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)) { | |
fde4d95a | 77 | ext4_error(sb, __func__, "Checksum bad for group %lu", |
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 * | |
e29d1cde | 100 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 101 | { |
617ba13b | 102 | struct ext4_group_desc *desc; |
ac27a0ec | 103 | struct buffer_head *bh = NULL; |
e29d1cde | 104 | ext4_fsblk_t bitmap_blk; |
ac27a0ec | 105 | |
617ba13b | 106 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 107 | if (!desc) |
e29d1cde ES |
108 | return NULL; |
109 | bitmap_blk = ext4_inode_bitmap(sb, desc); | |
110 | bh = sb_getblk(sb, bitmap_blk); | |
111 | if (unlikely(!bh)) { | |
112 | ext4_error(sb, __func__, | |
113 | "Cannot read inode bitmap - " | |
114 | "block_group = %lu, inode_bitmap = %llu", | |
115 | block_group, bitmap_blk); | |
116 | return NULL; | |
117 | } | |
c806e68f FB |
118 | if (buffer_uptodate(bh) && |
119 | !(desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
e29d1cde ES |
120 | return bh; |
121 | ||
c806e68f | 122 | lock_buffer(bh); |
b5f10eed | 123 | spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group)); |
717d50e4 | 124 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
e29d1cde ES |
125 | ext4_init_inode_bitmap(sb, bh, block_group, desc); |
126 | set_buffer_uptodate(bh); | |
127 | unlock_buffer(bh); | |
b5f10eed | 128 | spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group)); |
e29d1cde | 129 | return bh; |
717d50e4 | 130 | } |
b5f10eed | 131 | spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group)); |
e29d1cde ES |
132 | if (bh_submit_read(bh) < 0) { |
133 | put_bh(bh); | |
134 | ext4_error(sb, __func__, | |
ac27a0ec | 135 | "Cannot read inode bitmap - " |
bd81d8ee | 136 | "block_group = %lu, inode_bitmap = %llu", |
e29d1cde ES |
137 | block_group, bitmap_blk); |
138 | return NULL; | |
139 | } | |
ac27a0ec DK |
140 | return bh; |
141 | } | |
142 | ||
143 | /* | |
144 | * NOTE! When we get the inode, we're the only people | |
145 | * that have access to it, and as such there are no | |
146 | * race conditions we have to worry about. The inode | |
147 | * is not on the hash-lists, and it cannot be reached | |
148 | * through the filesystem because the directory entry | |
149 | * has been deleted earlier. | |
150 | * | |
151 | * HOWEVER: we must make sure that we get no aliases, | |
152 | * which means that we have to call "clear_inode()" | |
153 | * _before_ we mark the inode not in use in the inode | |
154 | * bitmaps. Otherwise a newly created file might use | |
155 | * the same inode number (not actually the same pointer | |
156 | * though), and then we'd have two inodes sharing the | |
157 | * same inode number and space on the harddisk. | |
158 | */ | |
af5bc92d | 159 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 160 | { |
af5bc92d | 161 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
162 | int is_directory; |
163 | unsigned long ino; | |
164 | struct buffer_head *bitmap_bh = NULL; | |
165 | struct buffer_head *bh2; | |
fd2d4291 | 166 | ext4_group_t block_group; |
ac27a0ec | 167 | unsigned long bit; |
af5bc92d TT |
168 | struct ext4_group_desc *gdp; |
169 | struct ext4_super_block *es; | |
617ba13b | 170 | struct ext4_sb_info *sbi; |
ac27a0ec | 171 | int fatal = 0, err; |
772cb7c8 | 172 | ext4_group_t flex_group; |
ac27a0ec DK |
173 | |
174 | if (atomic_read(&inode->i_count) > 1) { | |
4776004f TT |
175 | printk(KERN_ERR "ext4_free_inode: inode has count=%d\n", |
176 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
177 | return; |
178 | } | |
179 | if (inode->i_nlink) { | |
4776004f TT |
180 | printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n", |
181 | inode->i_nlink); | |
ac27a0ec DK |
182 | return; |
183 | } | |
184 | if (!sb) { | |
4776004f TT |
185 | printk(KERN_ERR "ext4_free_inode: inode on " |
186 | "nonexistent device\n"); | |
ac27a0ec DK |
187 | return; |
188 | } | |
617ba13b | 189 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
190 | |
191 | ino = inode->i_ino; | |
af5bc92d | 192 | ext4_debug("freeing inode %lu\n", ino); |
ac27a0ec DK |
193 | |
194 | /* | |
195 | * Note: we must free any quota before locking the superblock, | |
196 | * as writing the quota to disk may need the lock as well. | |
197 | */ | |
198 | DQUOT_INIT(inode); | |
617ba13b | 199 | ext4_xattr_delete_inode(handle, inode); |
ac27a0ec DK |
200 | DQUOT_FREE_INODE(inode); |
201 | DQUOT_DROP(inode); | |
202 | ||
203 | is_directory = S_ISDIR(inode->i_mode); | |
204 | ||
205 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
af5bc92d | 206 | clear_inode(inode); |
ac27a0ec | 207 | |
617ba13b MC |
208 | es = EXT4_SB(sb)->s_es; |
209 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
af5bc92d TT |
210 | ext4_error(sb, "ext4_free_inode", |
211 | "reserved or nonexistent inode %lu", ino); | |
ac27a0ec DK |
212 | goto error_return; |
213 | } | |
617ba13b MC |
214 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
215 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 216 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec DK |
217 | if (!bitmap_bh) |
218 | goto error_return; | |
219 | ||
220 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 221 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
222 | if (fatal) |
223 | goto error_return; | |
224 | ||
225 | /* Ok, now we can actually update the inode bitmaps.. */ | |
617ba13b | 226 | if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group), |
ac27a0ec | 227 | bit, bitmap_bh->b_data)) |
af5bc92d TT |
228 | ext4_error(sb, "ext4_free_inode", |
229 | "bit already cleared for inode %lu", ino); | |
ac27a0ec | 230 | else { |
af5bc92d | 231 | gdp = ext4_get_group_desc(sb, block_group, &bh2); |
ac27a0ec DK |
232 | |
233 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 234 | fatal = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec DK |
235 | if (fatal) goto error_return; |
236 | ||
237 | if (gdp) { | |
238 | spin_lock(sb_bgl_lock(sbi, block_group)); | |
e8546d06 | 239 | le16_add_cpu(&gdp->bg_free_inodes_count, 1); |
ac27a0ec | 240 | if (is_directory) |
e8546d06 | 241 | le16_add_cpu(&gdp->bg_used_dirs_count, -1); |
717d50e4 AD |
242 | gdp->bg_checksum = ext4_group_desc_csum(sbi, |
243 | block_group, gdp); | |
ac27a0ec DK |
244 | spin_unlock(sb_bgl_lock(sbi, block_group)); |
245 | percpu_counter_inc(&sbi->s_freeinodes_counter); | |
246 | if (is_directory) | |
247 | percpu_counter_dec(&sbi->s_dirs_counter); | |
248 | ||
772cb7c8 JS |
249 | if (sbi->s_log_groups_per_flex) { |
250 | flex_group = ext4_flex_group(sbi, block_group); | |
251 | spin_lock(sb_bgl_lock(sbi, flex_group)); | |
252 | sbi->s_flex_groups[flex_group].free_inodes++; | |
253 | spin_unlock(sb_bgl_lock(sbi, flex_group)); | |
254 | } | |
ac27a0ec | 255 | } |
0390131b FM |
256 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
257 | err = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
ac27a0ec DK |
258 | if (!fatal) fatal = err; |
259 | } | |
0390131b FM |
260 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); |
261 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
ac27a0ec DK |
262 | if (!fatal) |
263 | fatal = err; | |
264 | sb->s_dirt = 1; | |
265 | error_return: | |
266 | brelse(bitmap_bh); | |
617ba13b | 267 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
268 | } |
269 | ||
270 | /* | |
271 | * There are two policies for allocating an inode. If the new inode is | |
272 | * a directory, then a forward search is made for a block group with both | |
273 | * free space and a low directory-to-inode ratio; if that fails, then of | |
274 | * the groups with above-average free space, that group with the fewest | |
275 | * directories already is chosen. | |
276 | * | |
277 | * For other inodes, search forward from the parent directory\'s block | |
278 | * group to find a free inode. | |
279 | */ | |
2aa9fc4c AM |
280 | static int find_group_dir(struct super_block *sb, struct inode *parent, |
281 | ext4_group_t *best_group) | |
ac27a0ec | 282 | { |
fd2d4291 | 283 | ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; |
ac27a0ec | 284 | unsigned int freei, avefreei; |
617ba13b | 285 | struct ext4_group_desc *desc, *best_desc = NULL; |
2aa9fc4c AM |
286 | ext4_group_t group; |
287 | int ret = -1; | |
ac27a0ec | 288 | |
617ba13b | 289 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
290 | avefreei = freei / ngroups; |
291 | ||
292 | for (group = 0; group < ngroups; group++) { | |
af5bc92d | 293 | desc = ext4_get_group_desc(sb, group, NULL); |
ac27a0ec DK |
294 | if (!desc || !desc->bg_free_inodes_count) |
295 | continue; | |
296 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
297 | continue; | |
298 | if (!best_desc || | |
299 | (le16_to_cpu(desc->bg_free_blocks_count) > | |
300 | le16_to_cpu(best_desc->bg_free_blocks_count))) { | |
2aa9fc4c | 301 | *best_group = group; |
ac27a0ec | 302 | best_desc = desc; |
2aa9fc4c | 303 | ret = 0; |
ac27a0ec DK |
304 | } |
305 | } | |
2aa9fc4c | 306 | return ret; |
ac27a0ec DK |
307 | } |
308 | ||
772cb7c8 JS |
309 | #define free_block_ratio 10 |
310 | ||
311 | static int find_group_flex(struct super_block *sb, struct inode *parent, | |
312 | ext4_group_t *best_group) | |
313 | { | |
314 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
315 | struct ext4_group_desc *desc; | |
316 | struct buffer_head *bh; | |
317 | struct flex_groups *flex_group = sbi->s_flex_groups; | |
318 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; | |
319 | ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group); | |
320 | ext4_group_t ngroups = sbi->s_groups_count; | |
321 | int flex_size = ext4_flex_bg_size(sbi); | |
322 | ext4_group_t best_flex = parent_fbg_group; | |
323 | int blocks_per_flex = sbi->s_blocks_per_group * flex_size; | |
324 | int flexbg_free_blocks; | |
325 | int flex_freeb_ratio; | |
326 | ext4_group_t n_fbg_groups; | |
327 | ext4_group_t i; | |
328 | ||
329 | n_fbg_groups = (sbi->s_groups_count + flex_size - 1) >> | |
330 | sbi->s_log_groups_per_flex; | |
331 | ||
332 | find_close_to_parent: | |
333 | flexbg_free_blocks = flex_group[best_flex].free_blocks; | |
334 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; | |
335 | if (flex_group[best_flex].free_inodes && | |
336 | flex_freeb_ratio > free_block_ratio) | |
337 | goto found_flexbg; | |
338 | ||
339 | if (best_flex && best_flex == parent_fbg_group) { | |
340 | best_flex--; | |
341 | goto find_close_to_parent; | |
342 | } | |
343 | ||
344 | for (i = 0; i < n_fbg_groups; i++) { | |
345 | if (i == parent_fbg_group || i == parent_fbg_group - 1) | |
346 | continue; | |
347 | ||
348 | flexbg_free_blocks = flex_group[i].free_blocks; | |
349 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; | |
350 | ||
351 | if (flex_freeb_ratio > free_block_ratio && | |
352 | flex_group[i].free_inodes) { | |
353 | best_flex = i; | |
354 | goto found_flexbg; | |
355 | } | |
356 | ||
c001077f | 357 | if (flex_group[best_flex].free_inodes == 0 || |
772cb7c8 JS |
358 | (flex_group[i].free_blocks > |
359 | flex_group[best_flex].free_blocks && | |
360 | flex_group[i].free_inodes)) | |
361 | best_flex = i; | |
362 | } | |
363 | ||
364 | if (!flex_group[best_flex].free_inodes || | |
365 | !flex_group[best_flex].free_blocks) | |
366 | return -1; | |
367 | ||
368 | found_flexbg: | |
369 | for (i = best_flex * flex_size; i < ngroups && | |
370 | i < (best_flex + 1) * flex_size; i++) { | |
371 | desc = ext4_get_group_desc(sb, i, &bh); | |
372 | if (le16_to_cpu(desc->bg_free_inodes_count)) { | |
373 | *best_group = i; | |
374 | goto out; | |
375 | } | |
376 | } | |
377 | ||
378 | return -1; | |
379 | out: | |
380 | return 0; | |
381 | } | |
382 | ||
ac27a0ec DK |
383 | /* |
384 | * Orlov's allocator for directories. | |
385 | * | |
386 | * We always try to spread first-level directories. | |
387 | * | |
388 | * If there are blockgroups with both free inodes and free blocks counts | |
389 | * not worse than average we return one with smallest directory count. | |
390 | * Otherwise we simply return a random group. | |
391 | * | |
392 | * For the rest rules look so: | |
393 | * | |
394 | * It's OK to put directory into a group unless | |
395 | * it has too many directories already (max_dirs) or | |
396 | * it has too few free inodes left (min_inodes) or | |
397 | * it has too few free blocks left (min_blocks) or | |
398 | * it's already running too large debt (max_debt). | |
1cc8dcf5 | 399 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
400 | * conditions we search cyclically through the rest. If none |
401 | * of the groups look good we just look for a group with more | |
402 | * free inodes than average (starting at parent's group). | |
403 | * | |
404 | * Debt is incremented each time we allocate a directory and decremented | |
405 | * when we allocate an inode, within 0--255. | |
406 | */ | |
407 | ||
408 | #define INODE_COST 64 | |
409 | #define BLOCK_COST 256 | |
410 | ||
2aa9fc4c AM |
411 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
412 | ext4_group_t *group) | |
ac27a0ec | 413 | { |
fd2d4291 | 414 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b MC |
415 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
416 | struct ext4_super_block *es = sbi->s_es; | |
fd2d4291 | 417 | ext4_group_t ngroups = sbi->s_groups_count; |
617ba13b | 418 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 419 | unsigned int freei, avefreei; |
617ba13b MC |
420 | ext4_fsblk_t freeb, avefreeb; |
421 | ext4_fsblk_t blocks_per_dir; | |
ac27a0ec DK |
422 | unsigned int ndirs; |
423 | int max_debt, max_dirs, min_inodes; | |
617ba13b | 424 | ext4_grpblk_t min_blocks; |
2aa9fc4c | 425 | ext4_group_t i; |
617ba13b | 426 | struct ext4_group_desc *desc; |
ac27a0ec DK |
427 | |
428 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
429 | avefreei = freei / ngroups; | |
430 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3a5b2ecd | 431 | avefreeb = freeb; |
f4e5bc24 | 432 | do_div(avefreeb, ngroups); |
ac27a0ec DK |
433 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
434 | ||
435 | if ((parent == sb->s_root->d_inode) || | |
617ba13b | 436 | (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) { |
ac27a0ec | 437 | int best_ndir = inodes_per_group; |
2aa9fc4c AM |
438 | ext4_group_t grp; |
439 | int ret = -1; | |
ac27a0ec | 440 | |
2aa9fc4c AM |
441 | get_random_bytes(&grp, sizeof(grp)); |
442 | parent_group = (unsigned)grp % ngroups; | |
ac27a0ec | 443 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
444 | grp = (parent_group + i) % ngroups; |
445 | desc = ext4_get_group_desc(sb, grp, NULL); | |
ac27a0ec DK |
446 | if (!desc || !desc->bg_free_inodes_count) |
447 | continue; | |
448 | if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir) | |
449 | continue; | |
450 | if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei) | |
451 | continue; | |
452 | if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb) | |
453 | continue; | |
2aa9fc4c AM |
454 | *group = grp; |
455 | ret = 0; | |
ac27a0ec DK |
456 | best_ndir = le16_to_cpu(desc->bg_used_dirs_count); |
457 | } | |
2aa9fc4c AM |
458 | if (ret == 0) |
459 | return ret; | |
ac27a0ec DK |
460 | goto fallback; |
461 | } | |
462 | ||
bd81d8ee | 463 | blocks_per_dir = ext4_blocks_count(es) - freeb; |
f4e5bc24 | 464 | do_div(blocks_per_dir, ndirs); |
ac27a0ec DK |
465 | |
466 | max_dirs = ndirs / ngroups + inodes_per_group / 16; | |
467 | min_inodes = avefreei - inodes_per_group / 4; | |
617ba13b | 468 | min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4; |
ac27a0ec | 469 | |
3a5b2ecd | 470 | max_debt = EXT4_BLOCKS_PER_GROUP(sb); |
f4e5bc24 | 471 | max_debt /= max_t(int, blocks_per_dir, BLOCK_COST); |
ac27a0ec DK |
472 | if (max_debt * INODE_COST > inodes_per_group) |
473 | max_debt = inodes_per_group / INODE_COST; | |
474 | if (max_debt > 255) | |
475 | max_debt = 255; | |
476 | if (max_debt == 0) | |
477 | max_debt = 1; | |
478 | ||
479 | for (i = 0; i < ngroups; i++) { | |
2aa9fc4c AM |
480 | *group = (parent_group + i) % ngroups; |
481 | desc = ext4_get_group_desc(sb, *group, NULL); | |
ac27a0ec DK |
482 | if (!desc || !desc->bg_free_inodes_count) |
483 | continue; | |
484 | if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs) | |
485 | continue; | |
486 | if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes) | |
487 | continue; | |
488 | if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks) | |
489 | continue; | |
2aa9fc4c | 490 | return 0; |
ac27a0ec DK |
491 | } |
492 | ||
493 | fallback: | |
494 | for (i = 0; i < ngroups; i++) { | |
2aa9fc4c AM |
495 | *group = (parent_group + i) % ngroups; |
496 | desc = ext4_get_group_desc(sb, *group, NULL); | |
497 | if (desc && desc->bg_free_inodes_count && | |
498 | le16_to_cpu(desc->bg_free_inodes_count) >= avefreei) | |
499 | return 0; | |
ac27a0ec DK |
500 | } |
501 | ||
502 | if (avefreei) { | |
503 | /* | |
504 | * The free-inodes counter is approximate, and for really small | |
505 | * filesystems the above test can fail to find any blockgroups | |
506 | */ | |
507 | avefreei = 0; | |
508 | goto fallback; | |
509 | } | |
510 | ||
511 | return -1; | |
512 | } | |
513 | ||
2aa9fc4c AM |
514 | static int find_group_other(struct super_block *sb, struct inode *parent, |
515 | ext4_group_t *group) | |
ac27a0ec | 516 | { |
fd2d4291 AM |
517 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
518 | ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; | |
617ba13b | 519 | struct ext4_group_desc *desc; |
2aa9fc4c | 520 | ext4_group_t i; |
ac27a0ec DK |
521 | |
522 | /* | |
523 | * Try to place the inode in its parent directory | |
524 | */ | |
2aa9fc4c AM |
525 | *group = parent_group; |
526 | desc = ext4_get_group_desc(sb, *group, NULL); | |
ac27a0ec DK |
527 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
528 | le16_to_cpu(desc->bg_free_blocks_count)) | |
2aa9fc4c | 529 | return 0; |
ac27a0ec DK |
530 | |
531 | /* | |
532 | * We're going to place this inode in a different blockgroup from its | |
533 | * parent. We want to cause files in a common directory to all land in | |
534 | * the same blockgroup. But we want files which are in a different | |
535 | * directory which shares a blockgroup with our parent to land in a | |
536 | * different blockgroup. | |
537 | * | |
538 | * So add our directory's i_ino into the starting point for the hash. | |
539 | */ | |
2aa9fc4c | 540 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
541 | |
542 | /* | |
543 | * Use a quadratic hash to find a group with a free inode and some free | |
544 | * blocks. | |
545 | */ | |
546 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
547 | *group += i; |
548 | if (*group >= ngroups) | |
549 | *group -= ngroups; | |
550 | desc = ext4_get_group_desc(sb, *group, NULL); | |
ac27a0ec DK |
551 | if (desc && le16_to_cpu(desc->bg_free_inodes_count) && |
552 | le16_to_cpu(desc->bg_free_blocks_count)) | |
2aa9fc4c | 553 | return 0; |
ac27a0ec DK |
554 | } |
555 | ||
556 | /* | |
557 | * That failed: try linear search for a free inode, even if that group | |
558 | * has no free blocks. | |
559 | */ | |
2aa9fc4c | 560 | *group = parent_group; |
ac27a0ec | 561 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
562 | if (++*group >= ngroups) |
563 | *group = 0; | |
564 | desc = ext4_get_group_desc(sb, *group, NULL); | |
ac27a0ec | 565 | if (desc && le16_to_cpu(desc->bg_free_inodes_count)) |
2aa9fc4c | 566 | return 0; |
ac27a0ec DK |
567 | } |
568 | ||
569 | return -1; | |
570 | } | |
571 | ||
572 | /* | |
573 | * There are two policies for allocating an inode. If the new inode is | |
574 | * a directory, then a forward search is made for a block group with both | |
575 | * free space and a low directory-to-inode ratio; if that fails, then of | |
576 | * the groups with above-average free space, that group with the fewest | |
577 | * directories already is chosen. | |
578 | * | |
579 | * For other inodes, search forward from the parent directory's block | |
580 | * group to find a free inode. | |
581 | */ | |
af5bc92d | 582 | struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode) |
ac27a0ec DK |
583 | { |
584 | struct super_block *sb; | |
585 | struct buffer_head *bitmap_bh = NULL; | |
586 | struct buffer_head *bh2; | |
2aa9fc4c | 587 | ext4_group_t group = 0; |
ac27a0ec | 588 | unsigned long ino = 0; |
af5bc92d TT |
589 | struct inode *inode; |
590 | struct ext4_group_desc *gdp = NULL; | |
591 | struct ext4_super_block *es; | |
617ba13b MC |
592 | struct ext4_inode_info *ei; |
593 | struct ext4_sb_info *sbi; | |
2aa9fc4c | 594 | int ret2, err = 0; |
ac27a0ec | 595 | struct inode *ret; |
2aa9fc4c AM |
596 | ext4_group_t i; |
597 | int free = 0; | |
772cb7c8 | 598 | ext4_group_t flex_group; |
ac27a0ec DK |
599 | |
600 | /* Cannot create files in a deleted directory */ | |
601 | if (!dir || !dir->i_nlink) | |
602 | return ERR_PTR(-EPERM); | |
603 | ||
604 | sb = dir->i_sb; | |
605 | inode = new_inode(sb); | |
606 | if (!inode) | |
607 | return ERR_PTR(-ENOMEM); | |
617ba13b | 608 | ei = EXT4_I(inode); |
ac27a0ec | 609 | |
617ba13b | 610 | sbi = EXT4_SB(sb); |
ac27a0ec | 611 | es = sbi->s_es; |
772cb7c8 JS |
612 | |
613 | if (sbi->s_log_groups_per_flex) { | |
614 | ret2 = find_group_flex(sb, dir, &group); | |
615 | goto got_group; | |
616 | } | |
617 | ||
ac27a0ec | 618 | if (S_ISDIR(mode)) { |
af5bc92d | 619 | if (test_opt(sb, OLDALLOC)) |
2aa9fc4c | 620 | ret2 = find_group_dir(sb, dir, &group); |
ac27a0ec | 621 | else |
2aa9fc4c | 622 | ret2 = find_group_orlov(sb, dir, &group); |
ac27a0ec | 623 | } else |
2aa9fc4c | 624 | ret2 = find_group_other(sb, dir, &group); |
ac27a0ec | 625 | |
772cb7c8 | 626 | got_group: |
ac27a0ec | 627 | err = -ENOSPC; |
2aa9fc4c | 628 | if (ret2 == -1) |
ac27a0ec DK |
629 | goto out; |
630 | ||
631 | for (i = 0; i < sbi->s_groups_count; i++) { | |
632 | err = -EIO; | |
633 | ||
617ba13b | 634 | gdp = ext4_get_group_desc(sb, group, &bh2); |
ac27a0ec DK |
635 | if (!gdp) |
636 | goto fail; | |
637 | ||
638 | brelse(bitmap_bh); | |
e29d1cde | 639 | bitmap_bh = ext4_read_inode_bitmap(sb, group); |
ac27a0ec DK |
640 | if (!bitmap_bh) |
641 | goto fail; | |
642 | ||
643 | ino = 0; | |
644 | ||
645 | repeat_in_this_group: | |
617ba13b MC |
646 | ino = ext4_find_next_zero_bit((unsigned long *) |
647 | bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino); | |
648 | if (ino < EXT4_INODES_PER_GROUP(sb)) { | |
ac27a0ec DK |
649 | |
650 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 651 | err = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
652 | if (err) |
653 | goto fail; | |
654 | ||
617ba13b | 655 | if (!ext4_set_bit_atomic(sb_bgl_lock(sbi, group), |
ac27a0ec DK |
656 | ino, bitmap_bh->b_data)) { |
657 | /* we won it */ | |
658 | BUFFER_TRACE(bitmap_bh, | |
0390131b FM |
659 | "call ext4_handle_dirty_metadata"); |
660 | err = ext4_handle_dirty_metadata(handle, | |
661 | inode, | |
ac27a0ec DK |
662 | bitmap_bh); |
663 | if (err) | |
664 | goto fail; | |
665 | goto got; | |
666 | } | |
667 | /* we lost it */ | |
0390131b | 668 | ext4_handle_release_buffer(handle, bitmap_bh); |
ac27a0ec | 669 | |
617ba13b | 670 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
671 | goto repeat_in_this_group; |
672 | } | |
673 | ||
674 | /* | |
675 | * This case is possible in concurrent environment. It is very | |
676 | * rare. We cannot repeat the find_group_xxx() call because | |
677 | * that will simply return the same blockgroup, because the | |
678 | * group descriptor metadata has not yet been updated. | |
679 | * So we just go onto the next blockgroup. | |
680 | */ | |
681 | if (++group == sbi->s_groups_count) | |
682 | group = 0; | |
683 | } | |
684 | err = -ENOSPC; | |
685 | goto out; | |
686 | ||
687 | got: | |
717d50e4 AD |
688 | ino++; |
689 | if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || | |
690 | ino > EXT4_INODES_PER_GROUP(sb)) { | |
46e665e9 | 691 | ext4_error(sb, __func__, |
717d50e4 | 692 | "reserved inode or inode > inodes count - " |
fd2d4291 | 693 | "block_group = %lu, inode=%lu", group, |
717d50e4 | 694 | ino + group * EXT4_INODES_PER_GROUP(sb)); |
ac27a0ec DK |
695 | err = -EIO; |
696 | goto fail; | |
697 | } | |
698 | ||
699 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 700 | err = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec | 701 | if (err) goto fail; |
717d50e4 AD |
702 | |
703 | /* We may have to initialize the block bitmap if it isn't already */ | |
704 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && | |
705 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
574ca174 | 706 | struct buffer_head *block_bh = ext4_read_block_bitmap(sb, group); |
717d50e4 AD |
707 | |
708 | BUFFER_TRACE(block_bh, "get block bitmap access"); | |
709 | err = ext4_journal_get_write_access(handle, block_bh); | |
710 | if (err) { | |
711 | brelse(block_bh); | |
712 | goto fail; | |
713 | } | |
714 | ||
715 | free = 0; | |
716 | spin_lock(sb_bgl_lock(sbi, group)); | |
717 | /* recheck and clear flag under lock if we still need to */ | |
718 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
719 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); | |
720 | free = ext4_free_blocks_after_init(sb, group, gdp); | |
721 | gdp->bg_free_blocks_count = cpu_to_le16(free); | |
23712a9c FB |
722 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, |
723 | gdp); | |
717d50e4 AD |
724 | } |
725 | spin_unlock(sb_bgl_lock(sbi, group)); | |
726 | ||
727 | /* Don't need to dirty bitmap block if we didn't change it */ | |
728 | if (free) { | |
729 | BUFFER_TRACE(block_bh, "dirty block bitmap"); | |
0390131b FM |
730 | err = ext4_handle_dirty_metadata(handle, |
731 | NULL, block_bh); | |
717d50e4 AD |
732 | } |
733 | ||
734 | brelse(block_bh); | |
735 | if (err) | |
736 | goto fail; | |
737 | } | |
738 | ||
ac27a0ec | 739 | spin_lock(sb_bgl_lock(sbi, group)); |
717d50e4 AD |
740 | /* If we didn't allocate from within the initialized part of the inode |
741 | * table then we need to initialize up to this inode. */ | |
742 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { | |
743 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
744 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
745 | ||
746 | /* When marking the block group with | |
747 | * ~EXT4_BG_INODE_UNINIT we don't want to depend | |
b5f10eed | 748 | * on the value of bg_itable_unused even though |
717d50e4 AD |
749 | * mke2fs could have initialized the same for us. |
750 | * Instead we calculated the value below | |
751 | */ | |
752 | ||
753 | free = 0; | |
754 | } else { | |
755 | free = EXT4_INODES_PER_GROUP(sb) - | |
756 | le16_to_cpu(gdp->bg_itable_unused); | |
757 | } | |
758 | ||
759 | /* | |
760 | * Check the relative inode number against the last used | |
761 | * relative inode number in this group. if it is greater | |
762 | * we need to update the bg_itable_unused count | |
763 | * | |
764 | */ | |
765 | if (ino > free) | |
766 | gdp->bg_itable_unused = | |
767 | cpu_to_le16(EXT4_INODES_PER_GROUP(sb) - ino); | |
768 | } | |
769 | ||
e8546d06 | 770 | le16_add_cpu(&gdp->bg_free_inodes_count, -1); |
ac27a0ec | 771 | if (S_ISDIR(mode)) { |
e8546d06 | 772 | le16_add_cpu(&gdp->bg_used_dirs_count, 1); |
ac27a0ec | 773 | } |
717d50e4 | 774 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); |
ac27a0ec | 775 | spin_unlock(sb_bgl_lock(sbi, group)); |
0390131b FM |
776 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
777 | err = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
ac27a0ec DK |
778 | if (err) goto fail; |
779 | ||
780 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
781 | if (S_ISDIR(mode)) | |
782 | percpu_counter_inc(&sbi->s_dirs_counter); | |
783 | sb->s_dirt = 1; | |
784 | ||
772cb7c8 JS |
785 | if (sbi->s_log_groups_per_flex) { |
786 | flex_group = ext4_flex_group(sbi, group); | |
787 | spin_lock(sb_bgl_lock(sbi, flex_group)); | |
788 | sbi->s_flex_groups[flex_group].free_inodes--; | |
789 | spin_unlock(sb_bgl_lock(sbi, flex_group)); | |
790 | } | |
791 | ||
4c9c544e | 792 | inode->i_uid = current_fsuid(); |
af5bc92d | 793 | if (test_opt(sb, GRPID)) |
ac27a0ec DK |
794 | inode->i_gid = dir->i_gid; |
795 | else if (dir->i_mode & S_ISGID) { | |
796 | inode->i_gid = dir->i_gid; | |
797 | if (S_ISDIR(mode)) | |
798 | mode |= S_ISGID; | |
799 | } else | |
4c9c544e | 800 | inode->i_gid = current_fsgid(); |
ac27a0ec DK |
801 | inode->i_mode = mode; |
802 | ||
717d50e4 | 803 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
804 | /* This is the optimal IO size (for stat), not the fs block size */ |
805 | inode->i_blocks = 0; | |
ef7f3835 KS |
806 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
807 | ext4_current_time(inode); | |
ac27a0ec DK |
808 | |
809 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
810 | ei->i_dir_start_lookup = 0; | |
811 | ei->i_disksize = 0; | |
812 | ||
42bf0383 AK |
813 | /* |
814 | * Don't inherit extent flag from directory. We set extent flag on | |
815 | * newly created directory and file only if -o extent mount option is | |
816 | * specified | |
817 | */ | |
818 | ei->i_flags = EXT4_I(dir)->i_flags & ~(EXT4_INDEX_FL|EXT4_EXTENTS_FL); | |
ac27a0ec | 819 | if (S_ISLNK(mode)) |
617ba13b | 820 | ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL); |
ac27a0ec DK |
821 | /* dirsync only applies to directories */ |
822 | if (!S_ISDIR(mode)) | |
617ba13b | 823 | ei->i_flags &= ~EXT4_DIRSYNC_FL; |
ac27a0ec | 824 | ei->i_file_acl = 0; |
ac27a0ec | 825 | ei->i_dtime = 0; |
ac27a0ec DK |
826 | ei->i_block_group = group; |
827 | ||
617ba13b | 828 | ext4_set_inode_flags(inode); |
ac27a0ec | 829 | if (IS_DIRSYNC(inode)) |
0390131b | 830 | ext4_handle_sync(handle); |
6b38e842 AV |
831 | if (insert_inode_locked(inode) < 0) { |
832 | err = -EINVAL; | |
833 | goto fail_drop; | |
834 | } | |
ac27a0ec DK |
835 | spin_lock(&sbi->s_next_gen_lock); |
836 | inode->i_generation = sbi->s_next_generation++; | |
837 | spin_unlock(&sbi->s_next_gen_lock); | |
838 | ||
617ba13b | 839 | ei->i_state = EXT4_STATE_NEW; |
ef7f3835 KS |
840 | |
841 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
842 | |
843 | ret = inode; | |
af5bc92d | 844 | if (DQUOT_ALLOC_INODE(inode)) { |
ac27a0ec DK |
845 | err = -EDQUOT; |
846 | goto fail_drop; | |
847 | } | |
848 | ||
617ba13b | 849 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
850 | if (err) |
851 | goto fail_free_drop; | |
852 | ||
af5bc92d | 853 | err = ext4_init_security(handle, inode, dir); |
ac27a0ec DK |
854 | if (err) |
855 | goto fail_free_drop; | |
856 | ||
a86c6181 | 857 | if (test_opt(sb, EXTENTS)) { |
e4079a11 | 858 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 859 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
42bf0383 AK |
860 | EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL; |
861 | ext4_ext_tree_init(handle, inode); | |
42bf0383 | 862 | } |
a86c6181 | 863 | } |
ac27a0ec | 864 | |
8753e88f AK |
865 | err = ext4_mark_inode_dirty(handle, inode); |
866 | if (err) { | |
867 | ext4_std_error(sb, err); | |
868 | goto fail_free_drop; | |
869 | } | |
870 | ||
617ba13b | 871 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
ac27a0ec DK |
872 | goto really_out; |
873 | fail: | |
617ba13b | 874 | ext4_std_error(sb, err); |
ac27a0ec DK |
875 | out: |
876 | iput(inode); | |
877 | ret = ERR_PTR(err); | |
878 | really_out: | |
879 | brelse(bitmap_bh); | |
880 | return ret; | |
881 | ||
882 | fail_free_drop: | |
883 | DQUOT_FREE_INODE(inode); | |
884 | ||
885 | fail_drop: | |
886 | DQUOT_DROP(inode); | |
887 | inode->i_flags |= S_NOQUOTA; | |
888 | inode->i_nlink = 0; | |
6b38e842 | 889 | unlock_new_inode(inode); |
ac27a0ec DK |
890 | iput(inode); |
891 | brelse(bitmap_bh); | |
892 | return ERR_PTR(err); | |
893 | } | |
894 | ||
895 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 896 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 897 | { |
617ba13b | 898 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 899 | ext4_group_t block_group; |
ac27a0ec | 900 | int bit; |
1d1fe1ee | 901 | struct buffer_head *bitmap_bh; |
ac27a0ec | 902 | struct inode *inode = NULL; |
1d1fe1ee | 903 | long err = -EIO; |
ac27a0ec DK |
904 | |
905 | /* Error cases - e2fsck has already cleaned up for us */ | |
906 | if (ino > max_ino) { | |
46e665e9 | 907 | ext4_warning(sb, __func__, |
ac27a0ec | 908 | "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 909 | goto error; |
ac27a0ec DK |
910 | } |
911 | ||
617ba13b MC |
912 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
913 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 914 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 915 | if (!bitmap_bh) { |
46e665e9 | 916 | ext4_warning(sb, __func__, |
ac27a0ec | 917 | "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 918 | goto error; |
ac27a0ec DK |
919 | } |
920 | ||
921 | /* Having the inode bit set should be a 100% indicator that this | |
922 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
923 | * inodes that were being truncated, so we can't check i_nlink==0. | |
924 | */ | |
1d1fe1ee DH |
925 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
926 | goto bad_orphan; | |
927 | ||
928 | inode = ext4_iget(sb, ino); | |
929 | if (IS_ERR(inode)) | |
930 | goto iget_failed; | |
931 | ||
91ef4caf DG |
932 | /* |
933 | * If the orphans has i_nlinks > 0 then it should be able to be | |
934 | * truncated, otherwise it won't be removed from the orphan list | |
935 | * during processing and an infinite loop will result. | |
936 | */ | |
937 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
938 | goto bad_orphan; | |
939 | ||
1d1fe1ee DH |
940 | if (NEXT_ORPHAN(inode) > max_ino) |
941 | goto bad_orphan; | |
942 | brelse(bitmap_bh); | |
943 | return inode; | |
944 | ||
945 | iget_failed: | |
946 | err = PTR_ERR(inode); | |
947 | inode = NULL; | |
948 | bad_orphan: | |
46e665e9 | 949 | ext4_warning(sb, __func__, |
1d1fe1ee DH |
950 | "bad orphan inode %lu! e2fsck was run?", ino); |
951 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", | |
952 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
953 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
954 | printk(KERN_NOTICE "inode=%p\n", inode); | |
955 | if (inode) { | |
956 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
957 | is_bad_inode(inode)); | |
958 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
959 | NEXT_ORPHAN(inode)); | |
960 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
91ef4caf | 961 | printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink); |
ac27a0ec | 962 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 963 | if (inode->i_nlink == 0) |
ac27a0ec DK |
964 | inode->i_blocks = 0; |
965 | iput(inode); | |
ac27a0ec | 966 | } |
ac27a0ec | 967 | brelse(bitmap_bh); |
1d1fe1ee DH |
968 | error: |
969 | return ERR_PTR(err); | |
ac27a0ec DK |
970 | } |
971 | ||
af5bc92d | 972 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
973 | { |
974 | unsigned long desc_count; | |
617ba13b | 975 | struct ext4_group_desc *gdp; |
fd2d4291 | 976 | ext4_group_t i; |
617ba13b MC |
977 | #ifdef EXT4FS_DEBUG |
978 | struct ext4_super_block *es; | |
ac27a0ec DK |
979 | unsigned long bitmap_count, x; |
980 | struct buffer_head *bitmap_bh = NULL; | |
981 | ||
617ba13b | 982 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
983 | desc_count = 0; |
984 | bitmap_count = 0; | |
985 | gdp = NULL; | |
617ba13b | 986 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
af5bc92d | 987 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
988 | if (!gdp) |
989 | continue; | |
990 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
991 | brelse(bitmap_bh); | |
e29d1cde | 992 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
993 | if (!bitmap_bh) |
994 | continue; | |
995 | ||
617ba13b | 996 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
c549a95d | 997 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
ac27a0ec DK |
998 | i, le16_to_cpu(gdp->bg_free_inodes_count), x); |
999 | bitmap_count += x; | |
1000 | } | |
1001 | brelse(bitmap_bh); | |
4776004f TT |
1002 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1003 | "stored = %u, computed = %lu, %lu\n", | |
1004 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1005 | return desc_count; |
1006 | #else | |
1007 | desc_count = 0; | |
617ba13b | 1008 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
af5bc92d | 1009 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1010 | if (!gdp) |
1011 | continue; | |
1012 | desc_count += le16_to_cpu(gdp->bg_free_inodes_count); | |
1013 | cond_resched(); | |
1014 | } | |
1015 | return desc_count; | |
1016 | #endif | |
1017 | } | |
1018 | ||
1019 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1020 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1021 | { |
1022 | unsigned long count = 0; | |
fd2d4291 | 1023 | ext4_group_t i; |
ac27a0ec | 1024 | |
617ba13b | 1025 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
af5bc92d | 1026 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1027 | if (!gdp) |
1028 | continue; | |
1029 | count += le16_to_cpu(gdp->bg_used_dirs_count); | |
1030 | } | |
1031 | return count; | |
1032 | } |