | * To avoid calling the atomic setbit hundreds or thousands of times, we only | |
52 | * need to use it within a single byte (to ensure we get endianness right). | |
53 | * We can use memset for the rest of the bitmap as there are no other users. | |
54 | */ | |
61d08673 | 55 | void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap) |
717d50e4 AD |
56 | { |
57 | int i; | |
58 | ||
59 | if (start_bit >= end_bit) | |
60 | return; | |
61 | ||
62 | ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); | |
63 | for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) | |
64 | ext4_set_bit(i, bitmap); | |
65 | if (i < end_bit) | |
66 | memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); | |
67 | } | |
68 | ||
813e5727 TT |
69 | void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate) |
70 | { | |
71 | if (uptodate) { | |
72 | set_buffer_uptodate(bh); | |
73 | set_bitmap_uptodate(bh); | |
74 | } | |
75 | unlock_buffer(bh); | |
76 | put_bh(bh); | |
77 | } | |
78 | ||
9008a58e DW |
79 | static int ext4_validate_inode_bitmap(struct super_block *sb, |
80 | struct ext4_group_desc *desc, | |
81 | ext4_group_t block_group, | |
82 | struct buffer_head *bh) | |
83 | { | |
84 | ext4_fsblk_t blk; | |
85 | struct ext4_group_info *grp = ext4_get_group_info(sb, block_group); | |
9008a58e DW |
86 | |
87 | if (buffer_verified(bh)) | |
88 | return 0; | |
89 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) | |
90 | return -EFSCORRUPTED; | |
91 | ||
92 | ext4_lock_group(sb, block_group); | |
8d5a803c TT |
93 | if (buffer_verified(bh)) |
94 | goto verified; | |
9008a58e DW |
95 | blk = ext4_inode_bitmap(sb, desc); |
96 | if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh, | |
97 | EXT4_INODES_PER_GROUP(sb) / 8)) { | |
98 | ext4_unlock_group(sb, block_group); | |
99 | ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " | |
100 | "inode_bitmap = %llu", block_group, blk); | |
db79e6d1 WS |
101 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
102 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
9008a58e DW |
103 | return -EFSBADCRC; |
104 | } | |
105 | set_buffer_verified(bh); | |
8d5a803c | 106 | verified: |
9008a58e DW |
107 | ext4_unlock_group(sb, block_group); |
108 | return 0; | |
109 | } | |
110 | ||
ac27a0ec DK |
111 | /* |
112 | * Read the inode allocation bitmap for a given block_group, reading | |
113 | * into the specified slot in the superblock's bitmap cache. | |
114 | * | |
115 | * Return buffer_head of bitmap on success or NULL. | |
116 | */ | |
117 | static struct buffer_head * | |
e29d1cde | 118 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 119 | { |
617ba13b | 120 | struct ext4_group_desc *desc; |
7dac4a17 | 121 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
ac27a0ec | 122 | struct buffer_head *bh = NULL; |
e29d1cde | 123 | ext4_fsblk_t bitmap_blk; |
9008a58e | 124 | int err; |
ac27a0ec | 125 | |
617ba13b | 126 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 127 | if (!desc) |
9008a58e | 128 | return ERR_PTR(-EFSCORRUPTED); |
bfff6873 | 129 | |
e29d1cde | 130 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
7dac4a17 TT |
131 | if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || |
132 | (bitmap_blk >= ext4_blocks_count(sbi->s_es))) { | |
133 | ext4_error(sb, "Invalid inode bitmap blk %llu in " | |
134 | "block_group %u", bitmap_blk, block_group); | |
206f6d55 WS |
135 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
136 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
7dac4a17 TT |
137 | return ERR_PTR(-EFSCORRUPTED); |
138 | } | |
e29d1cde ES |
139 | bh = sb_getblk(sb, bitmap_blk); |
140 | if (unlikely(!bh)) { | |
5ef2a699 WS |
141 | ext4_warning(sb, "Cannot read inode bitmap - " |
142 | "block_group = %u, inode_bitmap = %llu", | |
143 | block_group, bitmap_blk); | |
0db9fdeb | 144 | return ERR_PTR(-ENOMEM); |
e29d1cde | 145 | } |
2ccb5fb9 | 146 | if (bitmap_uptodate(bh)) |
41a246d1 | 147 | goto verify; |
e29d1cde | 148 | |
c806e68f | 149 | lock_buffer(bh); |
2ccb5fb9 AK |
150 | if (bitmap_uptodate(bh)) { |
151 | unlock_buffer(bh); | |
41a246d1 | 152 | goto verify; |
2ccb5fb9 | 153 | } |
bfff6873 | 154 | |
955ce5f5 | 155 | ext4_lock_group(sb, block_group); |
8844618d TT |
156 | if (ext4_has_group_desc_csum(sb) && |
157 | (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) { | |
158 | if (block_group == 0) { | |
159 | ext4_unlock_group(sb, block_group); | |
160 | unlock_buffer(bh); | |
161 | ext4_error(sb, "Inode bitmap for bg 0 marked " | |
162 | "uninitialized"); | |
163 | err = -EFSCORRUPTED; | |
164 | goto out; | |
165 | } | |
044e6e3d TT |
166 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); |
167 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), | |
168 | sb->s_blocksize * 8, bh->b_data); | |
2ccb5fb9 | 169 | set_bitmap_uptodate(bh); |
e29d1cde | 170 | set_buffer_uptodate(bh); |
41a246d1 | 171 | set_buffer_verified(bh); |
955ce5f5 | 172 | ext4_unlock_group(sb, block_group); |
3300beda | 173 | unlock_buffer(bh); |
e29d1cde | 174 | return bh; |
717d50e4 | 175 | } |
955ce5f5 | 176 | ext4_unlock_group(sb, block_group); |
bfff6873 | 177 | |
2ccb5fb9 AK |
178 | if (buffer_uptodate(bh)) { |
179 | /* | |
180 | * if not uninit if bh is uptodate, | |
181 | * bitmap is also uptodate | |
182 | */ | |
183 | set_bitmap_uptodate(bh); | |
184 | unlock_buffer(bh); | |
41a246d1 | 185 | goto verify; |
2ccb5fb9 AK |
186 | } |
187 | /* | |
813e5727 | 188 | * submit the buffer_head for reading |
2ccb5fb9 | 189 | */ |
0562e0ba | 190 | trace_ext4_load_inode_bitmap(sb, block_group); |
813e5727 TT |
191 | bh->b_end_io = ext4_end_bitmap_read; |
192 | get_bh(bh); | |
2a222ca9 | 193 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
813e5727 TT |
194 | wait_on_buffer(bh); |
195 | if (!buffer_uptodate(bh)) { | |
e29d1cde | 196 | put_bh(bh); |
12062ddd | 197 | ext4_error(sb, "Cannot read inode bitmap - " |
813e5727 TT |
198 | "block_group = %u, inode_bitmap = %llu", |
199 | block_group, bitmap_blk); | |
206f6d55 WS |
200 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
201 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
9008a58e | 202 | return ERR_PTR(-EIO); |
e29d1cde | 203 | } |
41a246d1 DW |
204 | |
205 | verify: | |
9008a58e DW |
206 | err = ext4_validate_inode_bitmap(sb, desc, block_group, bh); |
207 | if (err) | |
208 | goto out; | |
ac27a0ec | 209 | return bh; |
9008a58e DW |
210 | out: |
211 | put_bh(bh); | |
212 | return ERR_PTR(err); | |
ac27a0ec DK |
213 | } |
214 | ||
215 | /* | |
216 | * NOTE! When we get the inode, we're the only people | |
217 | * that have access to it, and as such there are no | |
218 | * race conditions we have to worry about. The inode | |
219 | * is not on the hash-lists, and it cannot be reached | |
220 | * through the filesystem because the directory entry | |
221 | * has been deleted earlier. | |
222 | * | |
223 | * HOWEVER: we must make sure that we get no aliases, | |
224 | * which means that we have to call "clear_inode()" | |
225 | * _before_ we mark the inode not in use in the inode | |
226 | * bitmaps. Otherwise a newly created file might use | |
227 | * the same inode number (not actually the same pointer | |
228 | * though), and then we'd have two inodes sharing the | |
229 | * same inode number and space on the harddisk. | |
230 | */ | |
af5bc92d | 231 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 232 | { |
af5bc92d | 233 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
234 | int is_directory; |
235 | unsigned long ino; | |
236 | struct buffer_head *bitmap_bh = NULL; | |
237 | struct buffer_head *bh2; | |
fd2d4291 | 238 | ext4_group_t block_group; |
ac27a0ec | 239 | unsigned long bit; |
af5bc92d TT |
240 | struct ext4_group_desc *gdp; |
241 | struct ext4_super_block *es; | |
617ba13b | 242 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 243 | int fatal = 0, err, count, cleared; |
87a39389 | 244 | struct ext4_group_info *grp; |
ac27a0ec | 245 | |
92b97816 TT |
246 | if (!sb) { |
247 | printk(KERN_ERR "EXT4-fs: %s:%d: inode on " | |
248 | "nonexistent device\n", __func__, __LINE__); | |
ac27a0ec DK |
249 | return; |
250 | } | |
92b97816 TT |
251 | if (atomic_read(&inode->i_count) > 1) { |
252 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", | |
253 | __func__, __LINE__, inode->i_ino, | |
254 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
255 | return; |
256 | } | |
92b97816 TT |
257 | if (inode->i_nlink) { |
258 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", | |
259 | __func__, __LINE__, inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
260 | return; |
261 | } | |
617ba13b | 262 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
263 | |
264 | ino = inode->i_ino; | |
af5bc92d | 265 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 266 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
267 | |
268 | /* | |
269 | * Note: we must free any quota before locking the superblock, | |
270 | * as writing the quota to disk may need the lock as well. | |
271 | */ | |
871a2931 | 272 | dquot_initialize(inode); |
63936dda | 273 | dquot_free_inode(inode); |
9f754758 | 274 | dquot_drop(inode); |
ac27a0ec DK |
275 | |
276 | is_directory = S_ISDIR(inode->i_mode); | |
277 | ||
278 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 279 | ext4_clear_inode(inode); |
ac27a0ec | 280 | |
49598e04 | 281 | es = sbi->s_es; |
617ba13b | 282 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { |
12062ddd | 283 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
284 | goto error_return; |
285 | } | |
617ba13b MC |
286 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
287 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 288 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
87a39389 DW |
289 | /* Don't bother if the inode bitmap is corrupt. */ |
290 | grp = ext4_get_group_info(sb, block_group); | |
9008a58e DW |
291 | if (IS_ERR(bitmap_bh)) { |
292 | fatal = PTR_ERR(bitmap_bh); | |
293 | bitmap_bh = NULL; | |
294 | goto error_return; | |
295 | } | |
296 | if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) { | |
297 | fatal = -EFSCORRUPTED; | |
ac27a0ec | 298 | goto error_return; |
9008a58e | 299 | } |
ac27a0ec DK |
300 | |
301 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 302 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
303 | if (fatal) |
304 | goto error_return; | |
305 | ||
d17413c0 DM |
306 | fatal = -ESRCH; |
307 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
308 | if (gdp) { | |
ac27a0ec | 309 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 310 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
311 | } |
312 | ext4_lock_group(sb, block_group); | |
597d508c | 313 | cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); |
d17413c0 DM |
314 | if (fatal || !cleared) { |
315 | ext4_unlock_group(sb, block_group); | |
316 | goto out; | |
317 | } | |
7d39db14 | 318 | |
d17413c0 DM |
319 | count = ext4_free_inodes_count(sb, gdp) + 1; |
320 | ext4_free_inodes_set(sb, gdp, count); | |
321 | if (is_directory) { | |
322 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
323 | ext4_used_dirs_set(sb, gdp, count); | |
324 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 325 | } |
41a246d1 DW |
326 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, |
327 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 328 | ext4_group_desc_csum_set(sb, block_group, gdp); |
d17413c0 | 329 | ext4_unlock_group(sb, block_group); |
ac27a0ec | 330 | |
d17413c0 DM |
331 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
332 | if (sbi->s_log_groups_per_flex) { | |
333 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 334 | |
d17413c0 DM |
335 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
336 | if (is_directory) | |
337 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 338 | } |
d17413c0 DM |
339 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
340 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
341 | out: | |
342 | if (cleared) { | |
343 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
344 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
345 | if (!fatal) | |
346 | fatal = err; | |
87a39389 | 347 | } else { |
d17413c0 | 348 | ext4_error(sb, "bit already cleared for inode %lu", ino); |
db79e6d1 WS |
349 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
350 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
87a39389 | 351 | } |
d17413c0 | 352 | |
ac27a0ec DK |
353 | error_return: |
354 | brelse(bitmap_bh); | |
617ba13b | 355 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
356 | } |
357 | ||
a4912123 | 358 | struct orlov_stats { |
90ba983f | 359 | __u64 free_clusters; |
a4912123 | 360 | __u32 free_inodes; |
a4912123 TT |
361 | __u32 used_dirs; |
362 | }; | |
363 | ||
364 | /* | |
365 | * Helper function for Orlov's allocator; returns critical information | |
366 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
367 | * is a block group number; otherwise it is flex_bg number. | |
368 | */ | |
1f109d5a TT |
369 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
370 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
371 | { |
372 | struct ext4_group_desc *desc; | |
7d39db14 | 373 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 374 | |
7d39db14 TT |
375 | if (flex_size > 1) { |
376 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
90ba983f | 377 | stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); |
7d39db14 TT |
378 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); |
379 | return; | |
380 | } | |
a4912123 | 381 | |
7d39db14 TT |
382 | desc = ext4_get_group_desc(sb, g, NULL); |
383 | if (desc) { | |
384 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
021b65bb | 385 | stats->free_clusters = ext4_free_group_clusters(sb, desc); |
7d39db14 TT |
386 | stats->used_dirs = ext4_used_dirs_count(sb, desc); |
387 | } else { | |
388 | stats->free_inodes = 0; | |
24aaa8ef | 389 | stats->free_clusters = 0; |
7d39db14 | 390 | stats->used_dirs = 0; |
a4912123 TT |
391 | } |
392 | } | |
393 | ||
ac27a0ec DK |
394 | /* |
395 | * Orlov's allocator for directories. | |
396 | * | |
397 | * We always try to spread first-level directories. | |
398 | * | |
399 | * If there are blockgroups with both free inodes and free blocks counts | |
400 | * not worse than average we return one with smallest directory count. | |
401 | * Otherwise we simply return a random group. | |
402 | * | |
403 | * For the rest rules look so: | |
404 | * | |
405 | * It's OK to put directory into a group unless | |
406 | * it has too many directories already (max_dirs) or | |
407 | * it has too few free inodes left (min_inodes) or | |
408 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 409 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
410 | * conditions we search cyclically through the rest. If none |
411 | * of the groups look good we just look for a group with more | |
412 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
413 | */ |
414 | ||
2aa9fc4c | 415 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
dcca3fec | 416 | ext4_group_t *group, umode_t mode, |
f157a4aa | 417 | const struct qstr *qstr) |
ac27a0ec | 418 | { |
fd2d4291 | 419 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 420 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 421 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 422 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
14c83c9f | 423 | unsigned int freei, avefreei, grp_free; |
24aaa8ef | 424 | ext4_fsblk_t freeb, avefreec; |
ac27a0ec | 425 | unsigned int ndirs; |
a4912123 | 426 | int max_dirs, min_inodes; |
24aaa8ef | 427 | ext4_grpblk_t min_clusters; |
8df9675f | 428 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 429 | struct ext4_group_desc *desc; |
a4912123 TT |
430 | struct orlov_stats stats; |
431 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 432 | struct dx_hash_info hinfo; |
a4912123 | 433 | |
8df9675f | 434 | ngroups = real_ngroups; |
a4912123 | 435 | if (flex_size > 1) { |
8df9675f | 436 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
437 | sbi->s_log_groups_per_flex; |
438 | parent_group >>= sbi->s_log_groups_per_flex; | |
439 | } | |
ac27a0ec DK |
440 | |
441 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
442 | avefreei = freei / ngroups; | |
57042651 TT |
443 | freeb = EXT4_C2B(sbi, |
444 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
24aaa8ef TT |
445 | avefreec = freeb; |
446 | do_div(avefreec, ngroups); | |
ac27a0ec DK |
447 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
448 | ||
a4912123 | 449 | if (S_ISDIR(mode) && |
2b0143b5 | 450 | ((parent == d_inode(sb->s_root)) || |
12e9b892 | 451 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 452 | int best_ndir = inodes_per_group; |
2aa9fc4c | 453 | int ret = -1; |
ac27a0ec | 454 | |
f157a4aa TT |
455 | if (qstr) { |
456 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
457 | hinfo.seed = sbi->s_hash_seed; | |
b886ee3e | 458 | ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo); |
f157a4aa TT |
459 | grp = hinfo.hash; |
460 | } else | |
dd1f723b | 461 | grp = prandom_u32(); |
2aa9fc4c | 462 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 463 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
464 | g = (parent_group + i) % ngroups; |
465 | get_orlov_stats(sb, g, flex_size, &stats); | |
466 | if (!stats.free_inodes) | |
ac27a0ec | 467 | continue; |
a4912123 | 468 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 469 | continue; |
a4912123 | 470 | if (stats.free_inodes < avefreei) |
ac27a0ec | 471 | continue; |
24aaa8ef | 472 | if (stats.free_clusters < avefreec) |
ac27a0ec | 473 | continue; |
a4912123 | 474 | grp = g; |
2aa9fc4c | 475 | ret = 0; |
a4912123 TT |
476 | best_ndir = stats.used_dirs; |
477 | } | |
478 | if (ret) | |
479 | goto fallback; | |
480 | found_flex_bg: | |
481 | if (flex_size == 1) { | |
482 | *group = grp; | |
483 | return 0; | |
484 | } | |
485 | ||
486 | /* | |
487 | * We pack inodes at the beginning of the flexgroup's | |
488 | * inode tables. Block allocation decisions will do | |
489 | * something similar, although regular files will | |
490 | * start at 2nd block group of the flexgroup. See | |
491 | * ext4_ext_find_goal() and ext4_find_near(). | |
492 | */ | |
493 | grp *= flex_size; | |
494 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 495 | if (grp+i >= real_ngroups) |
a4912123 TT |
496 | break; |
497 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
498 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
499 | *group = grp+i; | |
500 | return 0; | |
501 | } | |
ac27a0ec | 502 | } |
ac27a0ec DK |
503 | goto fallback; |
504 | } | |
505 | ||
ac27a0ec | 506 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
507 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
508 | if (min_inodes < 1) | |
509 | min_inodes = 1; | |
24aaa8ef | 510 | min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; |
a4912123 TT |
511 | |
512 | /* | |
513 | * Start looking in the flex group where we last allocated an | |
514 | * inode for this parent directory | |
515 | */ | |
516 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
517 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
518 | if (flex_size > 1) | |
519 | parent_group >>= sbi->s_log_groups_per_flex; | |
520 | } | |
ac27a0ec DK |
521 | |
522 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
523 | grp = (parent_group + i) % ngroups; |
524 | get_orlov_stats(sb, grp, flex_size, &stats); | |
525 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 526 | continue; |
a4912123 | 527 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 528 | continue; |
24aaa8ef | 529 | if (stats.free_clusters < min_clusters) |
ac27a0ec | 530 | continue; |
a4912123 | 531 | goto found_flex_bg; |
ac27a0ec DK |
532 | } |
533 | ||
534 | fallback: | |
8df9675f | 535 | ngroups = real_ngroups; |
a4912123 | 536 | avefreei = freei / ngroups; |
b5451f7b | 537 | fallback_retry: |
a4912123 | 538 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 539 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
540 | grp = (parent_group + i) % ngroups; |
541 | desc = ext4_get_group_desc(sb, grp, NULL); | |
bb3d132a DC |
542 | if (desc) { |
543 | grp_free = ext4_free_inodes_count(sb, desc); | |
544 | if (grp_free && grp_free >= avefreei) { | |
545 | *group = grp; | |
546 | return 0; | |
547 | } | |
a4912123 | 548 | } |
ac27a0ec DK |
549 | } |
550 | ||
551 | if (avefreei) { | |
552 | /* | |
553 | * The free-inodes counter is approximate, and for really small | |
554 | * filesystems the above test can fail to find any blockgroups | |
555 | */ | |
556 | avefreei = 0; | |
b5451f7b | 557 | goto fallback_retry; |
ac27a0ec DK |
558 | } |
559 | ||
560 | return -1; | |
561 | } | |
562 | ||
2aa9fc4c | 563 | static int find_group_other(struct super_block *sb, struct inode *parent, |
dcca3fec | 564 | ext4_group_t *group, umode_t mode) |
ac27a0ec | 565 | { |
fd2d4291 | 566 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 567 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 568 | struct ext4_group_desc *desc; |
a4912123 TT |
569 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
570 | ||
571 | /* | |
572 | * Try to place the inode is the same flex group as its | |
573 | * parent. If we can't find space, use the Orlov algorithm to | |
574 | * find another flex group, and store that information in the | |
575 | * parent directory's inode information so that use that flex | |
576 | * group for future allocations. | |
577 | */ | |
578 | if (flex_size > 1) { | |
579 | int retry = 0; | |
580 | ||
581 | try_again: | |
582 | parent_group &= ~(flex_size-1); | |
583 | last = parent_group + flex_size; | |
584 | if (last > ngroups) | |
585 | last = ngroups; | |
586 | for (i = parent_group; i < last; i++) { | |
587 | desc = ext4_get_group_desc(sb, i, NULL); | |
588 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
589 | *group = i; | |
590 | return 0; | |
591 | } | |
592 | } | |
593 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
594 | retry = 1; | |
595 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
596 | goto try_again; | |
597 | } | |
598 | /* | |
599 | * If this didn't work, use the Orlov search algorithm | |
600 | * to find a new flex group; we pass in the mode to | |
601 | * avoid the topdir algorithms. | |
602 | */ | |
603 | *group = parent_group + flex_size; | |
604 | if (*group > ngroups) | |
605 | *group = 0; | |
7dc57615 | 606 | return find_group_orlov(sb, parent, group, mode, NULL); |
a4912123 | 607 | } |
ac27a0ec DK |
608 | |
609 | /* | |
610 | * Try to place the inode in its parent directory | |
611 | */ | |
2aa9fc4c AM |
612 | *group = parent_group; |
613 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 614 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 615 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 616 | return 0; |
ac27a0ec DK |
617 | |
618 | /* | |
619 | * We're going to place this inode in a different blockgroup from its | |
620 | * parent. We want to cause files in a common directory to all land in | |
621 | * the same blockgroup. But we want files which are in a different | |
622 | * directory which shares a blockgroup with our parent to land in a | |
623 | * different blockgroup. | |
624 | * | |
625 | * So add our directory's i_ino into the starting point for the hash. | |
626 | */ | |
2aa9fc4c | 627 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
628 | |
629 | /* | |
630 | * Use a quadratic hash to find a group with a free inode and some free | |
631 | * blocks. | |
632 | */ | |
633 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
634 | *group += i; |
635 | if (*group >= ngroups) | |
636 | *group -= ngroups; | |
637 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 638 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 639 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 640 | return 0; |
ac27a0ec DK |
641 | } |
642 | ||
643 | /* | |
644 | * That failed: try linear search for a free inode, even if that group | |
645 | * has no free blocks. | |
646 | */ | |
2aa9fc4c | 647 | *group = parent_group; |
ac27a0ec | 648 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
649 | if (++*group >= ngroups) |
650 | *group = 0; | |
651 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 652 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 653 | return 0; |
ac27a0ec DK |
654 | } |
655 | ||
656 | return -1; | |
657 | } | |
658 | ||
19883bd9 TT |
659 | /* |
660 | * In no journal mode, if an inode has recently been deleted, we want | |
661 | * to avoid reusing it until we're reasonably sure the inode table | |
662 | * block has been written back to disk. (Yes, these values are | |
663 | * somewhat arbitrary...) | |
664 | */ | |
665 | #define RECENTCY_MIN 5 | |
b5f51573 | 666 | #define RECENTCY_DIRTY 300 |
19883bd9 TT |
667 | |
668 | static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) | |
669 | { | |
670 | struct ext4_group_desc *gdp; | |
671 | struct ext4_inode *raw_inode; | |
672 | struct buffer_head *bh; | |
b5f51573 AD |
673 | int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
674 | int offset, ret = 0; | |
675 | int recentcy = RECENTCY_MIN; | |
676 | u32 dtime, now; | |
19883bd9 TT |
677 | |
678 | gdp = ext4_get_group_desc(sb, group, NULL); | |
679 | if (unlikely(!gdp)) | |
680 | return 0; | |
681 | ||
4f9d956d | 682 | bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) + |
19883bd9 | 683 | (ino / inodes_per_block)); |
4f9d956d | 684 | if (!bh || !buffer_uptodate(bh)) |
19883bd9 TT |
685 | /* |
686 | * If the block is not in the buffer cache, then it | |
687 | * must have been written out. | |
688 | */ | |
689 | goto out; | |
690 | ||
691 | offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
692 | raw_inode = (struct ext4_inode *) (bh->b_data + offset); | |
b5f51573 AD |
693 | |
694 | /* i_dtime is only 32 bits on disk, but we only care about relative | |
695 | * times in the range of a few minutes (i.e. long enough to sync a | |
696 | * recently-deleted inode to disk), so using the low 32 bits of the | |
697 | * clock (a 68 year range) is enough, see time_before32() */ | |
19883bd9 | 698 | dtime = le32_to_cpu(raw_inode->i_dtime); |
b5f51573 | 699 | now = ktime_get_real_seconds(); |
19883bd9 TT |
700 | if (buffer_dirty(bh)) |
701 | recentcy += RECENTCY_DIRTY; | |
702 | ||
b5f51573 AD |
703 | if (dtime && time_before32(dtime, now) && |
704 | time_before32(now, dtime + recentcy)) | |
19883bd9 TT |
705 | ret = 1; |
706 | out: | |
707 | brelse(bh); | |
708 | return ret; | |
709 | } | |
710 | ||
901ed070 WS |
711 | static int find_inode_bit(struct super_block *sb, ext4_group_t group, |
712 | struct buffer_head *bitmap, unsigned long *ino) | |
713 | { | |
714 | next: | |
715 | *ino = ext4_find_next_zero_bit((unsigned long *) | |
716 | bitmap->b_data, | |
717 | EXT4_INODES_PER_GROUP(sb), *ino); | |
718 | if (*ino >= EXT4_INODES_PER_GROUP(sb)) | |
719 | return 0; | |
720 | ||
721 | if ((EXT4_SB(sb)->s_journal == NULL) && | |
722 | recently_deleted(sb, group, *ino)) { | |
723 | *ino = *ino + 1; | |
724 | if (*ino < EXT4_INODES_PER_GROUP(sb)) | |
725 | goto next; | |
726 | return 0; | |
727 | } | |
728 | ||
729 | return 1; | |
730 | } | |
731 | ||
ac27a0ec DK |
732 | /* |
733 | * There are two policies for allocating an inode. If the new inode is | |
734 | * a directory, then a forward search is made for a block group with both | |
735 | * free space and a low directory-to-inode ratio; if that fails, then of | |
736 | * the groups with above-average free space, that group with the fewest | |
737 | * directories already is chosen. | |
738 | * | |
739 | * For other inodes, search forward from the parent directory's block | |
740 | * group to find a free inode. | |
741 | */ | |
1139575a TT |
742 | struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, |
743 | umode_t mode, const struct qstr *qstr, | |
1b917ed8 TE |
744 | __u32 goal, uid_t *owner, __u32 i_flags, |
745 | int handle_type, unsigned int line_no, | |
746 | int nblocks) | |
ac27a0ec DK |
747 | { |
748 | struct super_block *sb; | |
3300beda AK |
749 | struct buffer_head *inode_bitmap_bh = NULL; |
750 | struct buffer_head *group_desc_bh; | |
8df9675f | 751 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 752 | unsigned long ino = 0; |
af5bc92d TT |
753 | struct inode *inode; |
754 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
755 | struct ext4_inode_info *ei; |
756 | struct ext4_sb_info *sbi; | |
a7cdadee | 757 | int ret2, err; |
ac27a0ec | 758 | struct inode *ret; |
2aa9fc4c | 759 | ext4_group_t i; |
772cb7c8 | 760 | ext4_group_t flex_group; |
87a39389 | 761 | struct ext4_group_info *grp; |
e709e9df | 762 | int encrypt = 0; |
ac27a0ec DK |
763 | |
764 | /* Cannot create files in a deleted directory */ | |
765 | if (!dir || !dir->i_nlink) | |
766 | return ERR_PTR(-EPERM); | |
767 | ||
af65207c TE |
768 | sb = dir->i_sb; |
769 | sbi = EXT4_SB(sb); | |
770 | ||
771 | if (unlikely(ext4_forced_shutdown(sbi))) | |
0db1ff22 TT |
772 | return ERR_PTR(-EIO); |
773 | ||
0dcf2745 SF |
774 | /* Supplied owner must be valid */ |
775 | if (owner && (owner[0] == (uid_t)-1 || owner[1] == (uid_t)-1)) | |
776 | return ERR_PTR(-EOVERFLOW); | |
777 | ||
592ddec7 | 778 | if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) && |
ad47f953 TE |
779 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) && |
780 | !(i_flags & EXT4_EA_INODE_FL)) { | |
a7550b30 | 781 | err = fscrypt_get_encryption_info(dir); |
e709e9df TT |
782 | if (err) |
783 | return ERR_PTR(err); | |
a7550b30 | 784 | if (!fscrypt_has_encryption_key(dir)) |
54475f53 | 785 | return ERR_PTR(-ENOKEY); |
e709e9df TT |
786 | encrypt = 1; |
787 | } | |
788 | ||
af65207c TE |
789 | if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) { |
790 | #ifdef CONFIG_EXT4_FS_POSIX_ACL | |
791 | struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT); | |
792 | ||
996fc447 TT |
793 | if (IS_ERR(p)) |
794 | return ERR_CAST(p); | |
af65207c TE |
795 | if (p) { |
796 | int acl_size = p->a_count * sizeof(ext4_acl_entry); | |
797 | ||
798 | nblocks += (S_ISDIR(mode) ? 2 : 1) * | |
799 | __ext4_xattr_set_credits(sb, NULL /* inode */, | |
800 | NULL /* block_bh */, acl_size, | |
801 | true /* is_create */); | |
802 | posix_acl_release(p); | |
803 | } | |
804 | #endif | |
805 | ||
806 | #ifdef CONFIG_SECURITY | |
807 | { | |
808 | int num_security_xattrs = 1; | |
809 | ||
810 | #ifdef CONFIG_INTEGRITY | |
811 | num_security_xattrs++; | |
812 | #endif | |
813 | /* | |
814 | * We assume that security xattrs are never | |
815 | * more than 1k. In practice they are under | |
816 | * 128 bytes. | |
817 | */ | |
818 | nblocks += num_security_xattrs * | |
819 | __ext4_xattr_set_credits(sb, NULL /* inode */, | |
820 | NULL /* block_bh */, 1024, | |
821 | true /* is_create */); | |
822 | } | |
823 | #endif | |
824 | if (encrypt) | |
825 | nblocks += __ext4_xattr_set_credits(sb, | |
826 | NULL /* inode */, NULL /* block_bh */, | |
827 | FSCRYPT_SET_CONTEXT_MAX_SIZE, | |
828 | true /* is_create */); | |
829 | } | |
830 | ||
8df9675f | 831 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 832 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
833 | inode = new_inode(sb); |
834 | if (!inode) | |
835 | return ERR_PTR(-ENOMEM); | |
617ba13b | 836 | ei = EXT4_I(inode); |
772cb7c8 | 837 | |
eb9cc7e1 | 838 | /* |
b8a07463 | 839 | * Initialize owners and quota early so that we don't have to account |
eb9cc7e1 JK |
840 | * for quota initialization worst case in standard inode creating |
841 | * transaction | |
842 | */ | |
843 | if (owner) { | |
844 | inode->i_mode = mode; | |
845 | i_uid_write(inode, owner[0]); | |
846 | i_gid_write(inode, owner[1]); | |
847 | } else if (test_opt(sb, GRPID)) { | |
848 | inode->i_mode = mode; | |
849 | inode->i_uid = current_fsuid(); | |
850 | inode->i_gid = dir->i_gid; | |
851 | } else | |
852 | inode_init_owner(inode, dir, mode); | |
040cb378 | 853 | |
0b7b7779 | 854 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
855 | ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) |
856 | ei->i_projid = EXT4_I(dir)->i_projid; | |
857 | else | |
0dcf2745 | 858 | ei->i_projid = make_kprojid(sb->s_user_ns, EXT4_DEF_PROJID); |
040cb378 | 859 | |
a7cdadee JK |
860 | err = dquot_initialize(inode); |
861 | if (err) | |
862 | goto out; | |
eb9cc7e1 | 863 | |
11013911 AD |
864 | if (!goal) |
865 | goal = sbi->s_inode_goal; | |
866 | ||
e6462869 | 867 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
868 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
869 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
870 | ret2 = 0; | |
871 | goto got_group; | |
872 | } | |
873 | ||
4113c4ca LC |
874 | if (S_ISDIR(mode)) |
875 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); | |
876 | else | |
a4912123 | 877 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 878 | |
772cb7c8 | 879 | got_group: |
a4912123 | 880 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 881 | err = -ENOSPC; |
2aa9fc4c | 882 | if (ret2 == -1) |
ac27a0ec DK |
883 | goto out; |
884 | ||
119c0d44 TT |
885 | /* |
886 | * Normally we will only go through one pass of this loop, | |
887 | * unless we get unlucky and it turns out the group we selected | |
888 | * had its last inode grabbed by someone else. | |
889 | */ | |
11013911 | 890 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
891 | err = -EIO; |
892 | ||
3300beda | 893 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec | 894 | if (!gdp) |
eb9cc7e1 | 895 | goto out; |
ac27a0ec | 896 | |
f2a09af6 YY |
897 | /* |
898 | * Check free inodes count before loading bitmap. | |
899 | */ | |
2fe435d8 WS |
900 | if (ext4_free_inodes_count(sb, gdp) == 0) |
901 | goto next_group; | |
f2a09af6 | 902 | |
87a39389 DW |
903 | grp = ext4_get_group_info(sb, group); |
904 | /* Skip groups with already-known suspicious inode tables */ | |
2fe435d8 WS |
905 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) |
906 | goto next_group; | |
87a39389 | 907 | |
3300beda AK |
908 | brelse(inode_bitmap_bh); |
909 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
87a39389 | 910 | /* Skip groups with suspicious inode tables */ |
9008a58e DW |
911 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || |
912 | IS_ERR(inode_bitmap_bh)) { | |
913 | inode_bitmap_bh = NULL; | |
2fe435d8 | 914 | goto next_group; |
87a39389 | 915 | } |
ac27a0ec | 916 | |
ac27a0ec | 917 | repeat_in_this_group: |
901ed070 WS |
918 | ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); |
919 | if (!ret2) | |
a34eb503 | 920 | goto next_group; |
901ed070 WS |
921 | |
922 | if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) { | |
119c0d44 TT |
923 | ext4_error(sb, "reserved inode found cleared - " |
924 | "inode=%lu", ino + 1); | |
206f6d55 WS |
925 | ext4_mark_group_bitmap_corrupted(sb, group, |
926 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
2fe435d8 | 927 | goto next_group; |
119c0d44 | 928 | } |
901ed070 | 929 | |
1139575a TT |
930 | if (!handle) { |
931 | BUG_ON(nblocks <= 0); | |
932 | handle = __ext4_journal_start_sb(dir->i_sb, line_no, | |
5fe2fe89 JK |
933 | handle_type, nblocks, |
934 | 0); | |
1139575a TT |
935 | if (IS_ERR(handle)) { |
936 | err = PTR_ERR(handle); | |
eb9cc7e1 JK |
937 | ext4_std_error(sb, err); |
938 | goto out; | |
1139575a TT |
939 | } |
940 | } | |
ffb5387e ES |
941 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
942 | err = ext4_journal_get_write_access(handle, inode_bitmap_bh); | |
eb9cc7e1 JK |
943 | if (err) { |
944 | ext4_std_error(sb, err); | |
945 | goto out; | |
946 | } | |
119c0d44 TT |
947 | ext4_lock_group(sb, group); |
948 | ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); | |
901ed070 WS |
949 | if (ret2) { |
950 | /* Someone already took the bit. Repeat the search | |
951 | * with lock held. | |
952 | */ | |
953 | ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); | |
954 | if (ret2) { | |
955 | ext4_set_bit(ino, inode_bitmap_bh->b_data); | |
956 | ret2 = 0; | |
957 | } else { | |
958 | ret2 = 1; /* we didn't grab the inode */ | |
959 | } | |
960 | } | |
119c0d44 TT |
961 | ext4_unlock_group(sb, group); |
962 | ino++; /* the inode bitmap is zero-based */ | |
963 | if (!ret2) | |
964 | goto got; /* we grabbed the inode! */ | |
901ed070 | 965 | |
119c0d44 TT |
966 | if (ino < EXT4_INODES_PER_GROUP(sb)) |
967 | goto repeat_in_this_group; | |
a34eb503 TT |
968 | next_group: |
969 | if (++group == ngroups) | |
970 | group = 0; | |
ac27a0ec DK |
971 | } |
972 | err = -ENOSPC; | |
973 | goto out; | |
974 | ||
975 | got: | |
ffb5387e ES |
976 | BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); |
977 | err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); | |
eb9cc7e1 JK |
978 | if (err) { |
979 | ext4_std_error(sb, err); | |
980 | goto out; | |
981 | } | |
ffb5387e | 982 | |
61c219f5 TT |
983 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
984 | err = ext4_journal_get_write_access(handle, group_desc_bh); | |
985 | if (err) { | |
986 | ext4_std_error(sb, err); | |
987 | goto out; | |
988 | } | |
989 | ||
717d50e4 | 990 | /* We may have to initialize the block bitmap if it isn't already */ |
feb0ab32 | 991 | if (ext4_has_group_desc_csum(sb) && |
717d50e4 | 992 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 993 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 994 | |
3300beda | 995 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
9008a58e DW |
996 | if (IS_ERR(block_bitmap_bh)) { |
997 | err = PTR_ERR(block_bitmap_bh); | |
599a9b77 JK |
998 | goto out; |
999 | } | |
3300beda AK |
1000 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); |
1001 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 1002 | if (err) { |
3300beda | 1003 | brelse(block_bitmap_bh); |
eb9cc7e1 JK |
1004 | ext4_std_error(sb, err); |
1005 | goto out; | |
717d50e4 AD |
1006 | } |
1007 | ||
fd034a84 TT |
1008 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
1009 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
fd034a84 | 1010 | |
717d50e4 | 1011 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 1012 | ext4_lock_group(sb, group); |
8844618d TT |
1013 | if (ext4_has_group_desc_csum(sb) && |
1014 | (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) { | |
3300beda | 1015 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 1016 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 1017 | ext4_free_clusters_after_init(sb, group, gdp)); |
fa77dcfa | 1018 | ext4_block_bitmap_csum_set(sb, group, gdp, |
79f1ba49 | 1019 | block_bitmap_bh); |
feb0ab32 | 1020 | ext4_group_desc_csum_set(sb, group, gdp); |
717d50e4 | 1021 | } |
955ce5f5 | 1022 | ext4_unlock_group(sb, group); |
aeb1e5d6 | 1023 | brelse(block_bitmap_bh); |
717d50e4 | 1024 | |
eb9cc7e1 JK |
1025 | if (err) { |
1026 | ext4_std_error(sb, err); | |
1027 | goto out; | |
1028 | } | |
717d50e4 | 1029 | } |
119c0d44 | 1030 | |
119c0d44 | 1031 | /* Update the relevant bg descriptor fields */ |
41a246d1 | 1032 | if (ext4_has_group_desc_csum(sb)) { |
119c0d44 TT |
1033 | int free; |
1034 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1035 | ||
1036 | down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ | |
1037 | ext4_lock_group(sb, group); /* while we modify the bg desc */ | |
1038 | free = EXT4_INODES_PER_GROUP(sb) - | |
1039 | ext4_itable_unused_count(sb, gdp); | |
1040 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
1041 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
1042 | free = 0; | |
1043 | } | |
1044 | /* | |
1045 | * Check the relative inode number against the last used | |
1046 | * relative inode number in this group. if it is greater | |
1047 | * we need to update the bg_itable_unused count | |
1048 | */ | |
1049 | if (ino > free) | |
1050 | ext4_itable_unused_set(sb, gdp, | |
1051 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
1052 | up_read(&grp->alloc_sem); | |
6f2e9f0e TM |
1053 | } else { |
1054 | ext4_lock_group(sb, group); | |
119c0d44 | 1055 | } |
6f2e9f0e | 1056 | |
119c0d44 TT |
1057 | ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); |
1058 | if (S_ISDIR(mode)) { | |
1059 | ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); | |
1060 | if (sbi->s_log_groups_per_flex) { | |
1061 | ext4_group_t f = ext4_flex_group(sbi, group); | |
1062 | ||
1063 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); | |
1064 | } | |
1065 | } | |
41a246d1 DW |
1066 | if (ext4_has_group_desc_csum(sb)) { |
1067 | ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, | |
1068 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 1069 | ext4_group_desc_csum_set(sb, group, gdp); |
119c0d44 | 1070 | } |
6f2e9f0e | 1071 | ext4_unlock_group(sb, group); |
119c0d44 | 1072 | |
3300beda AK |
1073 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
1074 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
eb9cc7e1 JK |
1075 | if (err) { |
1076 | ext4_std_error(sb, err); | |
1077 | goto out; | |
1078 | } | |
ac27a0ec DK |
1079 | |
1080 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
1081 | if (S_ISDIR(mode)) | |
1082 | percpu_counter_inc(&sbi->s_dirs_counter); | |
ac27a0ec | 1083 | |
772cb7c8 JS |
1084 | if (sbi->s_log_groups_per_flex) { |
1085 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 1086 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 | 1087 | } |
ac27a0ec | 1088 | |
717d50e4 | 1089 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
1090 | /* This is the optimal IO size (for stat), not the fs block size */ |
1091 | inode->i_blocks = 0; | |
95582b00 | 1092 | inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode); |
7b62b293 | 1093 | ei->i_crtime = inode->i_mtime; |
ac27a0ec DK |
1094 | |
1095 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
1096 | ei->i_dir_start_lookup = 0; | |
1097 | ei->i_disksize = 0; | |
1098 | ||
4af83508 | 1099 | /* Don't inherit extent flag from directory, amongst others. */ |
2dc6b0d4 DG |
1100 | ei->i_flags = |
1101 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
1b917ed8 | 1102 | ei->i_flags |= i_flags; |
ac27a0ec | 1103 | ei->i_file_acl = 0; |
ac27a0ec | 1104 | ei->i_dtime = 0; |
ac27a0ec | 1105 | ei->i_block_group = group; |
a4912123 | 1106 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1107 | |
617ba13b | 1108 | ext4_set_inode_flags(inode); |
ac27a0ec | 1109 | if (IS_DIRSYNC(inode)) |
0390131b | 1110 | ext4_handle_sync(handle); |
6b38e842 | 1111 | if (insert_inode_locked(inode) < 0) { |
acd6ad83 JK |
1112 | /* |
1113 | * Likely a bitmap corruption causing inode to be allocated | |
1114 | * twice. | |
1115 | */ | |
1116 | err = -EIO; | |
eb9cc7e1 JK |
1117 | ext4_error(sb, "failed to insert inode %lu: doubly allocated?", |
1118 | inode->i_ino); | |
206f6d55 WS |
1119 | ext4_mark_group_bitmap_corrupted(sb, group, |
1120 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
eb9cc7e1 | 1121 | goto out; |
6b38e842 | 1122 | } |
23253068 | 1123 | inode->i_generation = prandom_u32(); |
ac27a0ec | 1124 | |
814525f4 | 1125 | /* Precompute checksum seed for inode metadata */ |
9aa5d32b | 1126 | if (ext4_has_metadata_csum(sb)) { |
814525f4 | 1127 | __u32 csum; |
814525f4 DW |
1128 | __le32 inum = cpu_to_le32(inode->i_ino); |
1129 | __le32 gen = cpu_to_le32(inode->i_generation); | |
1130 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
1131 | sizeof(inum)); | |
1132 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
1133 | sizeof(gen)); | |
1134 | } | |
1135 | ||
353eb83c | 1136 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 1137 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 | 1138 | |
49598e04 | 1139 | ei->i_extra_isize = sbi->s_want_extra_isize; |
f08225d1 | 1140 | ei->i_inline_off = 0; |
e2b911c5 | 1141 | if (ext4_has_feature_inline_data(sb)) |
f08225d1 | 1142 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
ac27a0ec | 1143 | ret = inode; |
63936dda CH |
1144 | err = dquot_alloc_inode(inode); |
1145 | if (err) | |
ac27a0ec | 1146 | goto fail_drop; |
ac27a0ec | 1147 | |
aa1dca3b EB |
1148 | /* |
1149 | * Since the encryption xattr will always be unique, create it first so | |
1150 | * that it's less likely to end up in an external xattr block and | |
1151 | * prevent its deduplication. | |
1152 | */ | |
1153 | if (encrypt) { | |
1154 | err = fscrypt_inherit_context(dir, inode, handle, true); | |
1155 | if (err) | |
1156 | goto fail_free_drop; | |
1157 | } | |
1158 | ||
1b917ed8 TE |
1159 | if (!(ei->i_flags & EXT4_EA_INODE_FL)) { |
1160 | err = ext4_init_acl(handle, inode, dir); | |
1161 | if (err) | |
1162 | goto fail_free_drop; | |
ac27a0ec | 1163 | |
ad47f953 TE |
1164 | err = ext4_init_security(handle, inode, dir, qstr); |
1165 | if (err) | |
1166 | goto fail_free_drop; | |
1167 | } | |
ac27a0ec | 1168 | |
e2b911c5 | 1169 | if (ext4_has_feature_extents(sb)) { |
e4079a11 | 1170 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1171 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1172 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1173 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1174 | } |
a86c6181 | 1175 | } |
ac27a0ec | 1176 | |
688f869c TT |
1177 | if (ext4_handle_valid(handle)) { |
1178 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
1179 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
1180 | } | |
1181 | ||
8753e88f AK |
1182 | err = ext4_mark_inode_dirty(handle, inode); |
1183 | if (err) { | |
1184 | ext4_std_error(sb, err); | |
1185 | goto fail_free_drop; | |
1186 | } | |
1187 | ||
617ba13b | 1188 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1189 | trace_ext4_allocate_inode(inode, dir, mode); |
3300beda | 1190 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1191 | return ret; |
1192 | ||
1193 | fail_free_drop: | |
63936dda | 1194 | dquot_free_inode(inode); |
ac27a0ec | 1195 | fail_drop: |
6d6b77f1 | 1196 | clear_nlink(inode); |
6b38e842 | 1197 | unlock_new_inode(inode); |
eb9cc7e1 JK |
1198 | out: |
1199 | dquot_drop(inode); | |
1200 | inode->i_flags |= S_NOQUOTA; | |
ac27a0ec | 1201 | iput(inode); |
3300beda | 1202 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1203 | return ERR_PTR(err); |
1204 | } | |
1205 | ||
1206 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1207 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1208 | { |
617ba13b | 1209 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1210 | ext4_group_t block_group; |
ac27a0ec | 1211 | int bit; |
7827a7f6 | 1212 | struct buffer_head *bitmap_bh = NULL; |
ac27a0ec | 1213 | struct inode *inode = NULL; |
7827a7f6 | 1214 | int err = -EFSCORRUPTED; |
ac27a0ec | 1215 | |
7827a7f6 TT |
1216 | if (ino < EXT4_FIRST_INO(sb) || ino > max_ino) |
1217 | goto bad_orphan; | |
ac27a0ec | 1218 | |
617ba13b MC |
1219 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1220 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1221 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
206f6d55 | 1222 | if (IS_ERR(bitmap_bh)) |
e884bce1 | 1223 | return ERR_CAST(bitmap_bh); |
ac27a0ec DK |
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 | ||
8a363970 | 1232 | inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL); |
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 */ | |
bc98a42c | 1362 | if (sb_rdonly(sb)) { |
bfff6873 LC |
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 | ||
50122847 TT |
1395 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) || |
1396 | ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) - | |
1397 | ext4_itable_unused_count(sb, gdp)) < | |
1398 | EXT4_FIRST_INO(sb)))) { | |
1084f252 TT |
1399 | ext4_error(sb, "Something is wrong with group %u: " |
1400 | "used itable blocks: %d; " | |
1401 | "itable unused count: %u", | |
857ac889 LC |
1402 | group, used_blks, |
1403 | ext4_itable_unused_count(sb, gdp)); | |
1404 | ret = 1; | |
33853a0d | 1405 | goto err_out; |
857ac889 LC |
1406 | } |
1407 | ||
bfff6873 LC |
1408 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1409 | num = sbi->s_itb_per_group - used_blks; | |
1410 | ||
1411 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1412 | ret = ext4_journal_get_write_access(handle, | |
1413 | group_desc_bh); | |
1414 | if (ret) | |
1415 | goto err_out; | |
1416 | ||
bfff6873 LC |
1417 | /* |
1418 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1419 | * flag anyway, because obviously, when it is full it does not need | |
1420 | * further zeroing. | |
1421 | */ | |
1422 | if (unlikely(num == 0)) | |
1423 | goto skip_zeroout; | |
1424 | ||
1425 | ext4_debug("going to zero out inode table in group %d\n", | |
1426 | group); | |
a107e5a3 | 1427 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1428 | if (ret < 0) |
1429 | goto err_out; | |
a107e5a3 TT |
1430 | if (barrier) |
1431 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); | |
bfff6873 LC |
1432 | |
1433 | skip_zeroout: | |
1434 | ext4_lock_group(sb, group); | |
1435 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
feb0ab32 | 1436 | ext4_group_desc_csum_set(sb, group, gdp); |
bfff6873 LC |
1437 | ext4_unlock_group(sb, group); |
1438 | ||
1439 | BUFFER_TRACE(group_desc_bh, | |
1440 | "call ext4_handle_dirty_metadata"); | |
1441 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1442 | group_desc_bh); | |
1443 | ||
1444 | err_out: | |
1445 | up_write(&grp->alloc_sem); | |
1446 | ext4_journal_stop(handle); | |
1447 | out: | |
1448 | return ret; | |
1449 | } |