Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/ialloc.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * BSD ufs-inspired inode and directory allocation by | |
10 | * Stephen Tweedie (sct@redhat.com), 1993 | |
11 | * Big-endian to little-endian byte-swapping/bitmaps by | |
12 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
13 | */ | |
14 | ||
15 | #include <linux/time.h> | |
16 | #include <linux/fs.h> | |
ac27a0ec DK |
17 | #include <linux/stat.h> |
18 | #include <linux/string.h> | |
19 | #include <linux/quotaops.h> | |
20 | #include <linux/buffer_head.h> | |
21 | #include <linux/random.h> | |
22 | #include <linux/bitops.h> | |
3a5b2ecd | 23 | #include <linux/blkdev.h> |
ac27a0ec | 24 | #include <asm/byteorder.h> |
9bffad1e | 25 | |
3dcf5451 CH |
26 | #include "ext4.h" |
27 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
28 | #include "xattr.h" |
29 | #include "acl.h" | |
30 | ||
9bffad1e TT |
31 | #include <trace/events/ext4.h> |
32 | ||
ac27a0ec DK |
33 | /* |
34 | * ialloc.c contains the inodes allocation and deallocation routines | |
35 | */ | |
36 | ||
37 | /* | |
38 | * The free inodes are managed by bitmaps. A file system contains several | |
39 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
40 | * block for inodes, N blocks for the inode table and data blocks. | |
41 | * | |
42 | * The file system contains group descriptors which are located after the | |
43 | * super block. Each descriptor contains the number of the bitmap block and | |
44 | * the free blocks count in the block. | |
45 | */ | |
46 | ||
717d50e4 AD |
47 | /* |
48 | * To avoid calling the atomic setbit hundreds or thousands of times, we only | |
49 | * need to use it within a single byte (to ensure we get endianness right). | |
50 | * We can use memset for the rest of the bitmap as there are no other users. | |
51 | */ | |
61d08673 | 52 | void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap) |
717d50e4 AD |
53 | { |
54 | int i; | |
55 | ||
56 | if (start_bit >= end_bit) | |
57 | return; | |
58 | ||
59 | ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); | |
60 | for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) | |
61 | ext4_set_bit(i, bitmap); | |
62 | if (i < end_bit) | |
63 | memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); | |
64 | } | |
65 | ||
66 | /* Initializes an uninitialized inode bitmap */ | |
1f109d5a TT |
67 | static unsigned ext4_init_inode_bitmap(struct super_block *sb, |
68 | struct buffer_head *bh, | |
69 | ext4_group_t block_group, | |
70 | struct ext4_group_desc *gdp) | |
717d50e4 | 71 | { |
bdfb6ff4 | 72 | struct ext4_group_info *grp; |
e43bb4e6 | 73 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
717d50e4 AD |
74 | J_ASSERT_BH(bh, buffer_locked(bh)); |
75 | ||
76 | /* If checksum is bad mark all blocks and inodes use to prevent | |
77 | * allocation, essentially implementing a per-group read-only flag. */ | |
feb0ab32 | 78 | if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) { |
12062ddd | 79 | ext4_error(sb, "Checksum bad for group %u", block_group); |
bdfb6ff4 | 80 | grp = ext4_get_group_info(sb, block_group); |
e43bb4e6 NJ |
81 | if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp)) |
82 | percpu_counter_sub(&sbi->s_freeclusters_counter, | |
83 | grp->bb_free); | |
bdfb6ff4 | 84 | set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state); |
e43bb4e6 NJ |
85 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
86 | int count; | |
87 | count = ext4_free_inodes_count(sb, gdp); | |
88 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
89 | count); | |
90 | } | |
bdfb6ff4 | 91 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
717d50e4 AD |
92 | return 0; |
93 | } | |
94 | ||
95 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
61d08673 | 96 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
717d50e4 | 97 | bh->b_data); |
41a246d1 DW |
98 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh, |
99 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 100 | ext4_group_desc_csum_set(sb, block_group, gdp); |
717d50e4 AD |
101 | |
102 | return EXT4_INODES_PER_GROUP(sb); | |
103 | } | |
ac27a0ec | 104 | |
813e5727 TT |
105 | void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate) |
106 | { | |
107 | if (uptodate) { | |
108 | set_buffer_uptodate(bh); | |
109 | set_bitmap_uptodate(bh); | |
110 | } | |
111 | unlock_buffer(bh); | |
112 | put_bh(bh); | |
113 | } | |
114 | ||
ac27a0ec DK |
115 | /* |
116 | * Read the inode allocation bitmap for a given block_group, reading | |
117 | * into the specified slot in the superblock's bitmap cache. | |
118 | * | |
119 | * Return buffer_head of bitmap on success or NULL. | |
120 | */ | |
121 | static struct buffer_head * | |
e29d1cde | 122 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 123 | { |
617ba13b | 124 | struct ext4_group_desc *desc; |
ac27a0ec | 125 | struct buffer_head *bh = NULL; |
e29d1cde | 126 | ext4_fsblk_t bitmap_blk; |
87a39389 | 127 | struct ext4_group_info *grp; |
e43bb4e6 | 128 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
ac27a0ec | 129 | |
617ba13b | 130 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 131 | if (!desc) |
e29d1cde | 132 | return NULL; |
bfff6873 | 133 | |
e29d1cde ES |
134 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
135 | bh = sb_getblk(sb, bitmap_blk); | |
136 | if (unlikely(!bh)) { | |
12062ddd | 137 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 138 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
139 | block_group, bitmap_blk); |
140 | return NULL; | |
141 | } | |
2ccb5fb9 | 142 | if (bitmap_uptodate(bh)) |
41a246d1 | 143 | goto verify; |
e29d1cde | 144 | |
c806e68f | 145 | lock_buffer(bh); |
2ccb5fb9 AK |
146 | if (bitmap_uptodate(bh)) { |
147 | unlock_buffer(bh); | |
41a246d1 | 148 | goto verify; |
2ccb5fb9 | 149 | } |
bfff6873 | 150 | |
955ce5f5 | 151 | ext4_lock_group(sb, block_group); |
717d50e4 | 152 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
e29d1cde | 153 | ext4_init_inode_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 154 | set_bitmap_uptodate(bh); |
e29d1cde | 155 | set_buffer_uptodate(bh); |
41a246d1 | 156 | set_buffer_verified(bh); |
955ce5f5 | 157 | ext4_unlock_group(sb, block_group); |
3300beda | 158 | unlock_buffer(bh); |
e29d1cde | 159 | return bh; |
717d50e4 | 160 | } |
955ce5f5 | 161 | ext4_unlock_group(sb, block_group); |
bfff6873 | 162 | |
2ccb5fb9 AK |
163 | if (buffer_uptodate(bh)) { |
164 | /* | |
165 | * if not uninit if bh is uptodate, | |
166 | * bitmap is also uptodate | |
167 | */ | |
168 | set_bitmap_uptodate(bh); | |
169 | unlock_buffer(bh); | |
41a246d1 | 170 | goto verify; |
2ccb5fb9 AK |
171 | } |
172 | /* | |
813e5727 | 173 | * submit the buffer_head for reading |
2ccb5fb9 | 174 | */ |
0562e0ba | 175 | trace_ext4_load_inode_bitmap(sb, block_group); |
813e5727 TT |
176 | bh->b_end_io = ext4_end_bitmap_read; |
177 | get_bh(bh); | |
9f203507 | 178 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
813e5727 TT |
179 | wait_on_buffer(bh); |
180 | if (!buffer_uptodate(bh)) { | |
e29d1cde | 181 | put_bh(bh); |
12062ddd | 182 | ext4_error(sb, "Cannot read inode bitmap - " |
813e5727 TT |
183 | "block_group = %u, inode_bitmap = %llu", |
184 | block_group, bitmap_blk); | |
e29d1cde ES |
185 | return NULL; |
186 | } | |
41a246d1 DW |
187 | |
188 | verify: | |
189 | ext4_lock_group(sb, block_group); | |
190 | if (!buffer_verified(bh) && | |
191 | !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh, | |
192 | EXT4_INODES_PER_GROUP(sb) / 8)) { | |
193 | ext4_unlock_group(sb, block_group); | |
194 | put_bh(bh); | |
195 | ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " | |
196 | "inode_bitmap = %llu", block_group, bitmap_blk); | |
87a39389 | 197 | grp = ext4_get_group_info(sb, block_group); |
e43bb4e6 NJ |
198 | if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
199 | int count; | |
200 | count = ext4_free_inodes_count(sb, desc); | |
201 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
202 | count); | |
203 | } | |
87a39389 | 204 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
41a246d1 DW |
205 | return NULL; |
206 | } | |
207 | ext4_unlock_group(sb, block_group); | |
208 | set_buffer_verified(bh); | |
ac27a0ec DK |
209 | return bh; |
210 | } | |
211 | ||
212 | /* | |
213 | * NOTE! When we get the inode, we're the only people | |
214 | * that have access to it, and as such there are no | |
215 | * race conditions we have to worry about. The inode | |
216 | * is not on the hash-lists, and it cannot be reached | |
217 | * through the filesystem because the directory entry | |
218 | * has been deleted earlier. | |
219 | * | |
220 | * HOWEVER: we must make sure that we get no aliases, | |
221 | * which means that we have to call "clear_inode()" | |
222 | * _before_ we mark the inode not in use in the inode | |
223 | * bitmaps. Otherwise a newly created file might use | |
224 | * the same inode number (not actually the same pointer | |
225 | * though), and then we'd have two inodes sharing the | |
226 | * same inode number and space on the harddisk. | |
227 | */ | |
af5bc92d | 228 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 229 | { |
af5bc92d | 230 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
231 | int is_directory; |
232 | unsigned long ino; | |
233 | struct buffer_head *bitmap_bh = NULL; | |
234 | struct buffer_head *bh2; | |
fd2d4291 | 235 | ext4_group_t block_group; |
ac27a0ec | 236 | unsigned long bit; |
af5bc92d TT |
237 | struct ext4_group_desc *gdp; |
238 | struct ext4_super_block *es; | |
617ba13b | 239 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 240 | int fatal = 0, err, count, cleared; |
87a39389 | 241 | struct ext4_group_info *grp; |
ac27a0ec | 242 | |
92b97816 TT |
243 | if (!sb) { |
244 | printk(KERN_ERR "EXT4-fs: %s:%d: inode on " | |
245 | "nonexistent device\n", __func__, __LINE__); | |
ac27a0ec DK |
246 | return; |
247 | } | |
92b97816 TT |
248 | if (atomic_read(&inode->i_count) > 1) { |
249 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", | |
250 | __func__, __LINE__, inode->i_ino, | |
251 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
252 | return; |
253 | } | |
92b97816 TT |
254 | if (inode->i_nlink) { |
255 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", | |
256 | __func__, __LINE__, inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
257 | return; |
258 | } | |
617ba13b | 259 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
260 | |
261 | ino = inode->i_ino; | |
af5bc92d | 262 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 263 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
264 | |
265 | /* | |
266 | * Note: we must free any quota before locking the superblock, | |
267 | * as writing the quota to disk may need the lock as well. | |
268 | */ | |
871a2931 | 269 | dquot_initialize(inode); |
617ba13b | 270 | ext4_xattr_delete_inode(handle, inode); |
63936dda | 271 | dquot_free_inode(inode); |
9f754758 | 272 | dquot_drop(inode); |
ac27a0ec DK |
273 | |
274 | is_directory = S_ISDIR(inode->i_mode); | |
275 | ||
276 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 277 | ext4_clear_inode(inode); |
ac27a0ec | 278 | |
617ba13b MC |
279 | es = EXT4_SB(sb)->s_es; |
280 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 281 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
282 | goto error_return; |
283 | } | |
617ba13b MC |
284 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
285 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 286 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
87a39389 DW |
287 | /* Don't bother if the inode bitmap is corrupt. */ |
288 | grp = ext4_get_group_info(sb, block_group); | |
289 | if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) || !bitmap_bh) | |
ac27a0ec DK |
290 | goto error_return; |
291 | ||
292 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 293 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
294 | if (fatal) |
295 | goto error_return; | |
296 | ||
d17413c0 DM |
297 | fatal = -ESRCH; |
298 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
299 | if (gdp) { | |
ac27a0ec | 300 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 301 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
302 | } |
303 | ext4_lock_group(sb, block_group); | |
597d508c | 304 | cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); |
d17413c0 DM |
305 | if (fatal || !cleared) { |
306 | ext4_unlock_group(sb, block_group); | |
307 | goto out; | |
308 | } | |
7d39db14 | 309 | |
d17413c0 DM |
310 | count = ext4_free_inodes_count(sb, gdp) + 1; |
311 | ext4_free_inodes_set(sb, gdp, count); | |
312 | if (is_directory) { | |
313 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
314 | ext4_used_dirs_set(sb, gdp, count); | |
315 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 316 | } |
41a246d1 DW |
317 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, |
318 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 319 | ext4_group_desc_csum_set(sb, block_group, gdp); |
d17413c0 | 320 | ext4_unlock_group(sb, block_group); |
ac27a0ec | 321 | |
d17413c0 DM |
322 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
323 | if (sbi->s_log_groups_per_flex) { | |
324 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 325 | |
d17413c0 DM |
326 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
327 | if (is_directory) | |
328 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 329 | } |
d17413c0 DM |
330 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
331 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
332 | out: | |
333 | if (cleared) { | |
334 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
335 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
336 | if (!fatal) | |
337 | fatal = err; | |
87a39389 | 338 | } else { |
d17413c0 | 339 | ext4_error(sb, "bit already cleared for inode %lu", ino); |
bf40c926 | 340 | if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { |
e43bb4e6 NJ |
341 | int count; |
342 | count = ext4_free_inodes_count(sb, gdp); | |
343 | percpu_counter_sub(&sbi->s_freeinodes_counter, | |
344 | count); | |
345 | } | |
87a39389 DW |
346 | set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state); |
347 | } | |
d17413c0 | 348 | |
ac27a0ec DK |
349 | error_return: |
350 | brelse(bitmap_bh); | |
617ba13b | 351 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
352 | } |
353 | ||
a4912123 | 354 | struct orlov_stats { |
90ba983f | 355 | __u64 free_clusters; |
a4912123 | 356 | __u32 free_inodes; |
a4912123 TT |
357 | __u32 used_dirs; |
358 | }; | |
359 | ||
360 | /* | |
361 | * Helper function for Orlov's allocator; returns critical information | |
362 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
363 | * is a block group number; otherwise it is flex_bg number. | |
364 | */ | |
1f109d5a TT |
365 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
366 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
367 | { |
368 | struct ext4_group_desc *desc; | |
7d39db14 | 369 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 370 | |
7d39db14 TT |
371 | if (flex_size > 1) { |
372 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
90ba983f | 373 | stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); |
7d39db14 TT |
374 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); |
375 | return; | |
376 | } | |
a4912123 | 377 | |
7d39db14 TT |
378 | desc = ext4_get_group_desc(sb, g, NULL); |
379 | if (desc) { | |
380 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
021b65bb | 381 | stats->free_clusters = ext4_free_group_clusters(sb, desc); |
7d39db14 TT |
382 | stats->used_dirs = ext4_used_dirs_count(sb, desc); |
383 | } else { | |
384 | stats->free_inodes = 0; | |
24aaa8ef | 385 | stats->free_clusters = 0; |
7d39db14 | 386 | stats->used_dirs = 0; |
a4912123 TT |
387 | } |
388 | } | |
389 | ||
ac27a0ec DK |
390 | /* |
391 | * Orlov's allocator for directories. | |
392 | * | |
393 | * We always try to spread first-level directories. | |
394 | * | |
395 | * If there are blockgroups with both free inodes and free blocks counts | |
396 | * not worse than average we return one with smallest directory count. | |
397 | * Otherwise we simply return a random group. | |
398 | * | |
399 | * For the rest rules look so: | |
400 | * | |
401 | * It's OK to put directory into a group unless | |
402 | * it has too many directories already (max_dirs) or | |
403 | * it has too few free inodes left (min_inodes) or | |
404 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 405 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
406 | * conditions we search cyclically through the rest. If none |
407 | * of the groups look good we just look for a group with more | |
408 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
409 | */ |
410 | ||
2aa9fc4c | 411 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
dcca3fec | 412 | ext4_group_t *group, umode_t mode, |
f157a4aa | 413 | const struct qstr *qstr) |
ac27a0ec | 414 | { |
fd2d4291 | 415 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 416 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 417 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 418 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
14c83c9f | 419 | unsigned int freei, avefreei, grp_free; |
24aaa8ef | 420 | ext4_fsblk_t freeb, avefreec; |
ac27a0ec | 421 | unsigned int ndirs; |
a4912123 | 422 | int max_dirs, min_inodes; |
24aaa8ef | 423 | ext4_grpblk_t min_clusters; |
8df9675f | 424 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 425 | struct ext4_group_desc *desc; |
a4912123 TT |
426 | struct orlov_stats stats; |
427 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 428 | struct dx_hash_info hinfo; |
a4912123 | 429 | |
8df9675f | 430 | ngroups = real_ngroups; |
a4912123 | 431 | if (flex_size > 1) { |
8df9675f | 432 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
433 | sbi->s_log_groups_per_flex; |
434 | parent_group >>= sbi->s_log_groups_per_flex; | |
435 | } | |
ac27a0ec DK |
436 | |
437 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
438 | avefreei = freei / ngroups; | |
57042651 TT |
439 | freeb = EXT4_C2B(sbi, |
440 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
24aaa8ef TT |
441 | avefreec = freeb; |
442 | do_div(avefreec, ngroups); | |
ac27a0ec DK |
443 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
444 | ||
a4912123 | 445 | if (S_ISDIR(mode) && |
2b0143b5 | 446 | ((parent == d_inode(sb->s_root)) || |
12e9b892 | 447 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 448 | int best_ndir = inodes_per_group; |
2aa9fc4c | 449 | int ret = -1; |
ac27a0ec | 450 | |
f157a4aa TT |
451 | if (qstr) { |
452 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
453 | hinfo.seed = sbi->s_hash_seed; | |
454 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
455 | grp = hinfo.hash; | |
456 | } else | |
dd1f723b | 457 | grp = prandom_u32(); |
2aa9fc4c | 458 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 459 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
460 | g = (parent_group + i) % ngroups; |
461 | get_orlov_stats(sb, g, flex_size, &stats); | |
462 | if (!stats.free_inodes) | |
ac27a0ec | 463 | continue; |
a4912123 | 464 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 465 | continue; |
a4912123 | 466 | if (stats.free_inodes < avefreei) |
ac27a0ec | 467 | continue; |
24aaa8ef | 468 | if (stats.free_clusters < avefreec) |
ac27a0ec | 469 | continue; |
a4912123 | 470 | grp = g; |
2aa9fc4c | 471 | ret = 0; |
a4912123 TT |
472 | best_ndir = stats.used_dirs; |
473 | } | |
474 | if (ret) | |
475 | goto fallback; | |
476 | found_flex_bg: | |
477 | if (flex_size == 1) { | |
478 | *group = grp; | |
479 | return 0; | |
480 | } | |
481 | ||
482 | /* | |
483 | * We pack inodes at the beginning of the flexgroup's | |
484 | * inode tables. Block allocation decisions will do | |
485 | * something similar, although regular files will | |
486 | * start at 2nd block group of the flexgroup. See | |
487 | * ext4_ext_find_goal() and ext4_find_near(). | |
488 | */ | |
489 | grp *= flex_size; | |
490 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 491 | if (grp+i >= real_ngroups) |
a4912123 TT |
492 | break; |
493 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
494 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
495 | *group = grp+i; | |
496 | return 0; | |
497 | } | |
ac27a0ec | 498 | } |
ac27a0ec DK |
499 | goto fallback; |
500 | } | |
501 | ||
ac27a0ec | 502 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
503 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
504 | if (min_inodes < 1) | |
505 | min_inodes = 1; | |
24aaa8ef | 506 | min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; |
a4912123 TT |
507 | |
508 | /* | |
509 | * Start looking in the flex group where we last allocated an | |
510 | * inode for this parent directory | |
511 | */ | |
512 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
513 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
514 | if (flex_size > 1) | |
515 | parent_group >>= sbi->s_log_groups_per_flex; | |
516 | } | |
ac27a0ec DK |
517 | |
518 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
519 | grp = (parent_group + i) % ngroups; |
520 | get_orlov_stats(sb, grp, flex_size, &stats); | |
521 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 522 | continue; |
a4912123 | 523 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 524 | continue; |
24aaa8ef | 525 | if (stats.free_clusters < min_clusters) |
ac27a0ec | 526 | continue; |
a4912123 | 527 | goto found_flex_bg; |
ac27a0ec DK |
528 | } |
529 | ||
530 | fallback: | |
8df9675f | 531 | ngroups = real_ngroups; |
a4912123 | 532 | avefreei = freei / ngroups; |
b5451f7b | 533 | fallback_retry: |
a4912123 | 534 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 535 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
536 | grp = (parent_group + i) % ngroups; |
537 | desc = ext4_get_group_desc(sb, grp, NULL); | |
bb3d132a DC |
538 | if (desc) { |
539 | grp_free = ext4_free_inodes_count(sb, desc); | |
540 | if (grp_free && grp_free >= avefreei) { | |
541 | *group = grp; | |
542 | return 0; | |
543 | } | |
a4912123 | 544 | } |
ac27a0ec DK |
545 | } |
546 | ||
547 | if (avefreei) { | |
548 | /* | |
549 | * The free-inodes counter is approximate, and for really small | |
550 | * filesystems the above test can fail to find any blockgroups | |
551 | */ | |
552 | avefreei = 0; | |
b5451f7b | 553 | goto fallback_retry; |
ac27a0ec DK |
554 | } |
555 | ||
556 | return -1; | |
557 | } | |
558 | ||
2aa9fc4c | 559 | static int find_group_other(struct super_block *sb, struct inode *parent, |
dcca3fec | 560 | ext4_group_t *group, umode_t mode) |
ac27a0ec | 561 | { |
fd2d4291 | 562 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 563 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 564 | struct ext4_group_desc *desc; |
a4912123 TT |
565 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
566 | ||
567 | /* | |
568 | * Try to place the inode is the same flex group as its | |
569 | * parent. If we can't find space, use the Orlov algorithm to | |
570 | * find another flex group, and store that information in the | |
571 | * parent directory's inode information so that use that flex | |
572 | * group for future allocations. | |
573 | */ | |
574 | if (flex_size > 1) { | |
575 | int retry = 0; | |
576 | ||
577 | try_again: | |
578 | parent_group &= ~(flex_size-1); | |
579 | last = parent_group + flex_size; | |
580 | if (last > ngroups) | |
581 | last = ngroups; | |
582 | for (i = parent_group; i < last; i++) { | |
583 | desc = ext4_get_group_desc(sb, i, NULL); | |
584 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
585 | *group = i; | |
586 | return 0; | |
587 | } | |
588 | } | |
589 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
590 | retry = 1; | |
591 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
592 | goto try_again; | |
593 | } | |
594 | /* | |
595 | * If this didn't work, use the Orlov search algorithm | |
596 | * to find a new flex group; we pass in the mode to | |
597 | * avoid the topdir algorithms. | |
598 | */ | |
599 | *group = parent_group + flex_size; | |
600 | if (*group > ngroups) | |
601 | *group = 0; | |
7dc57615 | 602 | return find_group_orlov(sb, parent, group, mode, NULL); |
a4912123 | 603 | } |
ac27a0ec DK |
604 | |
605 | /* | |
606 | * Try to place the inode in its parent directory | |
607 | */ | |
2aa9fc4c AM |
608 | *group = parent_group; |
609 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 610 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 611 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 612 | return 0; |
ac27a0ec DK |
613 | |
614 | /* | |
615 | * We're going to place this inode in a different blockgroup from its | |
616 | * parent. We want to cause files in a common directory to all land in | |
617 | * the same blockgroup. But we want files which are in a different | |
618 | * directory which shares a blockgroup with our parent to land in a | |
619 | * different blockgroup. | |
620 | * | |
621 | * So add our directory's i_ino into the starting point for the hash. | |
622 | */ | |
2aa9fc4c | 623 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
624 | |
625 | /* | |
626 | * Use a quadratic hash to find a group with a free inode and some free | |
627 | * blocks. | |
628 | */ | |
629 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
630 | *group += i; |
631 | if (*group >= ngroups) | |
632 | *group -= ngroups; | |
633 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 634 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 635 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 636 | return 0; |
ac27a0ec DK |
637 | } |
638 | ||
639 | /* | |
640 | * That failed: try linear search for a free inode, even if that group | |
641 | * has no free blocks. | |
642 | */ | |
2aa9fc4c | 643 | *group = parent_group; |
ac27a0ec | 644 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
645 | if (++*group >= ngroups) |
646 | *group = 0; | |
647 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 648 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 649 | return 0; |
ac27a0ec DK |
650 | } |
651 | ||
652 | return -1; | |
653 | } | |
654 | ||
19883bd9 TT |
655 | /* |
656 | * In no journal mode, if an inode has recently been deleted, we want | |
657 | * to avoid reusing it until we're reasonably sure the inode table | |
658 | * block has been written back to disk. (Yes, these values are | |
659 | * somewhat arbitrary...) | |
660 | */ | |
661 | #define RECENTCY_MIN 5 | |
662 | #define RECENTCY_DIRTY 30 | |
663 | ||
664 | static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) | |
665 | { | |
666 | struct ext4_group_desc *gdp; | |
667 | struct ext4_inode *raw_inode; | |
668 | struct buffer_head *bh; | |
669 | unsigned long dtime, now; | |
670 | int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
671 | int offset, ret = 0, recentcy = RECENTCY_MIN; | |
672 | ||
673 | gdp = ext4_get_group_desc(sb, group, NULL); | |
674 | if (unlikely(!gdp)) | |
675 | return 0; | |
676 | ||
677 | bh = sb_getblk(sb, ext4_inode_table(sb, gdp) + | |
678 | (ino / inodes_per_block)); | |
679 | if (unlikely(!bh) || !buffer_uptodate(bh)) | |
680 | /* | |
681 | * If the block is not in the buffer cache, then it | |
682 | * must have been written out. | |
683 | */ | |
684 | goto out; | |
685 | ||
686 | offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
687 | raw_inode = (struct ext4_inode *) (bh->b_data + offset); | |
688 | dtime = le32_to_cpu(raw_inode->i_dtime); | |
689 | now = get_seconds(); | |
690 | if (buffer_dirty(bh)) | |
691 | recentcy += RECENTCY_DIRTY; | |
692 | ||
693 | if (dtime && (dtime < now) && (now < dtime + recentcy)) | |
694 | ret = 1; | |
695 | out: | |
696 | brelse(bh); | |
697 | return ret; | |
698 | } | |
699 | ||
ac27a0ec DK |
700 | /* |
701 | * There are two policies for allocating an inode. If the new inode is | |
702 | * a directory, then a forward search is made for a block group with both | |
703 | * free space and a low directory-to-inode ratio; if that fails, then of | |
704 | * the groups with above-average free space, that group with the fewest | |
705 | * directories already is chosen. | |
706 | * | |
707 | * For other inodes, search forward from the parent directory's block | |
708 | * group to find a free inode. | |
709 | */ | |
1139575a TT |
710 | struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, |
711 | umode_t mode, const struct qstr *qstr, | |
712 | __u32 goal, uid_t *owner, int handle_type, | |
713 | unsigned int line_no, int nblocks) | |
ac27a0ec DK |
714 | { |
715 | struct super_block *sb; | |
3300beda AK |
716 | struct buffer_head *inode_bitmap_bh = NULL; |
717 | struct buffer_head *group_desc_bh; | |
8df9675f | 718 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 719 | unsigned long ino = 0; |
af5bc92d TT |
720 | struct inode *inode; |
721 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
722 | struct ext4_inode_info *ei; |
723 | struct ext4_sb_info *sbi; | |
39341867 | 724 | int ret2, err = 0; |
ac27a0ec | 725 | struct inode *ret; |
2aa9fc4c | 726 | ext4_group_t i; |
772cb7c8 | 727 | ext4_group_t flex_group; |
87a39389 | 728 | struct ext4_group_info *grp; |
ac27a0ec DK |
729 | |
730 | /* Cannot create files in a deleted directory */ | |
731 | if (!dir || !dir->i_nlink) | |
732 | return ERR_PTR(-EPERM); | |
733 | ||
734 | sb = dir->i_sb; | |
8df9675f | 735 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 736 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
737 | inode = new_inode(sb); |
738 | if (!inode) | |
739 | return ERR_PTR(-ENOMEM); | |
617ba13b | 740 | ei = EXT4_I(inode); |
617ba13b | 741 | sbi = EXT4_SB(sb); |
772cb7c8 | 742 | |
eb9cc7e1 JK |
743 | /* |
744 | * Initalize owners and quota early so that we don't have to account | |
745 | * for quota initialization worst case in standard inode creating | |
746 | * transaction | |
747 | */ | |
748 | if (owner) { | |
749 | inode->i_mode = mode; | |
750 | i_uid_write(inode, owner[0]); | |
751 | i_gid_write(inode, owner[1]); | |
752 | } else if (test_opt(sb, GRPID)) { | |
753 | inode->i_mode = mode; | |
754 | inode->i_uid = current_fsuid(); | |
755 | inode->i_gid = dir->i_gid; | |
756 | } else | |
757 | inode_init_owner(inode, dir, mode); | |
758 | dquot_initialize(inode); | |
759 | ||
11013911 AD |
760 | if (!goal) |
761 | goal = sbi->s_inode_goal; | |
762 | ||
e6462869 | 763 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
764 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
765 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
766 | ret2 = 0; | |
767 | goto got_group; | |
768 | } | |
769 | ||
4113c4ca LC |
770 | if (S_ISDIR(mode)) |
771 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); | |
772 | else | |
a4912123 | 773 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 774 | |
772cb7c8 | 775 | got_group: |
a4912123 | 776 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 777 | err = -ENOSPC; |
2aa9fc4c | 778 | if (ret2 == -1) |
ac27a0ec DK |
779 | goto out; |
780 | ||
119c0d44 TT |
781 | /* |
782 | * Normally we will only go through one pass of this loop, | |
783 | * unless we get unlucky and it turns out the group we selected | |
784 | * had its last inode grabbed by someone else. | |
785 | */ | |
11013911 | 786 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
787 | err = -EIO; |
788 | ||
3300beda | 789 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec | 790 | if (!gdp) |
eb9cc7e1 | 791 | goto out; |
ac27a0ec | 792 | |
f2a09af6 YY |
793 | /* |
794 | * Check free inodes count before loading bitmap. | |
795 | */ | |
796 | if (ext4_free_inodes_count(sb, gdp) == 0) { | |
797 | if (++group == ngroups) | |
798 | group = 0; | |
799 | continue; | |
800 | } | |
801 | ||
87a39389 DW |
802 | grp = ext4_get_group_info(sb, group); |
803 | /* Skip groups with already-known suspicious inode tables */ | |
804 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) { | |
805 | if (++group == ngroups) | |
806 | group = 0; | |
807 | continue; | |
808 | } | |
809 | ||
3300beda AK |
810 | brelse(inode_bitmap_bh); |
811 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
87a39389 DW |
812 | /* Skip groups with suspicious inode tables */ |
813 | if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || !inode_bitmap_bh) { | |
814 | if (++group == ngroups) | |
815 | group = 0; | |
816 | continue; | |
817 | } | |
ac27a0ec | 818 | |
ac27a0ec | 819 | repeat_in_this_group: |
617ba13b | 820 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
821 | inode_bitmap_bh->b_data, |
822 | EXT4_INODES_PER_GROUP(sb), ino); | |
a34eb503 TT |
823 | if (ino >= EXT4_INODES_PER_GROUP(sb)) |
824 | goto next_group; | |
119c0d44 TT |
825 | if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) { |
826 | ext4_error(sb, "reserved inode found cleared - " | |
827 | "inode=%lu", ino + 1); | |
828 | continue; | |
829 | } | |
19883bd9 TT |
830 | if ((EXT4_SB(sb)->s_journal == NULL) && |
831 | recently_deleted(sb, group, ino)) { | |
832 | ino++; | |
833 | goto next_inode; | |
834 | } | |
1139575a TT |
835 | if (!handle) { |
836 | BUG_ON(nblocks <= 0); | |
837 | handle = __ext4_journal_start_sb(dir->i_sb, line_no, | |
5fe2fe89 JK |
838 | handle_type, nblocks, |
839 | 0); | |
1139575a TT |
840 | if (IS_ERR(handle)) { |
841 | err = PTR_ERR(handle); | |
eb9cc7e1 JK |
842 | ext4_std_error(sb, err); |
843 | goto out; | |
1139575a TT |
844 | } |
845 | } | |
ffb5387e ES |
846 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
847 | err = ext4_journal_get_write_access(handle, inode_bitmap_bh); | |
eb9cc7e1 JK |
848 | if (err) { |
849 | ext4_std_error(sb, err); | |
850 | goto out; | |
851 | } | |
119c0d44 TT |
852 | ext4_lock_group(sb, group); |
853 | ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); | |
854 | ext4_unlock_group(sb, group); | |
855 | ino++; /* the inode bitmap is zero-based */ | |
856 | if (!ret2) | |
857 | goto got; /* we grabbed the inode! */ | |
19883bd9 | 858 | next_inode: |
119c0d44 TT |
859 | if (ino < EXT4_INODES_PER_GROUP(sb)) |
860 | goto repeat_in_this_group; | |
a34eb503 TT |
861 | next_group: |
862 | if (++group == ngroups) | |
863 | group = 0; | |
ac27a0ec DK |
864 | } |
865 | err = -ENOSPC; | |
866 | goto out; | |
867 | ||
868 | got: | |
ffb5387e ES |
869 | BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); |
870 | err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); | |
eb9cc7e1 JK |
871 | if (err) { |
872 | ext4_std_error(sb, err); | |
873 | goto out; | |
874 | } | |
ffb5387e | 875 | |
61c219f5 TT |
876 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
877 | err = ext4_journal_get_write_access(handle, group_desc_bh); | |
878 | if (err) { | |
879 | ext4_std_error(sb, err); | |
880 | goto out; | |
881 | } | |
882 | ||
717d50e4 | 883 | /* We may have to initialize the block bitmap if it isn't already */ |
feb0ab32 | 884 | if (ext4_has_group_desc_csum(sb) && |
717d50e4 | 885 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 886 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 887 | |
3300beda | 888 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
599a9b77 JK |
889 | if (!block_bitmap_bh) { |
890 | err = -EIO; | |
891 | goto out; | |
892 | } | |
3300beda AK |
893 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); |
894 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 895 | if (err) { |
3300beda | 896 | brelse(block_bitmap_bh); |
eb9cc7e1 JK |
897 | ext4_std_error(sb, err); |
898 | goto out; | |
717d50e4 AD |
899 | } |
900 | ||
fd034a84 TT |
901 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
902 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
fd034a84 | 903 | |
717d50e4 | 904 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 905 | ext4_lock_group(sb, group); |
717d50e4 | 906 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 907 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 908 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 909 | ext4_free_clusters_after_init(sb, group, gdp)); |
fa77dcfa | 910 | ext4_block_bitmap_csum_set(sb, group, gdp, |
79f1ba49 | 911 | block_bitmap_bh); |
feb0ab32 | 912 | ext4_group_desc_csum_set(sb, group, gdp); |
717d50e4 | 913 | } |
955ce5f5 | 914 | ext4_unlock_group(sb, group); |
aeb1e5d6 | 915 | brelse(block_bitmap_bh); |
717d50e4 | 916 | |
eb9cc7e1 JK |
917 | if (err) { |
918 | ext4_std_error(sb, err); | |
919 | goto out; | |
920 | } | |
717d50e4 | 921 | } |
119c0d44 | 922 | |
119c0d44 | 923 | /* Update the relevant bg descriptor fields */ |
41a246d1 | 924 | if (ext4_has_group_desc_csum(sb)) { |
119c0d44 TT |
925 | int free; |
926 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
927 | ||
928 | down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ | |
929 | ext4_lock_group(sb, group); /* while we modify the bg desc */ | |
930 | free = EXT4_INODES_PER_GROUP(sb) - | |
931 | ext4_itable_unused_count(sb, gdp); | |
932 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
933 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
934 | free = 0; | |
935 | } | |
936 | /* | |
937 | * Check the relative inode number against the last used | |
938 | * relative inode number in this group. if it is greater | |
939 | * we need to update the bg_itable_unused count | |
940 | */ | |
941 | if (ino > free) | |
942 | ext4_itable_unused_set(sb, gdp, | |
943 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
944 | up_read(&grp->alloc_sem); | |
6f2e9f0e TM |
945 | } else { |
946 | ext4_lock_group(sb, group); | |
119c0d44 | 947 | } |
6f2e9f0e | 948 | |
119c0d44 TT |
949 | ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); |
950 | if (S_ISDIR(mode)) { | |
951 | ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); | |
952 | if (sbi->s_log_groups_per_flex) { | |
953 | ext4_group_t f = ext4_flex_group(sbi, group); | |
954 | ||
955 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); | |
956 | } | |
957 | } | |
41a246d1 DW |
958 | if (ext4_has_group_desc_csum(sb)) { |
959 | ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, | |
960 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 961 | ext4_group_desc_csum_set(sb, group, gdp); |
119c0d44 | 962 | } |
6f2e9f0e | 963 | ext4_unlock_group(sb, group); |
119c0d44 | 964 | |
3300beda AK |
965 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
966 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
eb9cc7e1 JK |
967 | if (err) { |
968 | ext4_std_error(sb, err); | |
969 | goto out; | |
970 | } | |
ac27a0ec DK |
971 | |
972 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
973 | if (S_ISDIR(mode)) | |
974 | percpu_counter_inc(&sbi->s_dirs_counter); | |
ac27a0ec | 975 | |
772cb7c8 JS |
976 | if (sbi->s_log_groups_per_flex) { |
977 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 978 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 | 979 | } |
ac27a0ec | 980 | |
717d50e4 | 981 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
982 | /* This is the optimal IO size (for stat), not the fs block size */ |
983 | inode->i_blocks = 0; | |
ef7f3835 KS |
984 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
985 | ext4_current_time(inode); | |
ac27a0ec DK |
986 | |
987 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
988 | ei->i_dir_start_lookup = 0; | |
989 | ei->i_disksize = 0; | |
990 | ||
4af83508 | 991 | /* Don't inherit extent flag from directory, amongst others. */ |
2dc6b0d4 DG |
992 | ei->i_flags = |
993 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 994 | ei->i_file_acl = 0; |
ac27a0ec | 995 | ei->i_dtime = 0; |
ac27a0ec | 996 | ei->i_block_group = group; |
a4912123 | 997 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 998 | |
2058f83a MH |
999 | /* If the directory encrypted, then we should encrypt the inode. */ |
1000 | if ((S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) && | |
6ddb2447 TT |
1001 | (ext4_encrypted_inode(dir) || |
1002 | DUMMY_ENCRYPTION_ENABLED(sbi))) | |
2058f83a MH |
1003 | ext4_set_inode_flag(inode, EXT4_INODE_ENCRYPT); |
1004 | ||
617ba13b | 1005 | ext4_set_inode_flags(inode); |
ac27a0ec | 1006 | if (IS_DIRSYNC(inode)) |
0390131b | 1007 | ext4_handle_sync(handle); |
6b38e842 | 1008 | if (insert_inode_locked(inode) < 0) { |
acd6ad83 JK |
1009 | /* |
1010 | * Likely a bitmap corruption causing inode to be allocated | |
1011 | * twice. | |
1012 | */ | |
1013 | err = -EIO; | |
eb9cc7e1 JK |
1014 | ext4_error(sb, "failed to insert inode %lu: doubly allocated?", |
1015 | inode->i_ino); | |
1016 | goto out; | |
6b38e842 | 1017 | } |
ac27a0ec DK |
1018 | spin_lock(&sbi->s_next_gen_lock); |
1019 | inode->i_generation = sbi->s_next_generation++; | |
1020 | spin_unlock(&sbi->s_next_gen_lock); | |
1021 | ||
814525f4 | 1022 | /* Precompute checksum seed for inode metadata */ |
9aa5d32b | 1023 | if (ext4_has_metadata_csum(sb)) { |
814525f4 | 1024 | __u32 csum; |
814525f4 DW |
1025 | __le32 inum = cpu_to_le32(inode->i_ino); |
1026 | __le32 gen = cpu_to_le32(inode->i_generation); | |
1027 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
1028 | sizeof(inum)); | |
1029 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
1030 | sizeof(gen)); | |
1031 | } | |
1032 | ||
353eb83c | 1033 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 1034 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 KS |
1035 | |
1036 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
f08225d1 TM |
1037 | ei->i_inline_off = 0; |
1038 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA)) | |
1039 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
ac27a0ec | 1040 | ret = inode; |
63936dda CH |
1041 | err = dquot_alloc_inode(inode); |
1042 | if (err) | |
ac27a0ec | 1043 | goto fail_drop; |
ac27a0ec | 1044 | |
617ba13b | 1045 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
1046 | if (err) |
1047 | goto fail_free_drop; | |
1048 | ||
2a7dba39 | 1049 | err = ext4_init_security(handle, inode, dir, qstr); |
ac27a0ec DK |
1050 | if (err) |
1051 | goto fail_free_drop; | |
1052 | ||
83982b6f | 1053 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
e4079a11 | 1054 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1055 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1056 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1057 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1058 | } |
a86c6181 | 1059 | } |
ac27a0ec | 1060 | |
688f869c TT |
1061 | if (ext4_handle_valid(handle)) { |
1062 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
1063 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
1064 | } | |
1065 | ||
8753e88f AK |
1066 | err = ext4_mark_inode_dirty(handle, inode); |
1067 | if (err) { | |
1068 | ext4_std_error(sb, err); | |
1069 | goto fail_free_drop; | |
1070 | } | |
1071 | ||
617ba13b | 1072 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1073 | trace_ext4_allocate_inode(inode, dir, mode); |
3300beda | 1074 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1075 | return ret; |
1076 | ||
1077 | fail_free_drop: | |
63936dda | 1078 | dquot_free_inode(inode); |
ac27a0ec | 1079 | fail_drop: |
6d6b77f1 | 1080 | clear_nlink(inode); |
6b38e842 | 1081 | unlock_new_inode(inode); |
eb9cc7e1 JK |
1082 | out: |
1083 | dquot_drop(inode); | |
1084 | inode->i_flags |= S_NOQUOTA; | |
ac27a0ec | 1085 | iput(inode); |
3300beda | 1086 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1087 | return ERR_PTR(err); |
1088 | } | |
1089 | ||
1090 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1091 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1092 | { |
617ba13b | 1093 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1094 | ext4_group_t block_group; |
ac27a0ec | 1095 | int bit; |
1d1fe1ee | 1096 | struct buffer_head *bitmap_bh; |
ac27a0ec | 1097 | struct inode *inode = NULL; |
1d1fe1ee | 1098 | long err = -EIO; |
ac27a0ec DK |
1099 | |
1100 | /* Error cases - e2fsck has already cleaned up for us */ | |
1101 | if (ino > max_ino) { | |
12062ddd | 1102 | ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 1103 | goto error; |
ac27a0ec DK |
1104 | } |
1105 | ||
617ba13b MC |
1106 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1107 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1108 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 1109 | if (!bitmap_bh) { |
12062ddd | 1110 | ext4_warning(sb, "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 1111 | goto error; |
ac27a0ec DK |
1112 | } |
1113 | ||
1114 | /* Having the inode bit set should be a 100% indicator that this | |
1115 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1116 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1117 | */ | |
1d1fe1ee DH |
1118 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1119 | goto bad_orphan; | |
1120 | ||
1121 | inode = ext4_iget(sb, ino); | |
1122 | if (IS_ERR(inode)) | |
1123 | goto iget_failed; | |
1124 | ||
91ef4caf DG |
1125 | /* |
1126 | * If the orphans has i_nlinks > 0 then it should be able to be | |
1127 | * truncated, otherwise it won't be removed from the orphan list | |
1128 | * during processing and an infinite loop will result. | |
1129 | */ | |
1130 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
1131 | goto bad_orphan; | |
1132 | ||
1d1fe1ee DH |
1133 | if (NEXT_ORPHAN(inode) > max_ino) |
1134 | goto bad_orphan; | |
1135 | brelse(bitmap_bh); | |
1136 | return inode; | |
1137 | ||
1138 | iget_failed: | |
1139 | err = PTR_ERR(inode); | |
1140 | inode = NULL; | |
1141 | bad_orphan: | |
12062ddd | 1142 | ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino); |
8de5c325 | 1143 | printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n", |
1d1fe1ee DH |
1144 | bit, (unsigned long long)bitmap_bh->b_blocknr, |
1145 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
8de5c325 | 1146 | printk(KERN_WARNING "inode=%p\n", inode); |
1d1fe1ee | 1147 | if (inode) { |
8de5c325 | 1148 | printk(KERN_WARNING "is_bad_inode(inode)=%d\n", |
1d1fe1ee | 1149 | is_bad_inode(inode)); |
8de5c325 | 1150 | printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n", |
1d1fe1ee | 1151 | NEXT_ORPHAN(inode)); |
8de5c325 TT |
1152 | printk(KERN_WARNING "max_ino=%lu\n", max_ino); |
1153 | printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink); | |
ac27a0ec | 1154 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1155 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1156 | inode->i_blocks = 0; |
1157 | iput(inode); | |
ac27a0ec | 1158 | } |
ac27a0ec | 1159 | brelse(bitmap_bh); |
1d1fe1ee DH |
1160 | error: |
1161 | return ERR_PTR(err); | |
ac27a0ec DK |
1162 | } |
1163 | ||
af5bc92d | 1164 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1165 | { |
1166 | unsigned long desc_count; | |
617ba13b | 1167 | struct ext4_group_desc *gdp; |
8df9675f | 1168 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1169 | #ifdef EXT4FS_DEBUG |
1170 | struct ext4_super_block *es; | |
ac27a0ec DK |
1171 | unsigned long bitmap_count, x; |
1172 | struct buffer_head *bitmap_bh = NULL; | |
1173 | ||
617ba13b | 1174 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1175 | desc_count = 0; |
1176 | bitmap_count = 0; | |
1177 | gdp = NULL; | |
8df9675f | 1178 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1179 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1180 | if (!gdp) |
1181 | continue; | |
560671a0 | 1182 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1183 | brelse(bitmap_bh); |
e29d1cde | 1184 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
1185 | if (!bitmap_bh) |
1186 | continue; | |
1187 | ||
f6fb99ca TT |
1188 | x = ext4_count_free(bitmap_bh->b_data, |
1189 | EXT4_INODES_PER_GROUP(sb) / 8); | |
c549a95d | 1190 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1191 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1192 | bitmap_count += x; |
1193 | } | |
1194 | brelse(bitmap_bh); | |
4776004f TT |
1195 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1196 | "stored = %u, computed = %lu, %lu\n", | |
1197 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1198 | return desc_count; |
1199 | #else | |
1200 | desc_count = 0; | |
8df9675f | 1201 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1202 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1203 | if (!gdp) |
1204 | continue; | |
560671a0 | 1205 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1206 | cond_resched(); |
1207 | } | |
1208 | return desc_count; | |
1209 | #endif | |
1210 | } | |
1211 | ||
1212 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1213 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1214 | { |
1215 | unsigned long count = 0; | |
8df9675f | 1216 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1217 | |
8df9675f | 1218 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1219 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1220 | if (!gdp) |
1221 | continue; | |
560671a0 | 1222 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1223 | } |
1224 | return count; | |
1225 | } | |
bfff6873 LC |
1226 | |
1227 | /* | |
1228 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1229 | * inode table. Must be called without any spinlock held. The only place | |
1230 | * where it is called from on active part of filesystem is ext4lazyinit | |
1231 | * thread, so we do not need any special locks, however we have to prevent | |
1232 | * inode allocation from the current group, so we take alloc_sem lock, to | |
119c0d44 | 1233 | * block ext4_new_inode() until we are finished. |
bfff6873 | 1234 | */ |
e0cbee3e | 1235 | int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, |
bfff6873 LC |
1236 | int barrier) |
1237 | { | |
1238 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1239 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1240 | struct ext4_group_desc *gdp = NULL; | |
1241 | struct buffer_head *group_desc_bh; | |
1242 | handle_t *handle; | |
1243 | ext4_fsblk_t blk; | |
1244 | int num, ret = 0, used_blks = 0; | |
bfff6873 LC |
1245 | |
1246 | /* This should not happen, but just to be sure check this */ | |
1247 | if (sb->s_flags & MS_RDONLY) { | |
1248 | ret = 1; | |
1249 | goto out; | |
1250 | } | |
1251 | ||
1252 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1253 | if (!gdp) | |
1254 | goto out; | |
1255 | ||
1256 | /* | |
1257 | * We do not need to lock this, because we are the only one | |
1258 | * handling this flag. | |
1259 | */ | |
1260 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1261 | goto out; | |
1262 | ||
9924a92a | 1263 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
bfff6873 LC |
1264 | if (IS_ERR(handle)) { |
1265 | ret = PTR_ERR(handle); | |
1266 | goto out; | |
1267 | } | |
1268 | ||
1269 | down_write(&grp->alloc_sem); | |
1270 | /* | |
1271 | * If inode bitmap was already initialized there may be some | |
1272 | * used inodes so we need to skip blocks with used inodes in | |
1273 | * inode table. | |
1274 | */ | |
1275 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1276 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1277 | ext4_itable_unused_count(sb, gdp)), | |
1278 | sbi->s_inodes_per_block); | |
1279 | ||
857ac889 | 1280 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) { |
1084f252 TT |
1281 | ext4_error(sb, "Something is wrong with group %u: " |
1282 | "used itable blocks: %d; " | |
1283 | "itable unused count: %u", | |
857ac889 LC |
1284 | group, used_blks, |
1285 | ext4_itable_unused_count(sb, gdp)); | |
1286 | ret = 1; | |
33853a0d | 1287 | goto err_out; |
857ac889 LC |
1288 | } |
1289 | ||
bfff6873 LC |
1290 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1291 | num = sbi->s_itb_per_group - used_blks; | |
1292 | ||
1293 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1294 | ret = ext4_journal_get_write_access(handle, | |
1295 | group_desc_bh); | |
1296 | if (ret) | |
1297 | goto err_out; | |
1298 | ||
bfff6873 LC |
1299 | /* |
1300 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1301 | * flag anyway, because obviously, when it is full it does not need | |
1302 | * further zeroing. | |
1303 | */ | |
1304 | if (unlikely(num == 0)) | |
1305 | goto skip_zeroout; | |
1306 | ||
1307 | ext4_debug("going to zero out inode table in group %d\n", | |
1308 | group); | |
a107e5a3 | 1309 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1310 | if (ret < 0) |
1311 | goto err_out; | |
a107e5a3 TT |
1312 | if (barrier) |
1313 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); | |
bfff6873 LC |
1314 | |
1315 | skip_zeroout: | |
1316 | ext4_lock_group(sb, group); | |
1317 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
feb0ab32 | 1318 | ext4_group_desc_csum_set(sb, group, gdp); |
bfff6873 LC |
1319 | ext4_unlock_group(sb, group); |
1320 | ||
1321 | BUFFER_TRACE(group_desc_bh, | |
1322 | "call ext4_handle_dirty_metadata"); | |
1323 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1324 | group_desc_bh); | |
1325 | ||
1326 | err_out: | |
1327 | up_write(&grp->alloc_sem); | |
1328 | ext4_journal_stop(handle); | |
1329 | out: | |
1330 | return ret; | |
1331 | } |