Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/ext3/ialloc.c | |
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> | |
17 | #include <linux/jbd.h> | |
18 | #include <linux/ext3_fs.h> | |
19 | #include <linux/ext3_jbd.h> | |
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> | |
31bd2bc7 | 26 | #include <linux/extent_map.h> |
1da177e4 LT |
27 | |
28 | #include <asm/byteorder.h> | |
29 | ||
30 | #include "xattr.h" | |
31 | #include "acl.h" | |
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 | ||
47 | ||
48 | /* | |
49 | * Read the inode allocation bitmap for a given block_group, reading | |
50 | * into the specified slot in the superblock's bitmap cache. | |
51 | * | |
52 | * Return buffer_head of bitmap on success or NULL. | |
53 | */ | |
54 | static struct buffer_head * | |
55 | read_inode_bitmap(struct super_block * sb, unsigned long block_group) | |
56 | { | |
57 | struct ext3_group_desc *desc; | |
58 | struct buffer_head *bh = NULL; | |
59 | ||
60 | desc = ext3_get_group_desc(sb, block_group, NULL); | |
61 | if (!desc) | |
62 | goto error_out; | |
63 | ||
64 | bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap)); | |
65 | if (!bh) | |
66 | ext3_error(sb, "read_inode_bitmap", | |
67 | "Cannot read inode bitmap - " | |
68 | "block_group = %lu, inode_bitmap = %u", | |
69 | block_group, le32_to_cpu(desc->bg_inode_bitmap)); | |
70 | error_out: | |
71 | return bh; | |
72 | } | |
73 | ||
74 | /* | |
75 | * NOTE! When we get the inode, we're the only people | |
76 | * that have access to it, and as such there are no | |
77 | * race conditions we have to worry about. The inode | |
78 | * is not on the hash-lists, and it cannot be reached | |
79 | * through the filesystem because the directory entry | |
80 | * has been deleted earlier. | |
81 | * | |
82 | * HOWEVER: we must make sure that we get no aliases, | |
83 | * which means that we have to call "clear_inode()" | |
84 | * _before_ we mark the inode not in use in the inode | |
85 | * bitmaps. Otherwise a newly created file might use | |
86 | * the same inode number (not actually the same pointer | |
87 | * though), and then we'd have two inodes sharing the | |
88 | * same inode number and space on the harddisk. | |
89 | */ | |
90 | void ext3_free_inode (handle_t *handle, struct inode * inode) | |
91 | { | |
92 | struct super_block * sb = inode->i_sb; | |
93 | int is_directory; | |
94 | unsigned long ino; | |
95 | struct buffer_head *bitmap_bh = NULL; | |
96 | struct buffer_head *bh2; | |
97 | unsigned long block_group; | |
98 | unsigned long bit; | |
99 | struct ext3_group_desc * gdp; | |
100 | struct ext3_super_block * es; | |
101 | struct ext3_sb_info *sbi; | |
102 | int fatal = 0, err; | |
103 | ||
104 | if (atomic_read(&inode->i_count) > 1) { | |
105 | printk ("ext3_free_inode: inode has count=%d\n", | |
106 | atomic_read(&inode->i_count)); | |
107 | return; | |
108 | } | |
109 | if (inode->i_nlink) { | |
110 | printk ("ext3_free_inode: inode has nlink=%d\n", | |
111 | inode->i_nlink); | |
112 | return; | |
113 | } | |
114 | if (!sb) { | |
115 | printk("ext3_free_inode: inode on nonexistent device\n"); | |
116 | return; | |
117 | } | |
118 | sbi = EXT3_SB(sb); | |
119 | ||
120 | ino = inode->i_ino; | |
121 | ext3_debug ("freeing inode %lu\n", ino); | |
122 | ||
123 | /* | |
124 | * Note: we must free any quota before locking the superblock, | |
125 | * as writing the quota to disk may need the lock as well. | |
126 | */ | |
127 | DQUOT_INIT(inode); | |
128 | ext3_xattr_delete_inode(handle, inode); | |
129 | DQUOT_FREE_INODE(inode); | |
130 | DQUOT_DROP(inode); | |
131 | ||
132 | is_directory = S_ISDIR(inode->i_mode); | |
133 | ||
134 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
135 | clear_inode (inode); | |
136 | ||
137 | es = EXT3_SB(sb)->s_es; | |
138 | if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
139 | ext3_error (sb, "ext3_free_inode", | |
140 | "reserved or nonexistent inode %lu", ino); | |
141 | goto error_return; | |
142 | } | |
143 | block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); | |
144 | bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb); | |
145 | bitmap_bh = read_inode_bitmap(sb, block_group); | |
146 | if (!bitmap_bh) | |
147 | goto error_return; | |
148 | ||
149 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
150 | fatal = ext3_journal_get_write_access(handle, bitmap_bh); | |
151 | if (fatal) | |
152 | goto error_return; | |
153 | ||
154 | /* Ok, now we can actually update the inode bitmaps.. */ | |
155 | if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group), | |
156 | bit, bitmap_bh->b_data)) | |
157 | ext3_error (sb, "ext3_free_inode", | |
158 | "bit already cleared for inode %lu", ino); | |
159 | else { | |
160 | gdp = ext3_get_group_desc (sb, block_group, &bh2); | |
161 | ||
162 | BUFFER_TRACE(bh2, "get_write_access"); | |
163 | fatal = ext3_journal_get_write_access(handle, bh2); | |
164 | if (fatal) goto error_return; | |
165 | ||
166 | if (gdp) { | |
167 | spin_lock(sb_bgl_lock(sbi, block_group)); | |
50e8a289 | 168 | le16_add_cpu(&gdp->bg_free_inodes_count, 1); |
1da177e4 | 169 | if (is_directory) |
50e8a289 | 170 | le16_add_cpu(&gdp->bg_used_dirs_count, -1); |
1da177e4 LT |
171 | spin_unlock(sb_bgl_lock(sbi, block_group)); |
172 | percpu_counter_inc(&sbi->s_freeinodes_counter); | |
173 | if (is_directory) | |
174 | percpu_counter_dec(&sbi->s_dirs_counter); | |
175 | ||
176 | } | |
177 | BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata"); | |
178 | err = ext3_journal_dirty_metadata(handle, bh2); | |
179 | if (!fatal) fatal = err; | |
180 | } | |
181 | BUFFER_TRACE(bitmap_bh, "call ext3_journal_dirty_metadata"); | |
182 | err = ext3_journal_dirty_metadata(handle, bitmap_bh); | |
183 | if (!fatal) | |
184 | fatal = err; | |
185 | sb->s_dirt = 1; | |
186 | error_return: | |
187 | brelse(bitmap_bh); | |
188 | ext3_std_error(sb, fatal); | |
189 | } | |
190 | ||
191 | /* | |
192 | * There are two policies for allocating an inode. If the new inode is | |
193 | * a directory, then a forward search is made for a block group with both | |
194 | * free space and a low directory-to-inode ratio; if that fails, then of | |
195 | * the groups with above-average free space, that group with the fewest | |
196 | * directories already is chosen. | |
197 | * | |
198 | * For other inodes, search forward from the parent directory\'s block | |
199 | * group to find a free inode. | |
200 | */ | |
201 | static int find_group_dir(struct super_block *sb, struct inode *parent) | |
202 | { | |
203 | int ngroups = EXT3_SB(sb)->s_groups_count; | |
eee194e7 | 204 | unsigned int freei, avefreei; |
1da177e4 | 205 | struct ext3_group_desc *desc, *best_desc = NULL; |
1da177e4 LT |
206 | int group, best_group = -1; |
207 | ||
208 | freei = percpu_counter_read_positive(&EXT3_SB(sb)->s_freeinodes_counter); | |
209 | avefreei = freei / ngroups; | |
210 | ||
211 | for (group = 0; group < ngroups; group++) { | |
ef2fb679 | 212 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
213 | if (!desc || !desc->bg_free_inodes_count) |
214 | continue; | |
215 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
216 | continue; | |
ae6ddcc5 | 217 | if (!best_desc || |
1da177e4 LT |
218 | (le16_to_cpu(desc->bg_free_blocks_count) > |
219 | le16_to_cpu(best_desc->bg_free_blocks_count))) { | |
220 | best_group = group; | |
221 | best_desc = desc; | |
222 | } | |
223 | } | |
224 | return best_group; | |
225 | } | |
226 | ||
ae6ddcc5 MC |
227 | /* |
228 | * Orlov's allocator for directories. | |
229 | * | |
1da177e4 LT |
230 | * We always try to spread first-level directories. |
231 | * | |
ae6ddcc5 MC |
232 | * If there are blockgroups with both free inodes and free blocks counts |
233 | * not worse than average we return one with smallest directory count. | |
234 | * Otherwise we simply return a random group. | |
235 | * | |
236 | * For the rest rules look so: | |
237 | * | |
238 | * It's OK to put directory into a group unless | |
239 | * it has too many directories already (max_dirs) or | |
240 | * it has too few free inodes left (min_inodes) or | |
241 | * it has too few free blocks left (min_blocks) or | |
242 | * it's already running too large debt (max_debt). | |
1cc8dcf5 | 243 | * Parent's group is preferred, if it doesn't satisfy these |
ae6ddcc5 MC |
244 | * conditions we search cyclically through the rest. If none |
245 | * of the groups look good we just look for a group with more | |
246 | * free inodes than average (starting at parent's group). | |
247 | * | |
248 | * Debt is incremented each time we allocate a directory and decremented | |
249 | * when we allocate an inode, within 0--255. | |
250 | */ | |
1da177e4 LT |
251 | |
252 | #define INODE_COST 64 | |
253 | #define BLOCK_COST 256 | |
254 | ||
255 | static int find_group_orlov(struct super_block *sb, struct inode *parent) | |
256 | { | |
257 | int parent_group = EXT3_I(parent)->i_block_group; | |
258 | struct ext3_sb_info *sbi = EXT3_SB(sb); | |
259 | struct ext3_super_block *es = sbi->s_es; | |
260 | int ngroups = sbi->s_groups_count; | |
261 | int inodes_per_group = EXT3_INODES_PER_GROUP(sb); | |
eee194e7 | 262 | unsigned int freei, avefreei; |
1c2bf374 MC |
263 | ext3_fsblk_t freeb, avefreeb; |
264 | ext3_fsblk_t blocks_per_dir; | |
eee194e7 | 265 | unsigned int ndirs; |
1c2bf374 MC |
266 | int max_debt, max_dirs, min_inodes; |
267 | ext3_grpblk_t min_blocks; | |
1da177e4 LT |
268 | int group = -1, i; |
269 | struct ext3_group_desc *desc; | |
1da177e4 LT |
270 | |
271 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
272 | avefreei = freei / ngroups; | |
273 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
274 | avefreeb = freeb / ngroups; | |
275 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); | |
276 | ||
277 | if ((parent == sb->s_root->d_inode) || | |
278 | (EXT3_I(parent)->i_flags & EXT3_TOPDIR_FL)) { | |
279 | int best_ndir = inodes_per_group; | |
280 | int best_group = -1; | |
281 | ||
282 | get_random_bytes(&group, sizeof(group)); | |
283 | parent_group = (unsigned)group % ngroups; | |
284 | for (i = 0; i < ngroups; i++) { | |
285 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 286 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
287 | if (!desc || !desc->bg_free_inodes_count) |
288 | continue; | |
289 | if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) | |
290 | continue; | |
291 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
292 | continue; | |
293 | if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) | |
294 | continue; | |
295 | best_group = group; | |
296 | best_ndir = le16_to_cpu(desc->bg_used_dirs_count); | |
297 | } | |
298 | if (best_group >= 0) | |
299 | return best_group; | |
300 | goto fallback; | |
301 | } | |
302 | ||
303 | blocks_per_dir = (le32_to_cpu(es->s_blocks_count) - freeb) / ndirs; | |
304 | ||
305 | max_dirs = ndirs / ngroups + inodes_per_group / 16; | |
306 | min_inodes = avefreei - inodes_per_group / 4; | |
307 | min_blocks = avefreeb - EXT3_BLOCKS_PER_GROUP(sb) / 4; | |
308 | ||
1c2bf374 | 309 | max_debt = EXT3_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, (ext3_fsblk_t)BLOCK_COST); |
1da177e4 LT |
310 | if (max_debt * INODE_COST > inodes_per_group) |
311 | max_debt = inodes_per_group / INODE_COST; | |
312 | if (max_debt > 255) | |
313 | max_debt = 255; | |
314 | if (max_debt == 0) | |
315 | max_debt = 1; | |
316 | ||
317 | for (i = 0; i < ngroups; i++) { | |
318 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 319 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
320 | if (!desc || !desc->bg_free_inodes_count) |
321 | continue; | |
322 | if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs) | |
323 | continue; | |
324 | if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes) | |
325 | continue; | |
326 | if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks) | |
327 | continue; | |
328 | return group; | |
329 | } | |
330 | ||
331 | fallback: | |
332 | for (i = 0; i < ngroups; i++) { | |
333 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 334 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
335 | if (!desc || !desc->bg_free_inodes_count) |
336 | continue; | |
337 | if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei) | |
338 | return group; | |
339 | } | |
340 | ||
341 | if (avefreei) { | |
342 | /* | |
343 | * The free-inodes counter is approximate, and for really small | |
344 | * filesystems the above test can fail to find any blockgroups | |
345 | */ | |
346 | avefreei = 0; | |
347 | goto fallback; | |
348 | } | |
349 | ||
350 | return -1; | |
351 | } | |
352 | ||
353 | static int find_group_other(struct super_block *sb, struct inode *parent) | |
354 | { | |
355 | int parent_group = EXT3_I(parent)->i_block_group; | |
356 | int ngroups = EXT3_SB(sb)->s_groups_count; | |
357 | struct ext3_group_desc *desc; | |
1da177e4 LT |
358 | int group, i; |
359 | ||
360 | /* | |
361 | * Try to place the inode in its parent directory | |
362 | */ | |
363 | group = parent_group; | |
ef2fb679 | 364 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
365 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
366 | le16_to_cpu(desc->bg_free_blocks_count)) | |
367 | return group; | |
368 | ||
369 | /* | |
370 | * We're going to place this inode in a different blockgroup from its | |
371 | * parent. We want to cause files in a common directory to all land in | |
372 | * the same blockgroup. But we want files which are in a different | |
373 | * directory which shares a blockgroup with our parent to land in a | |
374 | * different blockgroup. | |
375 | * | |
376 | * So add our directory's i_ino into the starting point for the hash. | |
377 | */ | |
378 | group = (group + parent->i_ino) % ngroups; | |
379 | ||
380 | /* | |
381 | * Use a quadratic hash to find a group with a free inode and some free | |
382 | * blocks. | |
383 | */ | |
384 | for (i = 1; i < ngroups; i <<= 1) { | |
385 | group += i; | |
386 | if (group >= ngroups) | |
387 | group -= ngroups; | |
ef2fb679 | 388 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
389 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
390 | le16_to_cpu(desc->bg_free_blocks_count)) | |
391 | return group; | |
392 | } | |
393 | ||
394 | /* | |
395 | * That failed: try linear search for a free inode, even if that group | |
396 | * has no free blocks. | |
397 | */ | |
398 | group = parent_group; | |
399 | for (i = 0; i < ngroups; i++) { | |
400 | if (++group >= ngroups) | |
401 | group = 0; | |
ef2fb679 | 402 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
403 | if (desc && le16_to_cpu(desc->bg_free_inodes_count)) |
404 | return group; | |
405 | } | |
406 | ||
407 | return -1; | |
408 | } | |
409 | ||
410 | /* | |
411 | * There are two policies for allocating an inode. If the new inode is | |
412 | * a directory, then a forward search is made for a block group with both | |
413 | * free space and a low directory-to-inode ratio; if that fails, then of | |
414 | * the groups with above-average free space, that group with the fewest | |
415 | * directories already is chosen. | |
416 | * | |
417 | * For other inodes, search forward from the parent directory's block | |
418 | * group to find a free inode. | |
419 | */ | |
420 | struct inode *ext3_new_inode(handle_t *handle, struct inode * dir, int mode) | |
421 | { | |
422 | struct super_block *sb; | |
423 | struct buffer_head *bitmap_bh = NULL; | |
424 | struct buffer_head *bh2; | |
425 | int group; | |
426 | unsigned long ino = 0; | |
427 | struct inode * inode; | |
428 | struct ext3_group_desc * gdp = NULL; | |
429 | struct ext3_super_block * es; | |
430 | struct ext3_inode_info *ei; | |
431 | struct ext3_sb_info *sbi; | |
432 | int err = 0; | |
433 | struct inode *ret; | |
434 | int i; | |
435 | ||
436 | /* Cannot create files in a deleted directory */ | |
437 | if (!dir || !dir->i_nlink) | |
438 | return ERR_PTR(-EPERM); | |
439 | ||
440 | sb = dir->i_sb; | |
441 | inode = new_inode(sb); | |
442 | if (!inode) | |
443 | return ERR_PTR(-ENOMEM); | |
444 | ei = EXT3_I(inode); | |
445 | ||
446 | sbi = EXT3_SB(sb); | |
447 | es = sbi->s_es; | |
448 | if (S_ISDIR(mode)) { | |
449 | if (test_opt (sb, OLDALLOC)) | |
450 | group = find_group_dir(sb, dir); | |
451 | else | |
452 | group = find_group_orlov(sb, dir); | |
ae6ddcc5 | 453 | } else |
1da177e4 LT |
454 | group = find_group_other(sb, dir); |
455 | ||
456 | err = -ENOSPC; | |
457 | if (group == -1) | |
458 | goto out; | |
459 | ||
460 | for (i = 0; i < sbi->s_groups_count; i++) { | |
461 | err = -EIO; | |
462 | ||
463 | gdp = ext3_get_group_desc(sb, group, &bh2); | |
464 | if (!gdp) | |
465 | goto fail; | |
466 | ||
467 | brelse(bitmap_bh); | |
468 | bitmap_bh = read_inode_bitmap(sb, group); | |
469 | if (!bitmap_bh) | |
470 | goto fail; | |
471 | ||
472 | ino = 0; | |
473 | ||
474 | repeat_in_this_group: | |
475 | ino = ext3_find_next_zero_bit((unsigned long *) | |
476 | bitmap_bh->b_data, EXT3_INODES_PER_GROUP(sb), ino); | |
477 | if (ino < EXT3_INODES_PER_GROUP(sb)) { | |
478 | ||
479 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
480 | err = ext3_journal_get_write_access(handle, bitmap_bh); | |
481 | if (err) | |
482 | goto fail; | |
483 | ||
484 | if (!ext3_set_bit_atomic(sb_bgl_lock(sbi, group), | |
485 | ino, bitmap_bh->b_data)) { | |
486 | /* we won it */ | |
487 | BUFFER_TRACE(bitmap_bh, | |
488 | "call ext3_journal_dirty_metadata"); | |
489 | err = ext3_journal_dirty_metadata(handle, | |
490 | bitmap_bh); | |
491 | if (err) | |
492 | goto fail; | |
493 | goto got; | |
494 | } | |
495 | /* we lost it */ | |
496 | journal_release_buffer(handle, bitmap_bh); | |
497 | ||
498 | if (++ino < EXT3_INODES_PER_GROUP(sb)) | |
499 | goto repeat_in_this_group; | |
500 | } | |
501 | ||
502 | /* | |
503 | * This case is possible in concurrent environment. It is very | |
504 | * rare. We cannot repeat the find_group_xxx() call because | |
505 | * that will simply return the same blockgroup, because the | |
506 | * group descriptor metadata has not yet been updated. | |
507 | * So we just go onto the next blockgroup. | |
508 | */ | |
509 | if (++group == sbi->s_groups_count) | |
510 | group = 0; | |
511 | } | |
512 | err = -ENOSPC; | |
513 | goto out; | |
514 | ||
515 | got: | |
516 | ino += group * EXT3_INODES_PER_GROUP(sb) + 1; | |
517 | if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
518 | ext3_error (sb, "ext3_new_inode", | |
519 | "reserved inode or inode > inodes count - " | |
520 | "block_group = %d, inode=%lu", group, ino); | |
521 | err = -EIO; | |
522 | goto fail; | |
523 | } | |
524 | ||
525 | BUFFER_TRACE(bh2, "get_write_access"); | |
526 | err = ext3_journal_get_write_access(handle, bh2); | |
527 | if (err) goto fail; | |
528 | spin_lock(sb_bgl_lock(sbi, group)); | |
50e8a289 | 529 | le16_add_cpu(&gdp->bg_free_inodes_count, -1); |
1da177e4 | 530 | if (S_ISDIR(mode)) { |
50e8a289 | 531 | le16_add_cpu(&gdp->bg_used_dirs_count, 1); |
1da177e4 LT |
532 | } |
533 | spin_unlock(sb_bgl_lock(sbi, group)); | |
534 | BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata"); | |
535 | err = ext3_journal_dirty_metadata(handle, bh2); | |
536 | if (err) goto fail; | |
537 | ||
538 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
539 | if (S_ISDIR(mode)) | |
540 | percpu_counter_inc(&sbi->s_dirs_counter); | |
541 | sb->s_dirt = 1; | |
542 | ||
6a2f90e9 | 543 | inode->i_uid = current_fsuid(); |
1da177e4 LT |
544 | if (test_opt (sb, GRPID)) |
545 | inode->i_gid = dir->i_gid; | |
546 | else if (dir->i_mode & S_ISGID) { | |
547 | inode->i_gid = dir->i_gid; | |
548 | if (S_ISDIR(mode)) | |
549 | mode |= S_ISGID; | |
550 | } else | |
6a2f90e9 | 551 | inode->i_gid = current_fsgid(); |
1da177e4 LT |
552 | inode->i_mode = mode; |
553 | ||
554 | inode->i_ino = ino; | |
555 | /* This is the optimal IO size (for stat), not the fs block size */ | |
1da177e4 LT |
556 | inode->i_blocks = 0; |
557 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC; | |
558 | ||
559 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
560 | ei->i_dir_start_lookup = 0; | |
561 | ei->i_disksize = 0; | |
562 | ||
04143e2f DG |
563 | ei->i_flags = |
564 | ext3_mask_flags(mode, EXT3_I(dir)->i_flags & EXT3_FL_INHERITED); | |
1da177e4 LT |
565 | #ifdef EXT3_FRAGMENTS |
566 | ei->i_faddr = 0; | |
567 | ei->i_frag_no = 0; | |
568 | ei->i_frag_size = 0; | |
569 | #endif | |
570 | ei->i_file_acl = 0; | |
571 | ei->i_dir_acl = 0; | |
572 | ei->i_dtime = 0; | |
573 | ei->i_block_alloc_info = NULL; | |
574 | ei->i_block_group = group; | |
31bd2bc7 | 575 | extent_map_tree_init(&ei->extent_tree); |
1da177e4 LT |
576 | |
577 | ext3_set_inode_flags(inode); | |
578 | if (IS_DIRSYNC(inode)) | |
579 | handle->h_sync = 1; | |
c38012da AV |
580 | if (insert_inode_locked(inode) < 0) { |
581 | err = -EINVAL; | |
582 | goto fail_drop; | |
583 | } | |
1da177e4 LT |
584 | spin_lock(&sbi->s_next_gen_lock); |
585 | inode->i_generation = sbi->s_next_generation++; | |
586 | spin_unlock(&sbi->s_next_gen_lock); | |
587 | ||
588 | ei->i_state = EXT3_STATE_NEW; | |
589 | ei->i_extra_isize = | |
590 | (EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) ? | |
591 | sizeof(struct ext3_inode) - EXT3_GOOD_OLD_INODE_SIZE : 0; | |
592 | ||
593 | ret = inode; | |
594 | if(DQUOT_ALLOC_INODE(inode)) { | |
1da177e4 | 595 | err = -EDQUOT; |
dc7b5fd6 | 596 | goto fail_drop; |
1da177e4 | 597 | } |
dc7b5fd6 | 598 | |
1da177e4 | 599 | err = ext3_init_acl(handle, inode, dir); |
dc7b5fd6 CS |
600 | if (err) |
601 | goto fail_free_drop; | |
602 | ||
ac50960a | 603 | err = ext3_init_security(handle,inode, dir); |
dc7b5fd6 CS |
604 | if (err) |
605 | goto fail_free_drop; | |
606 | ||
1da177e4 LT |
607 | err = ext3_mark_inode_dirty(handle, inode); |
608 | if (err) { | |
609 | ext3_std_error(sb, err); | |
dc7b5fd6 | 610 | goto fail_free_drop; |
1da177e4 LT |
611 | } |
612 | ||
613 | ext3_debug("allocating inode %lu\n", inode->i_ino); | |
614 | goto really_out; | |
615 | fail: | |
616 | ext3_std_error(sb, err); | |
617 | out: | |
618 | iput(inode); | |
619 | ret = ERR_PTR(err); | |
620 | really_out: | |
621 | brelse(bitmap_bh); | |
622 | return ret; | |
623 | ||
dc7b5fd6 CS |
624 | fail_free_drop: |
625 | DQUOT_FREE_INODE(inode); | |
626 | ||
627 | fail_drop: | |
628 | DQUOT_DROP(inode); | |
1da177e4 LT |
629 | inode->i_flags |= S_NOQUOTA; |
630 | inode->i_nlink = 0; | |
c38012da | 631 | unlock_new_inode(inode); |
1da177e4 LT |
632 | iput(inode); |
633 | brelse(bitmap_bh); | |
634 | return ERR_PTR(err); | |
635 | } | |
636 | ||
637 | /* Verify that we are loading a valid orphan from disk */ | |
638 | struct inode *ext3_orphan_get(struct super_block *sb, unsigned long ino) | |
639 | { | |
640 | unsigned long max_ino = le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count); | |
641 | unsigned long block_group; | |
642 | int bit; | |
473043dc | 643 | struct buffer_head *bitmap_bh; |
1da177e4 | 644 | struct inode *inode = NULL; |
473043dc | 645 | long err = -EIO; |
1da177e4 LT |
646 | |
647 | /* Error cases - e2fsck has already cleaned up for us */ | |
648 | if (ino > max_ino) { | |
e05b6b52 | 649 | ext3_warning(sb, __func__, |
9f40668d | 650 | "bad orphan ino %lu! e2fsck was run?", ino); |
473043dc | 651 | goto error; |
1da177e4 LT |
652 | } |
653 | ||
654 | block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); | |
655 | bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb); | |
656 | bitmap_bh = read_inode_bitmap(sb, block_group); | |
657 | if (!bitmap_bh) { | |
e05b6b52 | 658 | ext3_warning(sb, __func__, |
9f40668d | 659 | "inode bitmap error for orphan %lu", ino); |
473043dc | 660 | goto error; |
1da177e4 LT |
661 | } |
662 | ||
663 | /* Having the inode bit set should be a 100% indicator that this | |
664 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
665 | * inodes that were being truncated, so we can't check i_nlink==0. | |
666 | */ | |
473043dc DH |
667 | if (!ext3_test_bit(bit, bitmap_bh->b_data)) |
668 | goto bad_orphan; | |
669 | ||
670 | inode = ext3_iget(sb, ino); | |
671 | if (IS_ERR(inode)) | |
672 | goto iget_failed; | |
673 | ||
ae76dd9a DG |
674 | /* |
675 | * If the orphans has i_nlinks > 0 then it should be able to be | |
676 | * truncated, otherwise it won't be removed from the orphan list | |
677 | * during processing and an infinite loop will result. | |
678 | */ | |
679 | if (inode->i_nlink && !ext3_can_truncate(inode)) | |
680 | goto bad_orphan; | |
681 | ||
473043dc DH |
682 | if (NEXT_ORPHAN(inode) > max_ino) |
683 | goto bad_orphan; | |
684 | brelse(bitmap_bh); | |
685 | return inode; | |
686 | ||
687 | iget_failed: | |
688 | err = PTR_ERR(inode); | |
689 | inode = NULL; | |
690 | bad_orphan: | |
e05b6b52 | 691 | ext3_warning(sb, __func__, |
473043dc DH |
692 | "bad orphan inode %lu! e2fsck was run?", ino); |
693 | printk(KERN_NOTICE "ext3_test_bit(bit=%d, block=%llu) = %d\n", | |
694 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
695 | ext3_test_bit(bit, bitmap_bh->b_data)); | |
696 | printk(KERN_NOTICE "inode=%p\n", inode); | |
697 | if (inode) { | |
698 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
699 | is_bad_inode(inode)); | |
700 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
701 | NEXT_ORPHAN(inode)); | |
702 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
ae76dd9a | 703 | printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink); |
1da177e4 | 704 | /* Avoid freeing blocks if we got a bad deleted inode */ |
473043dc | 705 | if (inode->i_nlink == 0) |
1da177e4 LT |
706 | inode->i_blocks = 0; |
707 | iput(inode); | |
1da177e4 | 708 | } |
1da177e4 | 709 | brelse(bitmap_bh); |
473043dc DH |
710 | error: |
711 | return ERR_PTR(err); | |
1da177e4 LT |
712 | } |
713 | ||
714 | unsigned long ext3_count_free_inodes (struct super_block * sb) | |
715 | { | |
716 | unsigned long desc_count; | |
717 | struct ext3_group_desc *gdp; | |
718 | int i; | |
719 | #ifdef EXT3FS_DEBUG | |
720 | struct ext3_super_block *es; | |
721 | unsigned long bitmap_count, x; | |
722 | struct buffer_head *bitmap_bh = NULL; | |
723 | ||
1da177e4 LT |
724 | es = EXT3_SB(sb)->s_es; |
725 | desc_count = 0; | |
726 | bitmap_count = 0; | |
727 | gdp = NULL; | |
728 | for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) { | |
729 | gdp = ext3_get_group_desc (sb, i, NULL); | |
730 | if (!gdp) | |
731 | continue; | |
732 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
733 | brelse(bitmap_bh); | |
734 | bitmap_bh = read_inode_bitmap(sb, i); | |
735 | if (!bitmap_bh) | |
736 | continue; | |
737 | ||
738 | x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8); | |
739 | printk("group %d: stored = %d, counted = %lu\n", | |
740 | i, le16_to_cpu(gdp->bg_free_inodes_count), x); | |
741 | bitmap_count += x; | |
742 | } | |
743 | brelse(bitmap_bh); | |
744 | printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n", | |
745 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
1da177e4 LT |
746 | return desc_count; |
747 | #else | |
748 | desc_count = 0; | |
749 | for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) { | |
750 | gdp = ext3_get_group_desc (sb, i, NULL); | |
751 | if (!gdp) | |
752 | continue; | |
753 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
754 | cond_resched(); | |
755 | } | |
756 | return desc_count; | |
757 | #endif | |
758 | } | |
759 | ||
760 | /* Called at mount-time, super-block is locked */ | |
761 | unsigned long ext3_count_dirs (struct super_block * sb) | |
762 | { | |
763 | unsigned long count = 0; | |
764 | int i; | |
765 | ||
766 | for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) { | |
767 | struct ext3_group_desc *gdp = ext3_get_group_desc (sb, i, NULL); | |
768 | if (!gdp) | |
769 | continue; | |
770 | count += le16_to_cpu(gdp->bg_used_dirs_count); | |
771 | } | |
772 | return count; | |
773 | } | |
774 |