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, | |
46f870d6 TT |
97 | EXT4_INODES_PER_GROUP(sb) / 8) || |
98 | ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) { | |
9008a58e DW |
99 | ext4_unlock_group(sb, block_group); |
100 | ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " | |
101 | "inode_bitmap = %llu", block_group, blk); | |
db79e6d1 WS |
102 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
103 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
9008a58e DW |
104 | return -EFSBADCRC; |
105 | } | |
106 | set_buffer_verified(bh); | |
8d5a803c | 107 | verified: |
9008a58e DW |
108 | ext4_unlock_group(sb, block_group); |
109 | return 0; | |
110 | } | |
111 | ||
ac27a0ec DK |
112 | /* |
113 | * Read the inode allocation bitmap for a given block_group, reading | |
114 | * into the specified slot in the superblock's bitmap cache. | |
115 | * | |
9033783c | 116 | * Return buffer_head of bitmap on success, or an ERR_PTR on error. |
ac27a0ec DK |
117 | */ |
118 | static struct buffer_head * | |
e29d1cde | 119 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 120 | { |
617ba13b | 121 | struct ext4_group_desc *desc; |
7dac4a17 | 122 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
ac27a0ec | 123 | struct buffer_head *bh = NULL; |
e29d1cde | 124 | ext4_fsblk_t bitmap_blk; |
9008a58e | 125 | int err; |
ac27a0ec | 126 | |
617ba13b | 127 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 128 | if (!desc) |
9008a58e | 129 | return ERR_PTR(-EFSCORRUPTED); |
bfff6873 | 130 | |
e29d1cde | 131 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
7dac4a17 TT |
132 | if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || |
133 | (bitmap_blk >= ext4_blocks_count(sbi->s_es))) { | |
134 | ext4_error(sb, "Invalid inode bitmap blk %llu in " | |
135 | "block_group %u", bitmap_blk, block_group); | |
206f6d55 WS |
136 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
137 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
7dac4a17 TT |
138 | return ERR_PTR(-EFSCORRUPTED); |
139 | } | |
e29d1cde ES |
140 | bh = sb_getblk(sb, bitmap_blk); |
141 | if (unlikely(!bh)) { | |
5ef2a699 WS |
142 | ext4_warning(sb, "Cannot read inode bitmap - " |
143 | "block_group = %u, inode_bitmap = %llu", | |
144 | block_group, bitmap_blk); | |
0db9fdeb | 145 | return ERR_PTR(-ENOMEM); |
e29d1cde | 146 | } |
2ccb5fb9 | 147 | if (bitmap_uptodate(bh)) |
41a246d1 | 148 | goto verify; |
e29d1cde | 149 | |
c806e68f | 150 | lock_buffer(bh); |
2ccb5fb9 AK |
151 | if (bitmap_uptodate(bh)) { |
152 | unlock_buffer(bh); | |
41a246d1 | 153 | goto verify; |
2ccb5fb9 | 154 | } |
bfff6873 | 155 | |
955ce5f5 | 156 | ext4_lock_group(sb, block_group); |
8844618d TT |
157 | if (ext4_has_group_desc_csum(sb) && |
158 | (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) { | |
159 | if (block_group == 0) { | |
160 | ext4_unlock_group(sb, block_group); | |
161 | unlock_buffer(bh); | |
162 | ext4_error(sb, "Inode bitmap for bg 0 marked " | |
163 | "uninitialized"); | |
164 | err = -EFSCORRUPTED; | |
165 | goto out; | |
166 | } | |
044e6e3d TT |
167 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); |
168 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), | |
169 | sb->s_blocksize * 8, bh->b_data); | |
2ccb5fb9 | 170 | set_bitmap_uptodate(bh); |
e29d1cde | 171 | set_buffer_uptodate(bh); |
41a246d1 | 172 | set_buffer_verified(bh); |
955ce5f5 | 173 | ext4_unlock_group(sb, block_group); |
3300beda | 174 | unlock_buffer(bh); |
e29d1cde | 175 | return bh; |
717d50e4 | 176 | } |
955ce5f5 | 177 | ext4_unlock_group(sb, block_group); |
bfff6873 | 178 | |
2ccb5fb9 AK |
179 | if (buffer_uptodate(bh)) { |
180 | /* | |
181 | * if not uninit if bh is uptodate, | |
182 | * bitmap is also uptodate | |
183 | */ | |
184 | set_bitmap_uptodate(bh); | |
185 | unlock_buffer(bh); | |
41a246d1 | 186 | goto verify; |
2ccb5fb9 AK |
187 | } |
188 | /* | |
813e5727 | 189 | * submit the buffer_head for reading |
2ccb5fb9 | 190 | */ |
0562e0ba | 191 | trace_ext4_load_inode_bitmap(sb, block_group); |
813e5727 TT |
192 | bh->b_end_io = ext4_end_bitmap_read; |
193 | get_bh(bh); | |
2a222ca9 | 194 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
813e5727 | 195 | wait_on_buffer(bh); |
46f870d6 | 196 | ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO); |
813e5727 | 197 | if (!buffer_uptodate(bh)) { |
e29d1cde | 198 | put_bh(bh); |
54d3adbc TT |
199 | ext4_error_err(sb, EIO, "Cannot read inode bitmap - " |
200 | "block_group = %u, inode_bitmap = %llu", | |
201 | block_group, bitmap_blk); | |
206f6d55 WS |
202 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
203 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
9008a58e | 204 | return ERR_PTR(-EIO); |
e29d1cde | 205 | } |
41a246d1 DW |
206 | |
207 | verify: | |
9008a58e DW |
208 | err = ext4_validate_inode_bitmap(sb, desc, block_group, bh); |
209 | if (err) | |
210 | goto out; | |
ac27a0ec | 211 | return bh; |
9008a58e DW |
212 | out: |
213 | put_bh(bh); | |
214 | return ERR_PTR(err); | |
ac27a0ec DK |
215 | } |
216 | ||
217 | /* | |
218 | * NOTE! When we get the inode, we're the only people | |
219 | * that have access to it, and as such there are no | |
220 | * race conditions we have to worry about. The inode | |
221 | * is not on the hash-lists, and it cannot be reached | |
222 | * through the filesystem because the directory entry | |
223 | * has been deleted earlier. | |
224 | * | |
225 | * HOWEVER: we must make sure that we get no aliases, | |
226 | * which means that we have to call "clear_inode()" | |
227 | * _before_ we mark the inode not in use in the inode | |
228 | * bitmaps. Otherwise a newly created file might use | |
229 | * the same inode number (not actually the same pointer | |
230 | * though), and then we'd have two inodes sharing the | |
231 | * same inode number and space on the harddisk. | |
232 | */ | |
af5bc92d | 233 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 234 | { |
af5bc92d | 235 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
236 | int is_directory; |
237 | unsigned long ino; | |
238 | struct buffer_head *bitmap_bh = NULL; | |
239 | struct buffer_head *bh2; | |
fd2d4291 | 240 | ext4_group_t block_group; |
ac27a0ec | 241 | unsigned long bit; |
af5bc92d TT |
242 | struct ext4_group_desc *gdp; |
243 | struct ext4_super_block *es; | |
617ba13b | 244 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 245 | int fatal = 0, err, count, cleared; |
87a39389 | 246 | struct ext4_group_info *grp; |
ac27a0ec | 247 | |
92b97816 TT |
248 | if (!sb) { |
249 | printk(KERN_ERR "EXT4-fs: %s:%d: inode on " | |
250 | "nonexistent device\n", __func__, __LINE__); | |
ac27a0ec DK |
251 | return; |
252 | } | |
92b97816 TT |
253 | if (atomic_read(&inode->i_count) > 1) { |
254 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", | |
255 | __func__, __LINE__, inode->i_ino, | |
256 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
257 | return; |
258 | } | |
92b97816 TT |
259 | if (inode->i_nlink) { |
260 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", | |
261 | __func__, __LINE__, inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
262 | return; |
263 | } | |
617ba13b | 264 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
265 | |
266 | ino = inode->i_ino; | |
af5bc92d | 267 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 268 | trace_ext4_free_inode(inode); |
ac27a0ec | 269 | |
871a2931 | 270 | dquot_initialize(inode); |
63936dda | 271 | dquot_free_inode(inode); |
ac27a0ec DK |
272 | |
273 | is_directory = S_ISDIR(inode->i_mode); | |
274 | ||
275 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 276 | ext4_clear_inode(inode); |
ac27a0ec | 277 | |
49598e04 | 278 | es = sbi->s_es; |
617ba13b | 279 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { |
12062ddd | 280 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
281 | goto error_return; |
282 | } | |
617ba13b MC |
283 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
284 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 285 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
87a39389 DW |
286 | /* Don't bother if the inode bitmap is corrupt. */ |
287 | grp = ext4_get_group_info(sb, block_group); | |
9008a58e DW |
288 | if (IS_ERR(bitmap_bh)) { |
289 | fatal = PTR_ERR(bitmap_bh); | |
290 | bitmap_bh = NULL; | |
291 | goto error_return; | |
292 | } | |
293 | if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) { | |
294 | fatal = -EFSCORRUPTED; | |
ac27a0ec | 295 | goto error_return; |
9008a58e | 296 | } |
ac27a0ec DK |
297 | |
298 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 299 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
300 | if (fatal) |
301 | goto error_return; | |
302 | ||
d17413c0 DM |
303 | fatal = -ESRCH; |
304 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
305 | if (gdp) { | |
ac27a0ec | 306 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 307 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
308 | } |
309 | ext4_lock_group(sb, block_group); | |
597d508c | 310 | cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); |
d17413c0 DM |
311 | if (fatal || !cleared) { |
312 | ext4_unlock_group(sb, block_group); | |
313 | goto out; | |
314 | } | |
7d39db14 | 315 | |
d17413c0 DM |
316 | count = ext4_free_inodes_count(sb, gdp) + 1; |
317 | ext4_free_inodes_set(sb, gdp, count); | |
318 | if (is_directory) { | |
319 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
320 | ext4_used_dirs_set(sb, gdp, count); | |
321 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 322 | } |
41a246d1 DW |
323 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, |
324 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 325 | ext4_group_desc_csum_set(sb, block_group, gdp); |
d17413c0 | 326 | ext4_unlock_group(sb, block_group); |
ac27a0ec | 327 | |
d17413c0 DM |
328 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
329 | if (sbi->s_log_groups_per_flex) { | |
7c990728 | 330 | struct flex_groups *fg; |
9f24e420 | 331 | |
7c990728 SJS |
332 | fg = sbi_array_rcu_deref(sbi, s_flex_groups, |
333 | ext4_flex_group(sbi, block_group)); | |
334 | atomic_inc(&fg->free_inodes); | |
d17413c0 | 335 | if (is_directory) |
7c990728 | 336 | atomic_dec(&fg->used_dirs); |
ac27a0ec | 337 | } |
d17413c0 DM |
338 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
339 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
340 | out: | |
341 | if (cleared) { | |
342 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
343 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
344 | if (!fatal) | |
345 | fatal = err; | |
87a39389 | 346 | } else { |
d17413c0 | 347 | ext4_error(sb, "bit already cleared for inode %lu", ino); |
db79e6d1 WS |
348 | ext4_mark_group_bitmap_corrupted(sb, block_group, |
349 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
87a39389 | 350 | } |
d17413c0 | 351 | |
ac27a0ec DK |
352 | error_return: |
353 | brelse(bitmap_bh); | |
617ba13b | 354 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
355 | } |
356 | ||
a4912123 | 357 | struct orlov_stats { |
90ba983f | 358 | __u64 free_clusters; |
a4912123 | 359 | __u32 free_inodes; |
a4912123 TT |
360 | __u32 used_dirs; |
361 | }; | |
362 | ||
363 | /* | |
364 | * Helper function for Orlov's allocator; returns critical information | |
365 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
366 | * is a block group number; otherwise it is flex_bg number. | |
367 | */ | |
1f109d5a TT |
368 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
369 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
370 | { |
371 | struct ext4_group_desc *desc; | |
a4912123 | 372 | |
7d39db14 | 373 | if (flex_size > 1) { |
7c990728 SJS |
374 | struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb), |
375 | s_flex_groups, g); | |
376 | stats->free_inodes = atomic_read(&fg->free_inodes); | |
377 | stats->free_clusters = atomic64_read(&fg->free_clusters); | |
378 | stats->used_dirs = atomic_read(&fg->used_dirs); | |
7d39db14 TT |
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 | */ | |
a17a9d93 | 665 | #define RECENTCY_MIN 60 |
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 | { | |
d05466b2 JK |
714 | bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL; |
715 | unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb); | |
716 | ||
901ed070 WS |
717 | next: |
718 | *ino = ext4_find_next_zero_bit((unsigned long *) | |
719 | bitmap->b_data, | |
720 | EXT4_INODES_PER_GROUP(sb), *ino); | |
721 | if (*ino >= EXT4_INODES_PER_GROUP(sb)) | |
d05466b2 | 722 | goto not_found; |
901ed070 | 723 | |
d05466b2 JK |
724 | if (check_recently_deleted && recently_deleted(sb, group, *ino)) { |
725 | recently_deleted_ino = *ino; | |
901ed070 WS |
726 | *ino = *ino + 1; |
727 | if (*ino < EXT4_INODES_PER_GROUP(sb)) | |
728 | goto next; | |
d05466b2 | 729 | goto not_found; |
901ed070 | 730 | } |
d05466b2 JK |
731 | return 1; |
732 | not_found: | |
733 | if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb)) | |
734 | return 0; | |
735 | /* | |
736 | * Not reusing recently deleted inodes is mostly a preference. We don't | |
737 | * want to report ENOSPC or skew allocation patterns because of that. | |
738 | * So return even recently deleted inode if we could find better in the | |
739 | * given range. | |
740 | */ | |
741 | *ino = recently_deleted_ino; | |
901ed070 WS |
742 | return 1; |
743 | } | |
744 | ||
177cc0e7 EB |
745 | static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode, |
746 | bool encrypt) | |
747 | { | |
748 | struct super_block *sb = dir->i_sb; | |
749 | int nblocks = 0; | |
750 | #ifdef CONFIG_EXT4_FS_POSIX_ACL | |
751 | struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT); | |
752 | ||
753 | if (IS_ERR(p)) | |
754 | return PTR_ERR(p); | |
755 | if (p) { | |
756 | int acl_size = p->a_count * sizeof(ext4_acl_entry); | |
757 | ||
758 | nblocks += (S_ISDIR(mode) ? 2 : 1) * | |
759 | __ext4_xattr_set_credits(sb, NULL /* inode */, | |
760 | NULL /* block_bh */, acl_size, | |
761 | true /* is_create */); | |
762 | posix_acl_release(p); | |
763 | } | |
764 | #endif | |
765 | ||
766 | #ifdef CONFIG_SECURITY | |
767 | { | |
768 | int num_security_xattrs = 1; | |
769 | ||
770 | #ifdef CONFIG_INTEGRITY | |
771 | num_security_xattrs++; | |
772 | #endif | |
773 | /* | |
774 | * We assume that security xattrs are never more than 1k. | |
775 | * In practice they are under 128 bytes. | |
776 | */ | |
777 | nblocks += num_security_xattrs * | |
778 | __ext4_xattr_set_credits(sb, NULL /* inode */, | |
779 | NULL /* block_bh */, 1024, | |
780 | true /* is_create */); | |
781 | } | |
782 | #endif | |
783 | if (encrypt) | |
784 | nblocks += __ext4_xattr_set_credits(sb, | |
785 | NULL /* inode */, | |
786 | NULL /* block_bh */, | |
787 | FSCRYPT_SET_CONTEXT_MAX_SIZE, | |
788 | true /* is_create */); | |
789 | return nblocks; | |
790 | } | |
791 | ||
ac27a0ec DK |
792 | /* |
793 | * There are two policies for allocating an inode. If the new inode is | |
794 | * a directory, then a forward search is made for a block group with both | |
795 | * free space and a low directory-to-inode ratio; if that fails, then of | |
796 | * the groups with above-average free space, that group with the fewest | |
797 | * directories already is chosen. | |
798 | * | |
799 | * For other inodes, search forward from the parent directory's block | |
800 | * group to find a free inode. | |
801 | */ | |
1139575a TT |
802 | struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, |
803 | umode_t mode, const struct qstr *qstr, | |
1b917ed8 TE |
804 | __u32 goal, uid_t *owner, __u32 i_flags, |
805 | int handle_type, unsigned int line_no, | |
806 | int nblocks) | |
ac27a0ec DK |
807 | { |
808 | struct super_block *sb; | |
3300beda AK |
809 | struct buffer_head *inode_bitmap_bh = NULL; |
810 | struct buffer_head *group_desc_bh; | |
8df9675f | 811 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 812 | unsigned long ino = 0; |
af5bc92d TT |
813 | struct inode *inode; |
814 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
815 | struct ext4_inode_info *ei; |
816 | struct ext4_sb_info *sbi; | |
a7cdadee | 817 | int ret2, err; |
ac27a0ec | 818 | struct inode *ret; |
2aa9fc4c | 819 | ext4_group_t i; |
772cb7c8 | 820 | ext4_group_t flex_group; |
87a39389 | 821 | struct ext4_group_info *grp; |
e709e9df | 822 | int encrypt = 0; |
ac27a0ec DK |
823 | |
824 | /* Cannot create files in a deleted directory */ | |
825 | if (!dir || !dir->i_nlink) | |
826 | return ERR_PTR(-EPERM); | |
827 | ||
af65207c TE |
828 | sb = dir->i_sb; |
829 | sbi = EXT4_SB(sb); | |
830 | ||
831 | if (unlikely(ext4_forced_shutdown(sbi))) | |
0db1ff22 TT |
832 | return ERR_PTR(-EIO); |
833 | ||
592ddec7 | 834 | if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) && |
ad47f953 TE |
835 | (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) && |
836 | !(i_flags & EXT4_EA_INODE_FL)) { | |
a7550b30 | 837 | err = fscrypt_get_encryption_info(dir); |
e709e9df TT |
838 | if (err) |
839 | return ERR_PTR(err); | |
a7550b30 | 840 | if (!fscrypt_has_encryption_key(dir)) |
54475f53 | 841 | return ERR_PTR(-ENOKEY); |
e709e9df TT |
842 | encrypt = 1; |
843 | } | |
844 | ||
af65207c | 845 | if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) { |
177cc0e7 EB |
846 | ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt); |
847 | if (ret2 < 0) | |
848 | return ERR_PTR(ret2); | |
849 | nblocks += ret2; | |
af65207c TE |
850 | } |
851 | ||
8df9675f | 852 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 853 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
854 | inode = new_inode(sb); |
855 | if (!inode) | |
856 | return ERR_PTR(-ENOMEM); | |
617ba13b | 857 | ei = EXT4_I(inode); |
772cb7c8 | 858 | |
eb9cc7e1 | 859 | /* |
b8a07463 | 860 | * Initialize owners and quota early so that we don't have to account |
eb9cc7e1 JK |
861 | * for quota initialization worst case in standard inode creating |
862 | * transaction | |
863 | */ | |
864 | if (owner) { | |
865 | inode->i_mode = mode; | |
866 | i_uid_write(inode, owner[0]); | |
867 | i_gid_write(inode, owner[1]); | |
868 | } else if (test_opt(sb, GRPID)) { | |
869 | inode->i_mode = mode; | |
870 | inode->i_uid = current_fsuid(); | |
871 | inode->i_gid = dir->i_gid; | |
872 | } else | |
873 | inode_init_owner(inode, dir, mode); | |
040cb378 | 874 | |
0b7b7779 | 875 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
876 | ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) |
877 | ei->i_projid = EXT4_I(dir)->i_projid; | |
878 | else | |
879 | ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID); | |
880 | ||
a7cdadee JK |
881 | err = dquot_initialize(inode); |
882 | if (err) | |
883 | goto out; | |
eb9cc7e1 | 884 | |
11013911 AD |
885 | if (!goal) |
886 | goal = sbi->s_inode_goal; | |
887 | ||
e6462869 | 888 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
889 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
890 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
891 | ret2 = 0; | |
892 | goto got_group; | |
893 | } | |
894 | ||
4113c4ca LC |
895 | if (S_ISDIR(mode)) |
896 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); | |
897 | else | |
a4912123 | 898 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 899 | |
772cb7c8 | 900 | got_group: |
a4912123 | 901 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 902 | err = -ENOSPC; |
2aa9fc4c | 903 | if (ret2 == -1) |
ac27a0ec DK |
904 | goto out; |
905 | ||
119c0d44 TT |
906 | /* |
907 | * Normally we will only go through one pass of this loop, | |
908 | * unless we get unlucky and it turns out the group we selected | |
909 | * had its last inode grabbed by someone else. | |
910 | */ | |
11013911 | 911 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
912 | err = -EIO; |
913 | ||
3300beda | 914 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec | 915 | if (!gdp) |
eb9cc7e1 | 916 | goto out; |
ac27a0ec | 917 | |
f2a09af6 YY |
918 | /* |
919 | * Check free inodes count before loading bitmap. | |
920 | */ | |
2fe435d8 WS |
921 | if (ext4_free_inodes_count(sb, gdp) == 0) |
922 | goto next_group; | |
f2a09af6 | 923 | |
87a39389 DW |
924 | grp = ext4_get_group_info(sb, group); |
925 | /* Skip groups with already-known suspicious inode tables */ | |
2fe435d8 WS |
926 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) |
927 | goto next_group; | |
87a39389 | 928 | |
3300beda AK |
929 | brelse(inode_bitmap_bh); |
930 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
87a39389 | 931 | /* Skip groups with suspicious inode tables */ |
9008a58e DW |
932 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || |
933 | IS_ERR(inode_bitmap_bh)) { | |
934 | inode_bitmap_bh = NULL; | |
2fe435d8 | 935 | goto next_group; |
87a39389 | 936 | } |
ac27a0ec | 937 | |
ac27a0ec | 938 | repeat_in_this_group: |
901ed070 WS |
939 | ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); |
940 | if (!ret2) | |
a34eb503 | 941 | goto next_group; |
901ed070 WS |
942 | |
943 | if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) { | |
119c0d44 TT |
944 | ext4_error(sb, "reserved inode found cleared - " |
945 | "inode=%lu", ino + 1); | |
206f6d55 WS |
946 | ext4_mark_group_bitmap_corrupted(sb, group, |
947 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
2fe435d8 | 948 | goto next_group; |
119c0d44 | 949 | } |
901ed070 | 950 | |
1139575a TT |
951 | if (!handle) { |
952 | BUG_ON(nblocks <= 0); | |
953 | handle = __ext4_journal_start_sb(dir->i_sb, line_no, | |
a70fd5ac | 954 | handle_type, nblocks, 0, |
955 | ext4_trans_default_revoke_credits(sb)); | |
1139575a TT |
956 | if (IS_ERR(handle)) { |
957 | err = PTR_ERR(handle); | |
eb9cc7e1 JK |
958 | ext4_std_error(sb, err); |
959 | goto out; | |
1139575a TT |
960 | } |
961 | } | |
ffb5387e ES |
962 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
963 | err = ext4_journal_get_write_access(handle, inode_bitmap_bh); | |
eb9cc7e1 JK |
964 | if (err) { |
965 | ext4_std_error(sb, err); | |
966 | goto out; | |
967 | } | |
119c0d44 TT |
968 | ext4_lock_group(sb, group); |
969 | ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); | |
901ed070 WS |
970 | if (ret2) { |
971 | /* Someone already took the bit. Repeat the search | |
972 | * with lock held. | |
973 | */ | |
974 | ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); | |
975 | if (ret2) { | |
976 | ext4_set_bit(ino, inode_bitmap_bh->b_data); | |
977 | ret2 = 0; | |
978 | } else { | |
979 | ret2 = 1; /* we didn't grab the inode */ | |
980 | } | |
981 | } | |
119c0d44 TT |
982 | ext4_unlock_group(sb, group); |
983 | ino++; /* the inode bitmap is zero-based */ | |
984 | if (!ret2) | |
985 | goto got; /* we grabbed the inode! */ | |
901ed070 | 986 | |
119c0d44 TT |
987 | if (ino < EXT4_INODES_PER_GROUP(sb)) |
988 | goto repeat_in_this_group; | |
a34eb503 TT |
989 | next_group: |
990 | if (++group == ngroups) | |
991 | group = 0; | |
ac27a0ec DK |
992 | } |
993 | err = -ENOSPC; | |
994 | goto out; | |
995 | ||
996 | got: | |
ffb5387e ES |
997 | BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); |
998 | err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); | |
eb9cc7e1 JK |
999 | if (err) { |
1000 | ext4_std_error(sb, err); | |
1001 | goto out; | |
1002 | } | |
ffb5387e | 1003 | |
61c219f5 TT |
1004 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
1005 | err = ext4_journal_get_write_access(handle, group_desc_bh); | |
1006 | if (err) { | |
1007 | ext4_std_error(sb, err); | |
1008 | goto out; | |
1009 | } | |
1010 | ||
717d50e4 | 1011 | /* We may have to initialize the block bitmap if it isn't already */ |
feb0ab32 | 1012 | if (ext4_has_group_desc_csum(sb) && |
717d50e4 | 1013 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 1014 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 1015 | |
3300beda | 1016 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
9008a58e DW |
1017 | if (IS_ERR(block_bitmap_bh)) { |
1018 | err = PTR_ERR(block_bitmap_bh); | |
599a9b77 JK |
1019 | goto out; |
1020 | } | |
3300beda AK |
1021 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); |
1022 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 1023 | if (err) { |
3300beda | 1024 | brelse(block_bitmap_bh); |
eb9cc7e1 JK |
1025 | ext4_std_error(sb, err); |
1026 | goto out; | |
717d50e4 AD |
1027 | } |
1028 | ||
fd034a84 TT |
1029 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
1030 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
fd034a84 | 1031 | |
717d50e4 | 1032 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 1033 | ext4_lock_group(sb, group); |
8844618d TT |
1034 | if (ext4_has_group_desc_csum(sb) && |
1035 | (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) { | |
3300beda | 1036 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 1037 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 1038 | ext4_free_clusters_after_init(sb, group, gdp)); |
fa77dcfa | 1039 | ext4_block_bitmap_csum_set(sb, group, gdp, |
79f1ba49 | 1040 | block_bitmap_bh); |
feb0ab32 | 1041 | ext4_group_desc_csum_set(sb, group, gdp); |
717d50e4 | 1042 | } |
955ce5f5 | 1043 | ext4_unlock_group(sb, group); |
aeb1e5d6 | 1044 | brelse(block_bitmap_bh); |
717d50e4 | 1045 | |
eb9cc7e1 JK |
1046 | if (err) { |
1047 | ext4_std_error(sb, err); | |
1048 | goto out; | |
1049 | } | |
717d50e4 | 1050 | } |
119c0d44 | 1051 | |
119c0d44 | 1052 | /* Update the relevant bg descriptor fields */ |
41a246d1 | 1053 | if (ext4_has_group_desc_csum(sb)) { |
119c0d44 TT |
1054 | int free; |
1055 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1056 | ||
1057 | down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ | |
1058 | ext4_lock_group(sb, group); /* while we modify the bg desc */ | |
1059 | free = EXT4_INODES_PER_GROUP(sb) - | |
1060 | ext4_itable_unused_count(sb, gdp); | |
1061 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
1062 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
1063 | free = 0; | |
1064 | } | |
1065 | /* | |
1066 | * Check the relative inode number against the last used | |
1067 | * relative inode number in this group. if it is greater | |
1068 | * we need to update the bg_itable_unused count | |
1069 | */ | |
1070 | if (ino > free) | |
1071 | ext4_itable_unused_set(sb, gdp, | |
1072 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
1073 | up_read(&grp->alloc_sem); | |
6f2e9f0e TM |
1074 | } else { |
1075 | ext4_lock_group(sb, group); | |
119c0d44 | 1076 | } |
6f2e9f0e | 1077 | |
119c0d44 TT |
1078 | ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); |
1079 | if (S_ISDIR(mode)) { | |
1080 | ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); | |
1081 | if (sbi->s_log_groups_per_flex) { | |
1082 | ext4_group_t f = ext4_flex_group(sbi, group); | |
1083 | ||
7c990728 SJS |
1084 | atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups, |
1085 | f)->used_dirs); | |
119c0d44 TT |
1086 | } |
1087 | } | |
41a246d1 DW |
1088 | if (ext4_has_group_desc_csum(sb)) { |
1089 | ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, | |
1090 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 1091 | ext4_group_desc_csum_set(sb, group, gdp); |
119c0d44 | 1092 | } |
6f2e9f0e | 1093 | ext4_unlock_group(sb, group); |
119c0d44 | 1094 | |
3300beda AK |
1095 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
1096 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
eb9cc7e1 JK |
1097 | if (err) { |
1098 | ext4_std_error(sb, err); | |
1099 | goto out; | |
1100 | } | |
ac27a0ec DK |
1101 | |
1102 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
1103 | if (S_ISDIR(mode)) | |
1104 | percpu_counter_inc(&sbi->s_dirs_counter); | |
ac27a0ec | 1105 | |
772cb7c8 JS |
1106 | if (sbi->s_log_groups_per_flex) { |
1107 | flex_group = ext4_flex_group(sbi, group); | |
7c990728 SJS |
1108 | atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups, |
1109 | flex_group)->free_inodes); | |
772cb7c8 | 1110 | } |
ac27a0ec | 1111 | |
717d50e4 | 1112 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
1113 | /* This is the optimal IO size (for stat), not the fs block size */ |
1114 | inode->i_blocks = 0; | |
95582b00 | 1115 | inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode); |
7b62b293 | 1116 | ei->i_crtime = inode->i_mtime; |
ac27a0ec DK |
1117 | |
1118 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
1119 | ei->i_dir_start_lookup = 0; | |
1120 | ei->i_disksize = 0; | |
1121 | ||
4af83508 | 1122 | /* Don't inherit extent flag from directory, amongst others. */ |
2dc6b0d4 DG |
1123 | ei->i_flags = |
1124 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
1b917ed8 | 1125 | ei->i_flags |= i_flags; |
ac27a0ec | 1126 | ei->i_file_acl = 0; |
ac27a0ec | 1127 | ei->i_dtime = 0; |
ac27a0ec | 1128 | ei->i_block_group = group; |
a4912123 | 1129 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1130 | |
043546e4 | 1131 | ext4_set_inode_flags(inode, true); |
ac27a0ec | 1132 | if (IS_DIRSYNC(inode)) |
0390131b | 1133 | ext4_handle_sync(handle); |
6b38e842 | 1134 | if (insert_inode_locked(inode) < 0) { |
acd6ad83 JK |
1135 | /* |
1136 | * Likely a bitmap corruption causing inode to be allocated | |
1137 | * twice. | |
1138 | */ | |
1139 | err = -EIO; | |
eb9cc7e1 JK |
1140 | ext4_error(sb, "failed to insert inode %lu: doubly allocated?", |
1141 | inode->i_ino); | |
206f6d55 WS |
1142 | ext4_mark_group_bitmap_corrupted(sb, group, |
1143 | EXT4_GROUP_INFO_IBITMAP_CORRUPT); | |
eb9cc7e1 | 1144 | goto out; |
6b38e842 | 1145 | } |
23253068 | 1146 | inode->i_generation = prandom_u32(); |
ac27a0ec | 1147 | |
814525f4 | 1148 | /* Precompute checksum seed for inode metadata */ |
9aa5d32b | 1149 | if (ext4_has_metadata_csum(sb)) { |
814525f4 | 1150 | __u32 csum; |
814525f4 DW |
1151 | __le32 inum = cpu_to_le32(inode->i_ino); |
1152 | __le32 gen = cpu_to_le32(inode->i_generation); | |
1153 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
1154 | sizeof(inum)); | |
1155 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
1156 | sizeof(gen)); | |
1157 | } | |
1158 | ||
353eb83c | 1159 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 1160 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 | 1161 | |
49598e04 | 1162 | ei->i_extra_isize = sbi->s_want_extra_isize; |
f08225d1 | 1163 | ei->i_inline_off = 0; |
e2b911c5 | 1164 | if (ext4_has_feature_inline_data(sb)) |
f08225d1 | 1165 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
ac27a0ec | 1166 | ret = inode; |
63936dda CH |
1167 | err = dquot_alloc_inode(inode); |
1168 | if (err) | |
ac27a0ec | 1169 | goto fail_drop; |
ac27a0ec | 1170 | |
aa1dca3b EB |
1171 | /* |
1172 | * Since the encryption xattr will always be unique, create it first so | |
1173 | * that it's less likely to end up in an external xattr block and | |
1174 | * prevent its deduplication. | |
1175 | */ | |
1176 | if (encrypt) { | |
1177 | err = fscrypt_inherit_context(dir, inode, handle, true); | |
1178 | if (err) | |
1179 | goto fail_free_drop; | |
1180 | } | |
1181 | ||
1b917ed8 TE |
1182 | if (!(ei->i_flags & EXT4_EA_INODE_FL)) { |
1183 | err = ext4_init_acl(handle, inode, dir); | |
1184 | if (err) | |
1185 | goto fail_free_drop; | |
ac27a0ec | 1186 | |
ad47f953 TE |
1187 | err = ext4_init_security(handle, inode, dir, qstr); |
1188 | if (err) | |
1189 | goto fail_free_drop; | |
1190 | } | |
ac27a0ec | 1191 | |
e2b911c5 | 1192 | if (ext4_has_feature_extents(sb)) { |
e4079a11 | 1193 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1194 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1195 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1196 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1197 | } |
a86c6181 | 1198 | } |
ac27a0ec | 1199 | |
688f869c TT |
1200 | if (ext4_handle_valid(handle)) { |
1201 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
1202 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
1203 | } | |
1204 | ||
8753e88f AK |
1205 | err = ext4_mark_inode_dirty(handle, inode); |
1206 | if (err) { | |
1207 | ext4_std_error(sb, err); | |
1208 | goto fail_free_drop; | |
1209 | } | |
1210 | ||
617ba13b | 1211 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1212 | trace_ext4_allocate_inode(inode, dir, mode); |
3300beda | 1213 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1214 | return ret; |
1215 | ||
1216 | fail_free_drop: | |
63936dda | 1217 | dquot_free_inode(inode); |
ac27a0ec | 1218 | fail_drop: |
6d6b77f1 | 1219 | clear_nlink(inode); |
6b38e842 | 1220 | unlock_new_inode(inode); |
eb9cc7e1 JK |
1221 | out: |
1222 | dquot_drop(inode); | |
1223 | inode->i_flags |= S_NOQUOTA; | |
ac27a0ec | 1224 | iput(inode); |
3300beda | 1225 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1226 | return ERR_PTR(err); |
1227 | } | |
1228 | ||
1229 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1230 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1231 | { |
617ba13b | 1232 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1233 | ext4_group_t block_group; |
ac27a0ec | 1234 | int bit; |
7827a7f6 | 1235 | struct buffer_head *bitmap_bh = NULL; |
ac27a0ec | 1236 | struct inode *inode = NULL; |
7827a7f6 | 1237 | int err = -EFSCORRUPTED; |
ac27a0ec | 1238 | |
7827a7f6 TT |
1239 | if (ino < EXT4_FIRST_INO(sb) || ino > max_ino) |
1240 | goto bad_orphan; | |
ac27a0ec | 1241 | |
617ba13b MC |
1242 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1243 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1244 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
206f6d55 | 1245 | if (IS_ERR(bitmap_bh)) |
e884bce1 | 1246 | return ERR_CAST(bitmap_bh); |
ac27a0ec DK |
1247 | |
1248 | /* Having the inode bit set should be a 100% indicator that this | |
1249 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1250 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1251 | */ | |
1d1fe1ee DH |
1252 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1253 | goto bad_orphan; | |
1254 | ||
8a363970 | 1255 | inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL); |
7827a7f6 TT |
1256 | if (IS_ERR(inode)) { |
1257 | err = PTR_ERR(inode); | |
54d3adbc TT |
1258 | ext4_error_err(sb, -err, |
1259 | "couldn't read orphan inode %lu (err %d)", | |
1260 | ino, err); | |
3bbd0ef2 | 1261 | brelse(bitmap_bh); |
7827a7f6 TT |
1262 | return inode; |
1263 | } | |
1d1fe1ee | 1264 | |
91ef4caf | 1265 | /* |
c9eb13a9 TT |
1266 | * If the orphans has i_nlinks > 0 then it should be able to |
1267 | * be truncated, otherwise it won't be removed from the orphan | |
1268 | * list during processing and an infinite loop will result. | |
1269 | * Similarly, it must not be a bad inode. | |
91ef4caf | 1270 | */ |
c9eb13a9 TT |
1271 | if ((inode->i_nlink && !ext4_can_truncate(inode)) || |
1272 | is_bad_inode(inode)) | |
91ef4caf DG |
1273 | goto bad_orphan; |
1274 | ||
1d1fe1ee DH |
1275 | if (NEXT_ORPHAN(inode) > max_ino) |
1276 | goto bad_orphan; | |
1277 | brelse(bitmap_bh); | |
1278 | return inode; | |
1279 | ||
1d1fe1ee | 1280 | bad_orphan: |
7827a7f6 TT |
1281 | ext4_error(sb, "bad orphan inode %lu", ino); |
1282 | if (bitmap_bh) | |
1283 | printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n", | |
1284 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
1285 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
1d1fe1ee | 1286 | if (inode) { |
7827a7f6 | 1287 | printk(KERN_ERR "is_bad_inode(inode)=%d\n", |
1d1fe1ee | 1288 | is_bad_inode(inode)); |
7827a7f6 | 1289 | printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n", |
1d1fe1ee | 1290 | NEXT_ORPHAN(inode)); |
7827a7f6 TT |
1291 | printk(KERN_ERR "max_ino=%lu\n", max_ino); |
1292 | printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink); | |
ac27a0ec | 1293 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1294 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1295 | inode->i_blocks = 0; |
1296 | iput(inode); | |
ac27a0ec | 1297 | } |
ac27a0ec | 1298 | brelse(bitmap_bh); |
1d1fe1ee | 1299 | return ERR_PTR(err); |
ac27a0ec DK |
1300 | } |
1301 | ||
af5bc92d | 1302 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1303 | { |
1304 | unsigned long desc_count; | |
617ba13b | 1305 | struct ext4_group_desc *gdp; |
8df9675f | 1306 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1307 | #ifdef EXT4FS_DEBUG |
1308 | struct ext4_super_block *es; | |
ac27a0ec DK |
1309 | unsigned long bitmap_count, x; |
1310 | struct buffer_head *bitmap_bh = NULL; | |
1311 | ||
617ba13b | 1312 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1313 | desc_count = 0; |
1314 | bitmap_count = 0; | |
1315 | gdp = NULL; | |
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 | 1321 | brelse(bitmap_bh); |
e29d1cde | 1322 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
9008a58e DW |
1323 | if (IS_ERR(bitmap_bh)) { |
1324 | bitmap_bh = NULL; | |
ac27a0ec | 1325 | continue; |
9008a58e | 1326 | } |
ac27a0ec | 1327 | |
f6fb99ca TT |
1328 | x = ext4_count_free(bitmap_bh->b_data, |
1329 | EXT4_INODES_PER_GROUP(sb) / 8); | |
c549a95d | 1330 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1331 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1332 | bitmap_count += x; |
1333 | } | |
1334 | brelse(bitmap_bh); | |
4776004f TT |
1335 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1336 | "stored = %u, computed = %lu, %lu\n", | |
1337 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1338 | return desc_count; |
1339 | #else | |
1340 | desc_count = 0; | |
8df9675f | 1341 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1342 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1343 | if (!gdp) |
1344 | continue; | |
560671a0 | 1345 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1346 | cond_resched(); |
1347 | } | |
1348 | return desc_count; | |
1349 | #endif | |
1350 | } | |
1351 | ||
1352 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1353 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1354 | { |
1355 | unsigned long count = 0; | |
8df9675f | 1356 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1357 | |
8df9675f | 1358 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1359 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1360 | if (!gdp) |
1361 | continue; | |
560671a0 | 1362 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1363 | } |
1364 | return count; | |
1365 | } | |
bfff6873 LC |
1366 | |
1367 | /* | |
1368 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1369 | * inode table. Must be called without any spinlock held. The only place | |
1370 | * where it is called from on active part of filesystem is ext4lazyinit | |
1371 | * thread, so we do not need any special locks, however we have to prevent | |
1372 | * inode allocation from the current group, so we take alloc_sem lock, to | |
119c0d44 | 1373 | * block ext4_new_inode() until we are finished. |
bfff6873 | 1374 | */ |
e0cbee3e | 1375 | int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, |
bfff6873 LC |
1376 | int barrier) |
1377 | { | |
1378 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1379 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1380 | struct ext4_group_desc *gdp = NULL; | |
1381 | struct buffer_head *group_desc_bh; | |
1382 | handle_t *handle; | |
1383 | ext4_fsblk_t blk; | |
1384 | int num, ret = 0, used_blks = 0; | |
bfff6873 LC |
1385 | |
1386 | /* This should not happen, but just to be sure check this */ | |
bc98a42c | 1387 | if (sb_rdonly(sb)) { |
bfff6873 LC |
1388 | ret = 1; |
1389 | goto out; | |
1390 | } | |
1391 | ||
1392 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1393 | if (!gdp) | |
1394 | goto out; | |
1395 | ||
1396 | /* | |
1397 | * We do not need to lock this, because we are the only one | |
1398 | * handling this flag. | |
1399 | */ | |
1400 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1401 | goto out; | |
1402 | ||
9924a92a | 1403 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
bfff6873 LC |
1404 | if (IS_ERR(handle)) { |
1405 | ret = PTR_ERR(handle); | |
1406 | goto out; | |
1407 | } | |
1408 | ||
1409 | down_write(&grp->alloc_sem); | |
1410 | /* | |
1411 | * If inode bitmap was already initialized there may be some | |
1412 | * used inodes so we need to skip blocks with used inodes in | |
1413 | * inode table. | |
1414 | */ | |
1415 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1416 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1417 | ext4_itable_unused_count(sb, gdp)), | |
1418 | sbi->s_inodes_per_block); | |
1419 | ||
50122847 TT |
1420 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) || |
1421 | ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) - | |
1422 | ext4_itable_unused_count(sb, gdp)) < | |
1423 | EXT4_FIRST_INO(sb)))) { | |
1084f252 TT |
1424 | ext4_error(sb, "Something is wrong with group %u: " |
1425 | "used itable blocks: %d; " | |
1426 | "itable unused count: %u", | |
857ac889 LC |
1427 | group, used_blks, |
1428 | ext4_itable_unused_count(sb, gdp)); | |
1429 | ret = 1; | |
33853a0d | 1430 | goto err_out; |
857ac889 LC |
1431 | } |
1432 | ||
bfff6873 LC |
1433 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1434 | num = sbi->s_itb_per_group - used_blks; | |
1435 | ||
1436 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1437 | ret = ext4_journal_get_write_access(handle, | |
1438 | group_desc_bh); | |
1439 | if (ret) | |
1440 | goto err_out; | |
1441 | ||
bfff6873 LC |
1442 | /* |
1443 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1444 | * flag anyway, because obviously, when it is full it does not need | |
1445 | * further zeroing. | |
1446 | */ | |
1447 | if (unlikely(num == 0)) | |
1448 | goto skip_zeroout; | |
1449 | ||
1450 | ext4_debug("going to zero out inode table in group %d\n", | |
1451 | group); | |
a107e5a3 | 1452 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1453 | if (ret < 0) |
1454 | goto err_out; | |
a107e5a3 | 1455 | if (barrier) |
9398554f | 1456 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS); |
bfff6873 LC |
1457 | |
1458 | skip_zeroout: | |
1459 | ext4_lock_group(sb, group); | |
1460 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
feb0ab32 | 1461 | ext4_group_desc_csum_set(sb, group, gdp); |
bfff6873 LC |
1462 | ext4_unlock_group(sb, group); |
1463 | ||
1464 | BUFFER_TRACE(group_desc_bh, | |
1465 | "call ext4_handle_dirty_metadata"); | |
1466 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1467 | group_desc_bh); | |
1468 | ||
1469 | err_out: | |
1470 | up_write(&grp->alloc_sem); | |
1471 | ext4_journal_stop(handle); | |
1472 | out: | |
1473 | return ret; | |
1474 | } |