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); | |
86 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
87 | ||
88 | if (buffer_verified(bh)) | |
89 | return 0; | |
90 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) | |
91 | return -EFSCORRUPTED; | |
92 | ||
93 | ext4_lock_group(sb, block_group); | |
dc1b4b71 TT |
94 | if (buffer_verified(bh)) |
95 | goto verified; | |
9008a58e DW |
96 | blk = ext4_inode_bitmap(sb, desc); |
97 | if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh, | |
98 | EXT4_INODES_PER_GROUP(sb) / 8)) { | |
99 | ext4_unlock_group(sb, block_group); | |
100 | ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " | |
101 | "inode_bitmap = %llu", block_group, blk); | |
102 | grp = ext4_get_group_info(sb, block_group); | |
103 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { | |
104 | int count; | |
105 | count = ext4_free_inodes_count(sb, desc); | |
106 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
107 | count); | |
108 | } | |
109 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); | |
110 | return -EFSBADCRC; | |
111 | } | |
112 | set_buffer_verified(bh); | |
dc1b4b71 | 113 | verified: |
9008a58e DW |
114 | ext4_unlock_group(sb, block_group); |
115 | return 0; | |
116 | } | |
117 | ||
ac27a0ec DK |
118 | /* |
119 | * Read the inode allocation bitmap for a given block_group, reading | |
120 | * into the specified slot in the superblock's bitmap cache. | |
121 | * | |
122 | * Return buffer_head of bitmap on success or NULL. | |
123 | */ | |
124 | static struct buffer_head * | |
e29d1cde | 125 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 126 | { |
617ba13b | 127 | struct ext4_group_desc *desc; |
b39430ea | 128 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
ac27a0ec | 129 | struct buffer_head *bh = NULL; |
e29d1cde | 130 | ext4_fsblk_t bitmap_blk; |
9008a58e | 131 | int err; |
ac27a0ec | 132 | |
617ba13b | 133 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 134 | if (!desc) |
9008a58e | 135 | return ERR_PTR(-EFSCORRUPTED); |
bfff6873 | 136 | |
e29d1cde | 137 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
b39430ea TT |
138 | if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || |
139 | (bitmap_blk >= ext4_blocks_count(sbi->s_es))) { | |
140 | ext4_error(sb, "Invalid inode bitmap blk %llu in " | |
141 | "block_group %u", bitmap_blk, block_group); | |
142 | return ERR_PTR(-EFSCORRUPTED); | |
143 | } | |
e29d1cde ES |
144 | bh = sb_getblk(sb, bitmap_blk); |
145 | if (unlikely(!bh)) { | |
12062ddd | 146 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 147 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde | 148 | block_group, bitmap_blk); |
9008a58e | 149 | return ERR_PTR(-EIO); |
e29d1cde | 150 | } |
2ccb5fb9 | 151 | if (bitmap_uptodate(bh)) |
41a246d1 | 152 | goto verify; |
e29d1cde | 153 | |
c806e68f | 154 | lock_buffer(bh); |
2ccb5fb9 AK |
155 | if (bitmap_uptodate(bh)) { |
156 | unlock_buffer(bh); | |
41a246d1 | 157 | goto verify; |
2ccb5fb9 | 158 | } |
bfff6873 | 159 | |
955ce5f5 | 160 | ext4_lock_group(sb, block_group); |
44a4bc97 TT |
161 | if (ext4_has_group_desc_csum(sb) && |
162 | (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) { | |
163 | if (block_group == 0) { | |
164 | ext4_unlock_group(sb, block_group); | |
165 | unlock_buffer(bh); | |
166 | ext4_error(sb, "Inode bitmap for bg 0 marked " | |
167 | "uninitialized"); | |
168 | err = -EFSCORRUPTED; | |
169 | goto out; | |
170 | } | |
bd499f55 TT |
171 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); |
172 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), | |
173 | sb->s_blocksize * 8, bh->b_data); | |
2ccb5fb9 | 174 | set_bitmap_uptodate(bh); |
e29d1cde | 175 | set_buffer_uptodate(bh); |
41a246d1 | 176 | set_buffer_verified(bh); |
955ce5f5 | 177 | ext4_unlock_group(sb, block_group); |
3300beda | 178 | unlock_buffer(bh); |
e29d1cde | 179 | return bh; |
717d50e4 | 180 | } |
955ce5f5 | 181 | ext4_unlock_group(sb, block_group); |
bfff6873 | 182 | |
2ccb5fb9 AK |
183 | if (buffer_uptodate(bh)) { |
184 | /* | |
185 | * if not uninit if bh is uptodate, | |
186 | * bitmap is also uptodate | |
187 | */ | |
188 | set_bitmap_uptodate(bh); | |
189 | unlock_buffer(bh); | |
41a246d1 | 190 | goto verify; |
2ccb5fb9 AK |
191 | } |
192 | /* | |
813e5727 | 193 | * submit the buffer_head for reading |
2ccb5fb9 | 194 | */ |
0562e0ba | 195 | trace_ext4_load_inode_bitmap(sb, block_group); |
813e5727 TT |
196 | bh->b_end_io = ext4_end_bitmap_read; |
197 | get_bh(bh); | |
2a222ca9 | 198 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
813e5727 TT |
199 | wait_on_buffer(bh); |
200 | if (!buffer_uptodate(bh)) { | |
e29d1cde | 201 | put_bh(bh); |
12062ddd | 202 | ext4_error(sb, "Cannot read inode bitmap - " |
813e5727 TT |
203 | "block_group = %u, inode_bitmap = %llu", |
204 | block_group, bitmap_blk); | |
9008a58e | 205 | return ERR_PTR(-EIO); |
e29d1cde | 206 | } |
41a246d1 DW |
207 | |
208 | verify: | |
9008a58e DW |
209 | err = ext4_validate_inode_bitmap(sb, desc, block_group, bh); |
210 | if (err) | |
211 | goto out; | |
ac27a0ec | 212 | return bh; |
9008a58e DW |
213 | out: |
214 | put_bh(bh); | |
215 | return ERR_PTR(err); | |
ac27a0ec DK |
216 | } |
217 | ||
218 | /* | |
219 | * NOTE! When we get the inode, we're the only people | |
220 | * that have access to it, and as such there are no | |
221 | * race conditions we have to worry about. The inode | |
222 | * is not on the hash-lists, and it cannot be reached | |
223 | * through the filesystem because the directory entry | |
224 | * has been deleted earlier. | |
225 | * | |
226 | * HOWEVER: we must make sure that we get no aliases, | |
227 | * which means that we have to call "clear_inode()" | |
228 | * _before_ we mark the inode not in use in the inode | |
229 | * bitmaps. Otherwise a newly created file might use | |
230 | * the same inode number (not actually the same pointer | |
231 | * though), and then we'd have two inodes sharing the | |
232 | * same inode number and space on the harddisk. | |
233 | */ | |
af5bc92d | 234 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 235 | { |
af5bc92d | 236 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
237 | int is_directory; |
238 | unsigned long ino; | |
239 | struct buffer_head *bitmap_bh = NULL; | |
240 | struct buffer_head *bh2; | |
fd2d4291 | 241 | ext4_group_t block_group; |
ac27a0ec | 242 | unsigned long bit; |
af5bc92d TT |
243 | struct ext4_group_desc *gdp; |
244 | struct ext4_super_block *es; | |
617ba13b | 245 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 246 | int fatal = 0, err, count, cleared; |
87a39389 | 247 | struct ext4_group_info *grp; |
ac27a0ec | 248 | |
92b97816 TT |
249 | if (!sb) { |
250 | printk(KERN_ERR "EXT4-fs: %s:%d: inode on " | |
251 | "nonexistent device\n", __func__, __LINE__); | |
ac27a0ec DK |
252 | return; |
253 | } | |
92b97816 TT |
254 | if (atomic_read(&inode->i_count) > 1) { |
255 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", | |
256 | __func__, __LINE__, inode->i_ino, | |
257 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
258 | return; |
259 | } | |
92b97816 TT |
260 | if (inode->i_nlink) { |
261 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", | |
262 | __func__, __LINE__, inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
263 | return; |
264 | } | |
617ba13b | 265 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
266 | |
267 | ino = inode->i_ino; | |
af5bc92d | 268 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 269 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
270 | |
271 | /* | |
272 | * Note: we must free any quota before locking the superblock, | |
273 | * as writing the quota to disk may need the lock as well. | |
274 | */ | |
871a2931 | 275 | dquot_initialize(inode); |
63936dda | 276 | dquot_free_inode(inode); |
9f754758 | 277 | dquot_drop(inode); |
ac27a0ec DK |
278 | |
279 | is_directory = S_ISDIR(inode->i_mode); | |
280 | ||
281 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 282 | ext4_clear_inode(inode); |
ac27a0ec | 283 | |
617ba13b MC |
284 | es = EXT4_SB(sb)->s_es; |
285 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 286 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
287 | goto error_return; |
288 | } | |
617ba13b MC |
289 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
290 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 291 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
87a39389 DW |
292 | /* Don't bother if the inode bitmap is corrupt. */ |
293 | grp = ext4_get_group_info(sb, block_group); | |
9008a58e DW |
294 | if (IS_ERR(bitmap_bh)) { |
295 | fatal = PTR_ERR(bitmap_bh); | |
296 | bitmap_bh = NULL; | |
297 | goto error_return; | |
298 | } | |
299 | if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) { | |
300 | fatal = -EFSCORRUPTED; | |
ac27a0ec | 301 | goto error_return; |
9008a58e | 302 | } |
ac27a0ec DK |
303 | |
304 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 305 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
306 | if (fatal) |
307 | goto error_return; | |
308 | ||
d17413c0 DM |
309 | fatal = -ESRCH; |
310 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
311 | if (gdp) { | |
ac27a0ec | 312 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 313 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
314 | } |
315 | ext4_lock_group(sb, block_group); | |
597d508c | 316 | cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); |
d17413c0 DM |
317 | if (fatal || !cleared) { |
318 | ext4_unlock_group(sb, block_group); | |
319 | goto out; | |
320 | } | |
7d39db14 | 321 | |
d17413c0 DM |
322 | count = ext4_free_inodes_count(sb, gdp) + 1; |
323 | ext4_free_inodes_set(sb, gdp, count); | |
324 | if (is_directory) { | |
325 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
326 | ext4_used_dirs_set(sb, gdp, count); | |
327 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 328 | } |
41a246d1 DW |
329 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, |
330 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 331 | ext4_group_desc_csum_set(sb, block_group, gdp); |
d17413c0 | 332 | ext4_unlock_group(sb, block_group); |
ac27a0ec | 333 | |
d17413c0 DM |
334 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
335 | if (sbi->s_log_groups_per_flex) { | |
336 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 337 | |
d17413c0 DM |
338 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
339 | if (is_directory) | |
340 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 341 | } |
d17413c0 DM |
342 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
343 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
344 | out: | |
345 | if (cleared) { | |
346 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
347 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
348 | if (!fatal) | |
349 | fatal = err; | |
87a39389 | 350 | } else { |
d17413c0 | 351 | ext4_error(sb, "bit already cleared for inode %lu", ino); |
bf40c926 | 352 | if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
e43bb4e6 NJ |
353 | int count; |
354 | count = ext4_free_inodes_count(sb, gdp); | |
355 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
356 | count); | |
357 | } | |
87a39389 DW |
358 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
359 | } | |
d17413c0 | 360 | |
ac27a0ec DK |
361 | error_return: |
362 | brelse(bitmap_bh); | |
617ba13b | 363 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
364 | } |
365 | ||
a4912123 | 366 | struct orlov_stats { |
90ba983f | 367 | __u64 free_clusters; |
a4912123 | 368 | __u32 free_inodes; |
a4912123 TT |
369 | __u32 used_dirs; |
370 | }; | |
371 | ||
372 | /* | |
373 | * Helper function for Orlov's allocator; returns critical information | |
374 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
375 | * is a block group number; otherwise it is flex_bg number. | |
376 | */ | |
1f109d5a TT |
377 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
378 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
379 | { |
380 | struct ext4_group_desc *desc; | |
7d39db14 | 381 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 382 | |
7d39db14 TT |
383 | if (flex_size > 1) { |
384 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
90ba983f | 385 | stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); |
7d39db14 TT |
386 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); |
387 | return; | |
388 | } | |
a4912123 | 389 | |
7d39db14 TT |
390 | desc = ext4_get_group_desc(sb, g, NULL); |
391 | if (desc) { | |
392 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
021b65bb | 393 | stats->free_clusters = ext4_free_group_clusters(sb, desc); |
7d39db14 TT |
394 | stats->used_dirs = ext4_used_dirs_count(sb, desc); |
395 | } else { | |
396 | stats->free_inodes = 0; | |
24aaa8ef | 397 | stats->free_clusters = 0; |
7d39db14 | 398 | stats->used_dirs = 0; |
a4912123 TT |
399 | } |
400 | } | |
401 | ||
ac27a0ec DK |
402 | /* |
403 | * Orlov's allocator for directories. | |
404 | * | |
405 | * We always try to spread first-level directories. | |
406 | * | |
407 | * If there are blockgroups with both free inodes and free blocks counts | |
408 | * not worse than average we return one with smallest directory count. | |
409 | * Otherwise we simply return a random group. | |
410 | * | |
411 | * For the rest rules look so: | |
412 | * | |
413 | * It's OK to put directory into a group unless | |
414 | * it has too many directories already (max_dirs) or | |
415 | * it has too few free inodes left (min_inodes) or | |
416 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 417 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
418 | * conditions we search cyclically through the rest. If none |
419 | * of the groups look good we just look for a group with more | |
420 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
421 | */ |
422 | ||
2aa9fc4c | 423 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
dcca3fec | 424 | ext4_group_t *group, umode_t mode, |
f157a4aa | 425 | const struct qstr *qstr) |
ac27a0ec | 426 | { |
fd2d4291 | 427 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 428 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 429 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 430 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
14c83c9f | 431 | unsigned int freei, avefreei, grp_free; |
24aaa8ef | 432 | ext4_fsblk_t freeb, avefreec; |
ac27a0ec | 433 | unsigned int ndirs; |
a4912123 | 434 | int max_dirs, min_inodes; |
24aaa8ef | 435 | ext4_grpblk_t min_clusters; |
8df9675f | 436 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 437 | struct ext4_group_desc *desc; |
a4912123 TT |
438 | struct orlov_stats stats; |
439 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 440 | struct dx_hash_info hinfo; |
a4912123 | 441 | |
8df9675f | 442 | ngroups = real_ngroups; |
a4912123 | 443 | if (flex_size > 1) { |
8df9675f | 444 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
445 | sbi->s_log_groups_per_flex; |
446 | parent_group >>= sbi->s_log_groups_per_flex; | |
447 | } | |
ac27a0ec DK |
448 | |
449 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
450 | avefreei = freei / ngroups; | |
57042651 TT |
451 | freeb = EXT4_C2B(sbi, |
452 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
24aaa8ef TT |
453 | avefreec = freeb; |
454 | do_div(avefreec, ngroups); | |
ac27a0ec DK |
455 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
456 | ||
a4912123 | 457 | if (S_ISDIR(mode) && |
2b0143b5 | 458 | ((parent == d_inode(sb->s_root)) || |
12e9b892 | 459 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 460 | int best_ndir = inodes_per_group; |
2aa9fc4c | 461 | int ret = -1; |
ac27a0ec | 462 | |
f157a4aa TT |
463 | if (qstr) { |
464 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
465 | hinfo.seed = sbi->s_hash_seed; | |
466 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
467 | grp = hinfo.hash; | |
468 | } else | |
dd1f723b | 469 | grp = prandom_u32(); |
2aa9fc4c | 470 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 471 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
472 | g = (parent_group + i) % ngroups; |
473 | get_orlov_stats(sb, g, flex_size, &stats); | |
474 | if (!stats.free_inodes) | |
ac27a0ec | 475 | continue; |
a4912123 | 476 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 477 | continue; |
a4912123 | 478 | if (stats.free_inodes < avefreei) |
ac27a0ec | 479 | continue; |
24aaa8ef | 480 | if (stats.free_clusters < avefreec) |
ac27a0ec | 481 | continue; |
a4912123 | 482 | grp = g; |
2aa9fc4c | 483 | ret = 0; |
a4912123 TT |
484 | best_ndir = stats.used_dirs; |
485 | } | |
486 | if (ret) | |
487 | goto fallback; | |
488 | found_flex_bg: | |
489 | if (flex_size == 1) { | |
490 | *group = grp; | |
491 | return 0; | |
492 | } | |
493 | ||
494 | /* | |
495 | * We pack inodes at the beginning of the flexgroup's | |
496 | * inode tables. Block allocation decisions will do | |
497 | * something similar, although regular files will | |
498 | * start at 2nd block group of the flexgroup. See | |
499 | * ext4_ext_find_goal() and ext4_find_near(). | |
500 | */ | |
501 | grp *= flex_size; | |
502 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 503 | if (grp+i >= real_ngroups) |
a4912123 TT |
504 | break; |
505 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
506 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
507 | *group = grp+i; | |
508 | return 0; | |
509 | } | |
ac27a0ec | 510 | } |
ac27a0ec DK |
511 | goto fallback; |
512 | } | |
513 | ||
ac27a0ec | 514 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
515 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
516 | if (min_inodes < 1) | |
517 | min_inodes = 1; | |
24aaa8ef | 518 | min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; |
a4912123 TT |
519 | |
520 | /* | |
521 | * Start looking in the flex group where we last allocated an | |
522 | * inode for this parent directory | |
523 | */ | |
524 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
525 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
526 | if (flex_size > 1) | |
527 | parent_group >>= sbi->s_log_groups_per_flex; | |
528 | } | |
ac27a0ec DK |
529 | |
530 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
531 | grp = (parent_group + i) % ngroups; |
532 | get_orlov_stats(sb, grp, flex_size, &stats); | |
533 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 534 | continue; |
a4912123 | 535 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 536 | continue; |
24aaa8ef | 537 | if (stats.free_clusters < min_clusters) |
ac27a0ec | 538 | continue; |
a4912123 | 539 | goto found_flex_bg; |
ac27a0ec DK |
540 | } |
541 | ||
542 | fallback: | |
8df9675f | 543 | ngroups = real_ngroups; |
a4912123 | 544 | avefreei = freei / ngroups; |
b5451f7b | 545 | fallback_retry: |
a4912123 | 546 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 547 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
548 | grp = (parent_group + i) % ngroups; |
549 | desc = ext4_get_group_desc(sb, grp, NULL); | |
bb3d132a DC |
550 | if (desc) { |
551 | grp_free = ext4_free_inodes_count(sb, desc); | |
552 | if (grp_free && grp_free >= avefreei) { | |
553 | *group = grp; | |
554 | return 0; | |
555 | } | |
a4912123 | 556 | } |
ac27a0ec DK |
557 | } |
558 | ||
559 | if (avefreei) { | |
560 | /* | |
561 | * The free-inodes counter is approximate, and for really small | |
562 | * filesystems the above test can fail to find any blockgroups | |
563 | */ | |
564 | avefreei = 0; | |
b5451f7b | 565 | goto fallback_retry; |
ac27a0ec DK |
566 | } |
567 | ||
568 | return -1; | |
569 | } | |
570 | ||
2aa9fc4c | 571 | static int find_group_other(struct super_block *sb, struct inode *parent, |
dcca3fec | 572 | ext4_group_t *group, umode_t mode) |
ac27a0ec | 573 | { |
fd2d4291 | 574 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 575 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 576 | struct ext4_group_desc *desc; |
a4912123 TT |
577 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
578 | ||
579 | /* | |
580 | * Try to place the inode is the same flex group as its | |
581 | * parent. If we can't find space, use the Orlov algorithm to | |
582 | * find another flex group, and store that information in the | |
583 | * parent directory's inode information so that use that flex | |
584 | * group for future allocations. | |
585 | */ | |
586 | if (flex_size > 1) { | |
587 | int retry = 0; | |
588 | ||
589 | try_again: | |
590 | parent_group &= ~(flex_size-1); | |
591 | last = parent_group + flex_size; | |
592 | if (last > ngroups) | |
593 | last = ngroups; | |
594 | for (i = parent_group; i < last; i++) { | |
595 | desc = ext4_get_group_desc(sb, i, NULL); | |
596 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
597 | *group = i; | |
598 | return 0; | |
599 | } | |
600 | } | |
601 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
602 | retry = 1; | |
603 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
604 | goto try_again; | |
605 | } | |
606 | /* | |
607 | * If this didn't work, use the Orlov search algorithm | |
608 | * to find a new flex group; we pass in the mode to | |
609 | * avoid the topdir algorithms. | |
610 | */ | |
611 | *group = parent_group + flex_size; | |
612 | if (*group > ngroups) | |
613 | *group = 0; | |
7dc57615 | 614 | return find_group_orlov(sb, parent, group, mode, NULL); |
a4912123 | 615 | } |
ac27a0ec DK |
616 | |
617 | /* | |
618 | * Try to place the inode in its parent directory | |
619 | */ | |
2aa9fc4c AM |
620 | *group = parent_group; |
621 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 622 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 623 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 624 | return 0; |
ac27a0ec DK |
625 | |
626 | /* | |
627 | * We're going to place this inode in a different blockgroup from its | |
628 | * parent. We want to cause files in a common directory to all land in | |
629 | * the same blockgroup. But we want files which are in a different | |
630 | * directory which shares a blockgroup with our parent to land in a | |
631 | * different blockgroup. | |
632 | * | |
633 | * So add our directory's i_ino into the starting point for the hash. | |
634 | */ | |
2aa9fc4c | 635 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
636 | |
637 | /* | |
638 | * Use a quadratic hash to find a group with a free inode and some free | |
639 | * blocks. | |
640 | */ | |
641 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
642 | *group += i; |
643 | if (*group >= ngroups) | |
644 | *group -= ngroups; | |
645 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 646 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 647 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 648 | return 0; |
ac27a0ec DK |
649 | } |
650 | ||
651 | /* | |
652 | * That failed: try linear search for a free inode, even if that group | |
653 | * has no free blocks. | |
654 | */ | |
2aa9fc4c | 655 | *group = parent_group; |
ac27a0ec | 656 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
657 | if (++*group >= ngroups) |
658 | *group = 0; | |
659 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 660 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 661 | return 0; |
ac27a0ec DK |
662 | } |
663 | ||
664 | return -1; | |
665 | } | |
666 | ||
19883bd9 TT |
667 | /* |
668 | * In no journal mode, if an inode has recently been deleted, we want | |
669 | * to avoid reusing it until we're reasonably sure the inode table | |
670 | * block has been written back to disk. (Yes, these values are | |
671 | * somewhat arbitrary...) | |
672 | */ | |
673 | #define RECENTCY_MIN 5 | |
b5f51573 | 674 | #define RECENTCY_DIRTY 300 |
19883bd9 TT |
675 | |
676 | static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) | |
677 | { | |
678 | struct ext4_group_desc *gdp; | |
679 | struct ext4_inode *raw_inode; | |
680 | struct buffer_head *bh; | |
b5f51573 AD |
681 | int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
682 | int offset, ret = 0; | |
683 | int recentcy = RECENTCY_MIN; | |
684 | u32 dtime, now; | |
19883bd9 TT |
685 | |
686 | gdp = ext4_get_group_desc(sb, group, NULL); | |
687 | if (unlikely(!gdp)) | |
688 | return 0; | |
689 | ||
4f9d956d | 690 | bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) + |
19883bd9 | 691 | (ino / inodes_per_block)); |
4f9d956d | 692 | if (!bh || !buffer_uptodate(bh)) |
19883bd9 TT |
693 | /* |
694 | * If the block is not in the buffer cache, then it | |
695 | * must have been written out. | |
696 | */ | |
697 | goto out; | |
698 | ||
699 | offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
700 | raw_inode = (struct ext4_inode *) (bh->b_data + offset); | |
b5f51573 AD |
701 | |
702 | /* i_dtime is only 32 bits on disk, but we only care about relative | |
703 | * times in the range of a few minutes (i.e. long enough to sync a | |
704 | * recently-deleted inode to disk), so using the low 32 bits of the | |
705 | * clock (a 68 year range) is enough, see time_before32() */ | |
19883bd9 | 706 | dtime = le32_to_cpu(raw_inode->i_dtime); |
b5f51573 | 707 | now = ktime_get_real_seconds(); |
19883bd9 TT |
708 | if (buffer_dirty(bh)) |
709 | recentcy += RECENTCY_DIRTY; | |
710 | ||
b5f51573 AD |
711 | if (dtime && time_before32(dtime, now) && |
712 | time_before32(now, dtime + recentcy)) | |
19883bd9 TT |
713 | ret = 1; |
714 | out: | |
715 | brelse(bh); | |
716 | return ret; | |
717 | } | |
718 | ||
901ed070 WS |
719 | static int find_inode_bit(struct super_block *sb, ext4_group_t group, |
720 | struct buffer_head *bitmap, unsigned long *ino) | |
721 | { | |
722 | next: | |
723 | *ino = ext4_find_next_zero_bit((unsigned long *) | |
724 | bitmap->b_data, | |
725 | EXT4_INODES_PER_GROUP(sb), *ino); | |
726 | if (*ino >= EXT4_INODES_PER_GROUP(sb)) | |
727 | return 0; | |
728 | ||
729 | if ((EXT4_SB(sb)->s_journal == NULL) && | |
730 | recently_deleted(sb, group, *ino)) { | |
731 | *ino = *ino + 1; | |
732 | if (*ino < EXT4_INODES_PER_GROUP(sb)) | |
733 | goto next; | |
734 | return 0; | |
735 | } | |
736 | ||
737 | return 1; | |
738 | } | |
739 | ||
ac27a0ec DK |
740 | /* |
741 | * There are two policies for allocating an inode. If the new inode is | |
742 | * a directory, then a forward search is made for a block group with both | |
743 | * free space and a low directory-to-inode ratio; if that fails, then of | |
744 | * the groups with above-average free space, that group with the fewest | |
745 | * directories already is chosen. | |
746 | * | |
747 | * For other inodes, search forward from the parent directory's block | |
748 | * group to find a free inode. | |
749 | */ | |
1139575a TT |
750 | struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, |
751 | umode_t mode, const struct qstr *qstr, | |
1b917ed8 TE |
752 | __u32 goal, uid_t *owner, __u32 i_flags, |
753 | int handle_type, unsigned int line_no, | |
754 | int nblocks) | |
ac27a0ec DK |
755 | { |
756 | struct super_block *sb; | |
3300beda AK |
757 | struct buffer_head *inode_bitmap_bh = NULL; |
758 | struct buffer_head *group_desc_bh; | |
8df9675f | 759 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 760 | unsigned long ino = 0; |
af5bc92d TT |
761 | struct inode *inode; |
762 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
763 | struct ext4_inode_info *ei; |
764 | struct ext4_sb_info *sbi; | |
a7cdadee | 765 | int ret2, err; |
ac27a0ec | 766 | struct inode *ret; |
2aa9fc4c | 767 | ext4_group_t i; |
772cb7c8 | 768 | ext4_group_t flex_group; |
87a39389 | 769 | struct ext4_group_info *grp; |
e709e9df | 770 | int encrypt = 0; |
ac27a0ec DK |
771 | |
772 | /* Cannot create files in a deleted directory */ | |
773 | if (!dir || !dir->i_nlink) | |
774 | return ERR_PTR(-EPERM); | |
775 | ||
af65207c TE |
776 | sb = dir->i_sb; |
777 | sbi = EXT4_SB(sb); | |
778 | ||
779 | if (unlikely(ext4_forced_shutdown(sbi))) | |
0db1ff22 TT |
780 | return ERR_PTR(-EIO); |
781 | ||
af65207c | 782 | if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) && |
ad47f953 TE |
783 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) && |
784 | !(i_flags & EXT4_EA_INODE_FL)) { | |
a7550b30 | 785 | err = fscrypt_get_encryption_info(dir); |
e709e9df TT |
786 | if (err) |
787 | return ERR_PTR(err); | |
a7550b30 | 788 | if (!fscrypt_has_encryption_key(dir)) |
54475f53 | 789 | return ERR_PTR(-ENOKEY); |
e709e9df TT |
790 | encrypt = 1; |
791 | } | |
792 | ||
af65207c TE |
793 | if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) { |
794 | #ifdef CONFIG_EXT4_FS_POSIX_ACL | |
795 | struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT); | |
796 | ||
0228af23 TT |
797 | if (IS_ERR(p)) |
798 | return ERR_CAST(p); | |
af65207c TE |
799 | if (p) { |
800 | int acl_size = p->a_count * sizeof(ext4_acl_entry); | |
801 | ||
802 | nblocks += (S_ISDIR(mode) ? 2 : 1) * | |
803 | __ext4_xattr_set_credits(sb, NULL /* inode */, | |
804 | NULL /* block_bh */, acl_size, | |
805 | true /* is_create */); | |
806 | posix_acl_release(p); | |
807 | } | |
808 | #endif | |
809 | ||
810 | #ifdef CONFIG_SECURITY | |
811 | { | |
812 | int num_security_xattrs = 1; | |
813 | ||
814 | #ifdef CONFIG_INTEGRITY | |
815 | num_security_xattrs++; | |
816 | #endif | |
817 | /* | |
818 | * We assume that security xattrs are never | |
819 | * more than 1k. In practice they are under | |
820 | * 128 bytes. | |
821 | */ | |
822 | nblocks += num_security_xattrs * | |
823 | __ext4_xattr_set_credits(sb, NULL /* inode */, | |
824 | NULL /* block_bh */, 1024, | |
825 | true /* is_create */); | |
826 | } | |
827 | #endif | |
828 | if (encrypt) | |
829 | nblocks += __ext4_xattr_set_credits(sb, | |
830 | NULL /* inode */, NULL /* block_bh */, | |
831 | FSCRYPT_SET_CONTEXT_MAX_SIZE, | |
832 | true /* is_create */); | |
833 | } | |
834 | ||
8df9675f | 835 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 836 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
837 | inode = new_inode(sb); |
838 | if (!inode) | |
839 | return ERR_PTR(-ENOMEM); | |
617ba13b | 840 | ei = EXT4_I(inode); |
772cb7c8 | 841 | |
eb9cc7e1 | 842 | /* |
b8a07463 | 843 | * Initialize owners and quota early so that we don't have to account |
eb9cc7e1 JK |
844 | * for quota initialization worst case in standard inode creating |
845 | * transaction | |
846 | */ | |
847 | if (owner) { | |
848 | inode->i_mode = mode; | |
849 | i_uid_write(inode, owner[0]); | |
850 | i_gid_write(inode, owner[1]); | |
851 | } else if (test_opt(sb, GRPID)) { | |
852 | inode->i_mode = mode; | |
853 | inode->i_uid = current_fsuid(); | |
854 | inode->i_gid = dir->i_gid; | |
855 | } else | |
856 | inode_init_owner(inode, dir, mode); | |
040cb378 | 857 | |
0b7b7779 | 858 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
859 | ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) |
860 | ei->i_projid = EXT4_I(dir)->i_projid; | |
861 | else | |
862 | ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID); | |
863 | ||
a7cdadee JK |
864 | err = dquot_initialize(inode); |
865 | if (err) | |
866 | goto out; | |
eb9cc7e1 | 867 | |
11013911 AD |
868 | if (!goal) |
869 | goal = sbi->s_inode_goal; | |
870 | ||
e6462869 | 871 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
872 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
873 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
874 | ret2 = 0; | |
875 | goto got_group; | |
876 | } | |
877 | ||
4113c4ca LC |
878 | if (S_ISDIR(mode)) |
879 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); | |
880 | else | |
a4912123 | 881 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 882 | |
772cb7c8 | 883 | got_group: |
a4912123 | 884 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 885 | err = -ENOSPC; |
2aa9fc4c | 886 | if (ret2 == -1) |
ac27a0ec DK |
887 | goto out; |
888 | ||
119c0d44 TT |
889 | /* |
890 | * Normally we will only go through one pass of this loop, | |
891 | * unless we get unlucky and it turns out the group we selected | |
892 | * had its last inode grabbed by someone else. | |
893 | */ | |
11013911 | 894 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
895 | err = -EIO; |
896 | ||
3300beda | 897 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec | 898 | if (!gdp) |
eb9cc7e1 | 899 | goto out; |
ac27a0ec | 900 | |
f2a09af6 YY |
901 | /* |
902 | * Check free inodes count before loading bitmap. | |
903 | */ | |
2fe435d8 WS |
904 | if (ext4_free_inodes_count(sb, gdp) == 0) |
905 | goto next_group; | |
f2a09af6 | 906 | |
87a39389 DW |
907 | grp = ext4_get_group_info(sb, group); |
908 | /* Skip groups with already-known suspicious inode tables */ | |
2fe435d8 WS |
909 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) |
910 | goto next_group; | |
87a39389 | 911 | |
3300beda AK |
912 | brelse(inode_bitmap_bh); |
913 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
87a39389 | 914 | /* Skip groups with suspicious inode tables */ |
9008a58e DW |
915 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || |
916 | IS_ERR(inode_bitmap_bh)) { | |
917 | inode_bitmap_bh = NULL; | |
2fe435d8 | 918 | goto next_group; |
87a39389 | 919 | } |
ac27a0ec | 920 | |
ac27a0ec | 921 | repeat_in_this_group: |
901ed070 WS |
922 | ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); |
923 | if (!ret2) | |
a34eb503 | 924 | goto next_group; |
901ed070 WS |
925 | |
926 | if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) { | |
119c0d44 TT |
927 | ext4_error(sb, "reserved inode found cleared - " |
928 | "inode=%lu", ino + 1); | |
2fe435d8 | 929 | goto next_group; |
119c0d44 | 930 | } |
901ed070 | 931 | |
1139575a TT |
932 | if (!handle) { |
933 | BUG_ON(nblocks <= 0); | |
934 | handle = __ext4_journal_start_sb(dir->i_sb, line_no, | |
5fe2fe89 JK |
935 | handle_type, nblocks, |
936 | 0); | |
1139575a TT |
937 | if (IS_ERR(handle)) { |
938 | err = PTR_ERR(handle); | |
eb9cc7e1 JK |
939 | ext4_std_error(sb, err); |
940 | goto out; | |
1139575a TT |
941 | } |
942 | } | |
ffb5387e ES |
943 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
944 | err = ext4_journal_get_write_access(handle, inode_bitmap_bh); | |
eb9cc7e1 JK |
945 | if (err) { |
946 | ext4_std_error(sb, err); | |
947 | goto out; | |
948 | } | |
119c0d44 TT |
949 | ext4_lock_group(sb, group); |
950 | ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); | |
901ed070 WS |
951 | if (ret2) { |
952 | /* Someone already took the bit. Repeat the search | |
953 | * with lock held. | |
954 | */ | |
955 | ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); | |
956 | if (ret2) { | |
957 | ext4_set_bit(ino, inode_bitmap_bh->b_data); | |
958 | ret2 = 0; | |
959 | } else { | |
960 | ret2 = 1; /* we didn't grab the inode */ | |
961 | } | |
962 | } | |
119c0d44 TT |
963 | ext4_unlock_group(sb, group); |
964 | ino++; /* the inode bitmap is zero-based */ | |
965 | if (!ret2) | |
966 | goto got; /* we grabbed the inode! */ | |
901ed070 | 967 | |
119c0d44 TT |
968 | if (ino < EXT4_INODES_PER_GROUP(sb)) |
969 | goto repeat_in_this_group; | |
a34eb503 TT |
970 | next_group: |
971 | if (++group == ngroups) | |
972 | group = 0; | |
ac27a0ec DK |
973 | } |
974 | err = -ENOSPC; | |
975 | goto out; | |
976 | ||
977 | got: | |
ffb5387e ES |
978 | BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); |
979 | err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); | |
eb9cc7e1 JK |
980 | if (err) { |
981 | ext4_std_error(sb, err); | |
982 | goto out; | |
983 | } | |
ffb5387e | 984 | |
61c219f5 TT |
985 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
986 | err = ext4_journal_get_write_access(handle, group_desc_bh); | |
987 | if (err) { | |
988 | ext4_std_error(sb, err); | |
989 | goto out; | |
990 | } | |
991 | ||
717d50e4 | 992 | /* We may have to initialize the block bitmap if it isn't already */ |
feb0ab32 | 993 | if (ext4_has_group_desc_csum(sb) && |
717d50e4 | 994 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 995 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 996 | |
3300beda | 997 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
9008a58e DW |
998 | if (IS_ERR(block_bitmap_bh)) { |
999 | err = PTR_ERR(block_bitmap_bh); | |
599a9b77 JK |
1000 | goto out; |
1001 | } | |
3300beda AK |
1002 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); |
1003 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 1004 | if (err) { |
3300beda | 1005 | brelse(block_bitmap_bh); |
eb9cc7e1 JK |
1006 | ext4_std_error(sb, err); |
1007 | goto out; | |
717d50e4 AD |
1008 | } |
1009 | ||
fd034a84 TT |
1010 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
1011 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
fd034a84 | 1012 | |
717d50e4 | 1013 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 1014 | ext4_lock_group(sb, group); |
44a4bc97 TT |
1015 | if (ext4_has_group_desc_csum(sb) && |
1016 | (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) { | |
3300beda | 1017 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 1018 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 1019 | ext4_free_clusters_after_init(sb, group, gdp)); |
fa77dcfa | 1020 | ext4_block_bitmap_csum_set(sb, group, gdp, |
79f1ba49 | 1021 | block_bitmap_bh); |
feb0ab32 | 1022 | ext4_group_desc_csum_set(sb, group, gdp); |
717d50e4 | 1023 | } |
955ce5f5 | 1024 | ext4_unlock_group(sb, group); |
aeb1e5d6 | 1025 | brelse(block_bitmap_bh); |
717d50e4 | 1026 | |
eb9cc7e1 JK |
1027 | if (err) { |
1028 | ext4_std_error(sb, err); | |
1029 | goto out; | |
1030 | } | |
717d50e4 | 1031 | } |
119c0d44 | 1032 | |
119c0d44 | 1033 | /* Update the relevant bg descriptor fields */ |
41a246d1 | 1034 | if (ext4_has_group_desc_csum(sb)) { |
119c0d44 TT |
1035 | int free; |
1036 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1037 | ||
1038 | down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ | |
1039 | ext4_lock_group(sb, group); /* while we modify the bg desc */ | |
1040 | free = EXT4_INODES_PER_GROUP(sb) - | |
1041 | ext4_itable_unused_count(sb, gdp); | |
1042 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
1043 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
1044 | free = 0; | |
1045 | } | |
1046 | /* | |
1047 | * Check the relative inode number against the last used | |
1048 | * relative inode number in this group. if it is greater | |
1049 | * we need to update the bg_itable_unused count | |
1050 | */ | |
1051 | if (ino > free) | |
1052 | ext4_itable_unused_set(sb, gdp, | |
1053 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
1054 | up_read(&grp->alloc_sem); | |
6f2e9f0e TM |
1055 | } else { |
1056 | ext4_lock_group(sb, group); | |
119c0d44 | 1057 | } |
6f2e9f0e | 1058 | |
119c0d44 TT |
1059 | ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); |
1060 | if (S_ISDIR(mode)) { | |
1061 | ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); | |
1062 | if (sbi->s_log_groups_per_flex) { | |
1063 | ext4_group_t f = ext4_flex_group(sbi, group); | |
1064 | ||
1065 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); | |
1066 | } | |
1067 | } | |
41a246d1 DW |
1068 | if (ext4_has_group_desc_csum(sb)) { |
1069 | ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, | |
1070 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 1071 | ext4_group_desc_csum_set(sb, group, gdp); |
119c0d44 | 1072 | } |
6f2e9f0e | 1073 | ext4_unlock_group(sb, group); |
119c0d44 | 1074 | |
3300beda AK |
1075 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
1076 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
eb9cc7e1 JK |
1077 | if (err) { |
1078 | ext4_std_error(sb, err); | |
1079 | goto out; | |
1080 | } | |
ac27a0ec DK |
1081 | |
1082 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
1083 | if (S_ISDIR(mode)) | |
1084 | percpu_counter_inc(&sbi->s_dirs_counter); | |
ac27a0ec | 1085 | |
772cb7c8 JS |
1086 | if (sbi->s_log_groups_per_flex) { |
1087 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 1088 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 | 1089 | } |
ac27a0ec | 1090 | |
717d50e4 | 1091 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
1092 | /* This is the optimal IO size (for stat), not the fs block size */ |
1093 | inode->i_blocks = 0; | |
ef7f3835 | 1094 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
eeca7ea1 | 1095 | current_time(inode); |
ac27a0ec DK |
1096 | |
1097 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
1098 | ei->i_dir_start_lookup = 0; | |
1099 | ei->i_disksize = 0; | |
1100 | ||
4af83508 | 1101 | /* Don't inherit extent flag from directory, amongst others. */ |
2dc6b0d4 DG |
1102 | ei->i_flags = |
1103 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
1b917ed8 | 1104 | ei->i_flags |= i_flags; |
ac27a0ec | 1105 | ei->i_file_acl = 0; |
ac27a0ec | 1106 | ei->i_dtime = 0; |
ac27a0ec | 1107 | ei->i_block_group = group; |
a4912123 | 1108 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1109 | |
617ba13b | 1110 | ext4_set_inode_flags(inode); |
ac27a0ec | 1111 | if (IS_DIRSYNC(inode)) |
0390131b | 1112 | ext4_handle_sync(handle); |
6b38e842 | 1113 | if (insert_inode_locked(inode) < 0) { |
acd6ad83 JK |
1114 | /* |
1115 | * Likely a bitmap corruption causing inode to be allocated | |
1116 | * twice. | |
1117 | */ | |
1118 | err = -EIO; | |
eb9cc7e1 JK |
1119 | ext4_error(sb, "failed to insert inode %lu: doubly allocated?", |
1120 | inode->i_ino); | |
1121 | goto out; | |
6b38e842 | 1122 | } |
ac27a0ec DK |
1123 | spin_lock(&sbi->s_next_gen_lock); |
1124 | inode->i_generation = sbi->s_next_generation++; | |
1125 | spin_unlock(&sbi->s_next_gen_lock); | |
1126 | ||
814525f4 | 1127 | /* Precompute checksum seed for inode metadata */ |
9aa5d32b | 1128 | if (ext4_has_metadata_csum(sb)) { |
814525f4 | 1129 | __u32 csum; |
814525f4 DW |
1130 | __le32 inum = cpu_to_le32(inode->i_ino); |
1131 | __le32 gen = cpu_to_le32(inode->i_generation); | |
1132 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
1133 | sizeof(inum)); | |
1134 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
1135 | sizeof(gen)); | |
1136 | } | |
1137 | ||
353eb83c | 1138 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 1139 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 KS |
1140 | |
1141 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
f08225d1 | 1142 | ei->i_inline_off = 0; |
e2b911c5 | 1143 | if (ext4_has_feature_inline_data(sb)) |
f08225d1 | 1144 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
ac27a0ec | 1145 | ret = inode; |
63936dda CH |
1146 | err = dquot_alloc_inode(inode); |
1147 | if (err) | |
ac27a0ec | 1148 | goto fail_drop; |
ac27a0ec | 1149 | |
aa1dca3b EB |
1150 | /* |
1151 | * Since the encryption xattr will always be unique, create it first so | |
1152 | * that it's less likely to end up in an external xattr block and | |
1153 | * prevent its deduplication. | |
1154 | */ | |
1155 | if (encrypt) { | |
1156 | err = fscrypt_inherit_context(dir, inode, handle, true); | |
1157 | if (err) | |
1158 | goto fail_free_drop; | |
1159 | } | |
1160 | ||
1b917ed8 TE |
1161 | if (!(ei->i_flags & EXT4_EA_INODE_FL)) { |
1162 | err = ext4_init_acl(handle, inode, dir); | |
1163 | if (err) | |
1164 | goto fail_free_drop; | |
ac27a0ec | 1165 | |
ad47f953 TE |
1166 | err = ext4_init_security(handle, inode, dir, qstr); |
1167 | if (err) | |
1168 | goto fail_free_drop; | |
1169 | } | |
ac27a0ec | 1170 | |
e2b911c5 | 1171 | if (ext4_has_feature_extents(sb)) { |
e4079a11 | 1172 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1173 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1174 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1175 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1176 | } |
a86c6181 | 1177 | } |
ac27a0ec | 1178 | |
688f869c TT |
1179 | if (ext4_handle_valid(handle)) { |
1180 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
1181 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
1182 | } | |
1183 | ||
8753e88f AK |
1184 | err = ext4_mark_inode_dirty(handle, inode); |
1185 | if (err) { | |
1186 | ext4_std_error(sb, err); | |
1187 | goto fail_free_drop; | |
1188 | } | |
1189 | ||
617ba13b | 1190 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1191 | trace_ext4_allocate_inode(inode, dir, mode); |
3300beda | 1192 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1193 | return ret; |
1194 | ||
1195 | fail_free_drop: | |
63936dda | 1196 | dquot_free_inode(inode); |
ac27a0ec | 1197 | fail_drop: |
6d6b77f1 | 1198 | clear_nlink(inode); |
6b38e842 | 1199 | unlock_new_inode(inode); |
eb9cc7e1 JK |
1200 | out: |
1201 | dquot_drop(inode); | |
1202 | inode->i_flags |= S_NOQUOTA; | |
ac27a0ec | 1203 | iput(inode); |
3300beda | 1204 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1205 | return ERR_PTR(err); |
1206 | } | |
1207 | ||
1208 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1209 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1210 | { |
617ba13b | 1211 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1212 | ext4_group_t block_group; |
ac27a0ec | 1213 | int bit; |
7827a7f6 | 1214 | struct buffer_head *bitmap_bh = NULL; |
ac27a0ec | 1215 | struct inode *inode = NULL; |
7827a7f6 | 1216 | int err = -EFSCORRUPTED; |
ac27a0ec | 1217 | |
7827a7f6 TT |
1218 | if (ino < EXT4_FIRST_INO(sb) || ino > max_ino) |
1219 | goto bad_orphan; | |
ac27a0ec | 1220 | |
617ba13b MC |
1221 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1222 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1223 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
9008a58e | 1224 | if (IS_ERR(bitmap_bh)) { |
7827a7f6 TT |
1225 | ext4_error(sb, "inode bitmap error %ld for orphan %lu", |
1226 | ino, PTR_ERR(bitmap_bh)); | |
1227 | return (struct inode *) bitmap_bh; | |
ac27a0ec DK |
1228 | } |
1229 | ||
1230 | /* Having the inode bit set should be a 100% indicator that this | |
1231 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1232 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1233 | */ | |
1d1fe1ee DH |
1234 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1235 | goto bad_orphan; | |
1236 | ||
1237 | inode = ext4_iget(sb, ino); | |
7827a7f6 TT |
1238 | if (IS_ERR(inode)) { |
1239 | err = PTR_ERR(inode); | |
1240 | ext4_error(sb, "couldn't read orphan inode %lu (err %d)", | |
1241 | ino, err); | |
1242 | return inode; | |
1243 | } | |
1d1fe1ee | 1244 | |
91ef4caf | 1245 | /* |
c9eb13a9 TT |
1246 | * If the orphans has i_nlinks > 0 then it should be able to |
1247 | * be truncated, otherwise it won't be removed from the orphan | |
1248 | * list during processing and an infinite loop will result. | |
1249 | * Similarly, it must not be a bad inode. | |
91ef4caf | 1250 | */ |
c9eb13a9 TT |
1251 | if ((inode->i_nlink && !ext4_can_truncate(inode)) || |
1252 | is_bad_inode(inode)) | |
91ef4caf DG |
1253 | goto bad_orphan; |
1254 | ||
1d1fe1ee DH |
1255 | if (NEXT_ORPHAN(inode) > max_ino) |
1256 | goto bad_orphan; | |
1257 | brelse(bitmap_bh); | |
1258 | return inode; | |
1259 | ||
1d1fe1ee | 1260 | bad_orphan: |
7827a7f6 TT |
1261 | ext4_error(sb, "bad orphan inode %lu", ino); |
1262 | if (bitmap_bh) | |
1263 | printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n", | |
1264 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
1265 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
1d1fe1ee | 1266 | if (inode) { |
7827a7f6 | 1267 | printk(KERN_ERR "is_bad_inode(inode)=%d\n", |
1d1fe1ee | 1268 | is_bad_inode(inode)); |
7827a7f6 | 1269 | printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n", |
1d1fe1ee | 1270 | NEXT_ORPHAN(inode)); |
7827a7f6 TT |
1271 | printk(KERN_ERR "max_ino=%lu\n", max_ino); |
1272 | printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink); | |
ac27a0ec | 1273 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1274 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1275 | inode->i_blocks = 0; |
1276 | iput(inode); | |
ac27a0ec | 1277 | } |
ac27a0ec | 1278 | brelse(bitmap_bh); |
1d1fe1ee | 1279 | return ERR_PTR(err); |
ac27a0ec DK |
1280 | } |
1281 | ||
af5bc92d | 1282 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1283 | { |
1284 | unsigned long desc_count; | |
617ba13b | 1285 | struct ext4_group_desc *gdp; |
8df9675f | 1286 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1287 | #ifdef EXT4FS_DEBUG |
1288 | struct ext4_super_block *es; | |
ac27a0ec DK |
1289 | unsigned long bitmap_count, x; |
1290 | struct buffer_head *bitmap_bh = NULL; | |
1291 | ||
617ba13b | 1292 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1293 | desc_count = 0; |
1294 | bitmap_count = 0; | |
1295 | gdp = NULL; | |
8df9675f | 1296 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1297 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1298 | if (!gdp) |
1299 | continue; | |
560671a0 | 1300 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1301 | brelse(bitmap_bh); |
e29d1cde | 1302 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
9008a58e DW |
1303 | if (IS_ERR(bitmap_bh)) { |
1304 | bitmap_bh = NULL; | |
ac27a0ec | 1305 | continue; |
9008a58e | 1306 | } |
ac27a0ec | 1307 | |
f6fb99ca TT |
1308 | x = ext4_count_free(bitmap_bh->b_data, |
1309 | EXT4_INODES_PER_GROUP(sb) / 8); | |
c549a95d | 1310 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1311 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1312 | bitmap_count += x; |
1313 | } | |
1314 | brelse(bitmap_bh); | |
4776004f TT |
1315 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1316 | "stored = %u, computed = %lu, %lu\n", | |
1317 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1318 | return desc_count; |
1319 | #else | |
1320 | desc_count = 0; | |
8df9675f | 1321 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1322 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1323 | if (!gdp) |
1324 | continue; | |
560671a0 | 1325 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1326 | cond_resched(); |
1327 | } | |
1328 | return desc_count; | |
1329 | #endif | |
1330 | } | |
1331 | ||
1332 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1333 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1334 | { |
1335 | unsigned long count = 0; | |
8df9675f | 1336 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1337 | |
8df9675f | 1338 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1339 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1340 | if (!gdp) |
1341 | continue; | |
560671a0 | 1342 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1343 | } |
1344 | return count; | |
1345 | } | |
bfff6873 LC |
1346 | |
1347 | /* | |
1348 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1349 | * inode table. Must be called without any spinlock held. The only place | |
1350 | * where it is called from on active part of filesystem is ext4lazyinit | |
1351 | * thread, so we do not need any special locks, however we have to prevent | |
1352 | * inode allocation from the current group, so we take alloc_sem lock, to | |
119c0d44 | 1353 | * block ext4_new_inode() until we are finished. |
bfff6873 | 1354 | */ |
e0cbee3e | 1355 | int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, |
bfff6873 LC |
1356 | int barrier) |
1357 | { | |
1358 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1359 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1360 | struct ext4_group_desc *gdp = NULL; | |
1361 | struct buffer_head *group_desc_bh; | |
1362 | handle_t *handle; | |
1363 | ext4_fsblk_t blk; | |
1364 | int num, ret = 0, used_blks = 0; | |
bfff6873 LC |
1365 | |
1366 | /* This should not happen, but just to be sure check this */ | |
bc98a42c | 1367 | if (sb_rdonly(sb)) { |
bfff6873 LC |
1368 | ret = 1; |
1369 | goto out; | |
1370 | } | |
1371 | ||
1372 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1373 | if (!gdp) | |
1374 | goto out; | |
1375 | ||
1376 | /* | |
1377 | * We do not need to lock this, because we are the only one | |
1378 | * handling this flag. | |
1379 | */ | |
1380 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1381 | goto out; | |
1382 | ||
9924a92a | 1383 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
bfff6873 LC |
1384 | if (IS_ERR(handle)) { |
1385 | ret = PTR_ERR(handle); | |
1386 | goto out; | |
1387 | } | |
1388 | ||
1389 | down_write(&grp->alloc_sem); | |
1390 | /* | |
1391 | * If inode bitmap was already initialized there may be some | |
1392 | * used inodes so we need to skip blocks with used inodes in | |
1393 | * inode table. | |
1394 | */ | |
1395 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1396 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1397 | ext4_itable_unused_count(sb, gdp)), | |
1398 | sbi->s_inodes_per_block); | |
1399 | ||
f547aa20 TT |
1400 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) || |
1401 | ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) - | |
1402 | ext4_itable_unused_count(sb, gdp)) < | |
1403 | EXT4_FIRST_INO(sb)))) { | |
1084f252 TT |
1404 | ext4_error(sb, "Something is wrong with group %u: " |
1405 | "used itable blocks: %d; " | |
1406 | "itable unused count: %u", | |
857ac889 LC |
1407 | group, used_blks, |
1408 | ext4_itable_unused_count(sb, gdp)); | |
1409 | ret = 1; | |
33853a0d | 1410 | goto err_out; |
857ac889 LC |
1411 | } |
1412 | ||
bfff6873 LC |
1413 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1414 | num = sbi->s_itb_per_group - used_blks; | |
1415 | ||
1416 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1417 | ret = ext4_journal_get_write_access(handle, | |
1418 | group_desc_bh); | |
1419 | if (ret) | |
1420 | goto err_out; | |
1421 | ||
bfff6873 LC |
1422 | /* |
1423 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1424 | * flag anyway, because obviously, when it is full it does not need | |
1425 | * further zeroing. | |
1426 | */ | |
1427 | if (unlikely(num == 0)) | |
1428 | goto skip_zeroout; | |
1429 | ||
1430 | ext4_debug("going to zero out inode table in group %d\n", | |
1431 | group); | |
a107e5a3 | 1432 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1433 | if (ret < 0) |
1434 | goto err_out; | |
a107e5a3 TT |
1435 | if (barrier) |
1436 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); | |
bfff6873 LC |
1437 | |
1438 | skip_zeroout: | |
1439 | ext4_lock_group(sb, group); | |
1440 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
feb0ab32 | 1441 | ext4_group_desc_csum_set(sb, group, gdp); |
bfff6873 LC |
1442 | ext4_unlock_group(sb, group); |
1443 | ||
1444 | BUFFER_TRACE(group_desc_bh, | |
1445 | "call ext4_handle_dirty_metadata"); | |
1446 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1447 | group_desc_bh); | |
1448 | ||
1449 | err_out: | |
1450 | up_write(&grp->alloc_sem); | |
1451 | ext4_journal_stop(handle); | |
1452 | out: | |
1453 | return ret; | |
1454 | } |