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> |
617ba13b | 18 | #include <linux/ext4_fs.h> |
dab291af | 19 | #include <linux/ext4_jbd2.h> |
ac27a0ec DK |
20 | #include <linux/stat.h> |
21 | #include <linux/string.h> | |
22 | #include <linux/quotaops.h> | |
23 | #include <linux/buffer_head.h> | |
24 | #include <linux/random.h> | |
25 | #include <linux/bitops.h> | |
3a5b2ecd | 26 | #include <linux/blkdev.h> |
ac27a0ec DK |
27 | #include <asm/byteorder.h> |
28 | ||
29 | #include "xattr.h" | |
30 | #include "acl.h" | |
717d50e4 | 31 | #include "group.h" |
ac27a0ec DK |
32 | |
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 | */ | |
52 | void mark_bitmap_end(int start_bit, int end_bit, char *bitmap) | |
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 */ | |
67 | unsigned ext4_init_inode_bitmap(struct super_block *sb, | |
68 | struct buffer_head *bh, int block_group, | |
69 | struct ext4_group_desc *gdp) | |
70 | { | |
71 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
72 | ||
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. */ | |
77 | if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) { | |
78 | ext4_error(sb, __FUNCTION__, "Checksum bad for group %u\n", | |
79 | block_group); | |
80 | gdp->bg_free_blocks_count = 0; | |
81 | gdp->bg_free_inodes_count = 0; | |
82 | gdp->bg_itable_unused = 0; | |
83 | memset(bh->b_data, 0xff, sb->s_blocksize); | |
84 | return 0; | |
85 | } | |
86 | ||
87 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
88 | mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb), | |
89 | bh->b_data); | |
90 | ||
91 | return EXT4_INODES_PER_GROUP(sb); | |
92 | } | |
ac27a0ec DK |
93 | |
94 | /* | |
95 | * Read the inode allocation bitmap for a given block_group, reading | |
96 | * into the specified slot in the superblock's bitmap cache. | |
97 | * | |
98 | * Return buffer_head of bitmap on success or NULL. | |
99 | */ | |
100 | static struct buffer_head * | |
101 | read_inode_bitmap(struct super_block * sb, unsigned long block_group) | |
102 | { | |
617ba13b | 103 | struct ext4_group_desc *desc; |
ac27a0ec DK |
104 | struct buffer_head *bh = NULL; |
105 | ||
617ba13b | 106 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec DK |
107 | if (!desc) |
108 | goto error_out; | |
717d50e4 AD |
109 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
110 | bh = sb_getblk(sb, ext4_inode_bitmap(sb, desc)); | |
111 | if (!buffer_uptodate(bh)) { | |
112 | lock_buffer(bh); | |
113 | if (!buffer_uptodate(bh)) { | |
114 | ext4_init_inode_bitmap(sb, bh, block_group, | |
115 | desc); | |
116 | set_buffer_uptodate(bh); | |
117 | } | |
118 | unlock_buffer(bh); | |
119 | } | |
120 | } else { | |
121 | bh = sb_bread(sb, ext4_inode_bitmap(sb, desc)); | |
122 | } | |
ac27a0ec | 123 | if (!bh) |
617ba13b | 124 | ext4_error(sb, "read_inode_bitmap", |
ac27a0ec | 125 | "Cannot read inode bitmap - " |
bd81d8ee | 126 | "block_group = %lu, inode_bitmap = %llu", |
8fadc143 | 127 | block_group, ext4_inode_bitmap(sb, desc)); |
ac27a0ec DK |
128 | error_out: |
129 | return bh; | |
130 | } | |
131 | ||
132 | /* | |
133 | * NOTE! When we get the inode, we're the only people | |
134 | * that have access to it, and as such there are no | |
135 | * race conditions we have to worry about. The inode | |
136 | * is not on the hash-lists, and it cannot be reached | |
137 | * through the filesystem because the directory entry | |
138 | * has been deleted earlier. | |
139 | * | |
140 | * HOWEVER: we must make sure that we get no aliases, | |
141 | * which means that we have to call "clear_inode()" | |
142 | * _before_ we mark the inode not in use in the inode | |
143 | * bitmaps. Otherwise a newly created file might use | |
144 | * the same inode number (not actually the same pointer | |
145 | * though), and then we'd have two inodes sharing the | |
146 | * same inode number and space on the harddisk. | |
147 | */ | |
617ba13b | 148 | void ext4_free_inode (handle_t *handle, struct inode * inode) |
ac27a0ec DK |
149 | { |
150 | struct super_block * sb = inode->i_sb; | |
151 | int is_directory; | |
152 | unsigned long ino; | |
153 | struct buffer_head *bitmap_bh = NULL; | |
154 | struct buffer_head *bh2; | |
155 | unsigned long block_group; | |
156 | unsigned long bit; | |
617ba13b MC |
157 | struct ext4_group_desc * gdp; |
158 | struct ext4_super_block * es; | |
159 | struct ext4_sb_info *sbi; | |
ac27a0ec DK |
160 | int fatal = 0, err; |
161 | ||
162 | if (atomic_read(&inode->i_count) > 1) { | |
617ba13b | 163 | printk ("ext4_free_inode: inode has count=%d\n", |
ac27a0ec DK |
164 | atomic_read(&inode->i_count)); |
165 | return; | |
166 | } | |
167 | if (inode->i_nlink) { | |
617ba13b | 168 | printk ("ext4_free_inode: inode has nlink=%d\n", |
ac27a0ec DK |
169 | inode->i_nlink); |
170 | return; | |
171 | } | |
172 | if (!sb) { | |
617ba13b | 173 | printk("ext4_free_inode: inode on nonexistent device\n"); |
ac27a0ec DK |
174 | return; |
175 | } | |
617ba13b | 176 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
177 | |
178 | ino = inode->i_ino; | |
617ba13b | 179 | ext4_debug ("freeing inode %lu\n", ino); |
ac27a0ec DK |
180 | |
181 | /* | |
182 | * Note: we must free any quota before locking the superblock, | |
183 | * as writing the quota to disk may need the lock as well. | |
184 | */ | |
185 | DQUOT_INIT(inode); | |
617ba13b | 186 | ext4_xattr_delete_inode(handle, inode); |
ac27a0ec DK |
187 | DQUOT_FREE_INODE(inode); |
188 | DQUOT_DROP(inode); | |
189 | ||
190 | is_directory = S_ISDIR(inode->i_mode); | |
191 | ||
192 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
193 | clear_inode (inode); | |
194 | ||
617ba13b MC |
195 | es = EXT4_SB(sb)->s_es; |
196 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
197 | ext4_error (sb, "ext4_free_inode", | |
ac27a0ec DK |
198 | "reserved or nonexistent inode %lu", ino); |
199 | goto error_return; | |
200 | } | |
617ba13b MC |
201 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
202 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
ac27a0ec DK |
203 | bitmap_bh = read_inode_bitmap(sb, block_group); |
204 | if (!bitmap_bh) | |
205 | goto error_return; | |
206 | ||
207 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 208 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
209 | if (fatal) |
210 | goto error_return; | |
211 | ||
212 | /* Ok, now we can actually update the inode bitmaps.. */ | |
617ba13b | 213 | if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
ac27a0ec | 214 | bit, bitmap_bh->b_data)) |
617ba13b | 215 | ext4_error (sb, "ext4_free_inode", |
ac27a0ec DK |
216 | "bit already cleared for inode %lu", ino); |
217 | else { | |
617ba13b | 218 | gdp = ext4_get_group_desc (sb, block_group, &bh2); |
ac27a0ec DK |
219 | |
220 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 221 | fatal = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec DK |
222 | if (fatal) goto error_return; |
223 | ||
224 | if (gdp) { | |
225 | spin_lock(sb_bgl_lock(sbi, block_group)); | |
226 | gdp->bg_free_inodes_count = cpu_to_le16( | |
227 | le16_to_cpu(gdp->bg_free_inodes_count) + 1); | |
228 | if (is_directory) | |
229 | gdp->bg_used_dirs_count = cpu_to_le16( | |
230 | le16_to_cpu(gdp->bg_used_dirs_count) - 1); | |
717d50e4 AD |
231 | gdp->bg_checksum = ext4_group_desc_csum(sbi, |
232 | block_group, gdp); | |
ac27a0ec DK |
233 | spin_unlock(sb_bgl_lock(sbi, block_group)); |
234 | percpu_counter_inc(&sbi->s_freeinodes_counter); | |
235 | if (is_directory) | |
236 | percpu_counter_dec(&sbi->s_dirs_counter); | |
237 | ||
238 | } | |
617ba13b MC |
239 | BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata"); |
240 | err = ext4_journal_dirty_metadata(handle, bh2); | |
ac27a0ec DK |
241 | if (!fatal) fatal = err; |
242 | } | |
617ba13b MC |
243 | BUFFER_TRACE(bitmap_bh, "call ext4_journal_dirty_metadata"); |
244 | err = ext4_journal_dirty_metadata(handle, bitmap_bh); | |
ac27a0ec DK |
245 | if (!fatal) |
246 | fatal = err; | |
247 | sb->s_dirt = 1; | |
248 | error_return: | |
249 | brelse(bitmap_bh); | |
617ba13b | 250 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
251 | } |
252 | ||
253 | /* | |
254 | * There are two policies for allocating an inode. If the new inode is | |
255 | * a directory, then a forward search is made for a block group with both | |
256 | * free space and a low directory-to-inode ratio; if that fails, then of | |
257 | * the groups with above-average free space, that group with the fewest | |
258 | * directories already is chosen. | |
259 | * | |
260 | * For other inodes, search forward from the parent directory\'s block | |
261 | * group to find a free inode. | |
262 | */ | |
263 | static int find_group_dir(struct super_block *sb, struct inode *parent) | |
264 | { | |
617ba13b | 265 | int ngroups = EXT4_SB(sb)->s_groups_count; |
ac27a0ec | 266 | unsigned int freei, avefreei; |
617ba13b | 267 | struct ext4_group_desc *desc, *best_desc = NULL; |
ac27a0ec DK |
268 | int group, best_group = -1; |
269 | ||
617ba13b | 270 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
271 | avefreei = freei / ngroups; |
272 | ||
273 | for (group = 0; group < ngroups; group++) { | |
ef2fb679 | 274 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
275 | if (!desc || !desc->bg_free_inodes_count) |
276 | continue; | |
277 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
278 | continue; | |
279 | if (!best_desc || | |
280 | (le16_to_cpu(desc->bg_free_blocks_count) > | |
281 | le16_to_cpu(best_desc->bg_free_blocks_count))) { | |
282 | best_group = group; | |
283 | best_desc = desc; | |
284 | } | |
285 | } | |
286 | return best_group; | |
287 | } | |
288 | ||
289 | /* | |
290 | * Orlov's allocator for directories. | |
291 | * | |
292 | * We always try to spread first-level directories. | |
293 | * | |
294 | * If there are blockgroups with both free inodes and free blocks counts | |
295 | * not worse than average we return one with smallest directory count. | |
296 | * Otherwise we simply return a random group. | |
297 | * | |
298 | * For the rest rules look so: | |
299 | * | |
300 | * It's OK to put directory into a group unless | |
301 | * it has too many directories already (max_dirs) or | |
302 | * it has too few free inodes left (min_inodes) or | |
303 | * it has too few free blocks left (min_blocks) or | |
304 | * it's already running too large debt (max_debt). | |
305 | * Parent's group is prefered, if it doesn't satisfy these | |
306 | * conditions we search cyclically through the rest. If none | |
307 | * of the groups look good we just look for a group with more | |
308 | * free inodes than average (starting at parent's group). | |
309 | * | |
310 | * Debt is incremented each time we allocate a directory and decremented | |
311 | * when we allocate an inode, within 0--255. | |
312 | */ | |
313 | ||
314 | #define INODE_COST 64 | |
315 | #define BLOCK_COST 256 | |
316 | ||
317 | static int find_group_orlov(struct super_block *sb, struct inode *parent) | |
318 | { | |
617ba13b MC |
319 | int parent_group = EXT4_I(parent)->i_block_group; |
320 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
321 | struct ext4_super_block *es = sbi->s_es; | |
ac27a0ec | 322 | int ngroups = sbi->s_groups_count; |
617ba13b | 323 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 324 | unsigned int freei, avefreei; |
617ba13b MC |
325 | ext4_fsblk_t freeb, avefreeb; |
326 | ext4_fsblk_t blocks_per_dir; | |
ac27a0ec DK |
327 | unsigned int ndirs; |
328 | int max_debt, max_dirs, min_inodes; | |
617ba13b | 329 | ext4_grpblk_t min_blocks; |
ac27a0ec | 330 | int group = -1, i; |
617ba13b | 331 | struct ext4_group_desc *desc; |
ac27a0ec DK |
332 | |
333 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
334 | avefreei = freei / ngroups; | |
335 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3a5b2ecd | 336 | avefreeb = freeb; |
f4e5bc24 | 337 | do_div(avefreeb, ngroups); |
ac27a0ec DK |
338 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
339 | ||
340 | if ((parent == sb->s_root->d_inode) || | |
617ba13b | 341 | (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) { |
ac27a0ec DK |
342 | int best_ndir = inodes_per_group; |
343 | int best_group = -1; | |
344 | ||
345 | get_random_bytes(&group, sizeof(group)); | |
346 | parent_group = (unsigned)group % ngroups; | |
347 | for (i = 0; i < ngroups; i++) { | |
348 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 349 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
350 | if (!desc || !desc->bg_free_inodes_count) |
351 | continue; | |
352 | if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) | |
353 | continue; | |
354 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
355 | continue; | |
356 | if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) | |
357 | continue; | |
358 | best_group = group; | |
359 | best_ndir = le16_to_cpu(desc->bg_used_dirs_count); | |
360 | } | |
361 | if (best_group >= 0) | |
362 | return best_group; | |
363 | goto fallback; | |
364 | } | |
365 | ||
bd81d8ee | 366 | blocks_per_dir = ext4_blocks_count(es) - freeb; |
f4e5bc24 | 367 | do_div(blocks_per_dir, ndirs); |
ac27a0ec DK |
368 | |
369 | max_dirs = ndirs / ngroups + inodes_per_group / 16; | |
370 | min_inodes = avefreei - inodes_per_group / 4; | |
617ba13b | 371 | min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4; |
ac27a0ec | 372 | |
3a5b2ecd | 373 | max_debt = EXT4_BLOCKS_PER_GROUP(sb); |
f4e5bc24 | 374 | max_debt /= max_t(int, blocks_per_dir, BLOCK_COST); |
ac27a0ec DK |
375 | if (max_debt * INODE_COST > inodes_per_group) |
376 | max_debt = inodes_per_group / INODE_COST; | |
377 | if (max_debt > 255) | |
378 | max_debt = 255; | |
379 | if (max_debt == 0) | |
380 | max_debt = 1; | |
381 | ||
382 | for (i = 0; i < ngroups; i++) { | |
383 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 384 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
385 | if (!desc || !desc->bg_free_inodes_count) |
386 | continue; | |
387 | if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs) | |
388 | continue; | |
389 | if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes) | |
390 | continue; | |
391 | if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks) | |
392 | continue; | |
393 | return group; | |
394 | } | |
395 | ||
396 | fallback: | |
397 | for (i = 0; i < ngroups; i++) { | |
398 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 399 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
400 | if (!desc || !desc->bg_free_inodes_count) |
401 | continue; | |
402 | if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei) | |
403 | return group; | |
404 | } | |
405 | ||
406 | if (avefreei) { | |
407 | /* | |
408 | * The free-inodes counter is approximate, and for really small | |
409 | * filesystems the above test can fail to find any blockgroups | |
410 | */ | |
411 | avefreei = 0; | |
412 | goto fallback; | |
413 | } | |
414 | ||
415 | return -1; | |
416 | } | |
417 | ||
418 | static int find_group_other(struct super_block *sb, struct inode *parent) | |
419 | { | |
617ba13b MC |
420 | int parent_group = EXT4_I(parent)->i_block_group; |
421 | int ngroups = EXT4_SB(sb)->s_groups_count; | |
422 | struct ext4_group_desc *desc; | |
ac27a0ec DK |
423 | int group, i; |
424 | ||
425 | /* | |
426 | * Try to place the inode in its parent directory | |
427 | */ | |
428 | group = parent_group; | |
ef2fb679 | 429 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
430 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
431 | le16_to_cpu(desc->bg_free_blocks_count)) | |
432 | return group; | |
433 | ||
434 | /* | |
435 | * We're going to place this inode in a different blockgroup from its | |
436 | * parent. We want to cause files in a common directory to all land in | |
437 | * the same blockgroup. But we want files which are in a different | |
438 | * directory which shares a blockgroup with our parent to land in a | |
439 | * different blockgroup. | |
440 | * | |
441 | * So add our directory's i_ino into the starting point for the hash. | |
442 | */ | |
443 | group = (group + parent->i_ino) % ngroups; | |
444 | ||
445 | /* | |
446 | * Use a quadratic hash to find a group with a free inode and some free | |
447 | * blocks. | |
448 | */ | |
449 | for (i = 1; i < ngroups; i <<= 1) { | |
450 | group += i; | |
451 | if (group >= ngroups) | |
452 | group -= ngroups; | |
ef2fb679 | 453 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
454 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
455 | le16_to_cpu(desc->bg_free_blocks_count)) | |
456 | return group; | |
457 | } | |
458 | ||
459 | /* | |
460 | * That failed: try linear search for a free inode, even if that group | |
461 | * has no free blocks. | |
462 | */ | |
463 | group = parent_group; | |
464 | for (i = 0; i < ngroups; i++) { | |
465 | if (++group >= ngroups) | |
466 | group = 0; | |
ef2fb679 | 467 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
468 | if (desc && le16_to_cpu(desc->bg_free_inodes_count)) |
469 | return group; | |
470 | } | |
471 | ||
472 | return -1; | |
473 | } | |
474 | ||
475 | /* | |
476 | * There are two policies for allocating an inode. If the new inode is | |
477 | * a directory, then a forward search is made for a block group with both | |
478 | * free space and a low directory-to-inode ratio; if that fails, then of | |
479 | * the groups with above-average free space, that group with the fewest | |
480 | * directories already is chosen. | |
481 | * | |
482 | * For other inodes, search forward from the parent directory's block | |
483 | * group to find a free inode. | |
484 | */ | |
617ba13b | 485 | struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode) |
ac27a0ec DK |
486 | { |
487 | struct super_block *sb; | |
488 | struct buffer_head *bitmap_bh = NULL; | |
489 | struct buffer_head *bh2; | |
490 | int group; | |
491 | unsigned long ino = 0; | |
492 | struct inode * inode; | |
617ba13b MC |
493 | struct ext4_group_desc * gdp = NULL; |
494 | struct ext4_super_block * es; | |
495 | struct ext4_inode_info *ei; | |
496 | struct ext4_sb_info *sbi; | |
ac27a0ec DK |
497 | int err = 0; |
498 | struct inode *ret; | |
717d50e4 | 499 | int i, free = 0; |
ac27a0ec DK |
500 | |
501 | /* Cannot create files in a deleted directory */ | |
502 | if (!dir || !dir->i_nlink) | |
503 | return ERR_PTR(-EPERM); | |
504 | ||
505 | sb = dir->i_sb; | |
506 | inode = new_inode(sb); | |
507 | if (!inode) | |
508 | return ERR_PTR(-ENOMEM); | |
617ba13b | 509 | ei = EXT4_I(inode); |
ac27a0ec | 510 | |
617ba13b | 511 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
512 | es = sbi->s_es; |
513 | if (S_ISDIR(mode)) { | |
514 | if (test_opt (sb, OLDALLOC)) | |
515 | group = find_group_dir(sb, dir); | |
516 | else | |
517 | group = find_group_orlov(sb, dir); | |
518 | } else | |
519 | group = find_group_other(sb, dir); | |
520 | ||
521 | err = -ENOSPC; | |
522 | if (group == -1) | |
523 | goto out; | |
524 | ||
525 | for (i = 0; i < sbi->s_groups_count; i++) { | |
526 | err = -EIO; | |
527 | ||
617ba13b | 528 | gdp = ext4_get_group_desc(sb, group, &bh2); |
ac27a0ec DK |
529 | if (!gdp) |
530 | goto fail; | |
531 | ||
532 | brelse(bitmap_bh); | |
533 | bitmap_bh = read_inode_bitmap(sb, group); | |
534 | if (!bitmap_bh) | |
535 | goto fail; | |
536 | ||
537 | ino = 0; | |
538 | ||
539 | repeat_in_this_group: | |
617ba13b MC |
540 | ino = ext4_find_next_zero_bit((unsigned long *) |
541 | bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino); | |
542 | if (ino < EXT4_INODES_PER_GROUP(sb)) { | |
ac27a0ec DK |
543 | |
544 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 545 | err = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
546 | if (err) |
547 | goto fail; | |
548 | ||
617ba13b | 549 | if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group), |
ac27a0ec DK |
550 | ino, bitmap_bh->b_data)) { |
551 | /* we won it */ | |
552 | BUFFER_TRACE(bitmap_bh, | |
617ba13b MC |
553 | "call ext4_journal_dirty_metadata"); |
554 | err = ext4_journal_dirty_metadata(handle, | |
ac27a0ec DK |
555 | bitmap_bh); |
556 | if (err) | |
557 | goto fail; | |
558 | goto got; | |
559 | } | |
560 | /* we lost it */ | |
dab291af | 561 | jbd2_journal_release_buffer(handle, bitmap_bh); |
ac27a0ec | 562 | |
617ba13b | 563 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
564 | goto repeat_in_this_group; |
565 | } | |
566 | ||
567 | /* | |
568 | * This case is possible in concurrent environment. It is very | |
569 | * rare. We cannot repeat the find_group_xxx() call because | |
570 | * that will simply return the same blockgroup, because the | |
571 | * group descriptor metadata has not yet been updated. | |
572 | * So we just go onto the next blockgroup. | |
573 | */ | |
574 | if (++group == sbi->s_groups_count) | |
575 | group = 0; | |
576 | } | |
577 | err = -ENOSPC; | |
578 | goto out; | |
579 | ||
580 | got: | |
717d50e4 AD |
581 | ino++; |
582 | if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || | |
583 | ino > EXT4_INODES_PER_GROUP(sb)) { | |
584 | ext4_error(sb, __FUNCTION__, | |
585 | "reserved inode or inode > inodes count - " | |
586 | "block_group = %d, inode=%lu", group, | |
587 | ino + group * EXT4_INODES_PER_GROUP(sb)); | |
ac27a0ec DK |
588 | err = -EIO; |
589 | goto fail; | |
590 | } | |
591 | ||
592 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 593 | err = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec | 594 | if (err) goto fail; |
717d50e4 AD |
595 | |
596 | /* We may have to initialize the block bitmap if it isn't already */ | |
597 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && | |
598 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
599 | struct buffer_head *block_bh = read_block_bitmap(sb, group); | |
600 | ||
601 | BUFFER_TRACE(block_bh, "get block bitmap access"); | |
602 | err = ext4_journal_get_write_access(handle, block_bh); | |
603 | if (err) { | |
604 | brelse(block_bh); | |
605 | goto fail; | |
606 | } | |
607 | ||
608 | free = 0; | |
609 | spin_lock(sb_bgl_lock(sbi, group)); | |
610 | /* recheck and clear flag under lock if we still need to */ | |
611 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
612 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); | |
613 | free = ext4_free_blocks_after_init(sb, group, gdp); | |
614 | gdp->bg_free_blocks_count = cpu_to_le16(free); | |
615 | } | |
616 | spin_unlock(sb_bgl_lock(sbi, group)); | |
617 | ||
618 | /* Don't need to dirty bitmap block if we didn't change it */ | |
619 | if (free) { | |
620 | BUFFER_TRACE(block_bh, "dirty block bitmap"); | |
621 | err = ext4_journal_dirty_metadata(handle, block_bh); | |
622 | } | |
623 | ||
624 | brelse(block_bh); | |
625 | if (err) | |
626 | goto fail; | |
627 | } | |
628 | ||
ac27a0ec | 629 | spin_lock(sb_bgl_lock(sbi, group)); |
717d50e4 AD |
630 | /* If we didn't allocate from within the initialized part of the inode |
631 | * table then we need to initialize up to this inode. */ | |
632 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { | |
633 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
634 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
635 | ||
636 | /* When marking the block group with | |
637 | * ~EXT4_BG_INODE_UNINIT we don't want to depend | |
638 | * on the value of bg_itable_unsed even though | |
639 | * mke2fs could have initialized the same for us. | |
640 | * Instead we calculated the value below | |
641 | */ | |
642 | ||
643 | free = 0; | |
644 | } else { | |
645 | free = EXT4_INODES_PER_GROUP(sb) - | |
646 | le16_to_cpu(gdp->bg_itable_unused); | |
647 | } | |
648 | ||
649 | /* | |
650 | * Check the relative inode number against the last used | |
651 | * relative inode number in this group. if it is greater | |
652 | * we need to update the bg_itable_unused count | |
653 | * | |
654 | */ | |
655 | if (ino > free) | |
656 | gdp->bg_itable_unused = | |
657 | cpu_to_le16(EXT4_INODES_PER_GROUP(sb) - ino); | |
658 | } | |
659 | ||
ac27a0ec DK |
660 | gdp->bg_free_inodes_count = |
661 | cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1); | |
662 | if (S_ISDIR(mode)) { | |
663 | gdp->bg_used_dirs_count = | |
664 | cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1); | |
665 | } | |
717d50e4 | 666 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); |
ac27a0ec | 667 | spin_unlock(sb_bgl_lock(sbi, group)); |
617ba13b MC |
668 | BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata"); |
669 | err = ext4_journal_dirty_metadata(handle, bh2); | |
ac27a0ec DK |
670 | if (err) goto fail; |
671 | ||
672 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
673 | if (S_ISDIR(mode)) | |
674 | percpu_counter_inc(&sbi->s_dirs_counter); | |
675 | sb->s_dirt = 1; | |
676 | ||
677 | inode->i_uid = current->fsuid; | |
678 | if (test_opt (sb, GRPID)) | |
679 | inode->i_gid = dir->i_gid; | |
680 | else if (dir->i_mode & S_ISGID) { | |
681 | inode->i_gid = dir->i_gid; | |
682 | if (S_ISDIR(mode)) | |
683 | mode |= S_ISGID; | |
684 | } else | |
685 | inode->i_gid = current->fsgid; | |
686 | inode->i_mode = mode; | |
687 | ||
717d50e4 | 688 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
689 | /* This is the optimal IO size (for stat), not the fs block size */ |
690 | inode->i_blocks = 0; | |
ef7f3835 KS |
691 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
692 | ext4_current_time(inode); | |
ac27a0ec DK |
693 | |
694 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
695 | ei->i_dir_start_lookup = 0; | |
696 | ei->i_disksize = 0; | |
697 | ||
617ba13b | 698 | ei->i_flags = EXT4_I(dir)->i_flags & ~EXT4_INDEX_FL; |
ac27a0ec | 699 | if (S_ISLNK(mode)) |
617ba13b | 700 | ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL); |
ac27a0ec DK |
701 | /* dirsync only applies to directories */ |
702 | if (!S_ISDIR(mode)) | |
617ba13b | 703 | ei->i_flags &= ~EXT4_DIRSYNC_FL; |
ac27a0ec DK |
704 | ei->i_file_acl = 0; |
705 | ei->i_dir_acl = 0; | |
706 | ei->i_dtime = 0; | |
707 | ei->i_block_alloc_info = NULL; | |
708 | ei->i_block_group = group; | |
709 | ||
617ba13b | 710 | ext4_set_inode_flags(inode); |
ac27a0ec DK |
711 | if (IS_DIRSYNC(inode)) |
712 | handle->h_sync = 1; | |
713 | insert_inode_hash(inode); | |
714 | spin_lock(&sbi->s_next_gen_lock); | |
715 | inode->i_generation = sbi->s_next_generation++; | |
716 | spin_unlock(&sbi->s_next_gen_lock); | |
717 | ||
617ba13b | 718 | ei->i_state = EXT4_STATE_NEW; |
ef7f3835 KS |
719 | |
720 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
721 | |
722 | ret = inode; | |
723 | if(DQUOT_ALLOC_INODE(inode)) { | |
724 | err = -EDQUOT; | |
725 | goto fail_drop; | |
726 | } | |
727 | ||
617ba13b | 728 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
729 | if (err) |
730 | goto fail_free_drop; | |
731 | ||
617ba13b | 732 | err = ext4_init_security(handle,inode, dir); |
ac27a0ec DK |
733 | if (err) |
734 | goto fail_free_drop; | |
735 | ||
617ba13b | 736 | err = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 737 | if (err) { |
617ba13b | 738 | ext4_std_error(sb, err); |
ac27a0ec DK |
739 | goto fail_free_drop; |
740 | } | |
a86c6181 AT |
741 | if (test_opt(sb, EXTENTS)) { |
742 | EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL; | |
743 | ext4_ext_tree_init(handle, inode); | |
744 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { | |
745 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); | |
746 | if (err) goto fail; | |
747 | EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS); | |
748 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "call ext4_journal_dirty_metadata"); | |
749 | err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh); | |
750 | } | |
751 | } | |
ac27a0ec | 752 | |
617ba13b | 753 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
ac27a0ec DK |
754 | goto really_out; |
755 | fail: | |
617ba13b | 756 | ext4_std_error(sb, err); |
ac27a0ec DK |
757 | out: |
758 | iput(inode); | |
759 | ret = ERR_PTR(err); | |
760 | really_out: | |
761 | brelse(bitmap_bh); | |
762 | return ret; | |
763 | ||
764 | fail_free_drop: | |
765 | DQUOT_FREE_INODE(inode); | |
766 | ||
767 | fail_drop: | |
768 | DQUOT_DROP(inode); | |
769 | inode->i_flags |= S_NOQUOTA; | |
770 | inode->i_nlink = 0; | |
771 | iput(inode); | |
772 | brelse(bitmap_bh); | |
773 | return ERR_PTR(err); | |
774 | } | |
775 | ||
776 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 777 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 778 | { |
617ba13b | 779 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
ac27a0ec DK |
780 | unsigned long block_group; |
781 | int bit; | |
782 | struct buffer_head *bitmap_bh = NULL; | |
783 | struct inode *inode = NULL; | |
784 | ||
785 | /* Error cases - e2fsck has already cleaned up for us */ | |
786 | if (ino > max_ino) { | |
617ba13b | 787 | ext4_warning(sb, __FUNCTION__, |
ac27a0ec DK |
788 | "bad orphan ino %lu! e2fsck was run?", ino); |
789 | goto out; | |
790 | } | |
791 | ||
617ba13b MC |
792 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
793 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
ac27a0ec DK |
794 | bitmap_bh = read_inode_bitmap(sb, block_group); |
795 | if (!bitmap_bh) { | |
617ba13b | 796 | ext4_warning(sb, __FUNCTION__, |
ac27a0ec DK |
797 | "inode bitmap error for orphan %lu", ino); |
798 | goto out; | |
799 | } | |
800 | ||
801 | /* Having the inode bit set should be a 100% indicator that this | |
802 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
803 | * inodes that were being truncated, so we can't check i_nlink==0. | |
804 | */ | |
617ba13b | 805 | if (!ext4_test_bit(bit, bitmap_bh->b_data) || |
ac27a0ec DK |
806 | !(inode = iget(sb, ino)) || is_bad_inode(inode) || |
807 | NEXT_ORPHAN(inode) > max_ino) { | |
617ba13b | 808 | ext4_warning(sb, __FUNCTION__, |
ac27a0ec | 809 | "bad orphan inode %lu! e2fsck was run?", ino); |
617ba13b | 810 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", |
ac27a0ec | 811 | bit, (unsigned long long)bitmap_bh->b_blocknr, |
617ba13b | 812 | ext4_test_bit(bit, bitmap_bh->b_data)); |
ac27a0ec DK |
813 | printk(KERN_NOTICE "inode=%p\n", inode); |
814 | if (inode) { | |
815 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
816 | is_bad_inode(inode)); | |
817 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
818 | NEXT_ORPHAN(inode)); | |
819 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
820 | } | |
821 | /* Avoid freeing blocks if we got a bad deleted inode */ | |
822 | if (inode && inode->i_nlink == 0) | |
823 | inode->i_blocks = 0; | |
824 | iput(inode); | |
825 | inode = NULL; | |
826 | } | |
827 | out: | |
828 | brelse(bitmap_bh); | |
829 | return inode; | |
830 | } | |
831 | ||
617ba13b | 832 | unsigned long ext4_count_free_inodes (struct super_block * sb) |
ac27a0ec DK |
833 | { |
834 | unsigned long desc_count; | |
617ba13b | 835 | struct ext4_group_desc *gdp; |
ac27a0ec | 836 | int i; |
617ba13b MC |
837 | #ifdef EXT4FS_DEBUG |
838 | struct ext4_super_block *es; | |
ac27a0ec DK |
839 | unsigned long bitmap_count, x; |
840 | struct buffer_head *bitmap_bh = NULL; | |
841 | ||
617ba13b | 842 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
843 | desc_count = 0; |
844 | bitmap_count = 0; | |
845 | gdp = NULL; | |
617ba13b MC |
846 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
847 | gdp = ext4_get_group_desc (sb, i, NULL); | |
ac27a0ec DK |
848 | if (!gdp) |
849 | continue; | |
850 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
851 | brelse(bitmap_bh); | |
852 | bitmap_bh = read_inode_bitmap(sb, i); | |
853 | if (!bitmap_bh) | |
854 | continue; | |
855 | ||
617ba13b | 856 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
ac27a0ec DK |
857 | printk("group %d: stored = %d, counted = %lu\n", |
858 | i, le16_to_cpu(gdp->bg_free_inodes_count), x); | |
859 | bitmap_count += x; | |
860 | } | |
861 | brelse(bitmap_bh); | |
617ba13b | 862 | printk("ext4_count_free_inodes: stored = %u, computed = %lu, %lu\n", |
ac27a0ec DK |
863 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); |
864 | return desc_count; | |
865 | #else | |
866 | desc_count = 0; | |
617ba13b MC |
867 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
868 | gdp = ext4_get_group_desc (sb, i, NULL); | |
ac27a0ec DK |
869 | if (!gdp) |
870 | continue; | |
871 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
872 | cond_resched(); | |
873 | } | |
874 | return desc_count; | |
875 | #endif | |
876 | } | |
877 | ||
878 | /* Called at mount-time, super-block is locked */ | |
617ba13b | 879 | unsigned long ext4_count_dirs (struct super_block * sb) |
ac27a0ec DK |
880 | { |
881 | unsigned long count = 0; | |
882 | int i; | |
883 | ||
617ba13b MC |
884 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
885 | struct ext4_group_desc *gdp = ext4_get_group_desc (sb, i, NULL); | |
ac27a0ec DK |
886 | if (!gdp) |
887 | continue; | |
888 | count += le16_to_cpu(gdp->bg_used_dirs_count); | |
889 | } | |
890 | return count; | |
891 | } | |
892 |