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 | { |
717d50e4 AD |
73 | J_ASSERT_BH(bh, buffer_locked(bh)); |
74 | ||
75 | /* If checksum is bad mark all blocks and inodes use to prevent | |
76 | * allocation, essentially implementing a per-group read-only flag. */ | |
feb0ab32 | 77 | if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) { |
12062ddd | 78 | ext4_error(sb, "Checksum bad for group %u", block_group); |
021b65bb | 79 | ext4_free_group_clusters_set(sb, gdp, 0); |
560671a0 AK |
80 | ext4_free_inodes_set(sb, gdp, 0); |
81 | ext4_itable_unused_set(sb, gdp, 0); | |
717d50e4 | 82 | memset(bh->b_data, 0xff, sb->s_blocksize); |
41a246d1 DW |
83 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh, |
84 | EXT4_INODES_PER_GROUP(sb) / 8); | |
717d50e4 AD |
85 | return 0; |
86 | } | |
87 | ||
88 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
61d08673 | 89 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
717d50e4 | 90 | bh->b_data); |
41a246d1 DW |
91 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh, |
92 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 93 | ext4_group_desc_csum_set(sb, block_group, gdp); |
717d50e4 AD |
94 | |
95 | return EXT4_INODES_PER_GROUP(sb); | |
96 | } | |
ac27a0ec | 97 | |
813e5727 TT |
98 | void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate) |
99 | { | |
100 | if (uptodate) { | |
101 | set_buffer_uptodate(bh); | |
102 | set_bitmap_uptodate(bh); | |
103 | } | |
104 | unlock_buffer(bh); | |
105 | put_bh(bh); | |
106 | } | |
107 | ||
ac27a0ec DK |
108 | /* |
109 | * Read the inode allocation bitmap for a given block_group, reading | |
110 | * into the specified slot in the superblock's bitmap cache. | |
111 | * | |
112 | * Return buffer_head of bitmap on success or NULL. | |
113 | */ | |
114 | static struct buffer_head * | |
e29d1cde | 115 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 116 | { |
617ba13b | 117 | struct ext4_group_desc *desc; |
ac27a0ec | 118 | struct buffer_head *bh = NULL; |
e29d1cde | 119 | ext4_fsblk_t bitmap_blk; |
ac27a0ec | 120 | |
617ba13b | 121 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 122 | if (!desc) |
e29d1cde | 123 | return NULL; |
bfff6873 | 124 | |
e29d1cde ES |
125 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
126 | bh = sb_getblk(sb, bitmap_blk); | |
127 | if (unlikely(!bh)) { | |
12062ddd | 128 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 129 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
130 | block_group, bitmap_blk); |
131 | return NULL; | |
132 | } | |
2ccb5fb9 | 133 | if (bitmap_uptodate(bh)) |
41a246d1 | 134 | goto verify; |
e29d1cde | 135 | |
c806e68f | 136 | lock_buffer(bh); |
2ccb5fb9 AK |
137 | if (bitmap_uptodate(bh)) { |
138 | unlock_buffer(bh); | |
41a246d1 | 139 | goto verify; |
2ccb5fb9 | 140 | } |
bfff6873 | 141 | |
955ce5f5 | 142 | ext4_lock_group(sb, block_group); |
717d50e4 | 143 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
e29d1cde | 144 | ext4_init_inode_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 145 | set_bitmap_uptodate(bh); |
e29d1cde | 146 | set_buffer_uptodate(bh); |
41a246d1 | 147 | set_buffer_verified(bh); |
955ce5f5 | 148 | ext4_unlock_group(sb, block_group); |
3300beda | 149 | unlock_buffer(bh); |
e29d1cde | 150 | return bh; |
717d50e4 | 151 | } |
955ce5f5 | 152 | ext4_unlock_group(sb, block_group); |
bfff6873 | 153 | |
2ccb5fb9 AK |
154 | if (buffer_uptodate(bh)) { |
155 | /* | |
156 | * if not uninit if bh is uptodate, | |
157 | * bitmap is also uptodate | |
158 | */ | |
159 | set_bitmap_uptodate(bh); | |
160 | unlock_buffer(bh); | |
41a246d1 | 161 | goto verify; |
2ccb5fb9 AK |
162 | } |
163 | /* | |
813e5727 | 164 | * submit the buffer_head for reading |
2ccb5fb9 | 165 | */ |
0562e0ba | 166 | trace_ext4_load_inode_bitmap(sb, block_group); |
813e5727 TT |
167 | bh->b_end_io = ext4_end_bitmap_read; |
168 | get_bh(bh); | |
9f203507 | 169 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
813e5727 TT |
170 | wait_on_buffer(bh); |
171 | if (!buffer_uptodate(bh)) { | |
e29d1cde | 172 | put_bh(bh); |
12062ddd | 173 | ext4_error(sb, "Cannot read inode bitmap - " |
813e5727 TT |
174 | "block_group = %u, inode_bitmap = %llu", |
175 | block_group, bitmap_blk); | |
e29d1cde ES |
176 | return NULL; |
177 | } | |
41a246d1 DW |
178 | |
179 | verify: | |
180 | ext4_lock_group(sb, block_group); | |
181 | if (!buffer_verified(bh) && | |
182 | !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh, | |
183 | EXT4_INODES_PER_GROUP(sb) / 8)) { | |
184 | ext4_unlock_group(sb, block_group); | |
185 | put_bh(bh); | |
186 | ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " | |
187 | "inode_bitmap = %llu", block_group, bitmap_blk); | |
188 | return NULL; | |
189 | } | |
190 | ext4_unlock_group(sb, block_group); | |
191 | set_buffer_verified(bh); | |
ac27a0ec DK |
192 | return bh; |
193 | } | |
194 | ||
195 | /* | |
196 | * NOTE! When we get the inode, we're the only people | |
197 | * that have access to it, and as such there are no | |
198 | * race conditions we have to worry about. The inode | |
199 | * is not on the hash-lists, and it cannot be reached | |
200 | * through the filesystem because the directory entry | |
201 | * has been deleted earlier. | |
202 | * | |
203 | * HOWEVER: we must make sure that we get no aliases, | |
204 | * which means that we have to call "clear_inode()" | |
205 | * _before_ we mark the inode not in use in the inode | |
206 | * bitmaps. Otherwise a newly created file might use | |
207 | * the same inode number (not actually the same pointer | |
208 | * though), and then we'd have two inodes sharing the | |
209 | * same inode number and space on the harddisk. | |
210 | */ | |
af5bc92d | 211 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 212 | { |
af5bc92d | 213 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
214 | int is_directory; |
215 | unsigned long ino; | |
216 | struct buffer_head *bitmap_bh = NULL; | |
217 | struct buffer_head *bh2; | |
fd2d4291 | 218 | ext4_group_t block_group; |
ac27a0ec | 219 | unsigned long bit; |
af5bc92d TT |
220 | struct ext4_group_desc *gdp; |
221 | struct ext4_super_block *es; | |
617ba13b | 222 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 223 | int fatal = 0, err, count, cleared; |
ac27a0ec | 224 | |
92b97816 TT |
225 | if (!sb) { |
226 | printk(KERN_ERR "EXT4-fs: %s:%d: inode on " | |
227 | "nonexistent device\n", __func__, __LINE__); | |
ac27a0ec DK |
228 | return; |
229 | } | |
92b97816 TT |
230 | if (atomic_read(&inode->i_count) > 1) { |
231 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", | |
232 | __func__, __LINE__, inode->i_ino, | |
233 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
234 | return; |
235 | } | |
92b97816 TT |
236 | if (inode->i_nlink) { |
237 | ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", | |
238 | __func__, __LINE__, inode->i_ino, inode->i_nlink); | |
ac27a0ec DK |
239 | return; |
240 | } | |
617ba13b | 241 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
242 | |
243 | ino = inode->i_ino; | |
af5bc92d | 244 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 245 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
246 | |
247 | /* | |
248 | * Note: we must free any quota before locking the superblock, | |
249 | * as writing the quota to disk may need the lock as well. | |
250 | */ | |
871a2931 | 251 | dquot_initialize(inode); |
617ba13b | 252 | ext4_xattr_delete_inode(handle, inode); |
63936dda | 253 | dquot_free_inode(inode); |
9f754758 | 254 | dquot_drop(inode); |
ac27a0ec DK |
255 | |
256 | is_directory = S_ISDIR(inode->i_mode); | |
257 | ||
258 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 259 | ext4_clear_inode(inode); |
ac27a0ec | 260 | |
617ba13b MC |
261 | es = EXT4_SB(sb)->s_es; |
262 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 263 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
264 | goto error_return; |
265 | } | |
617ba13b MC |
266 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
267 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 268 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec DK |
269 | if (!bitmap_bh) |
270 | goto error_return; | |
271 | ||
272 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 273 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
274 | if (fatal) |
275 | goto error_return; | |
276 | ||
d17413c0 DM |
277 | fatal = -ESRCH; |
278 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
279 | if (gdp) { | |
ac27a0ec | 280 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 281 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
282 | } |
283 | ext4_lock_group(sb, block_group); | |
597d508c | 284 | cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); |
d17413c0 DM |
285 | if (fatal || !cleared) { |
286 | ext4_unlock_group(sb, block_group); | |
287 | goto out; | |
288 | } | |
7d39db14 | 289 | |
d17413c0 DM |
290 | count = ext4_free_inodes_count(sb, gdp) + 1; |
291 | ext4_free_inodes_set(sb, gdp, count); | |
292 | if (is_directory) { | |
293 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
294 | ext4_used_dirs_set(sb, gdp, count); | |
295 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 296 | } |
41a246d1 DW |
297 | ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, |
298 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 299 | ext4_group_desc_csum_set(sb, block_group, gdp); |
d17413c0 | 300 | ext4_unlock_group(sb, block_group); |
ac27a0ec | 301 | |
d17413c0 DM |
302 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
303 | if (sbi->s_log_groups_per_flex) { | |
304 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 305 | |
d17413c0 DM |
306 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
307 | if (is_directory) | |
308 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 309 | } |
d17413c0 DM |
310 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
311 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
312 | out: | |
313 | if (cleared) { | |
314 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
315 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
316 | if (!fatal) | |
317 | fatal = err; | |
d17413c0 DM |
318 | } else |
319 | ext4_error(sb, "bit already cleared for inode %lu", ino); | |
320 | ||
ac27a0ec DK |
321 | error_return: |
322 | brelse(bitmap_bh); | |
617ba13b | 323 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
324 | } |
325 | ||
a4912123 | 326 | struct orlov_stats { |
90ba983f | 327 | __u64 free_clusters; |
a4912123 | 328 | __u32 free_inodes; |
a4912123 TT |
329 | __u32 used_dirs; |
330 | }; | |
331 | ||
332 | /* | |
333 | * Helper function for Orlov's allocator; returns critical information | |
334 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
335 | * is a block group number; otherwise it is flex_bg number. | |
336 | */ | |
1f109d5a TT |
337 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
338 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
339 | { |
340 | struct ext4_group_desc *desc; | |
7d39db14 | 341 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 342 | |
7d39db14 TT |
343 | if (flex_size > 1) { |
344 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
90ba983f | 345 | stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); |
7d39db14 TT |
346 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); |
347 | return; | |
348 | } | |
a4912123 | 349 | |
7d39db14 TT |
350 | desc = ext4_get_group_desc(sb, g, NULL); |
351 | if (desc) { | |
352 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
021b65bb | 353 | stats->free_clusters = ext4_free_group_clusters(sb, desc); |
7d39db14 TT |
354 | stats->used_dirs = ext4_used_dirs_count(sb, desc); |
355 | } else { | |
356 | stats->free_inodes = 0; | |
24aaa8ef | 357 | stats->free_clusters = 0; |
7d39db14 | 358 | stats->used_dirs = 0; |
a4912123 TT |
359 | } |
360 | } | |
361 | ||
ac27a0ec DK |
362 | /* |
363 | * Orlov's allocator for directories. | |
364 | * | |
365 | * We always try to spread first-level directories. | |
366 | * | |
367 | * If there are blockgroups with both free inodes and free blocks counts | |
368 | * not worse than average we return one with smallest directory count. | |
369 | * Otherwise we simply return a random group. | |
370 | * | |
371 | * For the rest rules look so: | |
372 | * | |
373 | * It's OK to put directory into a group unless | |
374 | * it has too many directories already (max_dirs) or | |
375 | * it has too few free inodes left (min_inodes) or | |
376 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 377 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
378 | * conditions we search cyclically through the rest. If none |
379 | * of the groups look good we just look for a group with more | |
380 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
381 | */ |
382 | ||
2aa9fc4c | 383 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
dcca3fec | 384 | ext4_group_t *group, umode_t mode, |
f157a4aa | 385 | const struct qstr *qstr) |
ac27a0ec | 386 | { |
fd2d4291 | 387 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 388 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 389 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 390 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
14c83c9f | 391 | unsigned int freei, avefreei, grp_free; |
24aaa8ef | 392 | ext4_fsblk_t freeb, avefreec; |
ac27a0ec | 393 | unsigned int ndirs; |
a4912123 | 394 | int max_dirs, min_inodes; |
24aaa8ef | 395 | ext4_grpblk_t min_clusters; |
8df9675f | 396 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 397 | struct ext4_group_desc *desc; |
a4912123 TT |
398 | struct orlov_stats stats; |
399 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 400 | struct dx_hash_info hinfo; |
a4912123 | 401 | |
8df9675f | 402 | ngroups = real_ngroups; |
a4912123 | 403 | if (flex_size > 1) { |
8df9675f | 404 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
405 | sbi->s_log_groups_per_flex; |
406 | parent_group >>= sbi->s_log_groups_per_flex; | |
407 | } | |
ac27a0ec DK |
408 | |
409 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
410 | avefreei = freei / ngroups; | |
57042651 TT |
411 | freeb = EXT4_C2B(sbi, |
412 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
24aaa8ef TT |
413 | avefreec = freeb; |
414 | do_div(avefreec, ngroups); | |
ac27a0ec DK |
415 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
416 | ||
a4912123 TT |
417 | if (S_ISDIR(mode) && |
418 | ((parent == sb->s_root->d_inode) || | |
12e9b892 | 419 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 420 | int best_ndir = inodes_per_group; |
2aa9fc4c | 421 | int ret = -1; |
ac27a0ec | 422 | |
f157a4aa TT |
423 | if (qstr) { |
424 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
425 | hinfo.seed = sbi->s_hash_seed; | |
426 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
427 | grp = hinfo.hash; | |
428 | } else | |
429 | get_random_bytes(&grp, sizeof(grp)); | |
2aa9fc4c | 430 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 431 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
432 | g = (parent_group + i) % ngroups; |
433 | get_orlov_stats(sb, g, flex_size, &stats); | |
434 | if (!stats.free_inodes) | |
ac27a0ec | 435 | continue; |
a4912123 | 436 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 437 | continue; |
a4912123 | 438 | if (stats.free_inodes < avefreei) |
ac27a0ec | 439 | continue; |
24aaa8ef | 440 | if (stats.free_clusters < avefreec) |
ac27a0ec | 441 | continue; |
a4912123 | 442 | grp = g; |
2aa9fc4c | 443 | ret = 0; |
a4912123 TT |
444 | best_ndir = stats.used_dirs; |
445 | } | |
446 | if (ret) | |
447 | goto fallback; | |
448 | found_flex_bg: | |
449 | if (flex_size == 1) { | |
450 | *group = grp; | |
451 | return 0; | |
452 | } | |
453 | ||
454 | /* | |
455 | * We pack inodes at the beginning of the flexgroup's | |
456 | * inode tables. Block allocation decisions will do | |
457 | * something similar, although regular files will | |
458 | * start at 2nd block group of the flexgroup. See | |
459 | * ext4_ext_find_goal() and ext4_find_near(). | |
460 | */ | |
461 | grp *= flex_size; | |
462 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 463 | if (grp+i >= real_ngroups) |
a4912123 TT |
464 | break; |
465 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
466 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
467 | *group = grp+i; | |
468 | return 0; | |
469 | } | |
ac27a0ec | 470 | } |
ac27a0ec DK |
471 | goto fallback; |
472 | } | |
473 | ||
ac27a0ec | 474 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
475 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
476 | if (min_inodes < 1) | |
477 | min_inodes = 1; | |
24aaa8ef | 478 | min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; |
a4912123 TT |
479 | |
480 | /* | |
481 | * Start looking in the flex group where we last allocated an | |
482 | * inode for this parent directory | |
483 | */ | |
484 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
485 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
486 | if (flex_size > 1) | |
487 | parent_group >>= sbi->s_log_groups_per_flex; | |
488 | } | |
ac27a0ec DK |
489 | |
490 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
491 | grp = (parent_group + i) % ngroups; |
492 | get_orlov_stats(sb, grp, flex_size, &stats); | |
493 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 494 | continue; |
a4912123 | 495 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 496 | continue; |
24aaa8ef | 497 | if (stats.free_clusters < min_clusters) |
ac27a0ec | 498 | continue; |
a4912123 | 499 | goto found_flex_bg; |
ac27a0ec DK |
500 | } |
501 | ||
502 | fallback: | |
8df9675f | 503 | ngroups = real_ngroups; |
a4912123 | 504 | avefreei = freei / ngroups; |
b5451f7b | 505 | fallback_retry: |
a4912123 | 506 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 507 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
508 | grp = (parent_group + i) % ngroups; |
509 | desc = ext4_get_group_desc(sb, grp, NULL); | |
bb3d132a DC |
510 | if (desc) { |
511 | grp_free = ext4_free_inodes_count(sb, desc); | |
512 | if (grp_free && grp_free >= avefreei) { | |
513 | *group = grp; | |
514 | return 0; | |
515 | } | |
a4912123 | 516 | } |
ac27a0ec DK |
517 | } |
518 | ||
519 | if (avefreei) { | |
520 | /* | |
521 | * The free-inodes counter is approximate, and for really small | |
522 | * filesystems the above test can fail to find any blockgroups | |
523 | */ | |
524 | avefreei = 0; | |
b5451f7b | 525 | goto fallback_retry; |
ac27a0ec DK |
526 | } |
527 | ||
528 | return -1; | |
529 | } | |
530 | ||
2aa9fc4c | 531 | static int find_group_other(struct super_block *sb, struct inode *parent, |
dcca3fec | 532 | ext4_group_t *group, umode_t mode) |
ac27a0ec | 533 | { |
fd2d4291 | 534 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 535 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 536 | struct ext4_group_desc *desc; |
a4912123 TT |
537 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
538 | ||
539 | /* | |
540 | * Try to place the inode is the same flex group as its | |
541 | * parent. If we can't find space, use the Orlov algorithm to | |
542 | * find another flex group, and store that information in the | |
543 | * parent directory's inode information so that use that flex | |
544 | * group for future allocations. | |
545 | */ | |
546 | if (flex_size > 1) { | |
547 | int retry = 0; | |
548 | ||
549 | try_again: | |
550 | parent_group &= ~(flex_size-1); | |
551 | last = parent_group + flex_size; | |
552 | if (last > ngroups) | |
553 | last = ngroups; | |
554 | for (i = parent_group; i < last; i++) { | |
555 | desc = ext4_get_group_desc(sb, i, NULL); | |
556 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
557 | *group = i; | |
558 | return 0; | |
559 | } | |
560 | } | |
561 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
562 | retry = 1; | |
563 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
564 | goto try_again; | |
565 | } | |
566 | /* | |
567 | * If this didn't work, use the Orlov search algorithm | |
568 | * to find a new flex group; we pass in the mode to | |
569 | * avoid the topdir algorithms. | |
570 | */ | |
571 | *group = parent_group + flex_size; | |
572 | if (*group > ngroups) | |
573 | *group = 0; | |
7dc57615 | 574 | return find_group_orlov(sb, parent, group, mode, NULL); |
a4912123 | 575 | } |
ac27a0ec DK |
576 | |
577 | /* | |
578 | * Try to place the inode in its parent directory | |
579 | */ | |
2aa9fc4c AM |
580 | *group = parent_group; |
581 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 582 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 583 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 584 | return 0; |
ac27a0ec DK |
585 | |
586 | /* | |
587 | * We're going to place this inode in a different blockgroup from its | |
588 | * parent. We want to cause files in a common directory to all land in | |
589 | * the same blockgroup. But we want files which are in a different | |
590 | * directory which shares a blockgroup with our parent to land in a | |
591 | * different blockgroup. | |
592 | * | |
593 | * So add our directory's i_ino into the starting point for the hash. | |
594 | */ | |
2aa9fc4c | 595 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
596 | |
597 | /* | |
598 | * Use a quadratic hash to find a group with a free inode and some free | |
599 | * blocks. | |
600 | */ | |
601 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
602 | *group += i; |
603 | if (*group >= ngroups) | |
604 | *group -= ngroups; | |
605 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 606 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 607 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 608 | return 0; |
ac27a0ec DK |
609 | } |
610 | ||
611 | /* | |
612 | * That failed: try linear search for a free inode, even if that group | |
613 | * has no free blocks. | |
614 | */ | |
2aa9fc4c | 615 | *group = parent_group; |
ac27a0ec | 616 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
617 | if (++*group >= ngroups) |
618 | *group = 0; | |
619 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 620 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 621 | return 0; |
ac27a0ec DK |
622 | } |
623 | ||
624 | return -1; | |
625 | } | |
626 | ||
627 | /* | |
628 | * There are two policies for allocating an inode. If the new inode is | |
629 | * a directory, then a forward search is made for a block group with both | |
630 | * free space and a low directory-to-inode ratio; if that fails, then of | |
631 | * the groups with above-average free space, that group with the fewest | |
632 | * directories already is chosen. | |
633 | * | |
634 | * For other inodes, search forward from the parent directory's block | |
635 | * group to find a free inode. | |
636 | */ | |
1139575a TT |
637 | struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, |
638 | umode_t mode, const struct qstr *qstr, | |
639 | __u32 goal, uid_t *owner, int handle_type, | |
640 | unsigned int line_no, int nblocks) | |
ac27a0ec DK |
641 | { |
642 | struct super_block *sb; | |
3300beda AK |
643 | struct buffer_head *inode_bitmap_bh = NULL; |
644 | struct buffer_head *group_desc_bh; | |
8df9675f | 645 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 646 | unsigned long ino = 0; |
af5bc92d TT |
647 | struct inode *inode; |
648 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
649 | struct ext4_inode_info *ei; |
650 | struct ext4_sb_info *sbi; | |
39341867 | 651 | int ret2, err = 0; |
ac27a0ec | 652 | struct inode *ret; |
2aa9fc4c | 653 | ext4_group_t i; |
772cb7c8 | 654 | ext4_group_t flex_group; |
ac27a0ec DK |
655 | |
656 | /* Cannot create files in a deleted directory */ | |
657 | if (!dir || !dir->i_nlink) | |
658 | return ERR_PTR(-EPERM); | |
659 | ||
660 | sb = dir->i_sb; | |
8df9675f | 661 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 662 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
663 | inode = new_inode(sb); |
664 | if (!inode) | |
665 | return ERR_PTR(-ENOMEM); | |
617ba13b | 666 | ei = EXT4_I(inode); |
617ba13b | 667 | sbi = EXT4_SB(sb); |
772cb7c8 | 668 | |
eb9cc7e1 JK |
669 | /* |
670 | * Initalize owners and quota early so that we don't have to account | |
671 | * for quota initialization worst case in standard inode creating | |
672 | * transaction | |
673 | */ | |
674 | if (owner) { | |
675 | inode->i_mode = mode; | |
676 | i_uid_write(inode, owner[0]); | |
677 | i_gid_write(inode, owner[1]); | |
678 | } else if (test_opt(sb, GRPID)) { | |
679 | inode->i_mode = mode; | |
680 | inode->i_uid = current_fsuid(); | |
681 | inode->i_gid = dir->i_gid; | |
682 | } else | |
683 | inode_init_owner(inode, dir, mode); | |
684 | dquot_initialize(inode); | |
685 | ||
11013911 AD |
686 | if (!goal) |
687 | goal = sbi->s_inode_goal; | |
688 | ||
e6462869 | 689 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
690 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
691 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
692 | ret2 = 0; | |
693 | goto got_group; | |
694 | } | |
695 | ||
4113c4ca LC |
696 | if (S_ISDIR(mode)) |
697 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); | |
698 | else | |
a4912123 | 699 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 700 | |
772cb7c8 | 701 | got_group: |
a4912123 | 702 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 703 | err = -ENOSPC; |
2aa9fc4c | 704 | if (ret2 == -1) |
ac27a0ec DK |
705 | goto out; |
706 | ||
119c0d44 TT |
707 | /* |
708 | * Normally we will only go through one pass of this loop, | |
709 | * unless we get unlucky and it turns out the group we selected | |
710 | * had its last inode grabbed by someone else. | |
711 | */ | |
11013911 | 712 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
713 | err = -EIO; |
714 | ||
3300beda | 715 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec | 716 | if (!gdp) |
eb9cc7e1 | 717 | goto out; |
ac27a0ec | 718 | |
f2a09af6 YY |
719 | /* |
720 | * Check free inodes count before loading bitmap. | |
721 | */ | |
722 | if (ext4_free_inodes_count(sb, gdp) == 0) { | |
723 | if (++group == ngroups) | |
724 | group = 0; | |
725 | continue; | |
726 | } | |
727 | ||
3300beda AK |
728 | brelse(inode_bitmap_bh); |
729 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
730 | if (!inode_bitmap_bh) | |
eb9cc7e1 | 731 | goto out; |
ac27a0ec | 732 | |
ac27a0ec | 733 | repeat_in_this_group: |
617ba13b | 734 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
735 | inode_bitmap_bh->b_data, |
736 | EXT4_INODES_PER_GROUP(sb), ino); | |
dc953e89 TT |
737 | if (ino >= EXT4_INODES_PER_GROUP(sb)) |
738 | goto next_group; | |
119c0d44 TT |
739 | if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) { |
740 | ext4_error(sb, "reserved inode found cleared - " | |
741 | "inode=%lu", ino + 1); | |
742 | continue; | |
743 | } | |
1139575a TT |
744 | if (!handle) { |
745 | BUG_ON(nblocks <= 0); | |
746 | handle = __ext4_journal_start_sb(dir->i_sb, line_no, | |
747 | handle_type, nblocks); | |
748 | if (IS_ERR(handle)) { | |
749 | err = PTR_ERR(handle); | |
eb9cc7e1 JK |
750 | ext4_std_error(sb, err); |
751 | goto out; | |
1139575a TT |
752 | } |
753 | } | |
ffb5387e ES |
754 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
755 | err = ext4_journal_get_write_access(handle, inode_bitmap_bh); | |
eb9cc7e1 JK |
756 | if (err) { |
757 | ext4_std_error(sb, err); | |
758 | goto out; | |
759 | } | |
119c0d44 TT |
760 | ext4_lock_group(sb, group); |
761 | ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); | |
762 | ext4_unlock_group(sb, group); | |
763 | ino++; /* the inode bitmap is zero-based */ | |
764 | if (!ret2) | |
765 | goto got; /* we grabbed the inode! */ | |
766 | if (ino < EXT4_INODES_PER_GROUP(sb)) | |
767 | goto repeat_in_this_group; | |
dc953e89 TT |
768 | next_group: |
769 | if (++group == ngroups) | |
770 | group = 0; | |
ac27a0ec DK |
771 | } |
772 | err = -ENOSPC; | |
773 | goto out; | |
774 | ||
775 | got: | |
ffb5387e ES |
776 | BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); |
777 | err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); | |
eb9cc7e1 JK |
778 | if (err) { |
779 | ext4_std_error(sb, err); | |
780 | goto out; | |
781 | } | |
ffb5387e | 782 | |
7cefa2c6 TT |
783 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
784 | err = ext4_journal_get_write_access(handle, group_desc_bh); | |
785 | if (err) { | |
786 | ext4_std_error(sb, err); | |
787 | goto out; | |
788 | } | |
789 | ||
717d50e4 | 790 | /* We may have to initialize the block bitmap if it isn't already */ |
feb0ab32 | 791 | if (ext4_has_group_desc_csum(sb) && |
717d50e4 | 792 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 793 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 794 | |
3300beda | 795 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
6ce75ebb JK |
796 | if (!block_bitmap_bh) { |
797 | err = -EIO; | |
798 | goto out; | |
799 | } | |
3300beda AK |
800 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); |
801 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 802 | if (err) { |
3300beda | 803 | brelse(block_bitmap_bh); |
eb9cc7e1 JK |
804 | ext4_std_error(sb, err); |
805 | goto out; | |
717d50e4 AD |
806 | } |
807 | ||
fd034a84 TT |
808 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
809 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
fd034a84 | 810 | |
717d50e4 | 811 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 812 | ext4_lock_group(sb, group); |
717d50e4 | 813 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 814 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 815 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 816 | ext4_free_clusters_after_init(sb, group, gdp)); |
fa77dcfa | 817 | ext4_block_bitmap_csum_set(sb, group, gdp, |
79f1ba49 | 818 | block_bitmap_bh); |
feb0ab32 | 819 | ext4_group_desc_csum_set(sb, group, gdp); |
717d50e4 | 820 | } |
955ce5f5 | 821 | ext4_unlock_group(sb, group); |
aeb1e5d6 | 822 | brelse(block_bitmap_bh); |
717d50e4 | 823 | |
eb9cc7e1 JK |
824 | if (err) { |
825 | ext4_std_error(sb, err); | |
826 | goto out; | |
827 | } | |
717d50e4 | 828 | } |
119c0d44 | 829 | |
119c0d44 | 830 | /* Update the relevant bg descriptor fields */ |
41a246d1 | 831 | if (ext4_has_group_desc_csum(sb)) { |
119c0d44 TT |
832 | int free; |
833 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
834 | ||
835 | down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ | |
836 | ext4_lock_group(sb, group); /* while we modify the bg desc */ | |
837 | free = EXT4_INODES_PER_GROUP(sb) - | |
838 | ext4_itable_unused_count(sb, gdp); | |
839 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
840 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
841 | free = 0; | |
842 | } | |
843 | /* | |
844 | * Check the relative inode number against the last used | |
845 | * relative inode number in this group. if it is greater | |
846 | * we need to update the bg_itable_unused count | |
847 | */ | |
848 | if (ino > free) | |
849 | ext4_itable_unused_set(sb, gdp, | |
850 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
851 | up_read(&grp->alloc_sem); | |
6f2e9f0e TM |
852 | } else { |
853 | ext4_lock_group(sb, group); | |
119c0d44 | 854 | } |
6f2e9f0e | 855 | |
119c0d44 TT |
856 | ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); |
857 | if (S_ISDIR(mode)) { | |
858 | ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); | |
859 | if (sbi->s_log_groups_per_flex) { | |
860 | ext4_group_t f = ext4_flex_group(sbi, group); | |
861 | ||
862 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); | |
863 | } | |
864 | } | |
41a246d1 DW |
865 | if (ext4_has_group_desc_csum(sb)) { |
866 | ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, | |
867 | EXT4_INODES_PER_GROUP(sb) / 8); | |
feb0ab32 | 868 | ext4_group_desc_csum_set(sb, group, gdp); |
119c0d44 | 869 | } |
6f2e9f0e | 870 | ext4_unlock_group(sb, group); |
119c0d44 | 871 | |
3300beda AK |
872 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
873 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
eb9cc7e1 JK |
874 | if (err) { |
875 | ext4_std_error(sb, err); | |
876 | goto out; | |
877 | } | |
ac27a0ec DK |
878 | |
879 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
880 | if (S_ISDIR(mode)) | |
881 | percpu_counter_inc(&sbi->s_dirs_counter); | |
ac27a0ec | 882 | |
772cb7c8 JS |
883 | if (sbi->s_log_groups_per_flex) { |
884 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 885 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 | 886 | } |
ac27a0ec | 887 | |
717d50e4 | 888 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
889 | /* This is the optimal IO size (for stat), not the fs block size */ |
890 | inode->i_blocks = 0; | |
ef7f3835 KS |
891 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
892 | ext4_current_time(inode); | |
ac27a0ec DK |
893 | |
894 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
895 | ei->i_dir_start_lookup = 0; | |
896 | ei->i_disksize = 0; | |
897 | ||
4af83508 | 898 | /* Don't inherit extent flag from directory, amongst others. */ |
2dc6b0d4 DG |
899 | ei->i_flags = |
900 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 901 | ei->i_file_acl = 0; |
ac27a0ec | 902 | ei->i_dtime = 0; |
ac27a0ec | 903 | ei->i_block_group = group; |
a4912123 | 904 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 905 | |
617ba13b | 906 | ext4_set_inode_flags(inode); |
ac27a0ec | 907 | if (IS_DIRSYNC(inode)) |
0390131b | 908 | ext4_handle_sync(handle); |
6b38e842 | 909 | if (insert_inode_locked(inode) < 0) { |
acd6ad83 JK |
910 | /* |
911 | * Likely a bitmap corruption causing inode to be allocated | |
912 | * twice. | |
913 | */ | |
914 | err = -EIO; | |
eb9cc7e1 JK |
915 | ext4_error(sb, "failed to insert inode %lu: doubly allocated?", |
916 | inode->i_ino); | |
917 | goto out; | |
6b38e842 | 918 | } |
ac27a0ec DK |
919 | spin_lock(&sbi->s_next_gen_lock); |
920 | inode->i_generation = sbi->s_next_generation++; | |
921 | spin_unlock(&sbi->s_next_gen_lock); | |
922 | ||
814525f4 DW |
923 | /* Precompute checksum seed for inode metadata */ |
924 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
925 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { | |
926 | __u32 csum; | |
814525f4 DW |
927 | __le32 inum = cpu_to_le32(inode->i_ino); |
928 | __le32 gen = cpu_to_le32(inode->i_generation); | |
929 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
930 | sizeof(inum)); | |
931 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
932 | sizeof(gen)); | |
933 | } | |
934 | ||
353eb83c | 935 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 936 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 KS |
937 | |
938 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec | 939 | |
f08225d1 TM |
940 | ei->i_inline_off = 0; |
941 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA)) | |
942 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
943 | ||
ac27a0ec | 944 | ret = inode; |
63936dda CH |
945 | err = dquot_alloc_inode(inode); |
946 | if (err) | |
ac27a0ec | 947 | goto fail_drop; |
ac27a0ec | 948 | |
617ba13b | 949 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
950 | if (err) |
951 | goto fail_free_drop; | |
952 | ||
2a7dba39 | 953 | err = ext4_init_security(handle, inode, dir, qstr); |
ac27a0ec DK |
954 | if (err) |
955 | goto fail_free_drop; | |
956 | ||
83982b6f | 957 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
e4079a11 | 958 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 959 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 960 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 961 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 962 | } |
a86c6181 | 963 | } |
ac27a0ec | 964 | |
688f869c TT |
965 | if (ext4_handle_valid(handle)) { |
966 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
967 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
968 | } | |
969 | ||
8753e88f AK |
970 | err = ext4_mark_inode_dirty(handle, inode); |
971 | if (err) { | |
972 | ext4_std_error(sb, err); | |
973 | goto fail_free_drop; | |
974 | } | |
975 | ||
617ba13b | 976 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 977 | trace_ext4_allocate_inode(inode, dir, mode); |
3300beda | 978 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
979 | return ret; |
980 | ||
981 | fail_free_drop: | |
63936dda | 982 | dquot_free_inode(inode); |
ac27a0ec | 983 | fail_drop: |
6d6b77f1 | 984 | clear_nlink(inode); |
6b38e842 | 985 | unlock_new_inode(inode); |
eb9cc7e1 JK |
986 | out: |
987 | dquot_drop(inode); | |
988 | inode->i_flags |= S_NOQUOTA; | |
ac27a0ec | 989 | iput(inode); |
3300beda | 990 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
991 | return ERR_PTR(err); |
992 | } | |
993 | ||
994 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 995 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 996 | { |
617ba13b | 997 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 998 | ext4_group_t block_group; |
ac27a0ec | 999 | int bit; |
1d1fe1ee | 1000 | struct buffer_head *bitmap_bh; |
ac27a0ec | 1001 | struct inode *inode = NULL; |
1d1fe1ee | 1002 | long err = -EIO; |
ac27a0ec DK |
1003 | |
1004 | /* Error cases - e2fsck has already cleaned up for us */ | |
1005 | if (ino > max_ino) { | |
12062ddd | 1006 | ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 1007 | goto error; |
ac27a0ec DK |
1008 | } |
1009 | ||
617ba13b MC |
1010 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1011 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1012 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 1013 | if (!bitmap_bh) { |
12062ddd | 1014 | ext4_warning(sb, "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 1015 | goto error; |
ac27a0ec DK |
1016 | } |
1017 | ||
1018 | /* Having the inode bit set should be a 100% indicator that this | |
1019 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1020 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1021 | */ | |
1d1fe1ee DH |
1022 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1023 | goto bad_orphan; | |
1024 | ||
1025 | inode = ext4_iget(sb, ino); | |
1026 | if (IS_ERR(inode)) | |
1027 | goto iget_failed; | |
1028 | ||
91ef4caf | 1029 | /* |
3d1658bb TT |
1030 | * If the orphans has i_nlinks > 0 then it should be able to |
1031 | * be truncated, otherwise it won't be removed from the orphan | |
1032 | * list during processing and an infinite loop will result. | |
1033 | * Similarly, it must not be a bad inode. | |
91ef4caf | 1034 | */ |
3d1658bb TT |
1035 | if ((inode->i_nlink && !ext4_can_truncate(inode)) || |
1036 | is_bad_inode(inode)) | |
91ef4caf DG |
1037 | goto bad_orphan; |
1038 | ||
1d1fe1ee DH |
1039 | if (NEXT_ORPHAN(inode) > max_ino) |
1040 | goto bad_orphan; | |
1041 | brelse(bitmap_bh); | |
1042 | return inode; | |
1043 | ||
1044 | iget_failed: | |
1045 | err = PTR_ERR(inode); | |
1046 | inode = NULL; | |
1047 | bad_orphan: | |
12062ddd | 1048 | ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino); |
8de5c325 | 1049 | printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n", |
1d1fe1ee DH |
1050 | bit, (unsigned long long)bitmap_bh->b_blocknr, |
1051 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
8de5c325 | 1052 | printk(KERN_WARNING "inode=%p\n", inode); |
1d1fe1ee | 1053 | if (inode) { |
8de5c325 | 1054 | printk(KERN_WARNING "is_bad_inode(inode)=%d\n", |
1d1fe1ee | 1055 | is_bad_inode(inode)); |
8de5c325 | 1056 | printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n", |
1d1fe1ee | 1057 | NEXT_ORPHAN(inode)); |
8de5c325 TT |
1058 | printk(KERN_WARNING "max_ino=%lu\n", max_ino); |
1059 | printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink); | |
ac27a0ec | 1060 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1061 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1062 | inode->i_blocks = 0; |
1063 | iput(inode); | |
ac27a0ec | 1064 | } |
ac27a0ec | 1065 | brelse(bitmap_bh); |
1d1fe1ee DH |
1066 | error: |
1067 | return ERR_PTR(err); | |
ac27a0ec DK |
1068 | } |
1069 | ||
af5bc92d | 1070 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1071 | { |
1072 | unsigned long desc_count; | |
617ba13b | 1073 | struct ext4_group_desc *gdp; |
8df9675f | 1074 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1075 | #ifdef EXT4FS_DEBUG |
1076 | struct ext4_super_block *es; | |
ac27a0ec DK |
1077 | unsigned long bitmap_count, x; |
1078 | struct buffer_head *bitmap_bh = NULL; | |
1079 | ||
617ba13b | 1080 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1081 | desc_count = 0; |
1082 | bitmap_count = 0; | |
1083 | gdp = NULL; | |
8df9675f | 1084 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1085 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1086 | if (!gdp) |
1087 | continue; | |
560671a0 | 1088 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1089 | brelse(bitmap_bh); |
e29d1cde | 1090 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
1091 | if (!bitmap_bh) |
1092 | continue; | |
1093 | ||
f6fb99ca TT |
1094 | x = ext4_count_free(bitmap_bh->b_data, |
1095 | EXT4_INODES_PER_GROUP(sb) / 8); | |
c549a95d | 1096 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1097 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1098 | bitmap_count += x; |
1099 | } | |
1100 | brelse(bitmap_bh); | |
4776004f TT |
1101 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1102 | "stored = %u, computed = %lu, %lu\n", | |
1103 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1104 | return desc_count; |
1105 | #else | |
1106 | desc_count = 0; | |
8df9675f | 1107 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1108 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1109 | if (!gdp) |
1110 | continue; | |
560671a0 | 1111 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1112 | cond_resched(); |
1113 | } | |
1114 | return desc_count; | |
1115 | #endif | |
1116 | } | |
1117 | ||
1118 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1119 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1120 | { |
1121 | unsigned long count = 0; | |
8df9675f | 1122 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1123 | |
8df9675f | 1124 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1125 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1126 | if (!gdp) |
1127 | continue; | |
560671a0 | 1128 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1129 | } |
1130 | return count; | |
1131 | } | |
bfff6873 LC |
1132 | |
1133 | /* | |
1134 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1135 | * inode table. Must be called without any spinlock held. The only place | |
1136 | * where it is called from on active part of filesystem is ext4lazyinit | |
1137 | * thread, so we do not need any special locks, however we have to prevent | |
1138 | * inode allocation from the current group, so we take alloc_sem lock, to | |
119c0d44 | 1139 | * block ext4_new_inode() until we are finished. |
bfff6873 | 1140 | */ |
e0cbee3e | 1141 | int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, |
bfff6873 LC |
1142 | int barrier) |
1143 | { | |
1144 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1145 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1146 | struct ext4_group_desc *gdp = NULL; | |
1147 | struct buffer_head *group_desc_bh; | |
1148 | handle_t *handle; | |
1149 | ext4_fsblk_t blk; | |
1150 | int num, ret = 0, used_blks = 0; | |
bfff6873 LC |
1151 | |
1152 | /* This should not happen, but just to be sure check this */ | |
1153 | if (sb->s_flags & MS_RDONLY) { | |
1154 | ret = 1; | |
1155 | goto out; | |
1156 | } | |
1157 | ||
1158 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1159 | if (!gdp) | |
1160 | goto out; | |
1161 | ||
1162 | /* | |
1163 | * We do not need to lock this, because we are the only one | |
1164 | * handling this flag. | |
1165 | */ | |
1166 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1167 | goto out; | |
1168 | ||
9924a92a | 1169 | handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); |
bfff6873 LC |
1170 | if (IS_ERR(handle)) { |
1171 | ret = PTR_ERR(handle); | |
1172 | goto out; | |
1173 | } | |
1174 | ||
1175 | down_write(&grp->alloc_sem); | |
1176 | /* | |
1177 | * If inode bitmap was already initialized there may be some | |
1178 | * used inodes so we need to skip blocks with used inodes in | |
1179 | * inode table. | |
1180 | */ | |
1181 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1182 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1183 | ext4_itable_unused_count(sb, gdp)), | |
1184 | sbi->s_inodes_per_block); | |
1185 | ||
857ac889 | 1186 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) { |
1084f252 TT |
1187 | ext4_error(sb, "Something is wrong with group %u: " |
1188 | "used itable blocks: %d; " | |
1189 | "itable unused count: %u", | |
857ac889 LC |
1190 | group, used_blks, |
1191 | ext4_itable_unused_count(sb, gdp)); | |
1192 | ret = 1; | |
33853a0d | 1193 | goto err_out; |
857ac889 LC |
1194 | } |
1195 | ||
bfff6873 LC |
1196 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1197 | num = sbi->s_itb_per_group - used_blks; | |
1198 | ||
1199 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1200 | ret = ext4_journal_get_write_access(handle, | |
1201 | group_desc_bh); | |
1202 | if (ret) | |
1203 | goto err_out; | |
1204 | ||
bfff6873 LC |
1205 | /* |
1206 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1207 | * flag anyway, because obviously, when it is full it does not need | |
1208 | * further zeroing. | |
1209 | */ | |
1210 | if (unlikely(num == 0)) | |
1211 | goto skip_zeroout; | |
1212 | ||
1213 | ext4_debug("going to zero out inode table in group %d\n", | |
1214 | group); | |
a107e5a3 | 1215 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1216 | if (ret < 0) |
1217 | goto err_out; | |
a107e5a3 TT |
1218 | if (barrier) |
1219 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); | |
bfff6873 LC |
1220 | |
1221 | skip_zeroout: | |
1222 | ext4_lock_group(sb, group); | |
1223 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
feb0ab32 | 1224 | ext4_group_desc_csum_set(sb, group, gdp); |
bfff6873 LC |
1225 | ext4_unlock_group(sb, group); |
1226 | ||
1227 | BUFFER_TRACE(group_desc_bh, | |
1228 | "call ext4_handle_dirty_metadata"); | |
1229 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1230 | group_desc_bh); | |
1231 | ||
1232 | err_out: | |
1233 | up_write(&grp->alloc_sem); | |
1234 | ext4_journal_stop(handle); | |
1235 | out: | |
1236 | return ret; | |
1237 | } |