]>
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" | |
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 | { | |
617ba13b | 56 | struct ext4_group_desc *desc; |
ac27a0ec DK |
57 | struct buffer_head *bh = NULL; |
58 | ||
617ba13b | 59 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec DK |
60 | if (!desc) |
61 | goto error_out; | |
62 | ||
8fadc143 | 63 | bh = sb_bread(sb, ext4_inode_bitmap(sb, desc)); |
ac27a0ec | 64 | if (!bh) |
617ba13b | 65 | ext4_error(sb, "read_inode_bitmap", |
ac27a0ec | 66 | "Cannot read inode bitmap - " |
bd81d8ee | 67 | "block_group = %lu, inode_bitmap = %llu", |
8fadc143 | 68 | block_group, ext4_inode_bitmap(sb, desc)); |
ac27a0ec DK |
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 | */ | |
617ba13b | 89 | void ext4_free_inode (handle_t *handle, struct inode * inode) |
ac27a0ec DK |
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; | |
617ba13b MC |
98 | struct ext4_group_desc * gdp; |
99 | struct ext4_super_block * es; | |
100 | struct ext4_sb_info *sbi; | |
ac27a0ec DK |
101 | int fatal = 0, err; |
102 | ||
103 | if (atomic_read(&inode->i_count) > 1) { | |
617ba13b | 104 | printk ("ext4_free_inode: inode has count=%d\n", |
ac27a0ec DK |
105 | atomic_read(&inode->i_count)); |
106 | return; | |
107 | } | |
108 | if (inode->i_nlink) { | |
617ba13b | 109 | printk ("ext4_free_inode: inode has nlink=%d\n", |
ac27a0ec DK |
110 | inode->i_nlink); |
111 | return; | |
112 | } | |
113 | if (!sb) { | |
617ba13b | 114 | printk("ext4_free_inode: inode on nonexistent device\n"); |
ac27a0ec DK |
115 | return; |
116 | } | |
617ba13b | 117 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
118 | |
119 | ino = inode->i_ino; | |
617ba13b | 120 | ext4_debug ("freeing inode %lu\n", ino); |
ac27a0ec DK |
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); | |
617ba13b | 127 | ext4_xattr_delete_inode(handle, inode); |
ac27a0ec DK |
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 | ||
617ba13b MC |
136 | es = EXT4_SB(sb)->s_es; |
137 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
138 | ext4_error (sb, "ext4_free_inode", | |
ac27a0ec DK |
139 | "reserved or nonexistent inode %lu", ino); |
140 | goto error_return; | |
141 | } | |
617ba13b MC |
142 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
143 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
ac27a0ec DK |
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"); | |
617ba13b | 149 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
150 | if (fatal) |
151 | goto error_return; | |
152 | ||
153 | /* Ok, now we can actually update the inode bitmaps.. */ | |
617ba13b | 154 | if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
ac27a0ec | 155 | bit, bitmap_bh->b_data)) |
617ba13b | 156 | ext4_error (sb, "ext4_free_inode", |
ac27a0ec DK |
157 | "bit already cleared for inode %lu", ino); |
158 | else { | |
617ba13b | 159 | gdp = ext4_get_group_desc (sb, block_group, &bh2); |
ac27a0ec DK |
160 | |
161 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 162 | fatal = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec DK |
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 | } | |
617ba13b MC |
178 | BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata"); |
179 | err = ext4_journal_dirty_metadata(handle, bh2); | |
ac27a0ec DK |
180 | if (!fatal) fatal = err; |
181 | } | |
617ba13b MC |
182 | BUFFER_TRACE(bitmap_bh, "call ext4_journal_dirty_metadata"); |
183 | err = ext4_journal_dirty_metadata(handle, bitmap_bh); | |
ac27a0ec DK |
184 | if (!fatal) |
185 | fatal = err; | |
186 | sb->s_dirt = 1; | |
187 | error_return: | |
188 | brelse(bitmap_bh); | |
617ba13b | 189 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
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 | { | |
617ba13b | 204 | int ngroups = EXT4_SB(sb)->s_groups_count; |
ac27a0ec | 205 | unsigned int freei, avefreei; |
617ba13b | 206 | struct ext4_group_desc *desc, *best_desc = NULL; |
ac27a0ec DK |
207 | int group, best_group = -1; |
208 | ||
617ba13b | 209 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
210 | avefreei = freei / ngroups; |
211 | ||
212 | for (group = 0; group < ngroups; group++) { | |
ef2fb679 | 213 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
214 | if (!desc || !desc->bg_free_inodes_count) |
215 | continue; | |
216 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
217 | continue; | |
218 | if (!best_desc || | |
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 | ||
228 | /* | |
229 | * Orlov's allocator for directories. | |
230 | * | |
231 | * We always try to spread first-level directories. | |
232 | * | |
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 | */ | |
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 | { | |
617ba13b MC |
258 | int parent_group = EXT4_I(parent)->i_block_group; |
259 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
260 | struct ext4_super_block *es = sbi->s_es; | |
ac27a0ec | 261 | int ngroups = sbi->s_groups_count; |
617ba13b | 262 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 263 | unsigned int freei, avefreei; |
617ba13b MC |
264 | ext4_fsblk_t freeb, avefreeb; |
265 | ext4_fsblk_t blocks_per_dir; | |
ac27a0ec DK |
266 | unsigned int ndirs; |
267 | int max_debt, max_dirs, min_inodes; | |
617ba13b | 268 | ext4_grpblk_t min_blocks; |
ac27a0ec | 269 | int group = -1, i; |
617ba13b | 270 | struct ext4_group_desc *desc; |
ac27a0ec DK |
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); | |
3a5b2ecd | 275 | avefreeb = freeb; |
f4e5bc24 | 276 | do_div(avefreeb, ngroups); |
ac27a0ec DK |
277 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
278 | ||
279 | if ((parent == sb->s_root->d_inode) || | |
617ba13b | 280 | (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) { |
ac27a0ec DK |
281 | int best_ndir = inodes_per_group; |
282 | int best_group = -1; | |
283 | ||
284 | get_random_bytes(&group, sizeof(group)); | |
285 | parent_group = (unsigned)group % ngroups; | |
286 | for (i = 0; i < ngroups; i++) { | |
287 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 288 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
289 | if (!desc || !desc->bg_free_inodes_count) |
290 | continue; | |
291 | if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) | |
292 | continue; | |
293 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
294 | continue; | |
295 | if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) | |
296 | continue; | |
297 | best_group = group; | |
298 | best_ndir = le16_to_cpu(desc->bg_used_dirs_count); | |
299 | } | |
300 | if (best_group >= 0) | |
301 | return best_group; | |
302 | goto fallback; | |
303 | } | |
304 | ||
bd81d8ee | 305 | blocks_per_dir = ext4_blocks_count(es) - freeb; |
f4e5bc24 | 306 | do_div(blocks_per_dir, ndirs); |
ac27a0ec DK |
307 | |
308 | max_dirs = ndirs / ngroups + inodes_per_group / 16; | |
309 | min_inodes = avefreei - inodes_per_group / 4; | |
617ba13b | 310 | min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4; |
ac27a0ec | 311 | |
3a5b2ecd | 312 | max_debt = EXT4_BLOCKS_PER_GROUP(sb); |
f4e5bc24 | 313 | max_debt /= max_t(int, blocks_per_dir, BLOCK_COST); |
ac27a0ec DK |
314 | if (max_debt * INODE_COST > inodes_per_group) |
315 | max_debt = inodes_per_group / INODE_COST; | |
316 | if (max_debt > 255) | |
317 | max_debt = 255; | |
318 | if (max_debt == 0) | |
319 | max_debt = 1; | |
320 | ||
321 | for (i = 0; i < ngroups; i++) { | |
322 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 323 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
324 | if (!desc || !desc->bg_free_inodes_count) |
325 | continue; | |
326 | if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs) | |
327 | continue; | |
328 | if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes) | |
329 | continue; | |
330 | if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks) | |
331 | continue; | |
332 | return group; | |
333 | } | |
334 | ||
335 | fallback: | |
336 | for (i = 0; i < ngroups; i++) { | |
337 | group = (parent_group + i) % ngroups; | |
ef2fb679 | 338 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
339 | if (!desc || !desc->bg_free_inodes_count) |
340 | continue; | |
341 | if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei) | |
342 | return group; | |
343 | } | |
344 | ||
345 | if (avefreei) { | |
346 | /* | |
347 | * The free-inodes counter is approximate, and for really small | |
348 | * filesystems the above test can fail to find any blockgroups | |
349 | */ | |
350 | avefreei = 0; | |
351 | goto fallback; | |
352 | } | |
353 | ||
354 | return -1; | |
355 | } | |
356 | ||
357 | static int find_group_other(struct super_block *sb, struct inode *parent) | |
358 | { | |
617ba13b MC |
359 | int parent_group = EXT4_I(parent)->i_block_group; |
360 | int ngroups = EXT4_SB(sb)->s_groups_count; | |
361 | struct ext4_group_desc *desc; | |
ac27a0ec DK |
362 | int group, i; |
363 | ||
364 | /* | |
365 | * Try to place the inode in its parent directory | |
366 | */ | |
367 | group = parent_group; | |
ef2fb679 | 368 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
369 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
370 | le16_to_cpu(desc->bg_free_blocks_count)) | |
371 | return group; | |
372 | ||
373 | /* | |
374 | * We're going to place this inode in a different blockgroup from its | |
375 | * parent. We want to cause files in a common directory to all land in | |
376 | * the same blockgroup. But we want files which are in a different | |
377 | * directory which shares a blockgroup with our parent to land in a | |
378 | * different blockgroup. | |
379 | * | |
380 | * So add our directory's i_ino into the starting point for the hash. | |
381 | */ | |
382 | group = (group + parent->i_ino) % ngroups; | |
383 | ||
384 | /* | |
385 | * Use a quadratic hash to find a group with a free inode and some free | |
386 | * blocks. | |
387 | */ | |
388 | for (i = 1; i < ngroups; i <<= 1) { | |
389 | group += i; | |
390 | if (group >= ngroups) | |
391 | group -= ngroups; | |
ef2fb679 | 392 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
393 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
394 | le16_to_cpu(desc->bg_free_blocks_count)) | |
395 | return group; | |
396 | } | |
397 | ||
398 | /* | |
399 | * That failed: try linear search for a free inode, even if that group | |
400 | * has no free blocks. | |
401 | */ | |
402 | group = parent_group; | |
403 | for (i = 0; i < ngroups; i++) { | |
404 | if (++group >= ngroups) | |
405 | group = 0; | |
ef2fb679 | 406 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
407 | if (desc && le16_to_cpu(desc->bg_free_inodes_count)) |
408 | return group; | |
409 | } | |
410 | ||
411 | return -1; | |
412 | } | |
413 | ||
414 | /* | |
415 | * There are two policies for allocating an inode. If the new inode is | |
416 | * a directory, then a forward search is made for a block group with both | |
417 | * free space and a low directory-to-inode ratio; if that fails, then of | |
418 | * the groups with above-average free space, that group with the fewest | |
419 | * directories already is chosen. | |
420 | * | |
421 | * For other inodes, search forward from the parent directory's block | |
422 | * group to find a free inode. | |
423 | */ | |
617ba13b | 424 | struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode) |
ac27a0ec DK |
425 | { |
426 | struct super_block *sb; | |
427 | struct buffer_head *bitmap_bh = NULL; | |
428 | struct buffer_head *bh2; | |
429 | int group; | |
430 | unsigned long ino = 0; | |
431 | struct inode * inode; | |
617ba13b MC |
432 | struct ext4_group_desc * gdp = NULL; |
433 | struct ext4_super_block * es; | |
434 | struct ext4_inode_info *ei; | |
435 | struct ext4_sb_info *sbi; | |
ac27a0ec DK |
436 | int err = 0; |
437 | struct inode *ret; | |
438 | int i; | |
439 | ||
440 | /* Cannot create files in a deleted directory */ | |
441 | if (!dir || !dir->i_nlink) | |
442 | return ERR_PTR(-EPERM); | |
443 | ||
444 | sb = dir->i_sb; | |
445 | inode = new_inode(sb); | |
446 | if (!inode) | |
447 | return ERR_PTR(-ENOMEM); | |
617ba13b | 448 | ei = EXT4_I(inode); |
ac27a0ec | 449 | |
617ba13b | 450 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
451 | es = sbi->s_es; |
452 | if (S_ISDIR(mode)) { | |
453 | if (test_opt (sb, OLDALLOC)) | |
454 | group = find_group_dir(sb, dir); | |
455 | else | |
456 | group = find_group_orlov(sb, dir); | |
457 | } else | |
458 | group = find_group_other(sb, dir); | |
459 | ||
460 | err = -ENOSPC; | |
461 | if (group == -1) | |
462 | goto out; | |
463 | ||
464 | for (i = 0; i < sbi->s_groups_count; i++) { | |
465 | err = -EIO; | |
466 | ||
617ba13b | 467 | gdp = ext4_get_group_desc(sb, group, &bh2); |
ac27a0ec DK |
468 | if (!gdp) |
469 | goto fail; | |
470 | ||
471 | brelse(bitmap_bh); | |
472 | bitmap_bh = read_inode_bitmap(sb, group); | |
473 | if (!bitmap_bh) | |
474 | goto fail; | |
475 | ||
476 | ino = 0; | |
477 | ||
478 | repeat_in_this_group: | |
617ba13b MC |
479 | ino = ext4_find_next_zero_bit((unsigned long *) |
480 | bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino); | |
481 | if (ino < EXT4_INODES_PER_GROUP(sb)) { | |
ac27a0ec DK |
482 | |
483 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 484 | err = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
485 | if (err) |
486 | goto fail; | |
487 | ||
617ba13b | 488 | if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group), |
ac27a0ec DK |
489 | ino, bitmap_bh->b_data)) { |
490 | /* we won it */ | |
491 | BUFFER_TRACE(bitmap_bh, | |
617ba13b MC |
492 | "call ext4_journal_dirty_metadata"); |
493 | err = ext4_journal_dirty_metadata(handle, | |
ac27a0ec DK |
494 | bitmap_bh); |
495 | if (err) | |
496 | goto fail; | |
497 | goto got; | |
498 | } | |
499 | /* we lost it */ | |
dab291af | 500 | jbd2_journal_release_buffer(handle, bitmap_bh); |
ac27a0ec | 501 | |
617ba13b | 502 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
503 | goto repeat_in_this_group; |
504 | } | |
505 | ||
506 | /* | |
507 | * This case is possible in concurrent environment. It is very | |
508 | * rare. We cannot repeat the find_group_xxx() call because | |
509 | * that will simply return the same blockgroup, because the | |
510 | * group descriptor metadata has not yet been updated. | |
511 | * So we just go onto the next blockgroup. | |
512 | */ | |
513 | if (++group == sbi->s_groups_count) | |
514 | group = 0; | |
515 | } | |
516 | err = -ENOSPC; | |
517 | goto out; | |
518 | ||
519 | got: | |
617ba13b MC |
520 | ino += group * EXT4_INODES_PER_GROUP(sb) + 1; |
521 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
522 | ext4_error (sb, "ext4_new_inode", | |
ac27a0ec DK |
523 | "reserved inode or inode > inodes count - " |
524 | "block_group = %d, inode=%lu", group, ino); | |
525 | err = -EIO; | |
526 | goto fail; | |
527 | } | |
528 | ||
529 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 530 | err = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec DK |
531 | if (err) goto fail; |
532 | spin_lock(sb_bgl_lock(sbi, group)); | |
533 | gdp->bg_free_inodes_count = | |
534 | cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1); | |
535 | if (S_ISDIR(mode)) { | |
536 | gdp->bg_used_dirs_count = | |
537 | cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1); | |
538 | } | |
539 | spin_unlock(sb_bgl_lock(sbi, group)); | |
617ba13b MC |
540 | BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata"); |
541 | err = ext4_journal_dirty_metadata(handle, bh2); | |
ac27a0ec DK |
542 | if (err) goto fail; |
543 | ||
544 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
545 | if (S_ISDIR(mode)) | |
546 | percpu_counter_inc(&sbi->s_dirs_counter); | |
547 | sb->s_dirt = 1; | |
548 | ||
549 | inode->i_uid = current->fsuid; | |
550 | if (test_opt (sb, GRPID)) | |
551 | inode->i_gid = dir->i_gid; | |
552 | else if (dir->i_mode & S_ISGID) { | |
553 | inode->i_gid = dir->i_gid; | |
554 | if (S_ISDIR(mode)) | |
555 | mode |= S_ISGID; | |
556 | } else | |
557 | inode->i_gid = current->fsgid; | |
558 | inode->i_mode = mode; | |
559 | ||
560 | inode->i_ino = ino; | |
561 | /* This is the optimal IO size (for stat), not the fs block size */ | |
562 | inode->i_blocks = 0; | |
ef7f3835 KS |
563 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
564 | ext4_current_time(inode); | |
ac27a0ec DK |
565 | |
566 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
567 | ei->i_dir_start_lookup = 0; | |
568 | ei->i_disksize = 0; | |
569 | ||
617ba13b | 570 | ei->i_flags = EXT4_I(dir)->i_flags & ~EXT4_INDEX_FL; |
ac27a0ec | 571 | if (S_ISLNK(mode)) |
617ba13b | 572 | ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL); |
ac27a0ec DK |
573 | /* dirsync only applies to directories */ |
574 | if (!S_ISDIR(mode)) | |
617ba13b | 575 | ei->i_flags &= ~EXT4_DIRSYNC_FL; |
ac27a0ec DK |
576 | ei->i_file_acl = 0; |
577 | ei->i_dir_acl = 0; | |
578 | ei->i_dtime = 0; | |
579 | ei->i_block_alloc_info = NULL; | |
580 | ei->i_block_group = group; | |
581 | ||
617ba13b | 582 | ext4_set_inode_flags(inode); |
ac27a0ec DK |
583 | if (IS_DIRSYNC(inode)) |
584 | handle->h_sync = 1; | |
585 | insert_inode_hash(inode); | |
586 | spin_lock(&sbi->s_next_gen_lock); | |
587 | inode->i_generation = sbi->s_next_generation++; | |
588 | spin_unlock(&sbi->s_next_gen_lock); | |
589 | ||
617ba13b | 590 | ei->i_state = EXT4_STATE_NEW; |
ef7f3835 KS |
591 | |
592 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
593 | |
594 | ret = inode; | |
595 | if(DQUOT_ALLOC_INODE(inode)) { | |
596 | err = -EDQUOT; | |
597 | goto fail_drop; | |
598 | } | |
599 | ||
617ba13b | 600 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
601 | if (err) |
602 | goto fail_free_drop; | |
603 | ||
617ba13b | 604 | err = ext4_init_security(handle,inode, dir); |
ac27a0ec DK |
605 | if (err) |
606 | goto fail_free_drop; | |
607 | ||
617ba13b | 608 | err = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 609 | if (err) { |
617ba13b | 610 | ext4_std_error(sb, err); |
ac27a0ec DK |
611 | goto fail_free_drop; |
612 | } | |
a86c6181 AT |
613 | if (test_opt(sb, EXTENTS)) { |
614 | EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL; | |
615 | ext4_ext_tree_init(handle, inode); | |
616 | if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { | |
617 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); | |
618 | if (err) goto fail; | |
619 | EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS); | |
620 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "call ext4_journal_dirty_metadata"); | |
621 | err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh); | |
622 | } | |
623 | } | |
ac27a0ec | 624 | |
617ba13b | 625 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
ac27a0ec DK |
626 | goto really_out; |
627 | fail: | |
617ba13b | 628 | ext4_std_error(sb, err); |
ac27a0ec DK |
629 | out: |
630 | iput(inode); | |
631 | ret = ERR_PTR(err); | |
632 | really_out: | |
633 | brelse(bitmap_bh); | |
634 | return ret; | |
635 | ||
636 | fail_free_drop: | |
637 | DQUOT_FREE_INODE(inode); | |
638 | ||
639 | fail_drop: | |
640 | DQUOT_DROP(inode); | |
641 | inode->i_flags |= S_NOQUOTA; | |
642 | inode->i_nlink = 0; | |
643 | iput(inode); | |
644 | brelse(bitmap_bh); | |
645 | return ERR_PTR(err); | |
646 | } | |
647 | ||
648 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 649 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 650 | { |
617ba13b | 651 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
ac27a0ec DK |
652 | unsigned long block_group; |
653 | int bit; | |
654 | struct buffer_head *bitmap_bh = NULL; | |
655 | struct inode *inode = NULL; | |
656 | ||
657 | /* Error cases - e2fsck has already cleaned up for us */ | |
658 | if (ino > max_ino) { | |
617ba13b | 659 | ext4_warning(sb, __FUNCTION__, |
ac27a0ec DK |
660 | "bad orphan ino %lu! e2fsck was run?", ino); |
661 | goto out; | |
662 | } | |
663 | ||
617ba13b MC |
664 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
665 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
ac27a0ec DK |
666 | bitmap_bh = read_inode_bitmap(sb, block_group); |
667 | if (!bitmap_bh) { | |
617ba13b | 668 | ext4_warning(sb, __FUNCTION__, |
ac27a0ec DK |
669 | "inode bitmap error for orphan %lu", ino); |
670 | goto out; | |
671 | } | |
672 | ||
673 | /* Having the inode bit set should be a 100% indicator that this | |
674 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
675 | * inodes that were being truncated, so we can't check i_nlink==0. | |
676 | */ | |
617ba13b | 677 | if (!ext4_test_bit(bit, bitmap_bh->b_data) || |
ac27a0ec DK |
678 | !(inode = iget(sb, ino)) || is_bad_inode(inode) || |
679 | NEXT_ORPHAN(inode) > max_ino) { | |
617ba13b | 680 | ext4_warning(sb, __FUNCTION__, |
ac27a0ec | 681 | "bad orphan inode %lu! e2fsck was run?", ino); |
617ba13b | 682 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", |
ac27a0ec | 683 | bit, (unsigned long long)bitmap_bh->b_blocknr, |
617ba13b | 684 | ext4_test_bit(bit, bitmap_bh->b_data)); |
ac27a0ec DK |
685 | printk(KERN_NOTICE "inode=%p\n", inode); |
686 | if (inode) { | |
687 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
688 | is_bad_inode(inode)); | |
689 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
690 | NEXT_ORPHAN(inode)); | |
691 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
692 | } | |
693 | /* Avoid freeing blocks if we got a bad deleted inode */ | |
694 | if (inode && inode->i_nlink == 0) | |
695 | inode->i_blocks = 0; | |
696 | iput(inode); | |
697 | inode = NULL; | |
698 | } | |
699 | out: | |
700 | brelse(bitmap_bh); | |
701 | return inode; | |
702 | } | |
703 | ||
617ba13b | 704 | unsigned long ext4_count_free_inodes (struct super_block * sb) |
ac27a0ec DK |
705 | { |
706 | unsigned long desc_count; | |
617ba13b | 707 | struct ext4_group_desc *gdp; |
ac27a0ec | 708 | int i; |
617ba13b MC |
709 | #ifdef EXT4FS_DEBUG |
710 | struct ext4_super_block *es; | |
ac27a0ec DK |
711 | unsigned long bitmap_count, x; |
712 | struct buffer_head *bitmap_bh = NULL; | |
713 | ||
617ba13b | 714 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
715 | desc_count = 0; |
716 | bitmap_count = 0; | |
717 | gdp = NULL; | |
617ba13b MC |
718 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
719 | gdp = ext4_get_group_desc (sb, i, NULL); | |
ac27a0ec DK |
720 | if (!gdp) |
721 | continue; | |
722 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
723 | brelse(bitmap_bh); | |
724 | bitmap_bh = read_inode_bitmap(sb, i); | |
725 | if (!bitmap_bh) | |
726 | continue; | |
727 | ||
617ba13b | 728 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
ac27a0ec DK |
729 | printk("group %d: stored = %d, counted = %lu\n", |
730 | i, le16_to_cpu(gdp->bg_free_inodes_count), x); | |
731 | bitmap_count += x; | |
732 | } | |
733 | brelse(bitmap_bh); | |
617ba13b | 734 | printk("ext4_count_free_inodes: stored = %u, computed = %lu, %lu\n", |
ac27a0ec DK |
735 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); |
736 | return desc_count; | |
737 | #else | |
738 | desc_count = 0; | |
617ba13b MC |
739 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
740 | gdp = ext4_get_group_desc (sb, i, NULL); | |
ac27a0ec DK |
741 | if (!gdp) |
742 | continue; | |
743 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
744 | cond_resched(); | |
745 | } | |
746 | return desc_count; | |
747 | #endif | |
748 | } | |
749 | ||
750 | /* Called at mount-time, super-block is locked */ | |
617ba13b | 751 | unsigned long ext4_count_dirs (struct super_block * sb) |
ac27a0ec DK |
752 | { |
753 | unsigned long count = 0; | |
754 | int i; | |
755 | ||
617ba13b MC |
756 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
757 | struct ext4_group_desc *gdp = ext4_get_group_desc (sb, i, NULL); | |
ac27a0ec DK |
758 | if (!gdp) |
759 | continue; | |
760 | count += le16_to_cpu(gdp->bg_used_dirs_count); | |
761 | } | |
762 | return count; | |
763 | } | |
764 |