Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/ialloc.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * BSD ufs-inspired inode and directory allocation by | |
10 | * Stephen Tweedie (sct@redhat.com), 1993 | |
11 | * Big-endian to little-endian byte-swapping/bitmaps by | |
12 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
13 | */ | |
14 | ||
15 | #include <linux/time.h> | |
16 | #include <linux/fs.h> | |
dab291af | 17 | #include <linux/jbd2.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> |
ac27a0ec | 25 | #include <asm/byteorder.h> |
9bffad1e | 26 | |
3dcf5451 CH |
27 | #include "ext4.h" |
28 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
29 | #include "xattr.h" |
30 | #include "acl.h" | |
31 | ||
9bffad1e TT |
32 | #include <trace/events/ext4.h> |
33 | ||
ac27a0ec DK |
34 | /* |
35 | * ialloc.c contains the inodes allocation and deallocation routines | |
36 | */ | |
37 | ||
38 | /* | |
39 | * The free inodes are managed by bitmaps. A file system contains several | |
40 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
41 | * block for inodes, N blocks for the inode table and data blocks. | |
42 | * | |
43 | * The file system contains group descriptors which are located after the | |
44 | * super block. Each descriptor contains the number of the bitmap block and | |
45 | * the free blocks count in the block. | |
46 | */ | |
47 | ||
717d50e4 AD |
48 | /* |
49 | * To avoid calling the atomic setbit hundreds or thousands of times, we only | |
50 | * need to use it within a single byte (to ensure we get endianness right). | |
51 | * We can use memset for the rest of the bitmap as there are no other users. | |
52 | */ | |
61d08673 | 53 | void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap) |
717d50e4 AD |
54 | { |
55 | int i; | |
56 | ||
57 | if (start_bit >= end_bit) | |
58 | return; | |
59 | ||
60 | ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); | |
61 | for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) | |
62 | ext4_set_bit(i, bitmap); | |
63 | if (i < end_bit) | |
64 | memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); | |
65 | } | |
66 | ||
67 | /* Initializes an uninitialized inode bitmap */ | |
1f109d5a TT |
68 | static unsigned ext4_init_inode_bitmap(struct super_block *sb, |
69 | struct buffer_head *bh, | |
70 | ext4_group_t block_group, | |
71 | struct ext4_group_desc *gdp) | |
717d50e4 | 72 | { |
bdfb6ff4 | 73 | struct ext4_group_info *grp; |
e43bb4e6 | 74 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
717d50e4 AD |
75 | J_ASSERT_BH(bh, buffer_locked(bh)); |
76 | ||
77 | /* If checksum is bad mark all blocks and inodes use to prevent | |
78 | * allocation, essentially implementing a per-group read-only flag. */ | |
feb0ab32 | 79 | if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) { |
12062ddd | 80 | ext4_error(sb, "Checksum bad for group %u", block_group); |
bdfb6ff4 | 81 | grp = ext4_get_group_info(sb, block_group); |
e43bb4e6 NJ |
82 | if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) |
83 | percpu_counter_sub(&sbi->s_freeclusters_counter, | |
84 | grp->bb_free); | |
bdfb6ff4 | 85 | set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state); |
e43bb4e6 NJ |
86 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
87 | int count; | |
88 | count = ext4_free_inodes_count(sb, gdp); | |
89 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
90 | count); | |
91 | } | |
bdfb6ff4 | 92 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
717d50e4 AD |
93 | return 0; |
94 | } | |
95 | ||
96 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
61d08673 | 97 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
717d50e4 | 98 | bh->b_data); |
41a246d1 DW |
99 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh, |
100 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 101 | ext4_group_desc_csum_set(sb, block_group, gdp); |
717d50e4 AD |
102 | |
103 | return EXT4_INODES_PER_GROUP(sb); | |
104 | } | |
ac27a0ec | 105 | |
813e5727 TT |
106 | void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate) |
107 | { | |
108 | if (uptodate) { | |
109 | set_buffer_uptodate(bh); | |
110 | set_bitmap_uptodate(bh); | |
111 | } | |
112 | unlock_buffer(bh); | |
113 | put_bh(bh); | |
114 | } | |
115 | ||
ac27a0ec DK |
116 | /* |
117 | * Read the inode allocation bitmap for a given block_group, reading | |
118 | * into the specified slot in the superblock's bitmap cache. | |
119 | * | |
120 | * Return buffer_head of bitmap on success or NULL. | |
121 | */ | |
122 | static struct buffer_head * | |
e29d1cde | 123 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 124 | { |
617ba13b | 125 | struct ext4_group_desc *desc; |
ac27a0ec | 126 | struct buffer_head *bh = NULL; |
e29d1cde | 127 | ext4_fsblk_t bitmap_blk; |
87a39389 | 128 | struct ext4_group_info *grp; |
e43bb4e6 | 129 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
ac27a0ec | 130 | |
617ba13b | 131 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 132 | if (!desc) |
e29d1cde | 133 | return NULL; |
bfff6873 | 134 | |
e29d1cde ES |
135 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
136 | bh = sb_getblk(sb, bitmap_blk); | |
137 | if (unlikely(!bh)) { | |
12062ddd | 138 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 139 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
140 | block_group, bitmap_blk); |
141 | return NULL; | |
142 | } | |
2ccb5fb9 | 143 | if (bitmap_uptodate(bh)) |
41a246d1 | 144 | goto verify; |
e29d1cde | 145 | |
c806e68f | 146 | lock_buffer(bh); |
2ccb5fb9 AK |
147 | if (bitmap_uptodate(bh)) { |
148 | unlock_buffer(bh); | |
41a246d1 | 149 | goto verify; |
2ccb5fb9 | 150 | } |
bfff6873 | 151 | |
955ce5f5 | 152 | ext4_lock_group(sb, block_group); |
717d50e4 | 153 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
e29d1cde | 154 | ext4_init_inode_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 155 | set_bitmap_uptodate(bh); |
e29d1cde | 156 | set_buffer_uptodate(bh); |
41a246d1 | 157 | set_buffer_verified(bh); |
955ce5f5 | 158 | ext4_unlock_group(sb, block_group); |
3300beda | 159 | unlock_buffer(bh); |
e29d1cde | 160 | return bh; |
717d50e4 | 161 | } |
955ce5f5 | 162 | ext4_unlock_group(sb, block_group); |
bfff6873 | 163 | |
2ccb5fb9 AK |
164 | if (buffer_uptodate(bh)) { |
165 | /* | |
166 | * if not uninit if bh is uptodate, | |
167 | * bitmap is also uptodate | |
168 | */ | |
169 | set_bitmap_uptodate(bh); | |
170 | unlock_buffer(bh); | |
41a246d1 | 171 | goto verify; |
2ccb5fb9 AK |
172 | } |
173 | /* | |
813e5727 | 174 | * submit the buffer_head for reading |
2ccb5fb9 | 175 | */ |
0562e0ba | 176 | trace_ext4_load_inode_bitmap(sb, block_group); |
813e5727 TT |
177 | bh->b_end_io = ext4_end_bitmap_read; |
178 | get_bh(bh); | |
9f203507 | 179 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
813e5727 TT |
180 | wait_on_buffer(bh); |
181 | if (!buffer_uptodate(bh)) { | |
e29d1cde | 182 | put_bh(bh); |
12062ddd | 183 | ext4_error(sb, "Cannot read inode bitmap - " |
813e5727 TT |
184 | "block_group = %u, inode_bitmap = %llu", |
185 | block_group, bitmap_blk); | |
e29d1cde ES |
186 | return NULL; |
187 | } | |
41a246d1 DW |
188 | |
189 | verify: | |
190 | ext4_lock_group(sb, block_group); | |
191 | if (!buffer_verified(bh) && | |
192 | !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh, | |
193 | EXT4_INODES_PER_GROUP(sb) / 8)) { | |
194 | ext4_unlock_group(sb, block_group); | |
195 | put_bh(bh); | |
196 | ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " | |
197 | "inode_bitmap = %llu", block_group, bitmap_blk); | |
87a39389 | 198 | grp = ext4_get_group_info(sb, block_group); |
e43bb4e6 NJ |
199 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
200 | int count; | |
201 | count = ext4_free_inodes_count(sb, desc); | |
202 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
203 | count); | |
204 | } | |
87a39389 | 205 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
41a246d1 DW |
206 | return NULL; |
207 | } | |
208 | ext4_unlock_group(sb, block_group); | |
209 | set_buffer_verified(bh); | |
ac27a0ec DK |
210 | return bh; |
211 | } | |
212 | ||
213 | /* | |
214 | * NOTE! When we get the inode, we're the only people | |
215 | * that have access to it, and as such there are no | |
216 | * race conditions we have to worry about. The inode | |
217 | * is not on the hash-lists, and it cannot be reached | |
218 | * through the filesystem because the directory entry | |
219 | * has been deleted earlier. | |
220 | * | |
221 | * HOWEVER: we must make sure that we get no aliases, | |
222 | * which means that we have to call "clear_inode()" | |
223 | * _before_ we mark the inode not in use in the inode | |
224 | * bitmaps. Otherwise a newly created file might use | |
225 | * the same inode number (not actually the same pointer | |
226 | * though), and then we'd have two inodes sharing the | |
227 | * same inode number and space on the harddisk. | |
228 | */ | |
af5bc92d | 229 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 230 | { |
af5bc92d | 231 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
232 | int is_directory; |
233 | unsigned long ino; | |
234 | struct buffer_head *bitmap_bh = NULL; | |
235 | struct buffer_head *bh2; | |
fd2d4291 | 236 | ext4_group_t block_group; |
ac27a0ec | 237 | unsigned long bit; |
af5bc92d TT |
238 | struct ext4_group_desc *gdp; |
239 | struct ext4_super_block *es; | |
617ba13b | 240 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 241 | int fatal = 0, err, count, cleared; |
87a39389 | 242 | struct ext4_group_info *grp; |
ac27a0ec | 243 | |
92b97816 TT |
244 | if (!sb) { |
245 | printk(KERN_ERR "EXT4-fs: %s:%d: inode on " | |
246 | "nonexistent device\n", __func__, __LINE__); | |
ac27a0ec DK |
247 | return; |
248 | } | |
92b97816 TT |
249 | if (atomic_read(&inode->i_count) > 1) { |
250 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", | |
251 | __func__, __LINE__, inode->i_ino, | |
252 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
253 | return; |
254 | } | |
92b97816 TT |
255 | if (inode->i_nlink) { |
256 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", | |
257 | __func__, __LINE__, inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
258 | return; |
259 | } | |
617ba13b | 260 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
261 | |
262 | ino = inode->i_ino; | |
af5bc92d | 263 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 264 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
265 | |
266 | /* | |
267 | * Note: we must free any quota before locking the superblock, | |
268 | * as writing the quota to disk may need the lock as well. | |
269 | */ | |
871a2931 | 270 | dquot_initialize(inode); |
617ba13b | 271 | ext4_xattr_delete_inode(handle, inode); |
63936dda | 272 | dquot_free_inode(inode); |
9f754758 | 273 | dquot_drop(inode); |
ac27a0ec DK |
274 | |
275 | is_directory = S_ISDIR(inode->i_mode); | |
276 | ||
277 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 278 | ext4_clear_inode(inode); |
ac27a0ec | 279 | |
617ba13b MC |
280 | es = EXT4_SB(sb)->s_es; |
281 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 282 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
283 | goto error_return; |
284 | } | |
617ba13b MC |
285 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
286 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 287 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
87a39389 DW |
288 | /* Don't bother if the inode bitmap is corrupt. */ |
289 | grp = ext4_get_group_info(sb, block_group); | |
290 | if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) || !bitmap_bh) | |
ac27a0ec DK |
291 | goto error_return; |
292 | ||
293 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 294 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
295 | if (fatal) |
296 | goto error_return; | |
297 | ||
d17413c0 DM |
298 | fatal = -ESRCH; |
299 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
300 | if (gdp) { | |
ac27a0ec | 301 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 302 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
303 | } |
304 | ext4_lock_group(sb, block_group); | |
597d508c | 305 | cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); |
d17413c0 DM |
306 | if (fatal || !cleared) { |
307 | ext4_unlock_group(sb, block_group); | |
308 | goto out; | |
309 | } | |
7d39db14 | 310 | |
d17413c0 DM |
311 | count = ext4_free_inodes_count(sb, gdp) + 1; |
312 | ext4_free_inodes_set(sb, gdp, count); | |
313 | if (is_directory) { | |
314 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
315 | ext4_used_dirs_set(sb, gdp, count); | |
316 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 317 | } |
41a246d1 DW |
318 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, |
319 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 320 | ext4_group_desc_csum_set(sb, block_group, gdp); |
d17413c0 | 321 | ext4_unlock_group(sb, block_group); |
ac27a0ec | 322 | |
d17413c0 DM |
323 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
324 | if (sbi->s_log_groups_per_flex) { | |
325 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 326 | |
d17413c0 DM |
327 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
328 | if (is_directory) | |
329 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 330 | } |
d17413c0 DM |
331 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
332 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
333 | out: | |
334 | if (cleared) { | |
335 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
336 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
337 | if (!fatal) | |
338 | fatal = err; | |
87a39389 | 339 | } else { |
d17413c0 | 340 | ext4_error(sb, "bit already cleared for inode %lu", ino); |
e43bb4e6 NJ |
341 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
342 | int count; | |
343 | count = ext4_free_inodes_count(sb, gdp); | |
344 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
345 | count); | |
346 | } | |
87a39389 DW |
347 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
348 | } | |
d17413c0 | 349 | |
ac27a0ec DK |
350 | error_return: |
351 | brelse(bitmap_bh); | |
617ba13b | 352 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
353 | } |
354 | ||
a4912123 | 355 | struct orlov_stats { |
90ba983f | 356 | __u64 free_clusters; |
a4912123 | 357 | __u32 free_inodes; |
a4912123 TT |
358 | __u32 used_dirs; |
359 | }; | |
360 | ||
361 | /* | |
362 | * Helper function for Orlov's allocator; returns critical information | |
363 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
364 | * is a block group number; otherwise it is flex_bg number. | |
365 | */ | |
1f109d5a TT |
366 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
367 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
368 | { |
369 | struct ext4_group_desc *desc; | |
7d39db14 | 370 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 371 | |
7d39db14 TT |
372 | if (flex_size > 1) { |
373 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
90ba983f | 374 | stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); |
7d39db14 TT |
375 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); |
376 | return; | |
377 | } | |
a4912123 | 378 | |
7d39db14 TT |
379 | desc = ext4_get_group_desc(sb, g, NULL); |
380 | if (desc) { | |
381 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
021b65bb | 382 | stats->free_clusters = ext4_free_group_clusters(sb, desc); |
7d39db14 TT |
383 | stats->used_dirs = ext4_used_dirs_count(sb, desc); |
384 | } else { | |
385 | stats->free_inodes = 0; | |
24aaa8ef | 386 | stats->free_clusters = 0; |
7d39db14 | 387 | stats->used_dirs = 0; |
a4912123 TT |
388 | } |
389 | } | |
390 | ||
ac27a0ec DK |
391 | /* |
392 | * Orlov's allocator for directories. | |
393 | * | |
394 | * We always try to spread first-level directories. | |
395 | * | |
396 | * If there are blockgroups with both free inodes and free blocks counts | |
397 | * not worse than average we return one with smallest directory count. | |
398 | * Otherwise we simply return a random group. | |
399 | * | |
400 | * For the rest rules look so: | |
401 | * | |
402 | * It's OK to put directory into a group unless | |
403 | * it has too many directories already (max_dirs) or | |
404 | * it has too few free inodes left (min_inodes) or | |
405 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 406 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
407 | * conditions we search cyclically through the rest. If none |
408 | * of the groups look good we just look for a group with more | |
409 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
410 | */ |
411 | ||
2aa9fc4c | 412 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
dcca3fec | 413 | ext4_group_t *group, umode_t mode, |
f157a4aa | 414 | const struct qstr *qstr) |
ac27a0ec | 415 | { |
fd2d4291 | 416 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 417 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 418 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 419 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
14c83c9f | 420 | unsigned int freei, avefreei, grp_free; |
24aaa8ef | 421 | ext4_fsblk_t freeb, avefreec; |
ac27a0ec | 422 | unsigned int ndirs; |
a4912123 | 423 | int max_dirs, min_inodes; |
24aaa8ef | 424 | ext4_grpblk_t min_clusters; |
8df9675f | 425 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 426 | struct ext4_group_desc *desc; |
a4912123 TT |
427 | struct orlov_stats stats; |
428 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 429 | struct dx_hash_info hinfo; |
a4912123 | 430 | |
8df9675f | 431 | ngroups = real_ngroups; |
a4912123 | 432 | if (flex_size > 1) { |
8df9675f | 433 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
434 | sbi->s_log_groups_per_flex; |
435 | parent_group >>= sbi->s_log_groups_per_flex; | |
436 | } | |
ac27a0ec DK |
437 | |
438 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
439 | avefreei = freei / ngroups; | |
57042651 TT |
440 | freeb = EXT4_C2B(sbi, |
441 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
24aaa8ef TT |
442 | avefreec = freeb; |
443 | do_div(avefreec, ngroups); | |
ac27a0ec DK |
444 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
445 | ||
a4912123 TT |
446 | if (S_ISDIR(mode) && |
447 | ((parent == sb->s_root->d_inode) || | |
12e9b892 | 448 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 449 | int best_ndir = inodes_per_group; |
2aa9fc4c | 450 | int ret = -1; |
ac27a0ec | 451 | |
f157a4aa TT |
452 | if (qstr) { |
453 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
454 | hinfo.seed = sbi->s_hash_seed; | |
455 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
456 | grp = hinfo.hash; | |
457 | } else | |
dd1f723b | 458 | grp = prandom_u32(); |
2aa9fc4c | 459 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 460 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
461 | g = (parent_group + i) % ngroups; |
462 | get_orlov_stats(sb, g, flex_size, &stats); | |
463 | if (!stats.free_inodes) | |
ac27a0ec | 464 | continue; |
a4912123 | 465 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 466 | continue; |
a4912123 | 467 | if (stats.free_inodes < avefreei) |
ac27a0ec | 468 | continue; |
24aaa8ef | 469 | if (stats.free_clusters < avefreec) |
ac27a0ec | 470 | continue; |
a4912123 | 471 | grp = g; |
2aa9fc4c | 472 | ret = 0; |
a4912123 TT |
473 | best_ndir = stats.used_dirs; |
474 | } | |
475 | if (ret) | |
476 | goto fallback; | |
477 | found_flex_bg: | |
478 | if (flex_size == 1) { | |
479 | *group = grp; | |
480 | return 0; | |
481 | } | |
482 | ||
483 | /* | |
484 | * We pack inodes at the beginning of the flexgroup's | |
485 | * inode tables. Block allocation decisions will do | |
486 | * something similar, although regular files will | |
487 | * start at 2nd block group of the flexgroup. See | |
488 | * ext4_ext_find_goal() and ext4_find_near(). | |
489 | */ | |
490 | grp *= flex_size; | |
491 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 492 | if (grp+i >= real_ngroups) |
a4912123 TT |
493 | break; |
494 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
495 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
496 | *group = grp+i; | |
497 | return 0; | |
498 | } | |
ac27a0ec | 499 | } |
ac27a0ec DK |
500 | goto fallback; |
501 | } | |
502 | ||
ac27a0ec | 503 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
504 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
505 | if (min_inodes < 1) | |
506 | min_inodes = 1; | |
24aaa8ef | 507 | min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; |
a4912123 TT |
508 | |
509 | /* | |
510 | * Start looking in the flex group where we last allocated an | |
511 | * inode for this parent directory | |
512 | */ | |
513 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
514 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
515 | if (flex_size > 1) | |
516 | parent_group >>= sbi->s_log_groups_per_flex; | |
517 | } | |
ac27a0ec DK |
518 | |
519 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
520 | grp = (parent_group + i) % ngroups; |
521 | get_orlov_stats(sb, grp, flex_size, &stats); | |
522 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 523 | continue; |
a4912123 | 524 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 525 | continue; |
24aaa8ef | 526 | if (stats.free_clusters < min_clusters) |
ac27a0ec | 527 | continue; |
a4912123 | 528 | goto found_flex_bg; |
ac27a0ec DK |
529 | } |
530 | ||
531 | fallback: | |
8df9675f | 532 | ngroups = real_ngroups; |
a4912123 | 533 | avefreei = freei / ngroups; |
b5451f7b | 534 | fallback_retry: |
a4912123 | 535 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 536 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
537 | grp = (parent_group + i) % ngroups; |
538 | desc = ext4_get_group_desc(sb, grp, NULL); | |
bb3d132a DC |
539 | if (desc) { |
540 | grp_free = ext4_free_inodes_count(sb, desc); | |
541 | if (grp_free && grp_free >= avefreei) { | |
542 | *group = grp; | |
543 | return 0; | |
544 | } | |
a4912123 | 545 | } |
ac27a0ec DK |
546 | } |
547 | ||
548 | if (avefreei) { | |
549 | /* | |
550 | * The free-inodes counter is approximate, and for really small | |
551 | * filesystems the above test can fail to find any blockgroups | |
552 | */ | |
553 | avefreei = 0; | |
b5451f7b | 554 | goto fallback_retry; |
ac27a0ec DK |
555 | } |
556 | ||
557 | return -1; | |
558 | } | |
559 | ||
2aa9fc4c | 560 | static int find_group_other(struct super_block *sb, struct inode *parent, |
dcca3fec | 561 | ext4_group_t *group, umode_t mode) |
ac27a0ec | 562 | { |
fd2d4291 | 563 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 564 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 565 | struct ext4_group_desc *desc; |
a4912123 TT |
566 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
567 | ||
568 | /* | |
569 | * Try to place the inode is the same flex group as its | |
570 | * parent. If we can't find space, use the Orlov algorithm to | |
571 | * find another flex group, and store that information in the | |
572 | * parent directory's inode information so that use that flex | |
573 | * group for future allocations. | |
574 | */ | |
575 | if (flex_size > 1) { | |
576 | int retry = 0; | |
577 | ||
578 | try_again: | |
579 | parent_group &= ~(flex_size-1); | |
580 | last = parent_group + flex_size; | |
581 | if (last > ngroups) | |
582 | last = ngroups; | |
583 | for (i = parent_group; i < last; i++) { | |
584 | desc = ext4_get_group_desc(sb, i, NULL); | |
585 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
586 | *group = i; | |
587 | return 0; | |
588 | } | |
589 | } | |
590 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
591 | retry = 1; | |
592 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
593 | goto try_again; | |
594 | } | |
595 | /* | |
596 | * If this didn't work, use the Orlov search algorithm | |
597 | * to find a new flex group; we pass in the mode to | |
598 | * avoid the topdir algorithms. | |
599 | */ | |
600 | *group = parent_group + flex_size; | |
601 | if (*group > ngroups) | |
602 | *group = 0; | |
7dc57615 | 603 | return find_group_orlov(sb, parent, group, mode, NULL); |
a4912123 | 604 | } |
ac27a0ec DK |
605 | |
606 | /* | |
607 | * Try to place the inode in its parent directory | |
608 | */ | |
2aa9fc4c AM |
609 | *group = parent_group; |
610 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 611 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 612 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 613 | return 0; |
ac27a0ec DK |
614 | |
615 | /* | |
616 | * We're going to place this inode in a different blockgroup from its | |
617 | * parent. We want to cause files in a common directory to all land in | |
618 | * the same blockgroup. But we want files which are in a different | |
619 | * directory which shares a blockgroup with our parent to land in a | |
620 | * different blockgroup. | |
621 | * | |
622 | * So add our directory's i_ino into the starting point for the hash. | |
623 | */ | |
2aa9fc4c | 624 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
625 | |
626 | /* | |
627 | * Use a quadratic hash to find a group with a free inode and some free | |
628 | * blocks. | |
629 | */ | |
630 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
631 | *group += i; |
632 | if (*group >= ngroups) | |
633 | *group -= ngroups; | |
634 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 635 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 636 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 637 | return 0; |
ac27a0ec DK |
638 | } |
639 | ||
640 | /* | |
641 | * That failed: try linear search for a free inode, even if that group | |
642 | * has no free blocks. | |
643 | */ | |
2aa9fc4c | 644 | *group = parent_group; |
ac27a0ec | 645 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
646 | if (++*group >= ngroups) |
647 | *group = 0; | |
648 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 649 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 650 | return 0; |
ac27a0ec DK |
651 | } |
652 | ||
653 | return -1; | |
654 | } | |
655 | ||
19883bd9 TT |
656 | /* |
657 | * In no journal mode, if an inode has recently been deleted, we want | |
658 | * to avoid reusing it until we're reasonably sure the inode table | |
659 | * block has been written back to disk. (Yes, these values are | |
660 | * somewhat arbitrary...) | |
661 | */ | |
662 | #define RECENTCY_MIN 5 | |
663 | #define RECENTCY_DIRTY 30 | |
664 | ||
665 | static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) | |
666 | { | |
667 | struct ext4_group_desc *gdp; | |
668 | struct ext4_inode *raw_inode; | |
669 | struct buffer_head *bh; | |
670 | unsigned long dtime, now; | |
671 | int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
672 | int offset, ret = 0, recentcy = RECENTCY_MIN; | |
673 | ||
674 | gdp = ext4_get_group_desc(sb, group, NULL); | |
675 | if (unlikely(!gdp)) | |
676 | return 0; | |
677 | ||
678 | bh = sb_getblk(sb, ext4_inode_table(sb, gdp) + | |
679 | (ino / inodes_per_block)); | |
680 | if (unlikely(!bh) || !buffer_uptodate(bh)) | |
681 | /* | |
682 | * If the block is not in the buffer cache, then it | |
683 | * must have been written out. | |
684 | */ | |
685 | goto out; | |
686 | ||
687 | offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
688 | raw_inode = (struct ext4_inode *) (bh->b_data + offset); | |
689 | dtime = le32_to_cpu(raw_inode->i_dtime); | |
690 | now = get_seconds(); | |
691 | if (buffer_dirty(bh)) | |
692 | recentcy += RECENTCY_DIRTY; | |
693 | ||
694 | if (dtime && (dtime < now) && (now < dtime + recentcy)) | |
695 | ret = 1; | |
696 | out: | |
697 | brelse(bh); | |
698 | return ret; | |
699 | } | |
700 | ||
ac27a0ec DK |
701 | /* |
702 | * There are two policies for allocating an inode. If the new inode is | |
703 | * a directory, then a forward search is made for a block group with both | |
704 | * free space and a low directory-to-inode ratio; if that fails, then of | |
705 | * the groups with above-average free space, that group with the fewest | |
706 | * directories already is chosen. | |
707 | * | |
708 | * For other inodes, search forward from the parent directory's block | |
709 | * group to find a free inode. | |
710 | */ | |
1139575a TT |
711 | struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, |
712 | umode_t mode, const struct qstr *qstr, | |
713 | __u32 goal, uid_t *owner, int handle_type, | |
714 | unsigned int line_no, int nblocks) | |
ac27a0ec DK |
715 | { |
716 | struct super_block *sb; | |
3300beda AK |
717 | struct buffer_head *inode_bitmap_bh = NULL; |
718 | struct buffer_head *group_desc_bh; | |
8df9675f | 719 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 720 | unsigned long ino = 0; |
af5bc92d TT |
721 | struct inode *inode; |
722 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
723 | struct ext4_inode_info *ei; |
724 | struct ext4_sb_info *sbi; | |
39341867 | 725 | int ret2, err = 0; |
ac27a0ec | 726 | struct inode *ret; |
2aa9fc4c | 727 | ext4_group_t i; |
772cb7c8 | 728 | ext4_group_t flex_group; |
87a39389 | 729 | struct ext4_group_info *grp; |
ac27a0ec DK |
730 | |
731 | /* Cannot create files in a deleted directory */ | |
732 | if (!dir || !dir->i_nlink) | |
733 | return ERR_PTR(-EPERM); | |
734 | ||
735 | sb = dir->i_sb; | |
8df9675f | 736 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 737 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
738 | inode = new_inode(sb); |
739 | if (!inode) | |
740 | return ERR_PTR(-ENOMEM); | |
617ba13b | 741 | ei = EXT4_I(inode); |
617ba13b | 742 | sbi = EXT4_SB(sb); |
772cb7c8 | 743 | |
eb9cc7e1 JK |
744 | /* |
745 | * Initalize owners and quota early so that we don't have to account | |
746 | * for quota initialization worst case in standard inode creating | |
747 | * transaction | |
748 | */ | |
749 | if (owner) { | |
750 | inode->i_mode = mode; | |
751 | i_uid_write(inode, owner[0]); | |
752 | i_gid_write(inode, owner[1]); | |
753 | } else if (test_opt(sb, GRPID)) { | |
754 | inode->i_mode = mode; | |
755 | inode->i_uid = current_fsuid(); | |
756 | inode->i_gid = dir->i_gid; | |
757 | } else | |
758 | inode_init_owner(inode, dir, mode); | |
759 | dquot_initialize(inode); | |
760 | ||
11013911 AD |
761 | if (!goal) |
762 | goal = sbi->s_inode_goal; | |
763 | ||
e6462869 | 764 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
765 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
766 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
767 | ret2 = 0; | |
768 | goto got_group; | |
769 | } | |
770 | ||
4113c4ca LC |
771 | if (S_ISDIR(mode)) |
772 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); | |
773 | else | |
a4912123 | 774 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 775 | |
772cb7c8 | 776 | got_group: |
a4912123 | 777 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 778 | err = -ENOSPC; |
2aa9fc4c | 779 | if (ret2 == -1) |
ac27a0ec DK |
780 | goto out; |
781 | ||
119c0d44 TT |
782 | /* |
783 | * Normally we will only go through one pass of this loop, | |
784 | * unless we get unlucky and it turns out the group we selected | |
785 | * had its last inode grabbed by someone else. | |
786 | */ | |
11013911 | 787 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
788 | err = -EIO; |
789 | ||
3300beda | 790 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec | 791 | if (!gdp) |
eb9cc7e1 | 792 | goto out; |
ac27a0ec | 793 | |
f2a09af6 YY |
794 | /* |
795 | * Check free inodes count before loading bitmap. | |
796 | */ | |
797 | if (ext4_free_inodes_count(sb, gdp) == 0) { | |
798 | if (++group == ngroups) | |
799 | group = 0; | |
800 | continue; | |
801 | } | |
802 | ||
87a39389 DW |
803 | grp = ext4_get_group_info(sb, group); |
804 | /* Skip groups with already-known suspicious inode tables */ | |
805 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { | |
806 | if (++group == ngroups) | |
807 | group = 0; | |
808 | continue; | |
809 | } | |
810 | ||
3300beda AK |
811 | brelse(inode_bitmap_bh); |
812 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
87a39389 DW |
813 | /* Skip groups with suspicious inode tables */ |
814 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || !inode_bitmap_bh) { | |
815 | if (++group == ngroups) | |
816 | group = 0; | |
817 | continue; | |
818 | } | |
ac27a0ec | 819 | |
ac27a0ec | 820 | repeat_in_this_group: |
617ba13b | 821 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
822 | inode_bitmap_bh->b_data, |
823 | EXT4_INODES_PER_GROUP(sb), ino); | |
a34eb503 TT |
824 | if (ino >= EXT4_INODES_PER_GROUP(sb)) |
825 | goto next_group; | |
119c0d44 TT |
826 | if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) { |
827 | ext4_error(sb, "reserved inode found cleared - " | |
828 | "inode=%lu", ino + 1); | |
829 | continue; | |
830 | } | |
19883bd9 TT |
831 | if ((EXT4_SB(sb)->s_journal == NULL) && |
832 | recently_deleted(sb, group, ino)) { | |
833 | ino++; | |
834 | goto next_inode; | |
835 | } | |
1139575a TT |
836 | if (!handle) { |
837 | BUG_ON(nblocks <= 0); | |
838 | handle = __ext4_journal_start_sb(dir->i_sb, line_no, | |
5fe2fe89 JK |
839 | handle_type, nblocks, |
840 | 0); | |
1139575a TT |
841 | if (IS_ERR(handle)) { |
842 | err = PTR_ERR(handle); | |
eb9cc7e1 JK |
843 | ext4_std_error(sb, err); |
844 | goto out; | |
1139575a TT |
845 | } |
846 | } | |
ffb5387e ES |
847 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
848 | err = ext4_journal_get_write_access(handle, inode_bitmap_bh); | |
eb9cc7e1 JK |
849 | if (err) { |
850 | ext4_std_error(sb, err); | |
851 | goto out; | |
852 | } | |
119c0d44 TT |
853 | ext4_lock_group(sb, group); |
854 | ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); | |
855 | ext4_unlock_group(sb, group); | |
856 | ino++; /* the inode bitmap is zero-based */ | |
857 | if (!ret2) | |
858 | goto got; /* we grabbed the inode! */ | |
19883bd9 | 859 | next_inode: |
119c0d44 TT |
860 | if (ino < EXT4_INODES_PER_GROUP(sb)) |
861 | goto repeat_in_this_group; | |
a34eb503 TT |
862 | next_group: |
863 | if (++group == ngroups) | |
864 | group = 0; | |
ac27a0ec DK |
865 | } |
866 | err = -ENOSPC; | |
867 | goto out; | |
868 | ||
869 | got: | |
ffb5387e ES |
870 | BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); |
871 | err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); | |
eb9cc7e1 JK |
872 | if (err) { |
873 | ext4_std_error(sb, err); | |
874 | goto out; | |
875 | } | |
ffb5387e | 876 | |
61c219f5 TT |
877 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
878 | err = ext4_journal_get_write_access(handle, group_desc_bh); | |
879 | if (err) { | |
880 | ext4_std_error(sb, err); | |
881 | goto out; | |
882 | } | |
883 | ||
717d50e4 | 884 | /* We may have to initialize the block bitmap if it isn't already */ |
feb0ab32 | 885 | if (ext4_has_group_desc_csum(sb) && |
717d50e4 | 886 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 887 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 888 | |
3300beda AK |
889 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
890 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); | |
891 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 892 | if (err) { |
3300beda | 893 | brelse(block_bitmap_bh); |
eb9cc7e1 JK |
894 | ext4_std_error(sb, err); |
895 | goto out; | |
717d50e4 AD |
896 | } |
897 | ||
fd034a84 TT |
898 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
899 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
fd034a84 | 900 | |
717d50e4 | 901 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 902 | ext4_lock_group(sb, group); |
717d50e4 | 903 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 904 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 905 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 906 | ext4_free_clusters_after_init(sb, group, gdp)); |
fa77dcfa | 907 | ext4_block_bitmap_csum_set(sb, group, gdp, |
79f1ba49 | 908 | block_bitmap_bh); |
feb0ab32 | 909 | ext4_group_desc_csum_set(sb, group, gdp); |
717d50e4 | 910 | } |
955ce5f5 | 911 | ext4_unlock_group(sb, group); |
aeb1e5d6 | 912 | brelse(block_bitmap_bh); |
717d50e4 | 913 | |
eb9cc7e1 JK |
914 | if (err) { |
915 | ext4_std_error(sb, err); | |
916 | goto out; | |
917 | } | |
717d50e4 | 918 | } |
119c0d44 | 919 | |
119c0d44 | 920 | /* Update the relevant bg descriptor fields */ |
41a246d1 | 921 | if (ext4_has_group_desc_csum(sb)) { |
119c0d44 TT |
922 | int free; |
923 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
924 | ||
925 | down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ | |
926 | ext4_lock_group(sb, group); /* while we modify the bg desc */ | |
927 | free = EXT4_INODES_PER_GROUP(sb) - | |
928 | ext4_itable_unused_count(sb, gdp); | |
929 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
930 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
931 | free = 0; | |
932 | } | |
933 | /* | |
934 | * Check the relative inode number against the last used | |
935 | * relative inode number in this group. if it is greater | |
936 | * we need to update the bg_itable_unused count | |
937 | */ | |
938 | if (ino > free) | |
939 | ext4_itable_unused_set(sb, gdp, | |
940 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
941 | up_read(&grp->alloc_sem); | |
6f2e9f0e TM |
942 | } else { |
943 | ext4_lock_group(sb, group); | |
119c0d44 | 944 | } |
6f2e9f0e | 945 | |
119c0d44 TT |
946 | ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); |
947 | if (S_ISDIR(mode)) { | |
948 | ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); | |
949 | if (sbi->s_log_groups_per_flex) { | |
950 | ext4_group_t f = ext4_flex_group(sbi, group); | |
951 | ||
952 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); | |
953 | } | |
954 | } | |
41a246d1 DW |
955 | if (ext4_has_group_desc_csum(sb)) { |
956 | ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, | |
957 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 958 | ext4_group_desc_csum_set(sb, group, gdp); |
119c0d44 | 959 | } |
6f2e9f0e | 960 | ext4_unlock_group(sb, group); |
119c0d44 | 961 | |
3300beda AK |
962 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
963 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
eb9cc7e1 JK |
964 | if (err) { |
965 | ext4_std_error(sb, err); | |
966 | goto out; | |
967 | } | |
ac27a0ec DK |
968 | |
969 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
970 | if (S_ISDIR(mode)) | |
971 | percpu_counter_inc(&sbi->s_dirs_counter); | |
ac27a0ec | 972 | |
772cb7c8 JS |
973 | if (sbi->s_log_groups_per_flex) { |
974 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 975 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 | 976 | } |
ac27a0ec | 977 | |
717d50e4 | 978 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
979 | /* This is the optimal IO size (for stat), not the fs block size */ |
980 | inode->i_blocks = 0; | |
ef7f3835 KS |
981 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
982 | ext4_current_time(inode); | |
ac27a0ec DK |
983 | |
984 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
985 | ei->i_dir_start_lookup = 0; | |
986 | ei->i_disksize = 0; | |
987 | ||
4af83508 | 988 | /* Don't inherit extent flag from directory, amongst others. */ |
2dc6b0d4 DG |
989 | ei->i_flags = |
990 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 991 | ei->i_file_acl = 0; |
ac27a0ec | 992 | ei->i_dtime = 0; |
ac27a0ec | 993 | ei->i_block_group = group; |
a4912123 | 994 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 995 | |
617ba13b | 996 | ext4_set_inode_flags(inode); |
ac27a0ec | 997 | if (IS_DIRSYNC(inode)) |
0390131b | 998 | ext4_handle_sync(handle); |
6b38e842 | 999 | if (insert_inode_locked(inode) < 0) { |
acd6ad83 JK |
1000 | /* |
1001 | * Likely a bitmap corruption causing inode to be allocated | |
1002 | * twice. | |
1003 | */ | |
1004 | err = -EIO; | |
eb9cc7e1 JK |
1005 | ext4_error(sb, "failed to insert inode %lu: doubly allocated?", |
1006 | inode->i_ino); | |
1007 | goto out; | |
6b38e842 | 1008 | } |
ac27a0ec DK |
1009 | spin_lock(&sbi->s_next_gen_lock); |
1010 | inode->i_generation = sbi->s_next_generation++; | |
1011 | spin_unlock(&sbi->s_next_gen_lock); | |
1012 | ||
814525f4 DW |
1013 | /* Precompute checksum seed for inode metadata */ |
1014 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
1015 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { | |
1016 | __u32 csum; | |
814525f4 DW |
1017 | __le32 inum = cpu_to_le32(inode->i_ino); |
1018 | __le32 gen = cpu_to_le32(inode->i_generation); | |
1019 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
1020 | sizeof(inum)); | |
1021 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
1022 | sizeof(gen)); | |
1023 | } | |
1024 | ||
353eb83c | 1025 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 1026 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 KS |
1027 | |
1028 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec | 1029 | |
f08225d1 TM |
1030 | ei->i_inline_off = 0; |
1031 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA)) | |
1032 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
1033 | ||
ac27a0ec | 1034 | ret = inode; |
63936dda CH |
1035 | err = dquot_alloc_inode(inode); |
1036 | if (err) | |
ac27a0ec | 1037 | goto fail_drop; |
ac27a0ec | 1038 | |
617ba13b | 1039 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
1040 | if (err) |
1041 | goto fail_free_drop; | |
1042 | ||
2a7dba39 | 1043 | err = ext4_init_security(handle, inode, dir, qstr); |
ac27a0ec DK |
1044 | if (err) |
1045 | goto fail_free_drop; | |
1046 | ||
83982b6f | 1047 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
e4079a11 | 1048 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1049 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1050 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1051 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1052 | } |
a86c6181 | 1053 | } |
ac27a0ec | 1054 | |
688f869c TT |
1055 | if (ext4_handle_valid(handle)) { |
1056 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
1057 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
1058 | } | |
1059 | ||
8753e88f AK |
1060 | err = ext4_mark_inode_dirty(handle, inode); |
1061 | if (err) { | |
1062 | ext4_std_error(sb, err); | |
1063 | goto fail_free_drop; | |
1064 | } | |
1065 | ||
617ba13b | 1066 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1067 | trace_ext4_allocate_inode(inode, dir, mode); |
3300beda | 1068 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1069 | return ret; |
1070 | ||
1071 | fail_free_drop: | |
63936dda | 1072 | dquot_free_inode(inode); |
ac27a0ec | 1073 | fail_drop: |
6d6b77f1 | 1074 | clear_nlink(inode); |
6b38e842 | 1075 | unlock_new_inode(inode); |
eb9cc7e1 JK |
1076 | out: |
1077 | dquot_drop(inode); | |
1078 | inode->i_flags |= S_NOQUOTA; | |
ac27a0ec | 1079 | iput(inode); |
3300beda | 1080 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1081 | return ERR_PTR(err); |
1082 | } | |
1083 | ||
1084 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1085 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1086 | { |
617ba13b | 1087 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1088 | ext4_group_t block_group; |
ac27a0ec | 1089 | int bit; |
1d1fe1ee | 1090 | struct buffer_head *bitmap_bh; |
ac27a0ec | 1091 | struct inode *inode = NULL; |
1d1fe1ee | 1092 | long err = -EIO; |
ac27a0ec DK |
1093 | |
1094 | /* Error cases - e2fsck has already cleaned up for us */ | |
1095 | if (ino > max_ino) { | |
12062ddd | 1096 | ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 1097 | goto error; |
ac27a0ec DK |
1098 | } |
1099 | ||
617ba13b MC |
1100 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1101 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1102 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 1103 | if (!bitmap_bh) { |
12062ddd | 1104 | ext4_warning(sb, "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 1105 | goto error; |
ac27a0ec DK |
1106 | } |
1107 | ||
1108 | /* Having the inode bit set should be a 100% indicator that this | |
1109 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1110 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1111 | */ | |
1d1fe1ee DH |
1112 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1113 | goto bad_orphan; | |
1114 | ||
1115 | inode = ext4_iget(sb, ino); | |
1116 | if (IS_ERR(inode)) | |
1117 | goto iget_failed; | |
1118 | ||
91ef4caf DG |
1119 | /* |
1120 | * If the orphans has i_nlinks > 0 then it should be able to be | |
1121 | * truncated, otherwise it won't be removed from the orphan list | |
1122 | * during processing and an infinite loop will result. | |
1123 | */ | |
1124 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
1125 | goto bad_orphan; | |
1126 | ||
1d1fe1ee DH |
1127 | if (NEXT_ORPHAN(inode) > max_ino) |
1128 | goto bad_orphan; | |
1129 | brelse(bitmap_bh); | |
1130 | return inode; | |
1131 | ||
1132 | iget_failed: | |
1133 | err = PTR_ERR(inode); | |
1134 | inode = NULL; | |
1135 | bad_orphan: | |
12062ddd | 1136 | ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino); |
8de5c325 | 1137 | printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n", |
1d1fe1ee DH |
1138 | bit, (unsigned long long)bitmap_bh->b_blocknr, |
1139 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
8de5c325 | 1140 | printk(KERN_WARNING "inode=%p\n", inode); |
1d1fe1ee | 1141 | if (inode) { |
8de5c325 | 1142 | printk(KERN_WARNING "is_bad_inode(inode)=%d\n", |
1d1fe1ee | 1143 | is_bad_inode(inode)); |
8de5c325 | 1144 | printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n", |
1d1fe1ee | 1145 | NEXT_ORPHAN(inode)); |
8de5c325 TT |
1146 | printk(KERN_WARNING "max_ino=%lu\n", max_ino); |
1147 | printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink); | |
ac27a0ec | 1148 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1149 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1150 | inode->i_blocks = 0; |
1151 | iput(inode); | |
ac27a0ec | 1152 | } |
ac27a0ec | 1153 | brelse(bitmap_bh); |
1d1fe1ee DH |
1154 | error: |
1155 | return ERR_PTR(err); | |
ac27a0ec DK |
1156 | } |
1157 | ||
af5bc92d | 1158 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1159 | { |
1160 | unsigned long desc_count; | |
617ba13b | 1161 | struct ext4_group_desc *gdp; |
8df9675f | 1162 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1163 | #ifdef EXT4FS_DEBUG |
1164 | struct ext4_super_block *es; | |
ac27a0ec DK |
1165 | unsigned long bitmap_count, x; |
1166 | struct buffer_head *bitmap_bh = NULL; | |
1167 | ||
617ba13b | 1168 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1169 | desc_count = 0; |
1170 | bitmap_count = 0; | |
1171 | gdp = NULL; | |
8df9675f | 1172 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1173 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1174 | if (!gdp) |
1175 | continue; | |
560671a0 | 1176 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1177 | brelse(bitmap_bh); |
e29d1cde | 1178 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
1179 | if (!bitmap_bh) |
1180 | continue; | |
1181 | ||
f6fb99ca TT |
1182 | x = ext4_count_free(bitmap_bh->b_data, |
1183 | EXT4_INODES_PER_GROUP(sb) / 8); | |
c549a95d | 1184 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1185 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1186 | bitmap_count += x; |
1187 | } | |
1188 | brelse(bitmap_bh); | |
4776004f TT |
1189 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1190 | "stored = %u, computed = %lu, %lu\n", | |
1191 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1192 | return desc_count; |
1193 | #else | |
1194 | desc_count = 0; | |
8df9675f | 1195 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1196 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1197 | if (!gdp) |
1198 | continue; | |
560671a0 | 1199 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1200 | cond_resched(); |
1201 | } | |
1202 | return desc_count; | |
1203 | #endif | |
1204 | } | |
1205 | ||
1206 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1207 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1208 | { |
1209 | unsigned long count = 0; | |
8df9675f | 1210 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1211 | |
8df9675f | 1212 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1213 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1214 | if (!gdp) |
1215 | continue; | |
560671a0 | 1216 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1217 | } |
1218 | return count; | |
1219 | } | |
bfff6873 LC |
1220 | |
1221 | /* | |
1222 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1223 | * inode table. Must be called without any spinlock held. The only place | |
1224 | * where it is called from on active part of filesystem is ext4lazyinit | |
1225 | * thread, so we do not need any special locks, however we have to prevent | |
1226 | * inode allocation from the current group, so we take alloc_sem lock, to | |
119c0d44 | 1227 | * block ext4_new_inode() until we are finished. |
bfff6873 | 1228 | */ |
e0cbee3e | 1229 | int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, |
bfff6873 LC |
1230 | int barrier) |
1231 | { | |
1232 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1233 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1234 | struct ext4_group_desc *gdp = NULL; | |
1235 | struct buffer_head *group_desc_bh; | |
1236 | handle_t *handle; | |
1237 | ext4_fsblk_t blk; | |
1238 | int num, ret = 0, used_blks = 0; | |
bfff6873 LC |
1239 | |
1240 | /* This should not happen, but just to be sure check this */ | |
1241 | if (sb->s_flags & MS_RDONLY) { | |
1242 | ret = 1; | |
1243 | goto out; | |
1244 | } | |
1245 | ||
1246 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1247 | if (!gdp) | |
1248 | goto out; | |
1249 | ||
1250 | /* | |
1251 | * We do not need to lock this, because we are the only one | |
1252 | * handling this flag. | |
1253 | */ | |
1254 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1255 | goto out; | |
1256 | ||
9924a92a | 1257 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
bfff6873 LC |
1258 | if (IS_ERR(handle)) { |
1259 | ret = PTR_ERR(handle); | |
1260 | goto out; | |
1261 | } | |
1262 | ||
1263 | down_write(&grp->alloc_sem); | |
1264 | /* | |
1265 | * If inode bitmap was already initialized there may be some | |
1266 | * used inodes so we need to skip blocks with used inodes in | |
1267 | * inode table. | |
1268 | */ | |
1269 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1270 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1271 | ext4_itable_unused_count(sb, gdp)), | |
1272 | sbi->s_inodes_per_block); | |
1273 | ||
857ac889 | 1274 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) { |
1084f252 TT |
1275 | ext4_error(sb, "Something is wrong with group %u: " |
1276 | "used itable blocks: %d; " | |
1277 | "itable unused count: %u", | |
857ac889 LC |
1278 | group, used_blks, |
1279 | ext4_itable_unused_count(sb, gdp)); | |
1280 | ret = 1; | |
33853a0d | 1281 | goto err_out; |
857ac889 LC |
1282 | } |
1283 | ||
bfff6873 LC |
1284 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1285 | num = sbi->s_itb_per_group - used_blks; | |
1286 | ||
1287 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1288 | ret = ext4_journal_get_write_access(handle, | |
1289 | group_desc_bh); | |
1290 | if (ret) | |
1291 | goto err_out; | |
1292 | ||
bfff6873 LC |
1293 | /* |
1294 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1295 | * flag anyway, because obviously, when it is full it does not need | |
1296 | * further zeroing. | |
1297 | */ | |
1298 | if (unlikely(num == 0)) | |
1299 | goto skip_zeroout; | |
1300 | ||
1301 | ext4_debug("going to zero out inode table in group %d\n", | |
1302 | group); | |
a107e5a3 | 1303 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1304 | if (ret < 0) |
1305 | goto err_out; | |
a107e5a3 TT |
1306 | if (barrier) |
1307 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); | |
bfff6873 LC |
1308 | |
1309 | skip_zeroout: | |
1310 | ext4_lock_group(sb, group); | |
1311 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
feb0ab32 | 1312 | ext4_group_desc_csum_set(sb, group, gdp); |
bfff6873 LC |
1313 | ext4_unlock_group(sb, group); |
1314 | ||
1315 | BUFFER_TRACE(group_desc_bh, | |
1316 | "call ext4_handle_dirty_metadata"); | |
1317 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1318 | group_desc_bh); | |
1319 | ||
1320 | err_out: | |
1321 | up_write(&grp->alloc_sem); | |
1322 | ext4_journal_stop(handle); | |
1323 | out: | |
1324 | return ret; | |
1325 | } |