]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/ialloc.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * BSD ufs-inspired inode and directory allocation by | |
10 | * Stephen Tweedie (sct@redhat.com), 1993 | |
11 | * Big-endian to little-endian byte-swapping/bitmaps by | |
12 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
13 | */ | |
14 | ||
15 | #include <linux/time.h> | |
16 | #include <linux/fs.h> | |
dab291af | 17 | #include <linux/jbd2.h> |
ac27a0ec DK |
18 | #include <linux/stat.h> |
19 | #include <linux/string.h> | |
20 | #include <linux/quotaops.h> | |
21 | #include <linux/buffer_head.h> | |
22 | #include <linux/random.h> | |
23 | #include <linux/bitops.h> | |
3a5b2ecd | 24 | #include <linux/blkdev.h> |
ac27a0ec | 25 | #include <asm/byteorder.h> |
3dcf5451 CH |
26 | #include "ext4.h" |
27 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
28 | #include "xattr.h" |
29 | #include "acl.h" | |
717d50e4 | 30 | #include "group.h" |
ac27a0ec DK |
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 | ||
717d50e4 AD |
46 | /* |
47 | * To avoid calling the atomic setbit hundreds or thousands of times, we only | |
48 | * need to use it within a single byte (to ensure we get endianness right). | |
49 | * We can use memset for the rest of the bitmap as there are no other users. | |
50 | */ | |
51 | void mark_bitmap_end(int start_bit, int end_bit, char *bitmap) | |
52 | { | |
53 | int i; | |
54 | ||
55 | if (start_bit >= end_bit) | |
56 | return; | |
57 | ||
58 | ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); | |
59 | for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) | |
60 | ext4_set_bit(i, bitmap); | |
61 | if (i < end_bit) | |
62 | memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); | |
63 | } | |
64 | ||
65 | /* Initializes an uninitialized inode bitmap */ | |
fd2d4291 AM |
66 | unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh, |
67 | ext4_group_t block_group, | |
717d50e4 AD |
68 | struct ext4_group_desc *gdp) |
69 | { | |
70 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
71 | ||
72 | J_ASSERT_BH(bh, buffer_locked(bh)); | |
73 | ||
74 | /* If checksum is bad mark all blocks and inodes use to prevent | |
75 | * allocation, essentially implementing a per-group read-only flag. */ | |
76 | if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) { | |
46e665e9 | 77 | ext4_error(sb, __func__, "Checksum bad for group %lu\n", |
717d50e4 AD |
78 | block_group); |
79 | gdp->bg_free_blocks_count = 0; | |
80 | gdp->bg_free_inodes_count = 0; | |
81 | gdp->bg_itable_unused = 0; | |
82 | memset(bh->b_data, 0xff, sb->s_blocksize); | |
83 | return 0; | |
84 | } | |
85 | ||
86 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
87 | mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb), | |
88 | bh->b_data); | |
89 | ||
90 | return EXT4_INODES_PER_GROUP(sb); | |
91 | } | |
ac27a0ec DK |
92 | |
93 | /* | |
94 | * Read the inode allocation bitmap for a given block_group, reading | |
95 | * into the specified slot in the superblock's bitmap cache. | |
96 | * | |
97 | * Return buffer_head of bitmap on success or NULL. | |
98 | */ | |
99 | static struct buffer_head * | |
fd2d4291 | 100 | read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 101 | { |
617ba13b | 102 | struct ext4_group_desc *desc; |
ac27a0ec DK |
103 | struct buffer_head *bh = NULL; |
104 | ||
617ba13b | 105 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec DK |
106 | if (!desc) |
107 | goto error_out; | |
717d50e4 AD |
108 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
109 | bh = sb_getblk(sb, ext4_inode_bitmap(sb, desc)); | |
110 | if (!buffer_uptodate(bh)) { | |
111 | lock_buffer(bh); | |
112 | if (!buffer_uptodate(bh)) { | |
113 | ext4_init_inode_bitmap(sb, bh, block_group, | |
114 | desc); | |
115 | set_buffer_uptodate(bh); | |
116 | } | |
117 | unlock_buffer(bh); | |
118 | } | |
119 | } else { | |
120 | bh = sb_bread(sb, ext4_inode_bitmap(sb, desc)); | |
121 | } | |
ac27a0ec | 122 | if (!bh) |
617ba13b | 123 | ext4_error(sb, "read_inode_bitmap", |
ac27a0ec | 124 | "Cannot read inode bitmap - " |
bd81d8ee | 125 | "block_group = %lu, inode_bitmap = %llu", |
8fadc143 | 126 | block_group, ext4_inode_bitmap(sb, desc)); |
ac27a0ec DK |
127 | error_out: |
128 | return bh; | |
129 | } | |
130 | ||
131 | /* | |
132 | * NOTE! When we get the inode, we're the only people | |
133 | * that have access to it, and as such there are no | |
134 | * race conditions we have to worry about. The inode | |
135 | * is not on the hash-lists, and it cannot be reached | |
136 | * through the filesystem because the directory entry | |
137 | * has been deleted earlier. | |
138 | * | |
139 | * HOWEVER: we must make sure that we get no aliases, | |
140 | * which means that we have to call "clear_inode()" | |
141 | * _before_ we mark the inode not in use in the inode | |
142 | * bitmaps. Otherwise a newly created file might use | |
143 | * the same inode number (not actually the same pointer | |
144 | * though), and then we'd have two inodes sharing the | |
145 | * same inode number and space on the harddisk. | |
146 | */ | |
617ba13b | 147 | void ext4_free_inode (handle_t *handle, struct inode * inode) |
ac27a0ec DK |
148 | { |
149 | struct super_block * sb = inode->i_sb; | |
150 | int is_directory; | |
151 | unsigned long ino; | |
152 | struct buffer_head *bitmap_bh = NULL; | |
153 | struct buffer_head *bh2; | |
fd2d4291 | 154 | ext4_group_t block_group; |
ac27a0ec | 155 | unsigned long bit; |
617ba13b MC |
156 | struct ext4_group_desc * gdp; |
157 | struct ext4_super_block * es; | |
158 | struct ext4_sb_info *sbi; | |
ac27a0ec DK |
159 | int fatal = 0, err; |
160 | ||
161 | if (atomic_read(&inode->i_count) > 1) { | |
617ba13b | 162 | printk ("ext4_free_inode: inode has count=%d\n", |
ac27a0ec DK |
163 | atomic_read(&inode->i_count)); |
164 | return; | |
165 | } | |
166 | if (inode->i_nlink) { | |
617ba13b | 167 | printk ("ext4_free_inode: inode has nlink=%d\n", |
ac27a0ec DK |
168 | inode->i_nlink); |
169 | return; | |
170 | } | |
171 | if (!sb) { | |
617ba13b | 172 | printk("ext4_free_inode: inode on nonexistent device\n"); |
ac27a0ec DK |
173 | return; |
174 | } | |
617ba13b | 175 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
176 | |
177 | ino = inode->i_ino; | |
617ba13b | 178 | ext4_debug ("freeing inode %lu\n", ino); |
ac27a0ec DK |
179 | |
180 | /* | |
181 | * Note: we must free any quota before locking the superblock, | |
182 | * as writing the quota to disk may need the lock as well. | |
183 | */ | |
184 | DQUOT_INIT(inode); | |
617ba13b | 185 | ext4_xattr_delete_inode(handle, inode); |
ac27a0ec DK |
186 | DQUOT_FREE_INODE(inode); |
187 | DQUOT_DROP(inode); | |
188 | ||
189 | is_directory = S_ISDIR(inode->i_mode); | |
190 | ||
191 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
192 | clear_inode (inode); | |
193 | ||
617ba13b MC |
194 | es = EXT4_SB(sb)->s_es; |
195 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
196 | ext4_error (sb, "ext4_free_inode", | |
ac27a0ec DK |
197 | "reserved or nonexistent inode %lu", ino); |
198 | goto error_return; | |
199 | } | |
617ba13b MC |
200 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
201 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
ac27a0ec DK |
202 | bitmap_bh = read_inode_bitmap(sb, block_group); |
203 | if (!bitmap_bh) | |
204 | goto error_return; | |
205 | ||
206 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 207 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
208 | if (fatal) |
209 | goto error_return; | |
210 | ||
211 | /* Ok, now we can actually update the inode bitmaps.. */ | |
617ba13b | 212 | if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
ac27a0ec | 213 | bit, bitmap_bh->b_data)) |
617ba13b | 214 | ext4_error (sb, "ext4_free_inode", |
ac27a0ec DK |
215 | "bit already cleared for inode %lu", ino); |
216 | else { | |
617ba13b | 217 | gdp = ext4_get_group_desc (sb, block_group, &bh2); |
ac27a0ec DK |
218 | |
219 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 220 | fatal = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec DK |
221 | if (fatal) goto error_return; |
222 | ||
223 | if (gdp) { | |
224 | spin_lock(sb_bgl_lock(sbi, block_group)); | |
e8546d06 | 225 | le16_add_cpu(&gdp->bg_free_inodes_count, 1); |
ac27a0ec | 226 | if (is_directory) |
e8546d06 | 227 | le16_add_cpu(&gdp->bg_used_dirs_count, -1); |
717d50e4 AD |
228 | gdp->bg_checksum = ext4_group_desc_csum(sbi, |
229 | block_group, gdp); | |
ac27a0ec DK |
230 | spin_unlock(sb_bgl_lock(sbi, block_group)); |
231 | percpu_counter_inc(&sbi->s_freeinodes_counter); | |
232 | if (is_directory) | |
233 | percpu_counter_dec(&sbi->s_dirs_counter); | |
234 | ||
235 | } | |
617ba13b MC |
236 | BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata"); |
237 | err = ext4_journal_dirty_metadata(handle, bh2); | |
ac27a0ec DK |
238 | if (!fatal) fatal = err; |
239 | } | |
617ba13b MC |
240 | BUFFER_TRACE(bitmap_bh, "call ext4_journal_dirty_metadata"); |
241 | err = ext4_journal_dirty_metadata(handle, bitmap_bh); | |
ac27a0ec DK |
242 | if (!fatal) |
243 | fatal = err; | |
244 | sb->s_dirt = 1; | |
245 | error_return: | |
246 | brelse(bitmap_bh); | |
617ba13b | 247 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
248 | } |
249 | ||
250 | /* | |
251 | * There are two policies for allocating an inode. If the new inode is | |
252 | * a directory, then a forward search is made for a block group with both | |
253 | * free space and a low directory-to-inode ratio; if that fails, then of | |
254 | * the groups with above-average free space, that group with the fewest | |
255 | * directories already is chosen. | |
256 | * | |
257 | * For other inodes, search forward from the parent directory\'s block | |
258 | * group to find a free inode. | |
259 | */ | |
2aa9fc4c AM |
260 | static int find_group_dir(struct super_block *sb, struct inode *parent, |
261 | ext4_group_t *best_group) | |
ac27a0ec | 262 | { |
fd2d4291 | 263 | ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; |
ac27a0ec | 264 | unsigned int freei, avefreei; |
617ba13b | 265 | struct ext4_group_desc *desc, *best_desc = NULL; |
2aa9fc4c AM |
266 | ext4_group_t group; |
267 | int ret = -1; | |
ac27a0ec | 268 | |
617ba13b | 269 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
270 | avefreei = freei / ngroups; |
271 | ||
272 | for (group = 0; group < ngroups; group++) { | |
ef2fb679 | 273 | desc = ext4_get_group_desc (sb, group, NULL); |
ac27a0ec DK |
274 | if (!desc || !desc->bg_free_inodes_count) |
275 | continue; | |
276 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
277 | continue; | |
278 | if (!best_desc || | |
279 | (le16_to_cpu(desc->bg_free_blocks_count) > | |
280 | le16_to_cpu(best_desc->bg_free_blocks_count))) { | |
2aa9fc4c | 281 | *best_group = group; |
ac27a0ec | 282 | best_desc = desc; |
2aa9fc4c | 283 | ret = 0; |
ac27a0ec DK |
284 | } |
285 | } | |
2aa9fc4c | 286 | return ret; |
ac27a0ec DK |
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). | |
1cc8dcf5 | 305 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
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 | ||
2aa9fc4c AM |
317 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
318 | ext4_group_t *group) | |
ac27a0ec | 319 | { |
fd2d4291 | 320 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b MC |
321 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
322 | struct ext4_super_block *es = sbi->s_es; | |
fd2d4291 | 323 | ext4_group_t ngroups = sbi->s_groups_count; |
617ba13b | 324 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 325 | unsigned int freei, avefreei; |
617ba13b MC |
326 | ext4_fsblk_t freeb, avefreeb; |
327 | ext4_fsblk_t blocks_per_dir; | |
ac27a0ec DK |
328 | unsigned int ndirs; |
329 | int max_debt, max_dirs, min_inodes; | |
617ba13b | 330 | ext4_grpblk_t min_blocks; |
2aa9fc4c | 331 | ext4_group_t i; |
617ba13b | 332 | struct ext4_group_desc *desc; |
ac27a0ec DK |
333 | |
334 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
335 | avefreei = freei / ngroups; | |
336 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3a5b2ecd | 337 | avefreeb = freeb; |
f4e5bc24 | 338 | do_div(avefreeb, ngroups); |
ac27a0ec DK |
339 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
340 | ||
341 | if ((parent == sb->s_root->d_inode) || | |
617ba13b | 342 | (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) { |
ac27a0ec | 343 | int best_ndir = inodes_per_group; |
2aa9fc4c AM |
344 | ext4_group_t grp; |
345 | int ret = -1; | |
ac27a0ec | 346 | |
2aa9fc4c AM |
347 | get_random_bytes(&grp, sizeof(grp)); |
348 | parent_group = (unsigned)grp % ngroups; | |
ac27a0ec | 349 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
350 | grp = (parent_group + i) % ngroups; |
351 | desc = ext4_get_group_desc(sb, grp, NULL); | |
ac27a0ec DK |
352 | if (!desc || !desc->bg_free_inodes_count) |
353 | continue; | |
354 | if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) | |
355 | continue; | |
356 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
357 | continue; | |
358 | if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) | |
359 | continue; | |
2aa9fc4c AM |
360 | *group = grp; |
361 | ret = 0; | |
ac27a0ec DK |
362 | best_ndir = le16_to_cpu(desc->bg_used_dirs_count); |
363 | } | |
2aa9fc4c AM |
364 | if (ret == 0) |
365 | return ret; | |
ac27a0ec DK |
366 | goto fallback; |
367 | } | |
368 | ||
bd81d8ee | 369 | blocks_per_dir = ext4_blocks_count(es) - freeb; |
f4e5bc24 | 370 | do_div(blocks_per_dir, ndirs); |
ac27a0ec DK |
371 | |
372 | max_dirs = ndirs / ngroups + inodes_per_group / 16; | |
373 | min_inodes = avefreei - inodes_per_group / 4; | |
617ba13b | 374 | min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4; |
ac27a0ec | 375 | |
3a5b2ecd | 376 | max_debt = EXT4_BLOCKS_PER_GROUP(sb); |
f4e5bc24 | 377 | max_debt /= max_t(int, blocks_per_dir, BLOCK_COST); |
ac27a0ec DK |
378 | if (max_debt * INODE_COST > inodes_per_group) |
379 | max_debt = inodes_per_group / INODE_COST; | |
380 | if (max_debt > 255) | |
381 | max_debt = 255; | |
382 | if (max_debt == 0) | |
383 | max_debt = 1; | |
384 | ||
385 | for (i = 0; i < ngroups; i++) { | |
2aa9fc4c AM |
386 | *group = (parent_group + i) % ngroups; |
387 | desc = ext4_get_group_desc(sb, *group, NULL); | |
ac27a0ec DK |
388 | if (!desc || !desc->bg_free_inodes_count) |
389 | continue; | |
390 | if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs) | |
391 | continue; | |
392 | if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes) | |
393 | continue; | |
394 | if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks) | |
395 | continue; | |
2aa9fc4c | 396 | return 0; |
ac27a0ec DK |
397 | } |
398 | ||
399 | fallback: | |
400 | for (i = 0; i < ngroups; i++) { | |
2aa9fc4c AM |
401 | *group = (parent_group + i) % ngroups; |
402 | desc = ext4_get_group_desc(sb, *group, NULL); | |
403 | if (desc && desc->bg_free_inodes_count && | |
404 | le16_to_cpu(desc->bg_free_inodes_count) >= avefreei) | |
405 | return 0; | |
ac27a0ec DK |
406 | } |
407 | ||
408 | if (avefreei) { | |
409 | /* | |
410 | * The free-inodes counter is approximate, and for really small | |
411 | * filesystems the above test can fail to find any blockgroups | |
412 | */ | |
413 | avefreei = 0; | |
414 | goto fallback; | |
415 | } | |
416 | ||
417 | return -1; | |
418 | } | |
419 | ||
2aa9fc4c AM |
420 | static int find_group_other(struct super_block *sb, struct inode *parent, |
421 | ext4_group_t *group) | |
ac27a0ec | 422 | { |
fd2d4291 AM |
423 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
424 | ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; | |
617ba13b | 425 | struct ext4_group_desc *desc; |
2aa9fc4c | 426 | ext4_group_t i; |
ac27a0ec DK |
427 | |
428 | /* | |
429 | * Try to place the inode in its parent directory | |
430 | */ | |
2aa9fc4c AM |
431 | *group = parent_group; |
432 | desc = ext4_get_group_desc(sb, *group, NULL); | |
ac27a0ec DK |
433 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
434 | le16_to_cpu(desc->bg_free_blocks_count)) | |
2aa9fc4c | 435 | return 0; |
ac27a0ec DK |
436 | |
437 | /* | |
438 | * We're going to place this inode in a different blockgroup from its | |
439 | * parent. We want to cause files in a common directory to all land in | |
440 | * the same blockgroup. But we want files which are in a different | |
441 | * directory which shares a blockgroup with our parent to land in a | |
442 | * different blockgroup. | |
443 | * | |
444 | * So add our directory's i_ino into the starting point for the hash. | |
445 | */ | |
2aa9fc4c | 446 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
447 | |
448 | /* | |
449 | * Use a quadratic hash to find a group with a free inode and some free | |
450 | * blocks. | |
451 | */ | |
452 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
453 | *group += i; |
454 | if (*group >= ngroups) | |
455 | *group -= ngroups; | |
456 | desc = ext4_get_group_desc(sb, *group, NULL); | |
ac27a0ec DK |
457 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
458 | le16_to_cpu(desc->bg_free_blocks_count)) | |
2aa9fc4c | 459 | return 0; |
ac27a0ec DK |
460 | } |
461 | ||
462 | /* | |
463 | * That failed: try linear search for a free inode, even if that group | |
464 | * has no free blocks. | |
465 | */ | |
2aa9fc4c | 466 | *group = parent_group; |
ac27a0ec | 467 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
468 | if (++*group >= ngroups) |
469 | *group = 0; | |
470 | desc = ext4_get_group_desc(sb, *group, NULL); | |
ac27a0ec | 471 | if (desc && le16_to_cpu(desc->bg_free_inodes_count)) |
2aa9fc4c | 472 | return 0; |
ac27a0ec DK |
473 | } |
474 | ||
475 | return -1; | |
476 | } | |
477 | ||
478 | /* | |
479 | * There are two policies for allocating an inode. If the new inode is | |
480 | * a directory, then a forward search is made for a block group with both | |
481 | * free space and a low directory-to-inode ratio; if that fails, then of | |
482 | * the groups with above-average free space, that group with the fewest | |
483 | * directories already is chosen. | |
484 | * | |
485 | * For other inodes, search forward from the parent directory's block | |
486 | * group to find a free inode. | |
487 | */ | |
617ba13b | 488 | struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode) |
ac27a0ec DK |
489 | { |
490 | struct super_block *sb; | |
491 | struct buffer_head *bitmap_bh = NULL; | |
492 | struct buffer_head *bh2; | |
2aa9fc4c | 493 | ext4_group_t group = 0; |
ac27a0ec DK |
494 | unsigned long ino = 0; |
495 | struct inode * inode; | |
617ba13b MC |
496 | struct ext4_group_desc * gdp = NULL; |
497 | struct ext4_super_block * es; | |
498 | struct ext4_inode_info *ei; | |
499 | struct ext4_sb_info *sbi; | |
2aa9fc4c | 500 | int ret2, err = 0; |
ac27a0ec | 501 | struct inode *ret; |
2aa9fc4c AM |
502 | ext4_group_t i; |
503 | int free = 0; | |
ac27a0ec DK |
504 | |
505 | /* Cannot create files in a deleted directory */ | |
506 | if (!dir || !dir->i_nlink) | |
507 | return ERR_PTR(-EPERM); | |
508 | ||
509 | sb = dir->i_sb; | |
510 | inode = new_inode(sb); | |
511 | if (!inode) | |
512 | return ERR_PTR(-ENOMEM); | |
617ba13b | 513 | ei = EXT4_I(inode); |
ac27a0ec | 514 | |
617ba13b | 515 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
516 | es = sbi->s_es; |
517 | if (S_ISDIR(mode)) { | |
518 | if (test_opt (sb, OLDALLOC)) | |
2aa9fc4c | 519 | ret2 = find_group_dir(sb, dir, &group); |
ac27a0ec | 520 | else |
2aa9fc4c | 521 | ret2 = find_group_orlov(sb, dir, &group); |
ac27a0ec | 522 | } else |
2aa9fc4c | 523 | ret2 = find_group_other(sb, dir, &group); |
ac27a0ec DK |
524 | |
525 | err = -ENOSPC; | |
2aa9fc4c | 526 | if (ret2 == -1) |
ac27a0ec DK |
527 | goto out; |
528 | ||
529 | for (i = 0; i < sbi->s_groups_count; i++) { | |
530 | err = -EIO; | |
531 | ||
617ba13b | 532 | gdp = ext4_get_group_desc(sb, group, &bh2); |
ac27a0ec DK |
533 | if (!gdp) |
534 | goto fail; | |
535 | ||
536 | brelse(bitmap_bh); | |
537 | bitmap_bh = read_inode_bitmap(sb, group); | |
538 | if (!bitmap_bh) | |
539 | goto fail; | |
540 | ||
541 | ino = 0; | |
542 | ||
543 | repeat_in_this_group: | |
617ba13b MC |
544 | ino = ext4_find_next_zero_bit((unsigned long *) |
545 | bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino); | |
546 | if (ino < EXT4_INODES_PER_GROUP(sb)) { | |
ac27a0ec DK |
547 | |
548 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 549 | err = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
550 | if (err) |
551 | goto fail; | |
552 | ||
617ba13b | 553 | if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group), |
ac27a0ec DK |
554 | ino, bitmap_bh->b_data)) { |
555 | /* we won it */ | |
556 | BUFFER_TRACE(bitmap_bh, | |
617ba13b MC |
557 | "call ext4_journal_dirty_metadata"); |
558 | err = ext4_journal_dirty_metadata(handle, | |
ac27a0ec DK |
559 | bitmap_bh); |
560 | if (err) | |
561 | goto fail; | |
562 | goto got; | |
563 | } | |
564 | /* we lost it */ | |
dab291af | 565 | jbd2_journal_release_buffer(handle, bitmap_bh); |
ac27a0ec | 566 | |
617ba13b | 567 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
568 | goto repeat_in_this_group; |
569 | } | |
570 | ||
571 | /* | |
572 | * This case is possible in concurrent environment. It is very | |
573 | * rare. We cannot repeat the find_group_xxx() call because | |
574 | * that will simply return the same blockgroup, because the | |
575 | * group descriptor metadata has not yet been updated. | |
576 | * So we just go onto the next blockgroup. | |
577 | */ | |
578 | if (++group == sbi->s_groups_count) | |
579 | group = 0; | |
580 | } | |
581 | err = -ENOSPC; | |
582 | goto out; | |
583 | ||
584 | got: | |
717d50e4 AD |
585 | ino++; |
586 | if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || | |
587 | ino > EXT4_INODES_PER_GROUP(sb)) { | |
46e665e9 | 588 | ext4_error(sb, __func__, |
717d50e4 | 589 | "reserved inode or inode > inodes count - " |
fd2d4291 | 590 | "block_group = %lu, inode=%lu", group, |
717d50e4 | 591 | ino + group * EXT4_INODES_PER_GROUP(sb)); |
ac27a0ec DK |
592 | err = -EIO; |
593 | goto fail; | |
594 | } | |
595 | ||
596 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 597 | err = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec | 598 | if (err) goto fail; |
717d50e4 AD |
599 | |
600 | /* We may have to initialize the block bitmap if it isn't already */ | |
601 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && | |
602 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
603 | struct buffer_head *block_bh = read_block_bitmap(sb, group); | |
604 | ||
605 | BUFFER_TRACE(block_bh, "get block bitmap access"); | |
606 | err = ext4_journal_get_write_access(handle, block_bh); | |
607 | if (err) { | |
608 | brelse(block_bh); | |
609 | goto fail; | |
610 | } | |
611 | ||
612 | free = 0; | |
613 | spin_lock(sb_bgl_lock(sbi, group)); | |
614 | /* recheck and clear flag under lock if we still need to */ | |
615 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
616 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); | |
617 | free = ext4_free_blocks_after_init(sb, group, gdp); | |
618 | gdp->bg_free_blocks_count = cpu_to_le16(free); | |
619 | } | |
620 | spin_unlock(sb_bgl_lock(sbi, group)); | |
621 | ||
622 | /* Don't need to dirty bitmap block if we didn't change it */ | |
623 | if (free) { | |
624 | BUFFER_TRACE(block_bh, "dirty block bitmap"); | |
625 | err = ext4_journal_dirty_metadata(handle, block_bh); | |
626 | } | |
627 | ||
628 | brelse(block_bh); | |
629 | if (err) | |
630 | goto fail; | |
631 | } | |
632 | ||
ac27a0ec | 633 | spin_lock(sb_bgl_lock(sbi, group)); |
717d50e4 AD |
634 | /* If we didn't allocate from within the initialized part of the inode |
635 | * table then we need to initialize up to this inode. */ | |
636 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { | |
637 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
638 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
639 | ||
640 | /* When marking the block group with | |
641 | * ~EXT4_BG_INODE_UNINIT we don't want to depend | |
642 | * on the value of bg_itable_unsed even though | |
643 | * mke2fs could have initialized the same for us. | |
644 | * Instead we calculated the value below | |
645 | */ | |
646 | ||
647 | free = 0; | |
648 | } else { | |
649 | free = EXT4_INODES_PER_GROUP(sb) - | |
650 | le16_to_cpu(gdp->bg_itable_unused); | |
651 | } | |
652 | ||
653 | /* | |
654 | * Check the relative inode number against the last used | |
655 | * relative inode number in this group. if it is greater | |
656 | * we need to update the bg_itable_unused count | |
657 | * | |
658 | */ | |
659 | if (ino > free) | |
660 | gdp->bg_itable_unused = | |
661 | cpu_to_le16(EXT4_INODES_PER_GROUP(sb) - ino); | |
662 | } | |
663 | ||
e8546d06 | 664 | le16_add_cpu(&gdp->bg_free_inodes_count, -1); |
ac27a0ec | 665 | if (S_ISDIR(mode)) { |
e8546d06 | 666 | le16_add_cpu(&gdp->bg_used_dirs_count, 1); |
ac27a0ec | 667 | } |
717d50e4 | 668 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); |
ac27a0ec | 669 | spin_unlock(sb_bgl_lock(sbi, group)); |
617ba13b MC |
670 | BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata"); |
671 | err = ext4_journal_dirty_metadata(handle, bh2); | |
ac27a0ec DK |
672 | if (err) goto fail; |
673 | ||
674 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
675 | if (S_ISDIR(mode)) | |
676 | percpu_counter_inc(&sbi->s_dirs_counter); | |
677 | sb->s_dirt = 1; | |
678 | ||
679 | inode->i_uid = current->fsuid; | |
680 | if (test_opt (sb, GRPID)) | |
681 | inode->i_gid = dir->i_gid; | |
682 | else if (dir->i_mode & S_ISGID) { | |
683 | inode->i_gid = dir->i_gid; | |
684 | if (S_ISDIR(mode)) | |
685 | mode |= S_ISGID; | |
686 | } else | |
687 | inode->i_gid = current->fsgid; | |
688 | inode->i_mode = mode; | |
689 | ||
717d50e4 | 690 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
691 | /* This is the optimal IO size (for stat), not the fs block size */ |
692 | inode->i_blocks = 0; | |
ef7f3835 KS |
693 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
694 | ext4_current_time(inode); | |
ac27a0ec DK |
695 | |
696 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
697 | ei->i_dir_start_lookup = 0; | |
698 | ei->i_disksize = 0; | |
699 | ||
42bf0383 AK |
700 | /* |
701 | * Don't inherit extent flag from directory. We set extent flag on | |
702 | * newly created directory and file only if -o extent mount option is | |
703 | * specified | |
704 | */ | |
705 | ei->i_flags = EXT4_I(dir)->i_flags & ~(EXT4_INDEX_FL|EXT4_EXTENTS_FL); | |
ac27a0ec | 706 | if (S_ISLNK(mode)) |
617ba13b | 707 | ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL); |
ac27a0ec DK |
708 | /* dirsync only applies to directories */ |
709 | if (!S_ISDIR(mode)) | |
617ba13b | 710 | ei->i_flags &= ~EXT4_DIRSYNC_FL; |
ac27a0ec | 711 | ei->i_file_acl = 0; |
ac27a0ec DK |
712 | ei->i_dtime = 0; |
713 | ei->i_block_alloc_info = NULL; | |
714 | ei->i_block_group = group; | |
715 | ||
617ba13b | 716 | ext4_set_inode_flags(inode); |
ac27a0ec DK |
717 | if (IS_DIRSYNC(inode)) |
718 | handle->h_sync = 1; | |
719 | insert_inode_hash(inode); | |
720 | spin_lock(&sbi->s_next_gen_lock); | |
721 | inode->i_generation = sbi->s_next_generation++; | |
722 | spin_unlock(&sbi->s_next_gen_lock); | |
723 | ||
617ba13b | 724 | ei->i_state = EXT4_STATE_NEW; |
ef7f3835 KS |
725 | |
726 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
727 | |
728 | ret = inode; | |
729 | if(DQUOT_ALLOC_INODE(inode)) { | |
730 | err = -EDQUOT; | |
731 | goto fail_drop; | |
732 | } | |
733 | ||
617ba13b | 734 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
735 | if (err) |
736 | goto fail_free_drop; | |
737 | ||
617ba13b | 738 | err = ext4_init_security(handle,inode, dir); |
ac27a0ec DK |
739 | if (err) |
740 | goto fail_free_drop; | |
741 | ||
617ba13b | 742 | err = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec | 743 | if (err) { |
617ba13b | 744 | ext4_std_error(sb, err); |
ac27a0ec DK |
745 | goto fail_free_drop; |
746 | } | |
a86c6181 | 747 | if (test_opt(sb, EXTENTS)) { |
e65187e6 AK |
748 | /* set extent flag only for diretory, file and normal symlink*/ |
749 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { | |
42bf0383 AK |
750 | EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL; |
751 | ext4_ext_tree_init(handle, inode); | |
752 | err = ext4_update_incompat_feature(handle, sb, | |
753 | EXT4_FEATURE_INCOMPAT_EXTENTS); | |
754 | if (err) | |
755 | goto fail; | |
756 | } | |
a86c6181 | 757 | } |
ac27a0ec | 758 | |
617ba13b | 759 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
ac27a0ec DK |
760 | goto really_out; |
761 | fail: | |
617ba13b | 762 | ext4_std_error(sb, err); |
ac27a0ec DK |
763 | out: |
764 | iput(inode); | |
765 | ret = ERR_PTR(err); | |
766 | really_out: | |
767 | brelse(bitmap_bh); | |
768 | return ret; | |
769 | ||
770 | fail_free_drop: | |
771 | DQUOT_FREE_INODE(inode); | |
772 | ||
773 | fail_drop: | |
774 | DQUOT_DROP(inode); | |
775 | inode->i_flags |= S_NOQUOTA; | |
776 | inode->i_nlink = 0; | |
777 | iput(inode); | |
778 | brelse(bitmap_bh); | |
779 | return ERR_PTR(err); | |
780 | } | |
781 | ||
782 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 783 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 784 | { |
617ba13b | 785 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 786 | ext4_group_t block_group; |
ac27a0ec | 787 | int bit; |
1d1fe1ee | 788 | struct buffer_head *bitmap_bh; |
ac27a0ec | 789 | struct inode *inode = NULL; |
1d1fe1ee | 790 | long err = -EIO; |
ac27a0ec DK |
791 | |
792 | /* Error cases - e2fsck has already cleaned up for us */ | |
793 | if (ino > max_ino) { | |
46e665e9 | 794 | ext4_warning(sb, __func__, |
ac27a0ec | 795 | "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 796 | goto error; |
ac27a0ec DK |
797 | } |
798 | ||
617ba13b MC |
799 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
800 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
ac27a0ec DK |
801 | bitmap_bh = read_inode_bitmap(sb, block_group); |
802 | if (!bitmap_bh) { | |
46e665e9 | 803 | ext4_warning(sb, __func__, |
ac27a0ec | 804 | "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 805 | goto error; |
ac27a0ec DK |
806 | } |
807 | ||
808 | /* Having the inode bit set should be a 100% indicator that this | |
809 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
810 | * inodes that were being truncated, so we can't check i_nlink==0. | |
811 | */ | |
1d1fe1ee DH |
812 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
813 | goto bad_orphan; | |
814 | ||
815 | inode = ext4_iget(sb, ino); | |
816 | if (IS_ERR(inode)) | |
817 | goto iget_failed; | |
818 | ||
819 | if (NEXT_ORPHAN(inode) > max_ino) | |
820 | goto bad_orphan; | |
821 | brelse(bitmap_bh); | |
822 | return inode; | |
823 | ||
824 | iget_failed: | |
825 | err = PTR_ERR(inode); | |
826 | inode = NULL; | |
827 | bad_orphan: | |
46e665e9 | 828 | ext4_warning(sb, __func__, |
1d1fe1ee DH |
829 | "bad orphan inode %lu! e2fsck was run?", ino); |
830 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", | |
831 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
832 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
833 | printk(KERN_NOTICE "inode=%p\n", inode); | |
834 | if (inode) { | |
835 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
836 | is_bad_inode(inode)); | |
837 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
838 | NEXT_ORPHAN(inode)); | |
839 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
ac27a0ec | 840 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 841 | if (inode->i_nlink == 0) |
ac27a0ec DK |
842 | inode->i_blocks = 0; |
843 | iput(inode); | |
ac27a0ec | 844 | } |
ac27a0ec | 845 | brelse(bitmap_bh); |
1d1fe1ee DH |
846 | error: |
847 | return ERR_PTR(err); | |
ac27a0ec DK |
848 | } |
849 | ||
617ba13b | 850 | unsigned long ext4_count_free_inodes (struct super_block * sb) |
ac27a0ec DK |
851 | { |
852 | unsigned long desc_count; | |
617ba13b | 853 | struct ext4_group_desc *gdp; |
fd2d4291 | 854 | ext4_group_t i; |
617ba13b MC |
855 | #ifdef EXT4FS_DEBUG |
856 | struct ext4_super_block *es; | |
ac27a0ec DK |
857 | unsigned long bitmap_count, x; |
858 | struct buffer_head *bitmap_bh = NULL; | |
859 | ||
617ba13b | 860 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
861 | desc_count = 0; |
862 | bitmap_count = 0; | |
863 | gdp = NULL; | |
617ba13b MC |
864 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
865 | gdp = ext4_get_group_desc (sb, i, NULL); | |
ac27a0ec DK |
866 | if (!gdp) |
867 | continue; | |
868 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
869 | brelse(bitmap_bh); | |
870 | bitmap_bh = read_inode_bitmap(sb, i); | |
871 | if (!bitmap_bh) | |
872 | continue; | |
873 | ||
617ba13b | 874 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
c549a95d | 875 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
ac27a0ec DK |
876 | i, le16_to_cpu(gdp->bg_free_inodes_count), x); |
877 | bitmap_count += x; | |
878 | } | |
879 | brelse(bitmap_bh); | |
617ba13b | 880 | printk("ext4_count_free_inodes: stored = %u, computed = %lu, %lu\n", |
ac27a0ec DK |
881 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); |
882 | return desc_count; | |
883 | #else | |
884 | desc_count = 0; | |
617ba13b MC |
885 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
886 | gdp = ext4_get_group_desc (sb, i, NULL); | |
ac27a0ec DK |
887 | if (!gdp) |
888 | continue; | |
889 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
890 | cond_resched(); | |
891 | } | |
892 | return desc_count; | |
893 | #endif | |
894 | } | |
895 | ||
896 | /* Called at mount-time, super-block is locked */ | |
617ba13b | 897 | unsigned long ext4_count_dirs (struct super_block * sb) |
ac27a0ec DK |
898 | { |
899 | unsigned long count = 0; | |
fd2d4291 | 900 | ext4_group_t i; |
ac27a0ec | 901 | |
617ba13b MC |
902 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
903 | struct ext4_group_desc *gdp = ext4_get_group_desc (sb, i, NULL); | |
ac27a0ec DK |
904 | if (!gdp) |
905 | continue; | |
906 | count += le16_to_cpu(gdp->bg_used_dirs_count); | |
907 | } | |
908 | return count; | |
909 | } | |
910 |