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> | |
26 | ||
27 | #include <asm/byteorder.h> | |
28 | ||
29 | #include "xattr.h" | |
30 | #include "acl.h" | |
31 | ||
32 | /* | |
33 | * ialloc.c contains the inodes allocation and deallocation routines | |
34 | */ | |
35 | ||
36 | /* | |
37 | * The free inodes are managed by bitmaps. A file system contains several | |
38 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
39 | * block for inodes, N blocks for the inode table and data blocks. | |
40 | * | |
41 | * The file system contains group descriptors which are located after the | |
42 | * super block. Each descriptor contains the number of the bitmap block and | |
43 | * the free blocks count in the block. | |
44 | */ | |
45 | ||
46 | ||
47 | /* | |
48 | * Read the inode allocation bitmap for a given block_group, reading | |
49 | * into the specified slot in the superblock's bitmap cache. | |
50 | * | |
51 | * Return buffer_head of bitmap on success or NULL. | |
52 | */ | |
53 | static struct buffer_head * | |
54 | read_inode_bitmap(struct super_block * sb, unsigned long block_group) | |
55 | { | |
56 | struct ext3_group_desc *desc; | |
57 | struct buffer_head *bh = NULL; | |
58 | ||
59 | desc = ext3_get_group_desc(sb, block_group, NULL); | |
60 | if (!desc) | |
61 | goto error_out; | |
62 | ||
63 | bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap)); | |
64 | if (!bh) | |
65 | ext3_error(sb, "read_inode_bitmap", | |
66 | "Cannot read inode bitmap - " | |
67 | "block_group = %lu, inode_bitmap = %u", | |
68 | block_group, le32_to_cpu(desc->bg_inode_bitmap)); | |
69 | error_out: | |
70 | return bh; | |
71 | } | |
72 | ||
73 | /* | |
74 | * NOTE! When we get the inode, we're the only people | |
75 | * that have access to it, and as such there are no | |
76 | * race conditions we have to worry about. The inode | |
77 | * is not on the hash-lists, and it cannot be reached | |
78 | * through the filesystem because the directory entry | |
79 | * has been deleted earlier. | |
80 | * | |
81 | * HOWEVER: we must make sure that we get no aliases, | |
82 | * which means that we have to call "clear_inode()" | |
83 | * _before_ we mark the inode not in use in the inode | |
84 | * bitmaps. Otherwise a newly created file might use | |
85 | * the same inode number (not actually the same pointer | |
86 | * though), and then we'd have two inodes sharing the | |
87 | * same inode number and space on the harddisk. | |
88 | */ | |
89 | void ext3_free_inode (handle_t *handle, struct inode * inode) | |
90 | { | |
91 | struct super_block * sb = inode->i_sb; | |
92 | int is_directory; | |
93 | unsigned long ino; | |
94 | struct buffer_head *bitmap_bh = NULL; | |
95 | struct buffer_head *bh2; | |
96 | unsigned long block_group; | |
97 | unsigned long bit; | |
98 | struct ext3_group_desc * gdp; | |
99 | struct ext3_super_block * es; | |
100 | struct ext3_sb_info *sbi; | |
101 | int fatal = 0, err; | |
102 | ||
103 | if (atomic_read(&inode->i_count) > 1) { | |
104 | printk ("ext3_free_inode: inode has count=%d\n", | |
105 | atomic_read(&inode->i_count)); | |
106 | return; | |
107 | } | |
108 | if (inode->i_nlink) { | |
109 | printk ("ext3_free_inode: inode has nlink=%d\n", | |
110 | inode->i_nlink); | |
111 | return; | |
112 | } | |
113 | if (!sb) { | |
114 | printk("ext3_free_inode: inode on nonexistent device\n"); | |
115 | return; | |
116 | } | |
117 | sbi = EXT3_SB(sb); | |
118 | ||
119 | ino = inode->i_ino; | |
120 | ext3_debug ("freeing inode %lu\n", ino); | |
121 | ||
122 | /* | |
123 | * Note: we must free any quota before locking the superblock, | |
124 | * as writing the quota to disk may need the lock as well. | |
125 | */ | |
126 | DQUOT_INIT(inode); | |
127 | ext3_xattr_delete_inode(handle, inode); | |
128 | DQUOT_FREE_INODE(inode); | |
129 | DQUOT_DROP(inode); | |
130 | ||
131 | is_directory = S_ISDIR(inode->i_mode); | |
132 | ||
133 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
134 | clear_inode (inode); | |
135 | ||
136 | es = EXT3_SB(sb)->s_es; | |
137 | if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
138 | ext3_error (sb, "ext3_free_inode", | |
139 | "reserved or nonexistent inode %lu", ino); | |
140 | goto error_return; | |
141 | } | |
142 | block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); | |
143 | bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb); | |
144 | bitmap_bh = read_inode_bitmap(sb, block_group); | |
145 | if (!bitmap_bh) | |
146 | goto error_return; | |
147 | ||
148 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
149 | fatal = ext3_journal_get_write_access(handle, bitmap_bh); | |
150 | if (fatal) | |
151 | goto error_return; | |
152 | ||
153 | /* Ok, now we can actually update the inode bitmaps.. */ | |
154 | if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group), | |
155 | bit, bitmap_bh->b_data)) | |
156 | ext3_error (sb, "ext3_free_inode", | |
157 | "bit already cleared for inode %lu", ino); | |
158 | else { | |
159 | gdp = ext3_get_group_desc (sb, block_group, &bh2); | |
160 | ||
161 | BUFFER_TRACE(bh2, "get_write_access"); | |
162 | fatal = ext3_journal_get_write_access(handle, bh2); | |
163 | if (fatal) goto error_return; | |
164 | ||
165 | if (gdp) { | |
166 | spin_lock(sb_bgl_lock(sbi, block_group)); | |
167 | gdp->bg_free_inodes_count = cpu_to_le16( | |
168 | le16_to_cpu(gdp->bg_free_inodes_count) + 1); | |
169 | if (is_directory) | |
170 | gdp->bg_used_dirs_count = cpu_to_le16( | |
171 | le16_to_cpu(gdp->bg_used_dirs_count) - 1); | |
172 | spin_unlock(sb_bgl_lock(sbi, block_group)); | |
173 | percpu_counter_inc(&sbi->s_freeinodes_counter); | |
174 | if (is_directory) | |
175 | percpu_counter_dec(&sbi->s_dirs_counter); | |
176 | ||
177 | } | |
178 | BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata"); | |
179 | err = ext3_journal_dirty_metadata(handle, bh2); | |
180 | if (!fatal) fatal = err; | |
181 | } | |
182 | BUFFER_TRACE(bitmap_bh, "call ext3_journal_dirty_metadata"); | |
183 | err = ext3_journal_dirty_metadata(handle, bitmap_bh); | |
184 | if (!fatal) | |
185 | fatal = err; | |
186 | sb->s_dirt = 1; | |
187 | error_return: | |
188 | brelse(bitmap_bh); | |
189 | ext3_std_error(sb, fatal); | |
190 | } | |
191 | ||
192 | /* | |
193 | * There are two policies for allocating an inode. If the new inode is | |
194 | * a directory, then a forward search is made for a block group with both | |
195 | * free space and a low directory-to-inode ratio; if that fails, then of | |
196 | * the groups with above-average free space, that group with the fewest | |
197 | * directories already is chosen. | |
198 | * | |
199 | * For other inodes, search forward from the parent directory\'s block | |
200 | * group to find a free inode. | |
201 | */ | |
202 | static int find_group_dir(struct super_block *sb, struct inode *parent) | |
203 | { | |
204 | int ngroups = EXT3_SB(sb)->s_groups_count; | |
eee194e7 | 205 | unsigned int freei, avefreei; |
1da177e4 | 206 | struct ext3_group_desc *desc, *best_desc = NULL; |
1da177e4 LT |
207 | int group, best_group = -1; |
208 | ||
209 | freei = percpu_counter_read_positive(&EXT3_SB(sb)->s_freeinodes_counter); | |
210 | avefreei = freei / ngroups; | |
211 | ||
212 | for (group = 0; group < ngroups; group++) { | |
ef2fb679 | 213 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
214 | if (!desc || !desc->bg_free_inodes_count) |
215 | continue; | |
216 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
217 | continue; | |
ae6ddcc5 | 218 | if (!best_desc || |
1da177e4 LT |
219 | (le16_to_cpu(desc->bg_free_blocks_count) > |
220 | le16_to_cpu(best_desc->bg_free_blocks_count))) { | |
221 | best_group = group; | |
222 | best_desc = desc; | |
223 | } | |
224 | } | |
225 | return best_group; | |
226 | } | |
227 | ||
ae6ddcc5 MC |
228 | /* |
229 | * Orlov's allocator for directories. | |
230 | * | |
1da177e4 LT |
231 | * We always try to spread first-level directories. |
232 | * | |
ae6ddcc5 MC |
233 | * If there are blockgroups with both free inodes and free blocks counts |
234 | * not worse than average we return one with smallest directory count. | |
235 | * Otherwise we simply return a random group. | |
236 | * | |
237 | * For the rest rules look so: | |
238 | * | |
239 | * It's OK to put directory into a group unless | |
240 | * it has too many directories already (max_dirs) or | |
241 | * it has too few free inodes left (min_inodes) or | |
242 | * it has too few free blocks left (min_blocks) or | |
243 | * it's already running too large debt (max_debt). | |
244 | * Parent's group is prefered, if it doesn't satisfy these | |
245 | * conditions we search cyclically through the rest. If none | |
246 | * of the groups look good we just look for a group with more | |
247 | * free inodes than average (starting at parent's group). | |
248 | * | |
249 | * Debt is incremented each time we allocate a directory and decremented | |
250 | * when we allocate an inode, within 0--255. | |
251 | */ | |
1da177e4 LT |
252 | |
253 | #define INODE_COST 64 | |
254 | #define BLOCK_COST 256 | |
255 | ||
256 | static int find_group_orlov(struct super_block *sb, struct inode *parent) | |
257 | { | |
258 | int parent_group = EXT3_I(parent)->i_block_group; | |
259 | struct ext3_sb_info *sbi = EXT3_SB(sb); | |
260 | struct ext3_super_block *es = sbi->s_es; | |
261 | int ngroups = sbi->s_groups_count; | |
262 | int inodes_per_group = EXT3_INODES_PER_GROUP(sb); | |
eee194e7 | 263 | unsigned int freei, avefreei; |
1c2bf374 MC |
264 | ext3_fsblk_t freeb, avefreeb; |
265 | ext3_fsblk_t blocks_per_dir; | |
eee194e7 | 266 | unsigned int ndirs; |
1c2bf374 MC |
267 | int max_debt, max_dirs, min_inodes; |
268 | ext3_grpblk_t min_blocks; | |
1da177e4 LT |
269 | int group = -1, i; |
270 | struct ext3_group_desc *desc; | |
1da177e4 LT |
271 | |
272 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
273 | avefreei = freei / ngroups; | |
274 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
275 | avefreeb = freeb / ngroups; | |
276 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); | |
277 | ||
278 | if ((parent == sb->s_root->d_inode) || | |
279 | (EXT3_I(parent)->i_flags & EXT3_TOPDIR_FL)) { | |
280 | int best_ndir = inodes_per_group; | |
281 | int best_group = -1; | |
282 | ||
283 | get_random_bytes(&group, sizeof(group)); | |
284 | parent_group = (unsigned)group % ngroups; | |
285 | for (i = 0; i < ngroups; i++) { | |
286 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 287 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
288 | if (!desc || !desc->bg_free_inodes_count) |
289 | continue; | |
290 | if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) | |
291 | continue; | |
292 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
293 | continue; | |
294 | if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) | |
295 | continue; | |
296 | best_group = group; | |
297 | best_ndir = le16_to_cpu(desc->bg_used_dirs_count); | |
298 | } | |
299 | if (best_group >= 0) | |
300 | return best_group; | |
301 | goto fallback; | |
302 | } | |
303 | ||
304 | blocks_per_dir = (le32_to_cpu(es->s_blocks_count) - freeb) / ndirs; | |
305 | ||
306 | max_dirs = ndirs / ngroups + inodes_per_group / 16; | |
307 | min_inodes = avefreei - inodes_per_group / 4; | |
308 | min_blocks = avefreeb - EXT3_BLOCKS_PER_GROUP(sb) / 4; | |
309 | ||
1c2bf374 | 310 | max_debt = EXT3_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, (ext3_fsblk_t)BLOCK_COST); |
1da177e4 LT |
311 | if (max_debt * INODE_COST > inodes_per_group) |
312 | max_debt = inodes_per_group / INODE_COST; | |
313 | if (max_debt > 255) | |
314 | max_debt = 255; | |
315 | if (max_debt == 0) | |
316 | max_debt = 1; | |
317 | ||
318 | for (i = 0; i < ngroups; i++) { | |
319 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 320 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
321 | if (!desc || !desc->bg_free_inodes_count) |
322 | continue; | |
323 | if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs) | |
324 | continue; | |
325 | if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes) | |
326 | continue; | |
327 | if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks) | |
328 | continue; | |
329 | return group; | |
330 | } | |
331 | ||
332 | fallback: | |
333 | for (i = 0; i < ngroups; i++) { | |
334 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 335 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
336 | if (!desc || !desc->bg_free_inodes_count) |
337 | continue; | |
338 | if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei) | |
339 | return group; | |
340 | } | |
341 | ||
342 | if (avefreei) { | |
343 | /* | |
344 | * The free-inodes counter is approximate, and for really small | |
345 | * filesystems the above test can fail to find any blockgroups | |
346 | */ | |
347 | avefreei = 0; | |
348 | goto fallback; | |
349 | } | |
350 | ||
351 | return -1; | |
352 | } | |
353 | ||
354 | static int find_group_other(struct super_block *sb, struct inode *parent) | |
355 | { | |
356 | int parent_group = EXT3_I(parent)->i_block_group; | |
357 | int ngroups = EXT3_SB(sb)->s_groups_count; | |
358 | struct ext3_group_desc *desc; | |
1da177e4 LT |
359 | int group, i; |
360 | ||
361 | /* | |
362 | * Try to place the inode in its parent directory | |
363 | */ | |
364 | group = parent_group; | |
ef2fb679 | 365 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
366 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
367 | le16_to_cpu(desc->bg_free_blocks_count)) | |
368 | return group; | |
369 | ||
370 | /* | |
371 | * We're going to place this inode in a different blockgroup from its | |
372 | * parent. We want to cause files in a common directory to all land in | |
373 | * the same blockgroup. But we want files which are in a different | |
374 | * directory which shares a blockgroup with our parent to land in a | |
375 | * different blockgroup. | |
376 | * | |
377 | * So add our directory's i_ino into the starting point for the hash. | |
378 | */ | |
379 | group = (group + parent->i_ino) % ngroups; | |
380 | ||
381 | /* | |
382 | * Use a quadratic hash to find a group with a free inode and some free | |
383 | * blocks. | |
384 | */ | |
385 | for (i = 1; i < ngroups; i <<= 1) { | |
386 | group += i; | |
387 | if (group >= ngroups) | |
388 | group -= ngroups; | |
ef2fb679 | 389 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
390 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
391 | le16_to_cpu(desc->bg_free_blocks_count)) | |
392 | return group; | |
393 | } | |
394 | ||
395 | /* | |
396 | * That failed: try linear search for a free inode, even if that group | |
397 | * has no free blocks. | |
398 | */ | |
399 | group = parent_group; | |
400 | for (i = 0; i < ngroups; i++) { | |
401 | if (++group >= ngroups) | |
402 | group = 0; | |
ef2fb679 | 403 | desc = ext3_get_group_desc (sb, group, NULL); |
1da177e4 LT |
404 | if (desc && le16_to_cpu(desc->bg_free_inodes_count)) |
405 | return group; | |
406 | } | |
407 | ||
408 | return -1; | |
409 | } | |
410 | ||
411 | /* | |
412 | * There are two policies for allocating an inode. If the new inode is | |
413 | * a directory, then a forward search is made for a block group with both | |
414 | * free space and a low directory-to-inode ratio; if that fails, then of | |
415 | * the groups with above-average free space, that group with the fewest | |
416 | * directories already is chosen. | |
417 | * | |
418 | * For other inodes, search forward from the parent directory's block | |
419 | * group to find a free inode. | |
420 | */ | |
421 | struct inode *ext3_new_inode(handle_t *handle, struct inode * dir, int mode) | |
422 | { | |
423 | struct super_block *sb; | |
424 | struct buffer_head *bitmap_bh = NULL; | |
425 | struct buffer_head *bh2; | |
426 | int group; | |
427 | unsigned long ino = 0; | |
428 | struct inode * inode; | |
429 | struct ext3_group_desc * gdp = NULL; | |
430 | struct ext3_super_block * es; | |
431 | struct ext3_inode_info *ei; | |
432 | struct ext3_sb_info *sbi; | |
433 | int err = 0; | |
434 | struct inode *ret; | |
435 | int i; | |
436 | ||
437 | /* Cannot create files in a deleted directory */ | |
438 | if (!dir || !dir->i_nlink) | |
439 | return ERR_PTR(-EPERM); | |
440 | ||
441 | sb = dir->i_sb; | |
442 | inode = new_inode(sb); | |
443 | if (!inode) | |
444 | return ERR_PTR(-ENOMEM); | |
445 | ei = EXT3_I(inode); | |
446 | ||
447 | sbi = EXT3_SB(sb); | |
448 | es = sbi->s_es; | |
449 | if (S_ISDIR(mode)) { | |
450 | if (test_opt (sb, OLDALLOC)) | |
451 | group = find_group_dir(sb, dir); | |
452 | else | |
453 | group = find_group_orlov(sb, dir); | |
ae6ddcc5 | 454 | } else |
1da177e4 LT |
455 | group = find_group_other(sb, dir); |
456 | ||
457 | err = -ENOSPC; | |
458 | if (group == -1) | |
459 | goto out; | |
460 | ||
461 | for (i = 0; i < sbi->s_groups_count; i++) { | |
462 | err = -EIO; | |
463 | ||
464 | gdp = ext3_get_group_desc(sb, group, &bh2); | |
465 | if (!gdp) | |
466 | goto fail; | |
467 | ||
468 | brelse(bitmap_bh); | |
469 | bitmap_bh = read_inode_bitmap(sb, group); | |
470 | if (!bitmap_bh) | |
471 | goto fail; | |
472 | ||
473 | ino = 0; | |
474 | ||
475 | repeat_in_this_group: | |
476 | ino = ext3_find_next_zero_bit((unsigned long *) | |
477 | bitmap_bh->b_data, EXT3_INODES_PER_GROUP(sb), ino); | |
478 | if (ino < EXT3_INODES_PER_GROUP(sb)) { | |
479 | ||
480 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
481 | err = ext3_journal_get_write_access(handle, bitmap_bh); | |
482 | if (err) | |
483 | goto fail; | |
484 | ||
485 | if (!ext3_set_bit_atomic(sb_bgl_lock(sbi, group), | |
486 | ino, bitmap_bh->b_data)) { | |
487 | /* we won it */ | |
488 | BUFFER_TRACE(bitmap_bh, | |
489 | "call ext3_journal_dirty_metadata"); | |
490 | err = ext3_journal_dirty_metadata(handle, | |
491 | bitmap_bh); | |
492 | if (err) | |
493 | goto fail; | |
494 | goto got; | |
495 | } | |
496 | /* we lost it */ | |
497 | journal_release_buffer(handle, bitmap_bh); | |
498 | ||
499 | if (++ino < EXT3_INODES_PER_GROUP(sb)) | |
500 | goto repeat_in_this_group; | |
501 | } | |
502 | ||
503 | /* | |
504 | * This case is possible in concurrent environment. It is very | |
505 | * rare. We cannot repeat the find_group_xxx() call because | |
506 | * that will simply return the same blockgroup, because the | |
507 | * group descriptor metadata has not yet been updated. | |
508 | * So we just go onto the next blockgroup. | |
509 | */ | |
510 | if (++group == sbi->s_groups_count) | |
511 | group = 0; | |
512 | } | |
513 | err = -ENOSPC; | |
514 | goto out; | |
515 | ||
516 | got: | |
517 | ino += group * EXT3_INODES_PER_GROUP(sb) + 1; | |
518 | if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
519 | ext3_error (sb, "ext3_new_inode", | |
520 | "reserved inode or inode > inodes count - " | |
521 | "block_group = %d, inode=%lu", group, ino); | |
522 | err = -EIO; | |
523 | goto fail; | |
524 | } | |
525 | ||
526 | BUFFER_TRACE(bh2, "get_write_access"); | |
527 | err = ext3_journal_get_write_access(handle, bh2); | |
528 | if (err) goto fail; | |
529 | spin_lock(sb_bgl_lock(sbi, group)); | |
530 | gdp->bg_free_inodes_count = | |
531 | cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1); | |
532 | if (S_ISDIR(mode)) { | |
533 | gdp->bg_used_dirs_count = | |
534 | cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1); | |
535 | } | |
536 | spin_unlock(sb_bgl_lock(sbi, group)); | |
537 | BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata"); | |
538 | err = ext3_journal_dirty_metadata(handle, bh2); | |
539 | if (err) goto fail; | |
540 | ||
541 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
542 | if (S_ISDIR(mode)) | |
543 | percpu_counter_inc(&sbi->s_dirs_counter); | |
544 | sb->s_dirt = 1; | |
545 | ||
546 | inode->i_uid = current->fsuid; | |
547 | if (test_opt (sb, GRPID)) | |
548 | inode->i_gid = dir->i_gid; | |
549 | else if (dir->i_mode & S_ISGID) { | |
550 | inode->i_gid = dir->i_gid; | |
551 | if (S_ISDIR(mode)) | |
552 | mode |= S_ISGID; | |
553 | } else | |
554 | inode->i_gid = current->fsgid; | |
555 | inode->i_mode = mode; | |
556 | ||
557 | inode->i_ino = ino; | |
558 | /* This is the optimal IO size (for stat), not the fs block size */ | |
1da177e4 LT |
559 | inode->i_blocks = 0; |
560 | inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC; | |
561 | ||
562 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
563 | ei->i_dir_start_lookup = 0; | |
564 | ei->i_disksize = 0; | |
565 | ||
566 | ei->i_flags = EXT3_I(dir)->i_flags & ~EXT3_INDEX_FL; | |
567 | if (S_ISLNK(mode)) | |
568 | ei->i_flags &= ~(EXT3_IMMUTABLE_FL|EXT3_APPEND_FL); | |
569 | /* dirsync only applies to directories */ | |
570 | if (!S_ISDIR(mode)) | |
571 | ei->i_flags &= ~EXT3_DIRSYNC_FL; | |
572 | #ifdef EXT3_FRAGMENTS | |
573 | ei->i_faddr = 0; | |
574 | ei->i_frag_no = 0; | |
575 | ei->i_frag_size = 0; | |
576 | #endif | |
577 | ei->i_file_acl = 0; | |
578 | ei->i_dir_acl = 0; | |
579 | ei->i_dtime = 0; | |
580 | ei->i_block_alloc_info = NULL; | |
581 | ei->i_block_group = group; | |
582 | ||
583 | ext3_set_inode_flags(inode); | |
584 | if (IS_DIRSYNC(inode)) | |
585 | handle->h_sync = 1; | |
586 | insert_inode_hash(inode); | |
587 | spin_lock(&sbi->s_next_gen_lock); | |
588 | inode->i_generation = sbi->s_next_generation++; | |
589 | spin_unlock(&sbi->s_next_gen_lock); | |
590 | ||
591 | ei->i_state = EXT3_STATE_NEW; | |
592 | ei->i_extra_isize = | |
593 | (EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) ? | |
594 | sizeof(struct ext3_inode) - EXT3_GOOD_OLD_INODE_SIZE : 0; | |
595 | ||
596 | ret = inode; | |
597 | if(DQUOT_ALLOC_INODE(inode)) { | |
1da177e4 | 598 | err = -EDQUOT; |
dc7b5fd6 | 599 | goto fail_drop; |
1da177e4 | 600 | } |
dc7b5fd6 | 601 | |
1da177e4 | 602 | err = ext3_init_acl(handle, inode, dir); |
dc7b5fd6 CS |
603 | if (err) |
604 | goto fail_free_drop; | |
605 | ||
ac50960a | 606 | err = ext3_init_security(handle,inode, dir); |
dc7b5fd6 CS |
607 | if (err) |
608 | goto fail_free_drop; | |
609 | ||
1da177e4 LT |
610 | err = ext3_mark_inode_dirty(handle, inode); |
611 | if (err) { | |
612 | ext3_std_error(sb, err); | |
dc7b5fd6 | 613 | goto fail_free_drop; |
1da177e4 LT |
614 | } |
615 | ||
616 | ext3_debug("allocating inode %lu\n", inode->i_ino); | |
617 | goto really_out; | |
618 | fail: | |
619 | ext3_std_error(sb, err); | |
620 | out: | |
621 | iput(inode); | |
622 | ret = ERR_PTR(err); | |
623 | really_out: | |
624 | brelse(bitmap_bh); | |
625 | return ret; | |
626 | ||
dc7b5fd6 CS |
627 | fail_free_drop: |
628 | DQUOT_FREE_INODE(inode); | |
629 | ||
630 | fail_drop: | |
631 | DQUOT_DROP(inode); | |
1da177e4 LT |
632 | inode->i_flags |= S_NOQUOTA; |
633 | inode->i_nlink = 0; | |
634 | iput(inode); | |
635 | brelse(bitmap_bh); | |
636 | return ERR_PTR(err); | |
637 | } | |
638 | ||
639 | /* Verify that we are loading a valid orphan from disk */ | |
640 | struct inode *ext3_orphan_get(struct super_block *sb, unsigned long ino) | |
641 | { | |
642 | unsigned long max_ino = le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count); | |
643 | unsigned long block_group; | |
644 | int bit; | |
473043dc | 645 | struct buffer_head *bitmap_bh; |
1da177e4 | 646 | struct inode *inode = NULL; |
473043dc | 647 | long err = -EIO; |
1da177e4 LT |
648 | |
649 | /* Error cases - e2fsck has already cleaned up for us */ | |
650 | if (ino > max_ino) { | |
651 | ext3_warning(sb, __FUNCTION__, | |
9f40668d | 652 | "bad orphan ino %lu! e2fsck was run?", ino); |
473043dc | 653 | goto error; |
1da177e4 LT |
654 | } |
655 | ||
656 | block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); | |
657 | bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb); | |
658 | bitmap_bh = read_inode_bitmap(sb, block_group); | |
659 | if (!bitmap_bh) { | |
660 | ext3_warning(sb, __FUNCTION__, | |
9f40668d | 661 | "inode bitmap error for orphan %lu", ino); |
473043dc | 662 | goto error; |
1da177e4 LT |
663 | } |
664 | ||
665 | /* Having the inode bit set should be a 100% indicator that this | |
666 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
667 | * inodes that were being truncated, so we can't check i_nlink==0. | |
668 | */ | |
473043dc DH |
669 | if (!ext3_test_bit(bit, bitmap_bh->b_data)) |
670 | goto bad_orphan; | |
671 | ||
672 | inode = ext3_iget(sb, ino); | |
673 | if (IS_ERR(inode)) | |
674 | goto iget_failed; | |
675 | ||
676 | if (NEXT_ORPHAN(inode) > max_ino) | |
677 | goto bad_orphan; | |
678 | brelse(bitmap_bh); | |
679 | return inode; | |
680 | ||
681 | iget_failed: | |
682 | err = PTR_ERR(inode); | |
683 | inode = NULL; | |
684 | bad_orphan: | |
685 | ext3_warning(sb, __FUNCTION__, | |
686 | "bad orphan inode %lu! e2fsck was run?", ino); | |
687 | printk(KERN_NOTICE "ext3_test_bit(bit=%d, block=%llu) = %d\n", | |
688 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
689 | ext3_test_bit(bit, bitmap_bh->b_data)); | |
690 | printk(KERN_NOTICE "inode=%p\n", inode); | |
691 | if (inode) { | |
692 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
693 | is_bad_inode(inode)); | |
694 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
695 | NEXT_ORPHAN(inode)); | |
696 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
1da177e4 | 697 | /* Avoid freeing blocks if we got a bad deleted inode */ |
473043dc | 698 | if (inode->i_nlink == 0) |
1da177e4 LT |
699 | inode->i_blocks = 0; |
700 | iput(inode); | |
1da177e4 | 701 | } |
1da177e4 | 702 | brelse(bitmap_bh); |
473043dc DH |
703 | error: |
704 | return ERR_PTR(err); | |
1da177e4 LT |
705 | } |
706 | ||
707 | unsigned long ext3_count_free_inodes (struct super_block * sb) | |
708 | { | |
709 | unsigned long desc_count; | |
710 | struct ext3_group_desc *gdp; | |
711 | int i; | |
712 | #ifdef EXT3FS_DEBUG | |
713 | struct ext3_super_block *es; | |
714 | unsigned long bitmap_count, x; | |
715 | struct buffer_head *bitmap_bh = NULL; | |
716 | ||
1da177e4 LT |
717 | es = EXT3_SB(sb)->s_es; |
718 | desc_count = 0; | |
719 | bitmap_count = 0; | |
720 | gdp = NULL; | |
721 | for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) { | |
722 | gdp = ext3_get_group_desc (sb, i, NULL); | |
723 | if (!gdp) | |
724 | continue; | |
725 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
726 | brelse(bitmap_bh); | |
727 | bitmap_bh = read_inode_bitmap(sb, i); | |
728 | if (!bitmap_bh) | |
729 | continue; | |
730 | ||
731 | x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8); | |
732 | printk("group %d: stored = %d, counted = %lu\n", | |
733 | i, le16_to_cpu(gdp->bg_free_inodes_count), x); | |
734 | bitmap_count += x; | |
735 | } | |
736 | brelse(bitmap_bh); | |
737 | printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n", | |
738 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
1da177e4 LT |
739 | return desc_count; |
740 | #else | |
741 | desc_count = 0; | |
742 | for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) { | |
743 | gdp = ext3_get_group_desc (sb, i, NULL); | |
744 | if (!gdp) | |
745 | continue; | |
746 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
747 | cond_resched(); | |
748 | } | |
749 | return desc_count; | |
750 | #endif | |
751 | } | |
752 | ||
753 | /* Called at mount-time, super-block is locked */ | |
754 | unsigned long ext3_count_dirs (struct super_block * sb) | |
755 | { | |
756 | unsigned long count = 0; | |
757 | int i; | |
758 | ||
759 | for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) { | |
760 | struct ext3_group_desc *gdp = ext3_get_group_desc (sb, i, NULL); | |
761 | if (!gdp) | |
762 | continue; | |
763 | count += le16_to_cpu(gdp->bg_used_dirs_count); | |
764 | } | |
765 | return count; | |
766 | } | |
767 |