]>
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)) { | |
a9df9a49 | 77 | ext4_error(sb, __func__, "Checksum bad for group %u", |
717d50e4 | 78 | block_group); |
560671a0 AK |
79 | ext4_free_blks_set(sb, gdp, 0); |
80 | ext4_free_inodes_set(sb, gdp, 0); | |
81 | ext4_itable_unused_set(sb, gdp, 0); | |
717d50e4 AD |
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 - " | |
a9df9a49 | 114 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
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); | |
b5f10eed | 127 | spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group)); |
3300beda | 128 | unlock_buffer(bh); |
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 - " |
a9df9a49 | 136 | "block_group = %u, 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; |
560671a0 | 171 | int fatal = 0, err, count; |
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)); | |
560671a0 AK |
239 | count = ext4_free_inodes_count(sb, gdp) + 1; |
240 | ext4_free_inodes_set(sb, gdp, count); | |
241 | if (is_directory) { | |
242 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
243 | ext4_used_dirs_set(sb, gdp, count); | |
244 | } | |
717d50e4 AD |
245 | gdp->bg_checksum = ext4_group_desc_csum(sbi, |
246 | block_group, gdp); | |
ac27a0ec DK |
247 | spin_unlock(sb_bgl_lock(sbi, block_group)); |
248 | percpu_counter_inc(&sbi->s_freeinodes_counter); | |
249 | if (is_directory) | |
250 | percpu_counter_dec(&sbi->s_dirs_counter); | |
251 | ||
772cb7c8 JS |
252 | if (sbi->s_log_groups_per_flex) { |
253 | flex_group = ext4_flex_group(sbi, block_group); | |
254 | spin_lock(sb_bgl_lock(sbi, flex_group)); | |
255 | sbi->s_flex_groups[flex_group].free_inodes++; | |
256 | spin_unlock(sb_bgl_lock(sbi, flex_group)); | |
257 | } | |
ac27a0ec | 258 | } |
0390131b FM |
259 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
260 | err = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
ac27a0ec DK |
261 | if (!fatal) fatal = err; |
262 | } | |
0390131b FM |
263 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); |
264 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
ac27a0ec DK |
265 | if (!fatal) |
266 | fatal = err; | |
267 | sb->s_dirt = 1; | |
268 | error_return: | |
269 | brelse(bitmap_bh); | |
617ba13b | 270 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
271 | } |
272 | ||
273 | /* | |
274 | * There are two policies for allocating an inode. If the new inode is | |
275 | * a directory, then a forward search is made for a block group with both | |
276 | * free space and a low directory-to-inode ratio; if that fails, then of | |
277 | * the groups with above-average free space, that group with the fewest | |
278 | * directories already is chosen. | |
279 | * | |
280 | * For other inodes, search forward from the parent directory\'s block | |
281 | * group to find a free inode. | |
282 | */ | |
2aa9fc4c AM |
283 | static int find_group_dir(struct super_block *sb, struct inode *parent, |
284 | ext4_group_t *best_group) | |
ac27a0ec | 285 | { |
fd2d4291 | 286 | ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; |
ac27a0ec | 287 | unsigned int freei, avefreei; |
617ba13b | 288 | struct ext4_group_desc *desc, *best_desc = NULL; |
2aa9fc4c AM |
289 | ext4_group_t group; |
290 | int ret = -1; | |
ac27a0ec | 291 | |
617ba13b | 292 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
293 | avefreei = freei / ngroups; |
294 | ||
295 | for (group = 0; group < ngroups; group++) { | |
af5bc92d | 296 | desc = ext4_get_group_desc(sb, group, NULL); |
560671a0 | 297 | if (!desc || !ext4_free_inodes_count(sb, desc)) |
ac27a0ec | 298 | continue; |
560671a0 | 299 | if (ext4_free_inodes_count(sb, desc) < avefreei) |
ac27a0ec DK |
300 | continue; |
301 | if (!best_desc || | |
560671a0 AK |
302 | (ext4_free_blks_count(sb, desc) > |
303 | ext4_free_blks_count(sb, best_desc))) { | |
2aa9fc4c | 304 | *best_group = group; |
ac27a0ec | 305 | best_desc = desc; |
2aa9fc4c | 306 | ret = 0; |
ac27a0ec DK |
307 | } |
308 | } | |
2aa9fc4c | 309 | return ret; |
ac27a0ec DK |
310 | } |
311 | ||
772cb7c8 JS |
312 | #define free_block_ratio 10 |
313 | ||
314 | static int find_group_flex(struct super_block *sb, struct inode *parent, | |
315 | ext4_group_t *best_group) | |
316 | { | |
317 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
318 | struct ext4_group_desc *desc; | |
319 | struct buffer_head *bh; | |
320 | struct flex_groups *flex_group = sbi->s_flex_groups; | |
321 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; | |
322 | ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group); | |
323 | ext4_group_t ngroups = sbi->s_groups_count; | |
324 | int flex_size = ext4_flex_bg_size(sbi); | |
325 | ext4_group_t best_flex = parent_fbg_group; | |
326 | int blocks_per_flex = sbi->s_blocks_per_group * flex_size; | |
327 | int flexbg_free_blocks; | |
328 | int flex_freeb_ratio; | |
329 | ext4_group_t n_fbg_groups; | |
330 | ext4_group_t i; | |
331 | ||
332 | n_fbg_groups = (sbi->s_groups_count + flex_size - 1) >> | |
333 | sbi->s_log_groups_per_flex; | |
334 | ||
335 | find_close_to_parent: | |
336 | flexbg_free_blocks = flex_group[best_flex].free_blocks; | |
337 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; | |
338 | if (flex_group[best_flex].free_inodes && | |
339 | flex_freeb_ratio > free_block_ratio) | |
340 | goto found_flexbg; | |
341 | ||
342 | if (best_flex && best_flex == parent_fbg_group) { | |
343 | best_flex--; | |
344 | goto find_close_to_parent; | |
345 | } | |
346 | ||
347 | for (i = 0; i < n_fbg_groups; i++) { | |
348 | if (i == parent_fbg_group || i == parent_fbg_group - 1) | |
349 | continue; | |
350 | ||
351 | flexbg_free_blocks = flex_group[i].free_blocks; | |
352 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; | |
353 | ||
354 | if (flex_freeb_ratio > free_block_ratio && | |
355 | flex_group[i].free_inodes) { | |
356 | best_flex = i; | |
357 | goto found_flexbg; | |
358 | } | |
359 | ||
c001077f | 360 | if (flex_group[best_flex].free_inodes == 0 || |
772cb7c8 JS |
361 | (flex_group[i].free_blocks > |
362 | flex_group[best_flex].free_blocks && | |
363 | flex_group[i].free_inodes)) | |
364 | best_flex = i; | |
365 | } | |
366 | ||
367 | if (!flex_group[best_flex].free_inodes || | |
368 | !flex_group[best_flex].free_blocks) | |
369 | return -1; | |
370 | ||
371 | found_flexbg: | |
372 | for (i = best_flex * flex_size; i < ngroups && | |
373 | i < (best_flex + 1) * flex_size; i++) { | |
374 | desc = ext4_get_group_desc(sb, i, &bh); | |
560671a0 | 375 | if (ext4_free_inodes_count(sb, desc)) { |
772cb7c8 JS |
376 | *best_group = i; |
377 | goto out; | |
378 | } | |
379 | } | |
380 | ||
381 | return -1; | |
382 | out: | |
383 | return 0; | |
384 | } | |
385 | ||
ac27a0ec DK |
386 | /* |
387 | * Orlov's allocator for directories. | |
388 | * | |
389 | * We always try to spread first-level directories. | |
390 | * | |
391 | * If there are blockgroups with both free inodes and free blocks counts | |
392 | * not worse than average we return one with smallest directory count. | |
393 | * Otherwise we simply return a random group. | |
394 | * | |
395 | * For the rest rules look so: | |
396 | * | |
397 | * It's OK to put directory into a group unless | |
398 | * it has too many directories already (max_dirs) or | |
399 | * it has too few free inodes left (min_inodes) or | |
400 | * it has too few free blocks left (min_blocks) or | |
401 | * it's already running too large debt (max_debt). | |
1cc8dcf5 | 402 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
403 | * conditions we search cyclically through the rest. If none |
404 | * of the groups look good we just look for a group with more | |
405 | * free inodes than average (starting at parent's group). | |
406 | * | |
407 | * Debt is incremented each time we allocate a directory and decremented | |
408 | * when we allocate an inode, within 0--255. | |
409 | */ | |
410 | ||
411 | #define INODE_COST 64 | |
412 | #define BLOCK_COST 256 | |
413 | ||
2aa9fc4c AM |
414 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
415 | ext4_group_t *group) | |
ac27a0ec | 416 | { |
fd2d4291 | 417 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b MC |
418 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
419 | struct ext4_super_block *es = sbi->s_es; | |
fd2d4291 | 420 | ext4_group_t ngroups = sbi->s_groups_count; |
617ba13b | 421 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 422 | unsigned int freei, avefreei; |
617ba13b MC |
423 | ext4_fsblk_t freeb, avefreeb; |
424 | ext4_fsblk_t blocks_per_dir; | |
ac27a0ec DK |
425 | unsigned int ndirs; |
426 | int max_debt, max_dirs, min_inodes; | |
617ba13b | 427 | ext4_grpblk_t min_blocks; |
2aa9fc4c | 428 | ext4_group_t i; |
617ba13b | 429 | struct ext4_group_desc *desc; |
ac27a0ec DK |
430 | |
431 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
432 | avefreei = freei / ngroups; | |
433 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3a5b2ecd | 434 | avefreeb = freeb; |
f4e5bc24 | 435 | do_div(avefreeb, ngroups); |
ac27a0ec DK |
436 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
437 | ||
438 | if ((parent == sb->s_root->d_inode) || | |
617ba13b | 439 | (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) { |
ac27a0ec | 440 | int best_ndir = inodes_per_group; |
2aa9fc4c AM |
441 | ext4_group_t grp; |
442 | int ret = -1; | |
ac27a0ec | 443 | |
2aa9fc4c AM |
444 | get_random_bytes(&grp, sizeof(grp)); |
445 | parent_group = (unsigned)grp % ngroups; | |
ac27a0ec | 446 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
447 | grp = (parent_group + i) % ngroups; |
448 | desc = ext4_get_group_desc(sb, grp, NULL); | |
560671a0 | 449 | if (!desc || !ext4_free_inodes_count(sb, desc)) |
ac27a0ec | 450 | continue; |
560671a0 | 451 | if (ext4_used_dirs_count(sb, desc) >= best_ndir) |
ac27a0ec | 452 | continue; |
560671a0 | 453 | if (ext4_free_inodes_count(sb, desc) < avefreei) |
ac27a0ec | 454 | continue; |
560671a0 | 455 | if (ext4_free_blks_count(sb, desc) < avefreeb) |
ac27a0ec | 456 | continue; |
2aa9fc4c AM |
457 | *group = grp; |
458 | ret = 0; | |
560671a0 | 459 | best_ndir = ext4_used_dirs_count(sb, desc); |
ac27a0ec | 460 | } |
2aa9fc4c AM |
461 | if (ret == 0) |
462 | return ret; | |
ac27a0ec DK |
463 | goto fallback; |
464 | } | |
465 | ||
bd81d8ee | 466 | blocks_per_dir = ext4_blocks_count(es) - freeb; |
f4e5bc24 | 467 | do_div(blocks_per_dir, ndirs); |
ac27a0ec DK |
468 | |
469 | max_dirs = ndirs / ngroups + inodes_per_group / 16; | |
470 | min_inodes = avefreei - inodes_per_group / 4; | |
617ba13b | 471 | min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4; |
ac27a0ec | 472 | |
3a5b2ecd | 473 | max_debt = EXT4_BLOCKS_PER_GROUP(sb); |
f4e5bc24 | 474 | max_debt /= max_t(int, blocks_per_dir, BLOCK_COST); |
ac27a0ec DK |
475 | if (max_debt * INODE_COST > inodes_per_group) |
476 | max_debt = inodes_per_group / INODE_COST; | |
477 | if (max_debt > 255) | |
478 | max_debt = 255; | |
479 | if (max_debt == 0) | |
480 | max_debt = 1; | |
481 | ||
482 | for (i = 0; i < ngroups; i++) { | |
2aa9fc4c AM |
483 | *group = (parent_group + i) % ngroups; |
484 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 485 | if (!desc || !ext4_free_inodes_count(sb, desc)) |
ac27a0ec | 486 | continue; |
560671a0 | 487 | if (ext4_used_dirs_count(sb, desc) >= max_dirs) |
ac27a0ec | 488 | continue; |
560671a0 | 489 | if (ext4_free_inodes_count(sb, desc) < min_inodes) |
ac27a0ec | 490 | continue; |
560671a0 | 491 | if (ext4_free_blks_count(sb, desc) < min_blocks) |
ac27a0ec | 492 | continue; |
2aa9fc4c | 493 | return 0; |
ac27a0ec DK |
494 | } |
495 | ||
496 | fallback: | |
497 | for (i = 0; i < ngroups; i++) { | |
2aa9fc4c AM |
498 | *group = (parent_group + i) % ngroups; |
499 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
500 | if (desc && ext4_free_inodes_count(sb, desc) && |
501 | ext4_free_inodes_count(sb, desc) >= avefreei) | |
2aa9fc4c | 502 | return 0; |
ac27a0ec DK |
503 | } |
504 | ||
505 | if (avefreei) { | |
506 | /* | |
507 | * The free-inodes counter is approximate, and for really small | |
508 | * filesystems the above test can fail to find any blockgroups | |
509 | */ | |
510 | avefreei = 0; | |
511 | goto fallback; | |
512 | } | |
513 | ||
514 | return -1; | |
515 | } | |
516 | ||
2aa9fc4c AM |
517 | static int find_group_other(struct super_block *sb, struct inode *parent, |
518 | ext4_group_t *group) | |
ac27a0ec | 519 | { |
fd2d4291 AM |
520 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
521 | ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count; | |
617ba13b | 522 | struct ext4_group_desc *desc; |
2aa9fc4c | 523 | ext4_group_t i; |
ac27a0ec DK |
524 | |
525 | /* | |
526 | * Try to place the inode in its parent directory | |
527 | */ | |
2aa9fc4c AM |
528 | *group = parent_group; |
529 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
530 | if (desc && ext4_free_inodes_count(sb, desc) && |
531 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 532 | return 0; |
ac27a0ec DK |
533 | |
534 | /* | |
535 | * We're going to place this inode in a different blockgroup from its | |
536 | * parent. We want to cause files in a common directory to all land in | |
537 | * the same blockgroup. But we want files which are in a different | |
538 | * directory which shares a blockgroup with our parent to land in a | |
539 | * different blockgroup. | |
540 | * | |
541 | * So add our directory's i_ino into the starting point for the hash. | |
542 | */ | |
2aa9fc4c | 543 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
544 | |
545 | /* | |
546 | * Use a quadratic hash to find a group with a free inode and some free | |
547 | * blocks. | |
548 | */ | |
549 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
550 | *group += i; |
551 | if (*group >= ngroups) | |
552 | *group -= ngroups; | |
553 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
554 | if (desc && ext4_free_inodes_count(sb, desc) && |
555 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 556 | return 0; |
ac27a0ec DK |
557 | } |
558 | ||
559 | /* | |
560 | * That failed: try linear search for a free inode, even if that group | |
561 | * has no free blocks. | |
562 | */ | |
2aa9fc4c | 563 | *group = parent_group; |
ac27a0ec | 564 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
565 | if (++*group >= ngroups) |
566 | *group = 0; | |
567 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 568 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 569 | return 0; |
ac27a0ec DK |
570 | } |
571 | ||
572 | return -1; | |
573 | } | |
574 | ||
39341867 AK |
575 | /* |
576 | * claim the inode from the inode bitmap. If the group | |
577 | * is uninit we need to take the groups's sb_bgl_lock | |
578 | * and clear the uninit flag. The inode bitmap update | |
579 | * and group desc uninit flag clear should be done | |
580 | * after holding sb_bgl_lock so that ext4_read_inode_bitmap | |
581 | * doesn't race with the ext4_claim_inode | |
582 | */ | |
583 | static int ext4_claim_inode(struct super_block *sb, | |
584 | struct buffer_head *inode_bitmap_bh, | |
585 | unsigned long ino, ext4_group_t group, int mode) | |
586 | { | |
587 | int free = 0, retval = 0, count; | |
588 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
589 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); | |
590 | ||
591 | spin_lock(sb_bgl_lock(sbi, group)); | |
592 | if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) { | |
593 | /* not a free inode */ | |
594 | retval = 1; | |
595 | goto err_ret; | |
596 | } | |
597 | ino++; | |
598 | if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || | |
599 | ino > EXT4_INODES_PER_GROUP(sb)) { | |
600 | spin_unlock(sb_bgl_lock(sbi, group)); | |
601 | ext4_error(sb, __func__, | |
602 | "reserved inode or inode > inodes count - " | |
603 | "block_group = %u, inode=%lu", group, | |
604 | ino + group * EXT4_INODES_PER_GROUP(sb)); | |
605 | return 1; | |
606 | } | |
607 | /* If we didn't allocate from within the initialized part of the inode | |
608 | * table then we need to initialize up to this inode. */ | |
609 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { | |
610 | ||
611 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
612 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
613 | /* When marking the block group with | |
614 | * ~EXT4_BG_INODE_UNINIT we don't want to depend | |
615 | * on the value of bg_itable_unused even though | |
616 | * mke2fs could have initialized the same for us. | |
617 | * Instead we calculated the value below | |
618 | */ | |
619 | ||
620 | free = 0; | |
621 | } else { | |
622 | free = EXT4_INODES_PER_GROUP(sb) - | |
623 | ext4_itable_unused_count(sb, gdp); | |
624 | } | |
625 | ||
626 | /* | |
627 | * Check the relative inode number against the last used | |
628 | * relative inode number in this group. if it is greater | |
629 | * we need to update the bg_itable_unused count | |
630 | * | |
631 | */ | |
632 | if (ino > free) | |
633 | ext4_itable_unused_set(sb, gdp, | |
634 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
635 | } | |
636 | count = ext4_free_inodes_count(sb, gdp) - 1; | |
637 | ext4_free_inodes_set(sb, gdp, count); | |
638 | if (S_ISDIR(mode)) { | |
639 | count = ext4_used_dirs_count(sb, gdp) + 1; | |
640 | ext4_used_dirs_set(sb, gdp, count); | |
641 | } | |
642 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); | |
643 | err_ret: | |
644 | spin_unlock(sb_bgl_lock(sbi, group)); | |
645 | return retval; | |
646 | } | |
647 | ||
ac27a0ec DK |
648 | /* |
649 | * There are two policies for allocating an inode. If the new inode is | |
650 | * a directory, then a forward search is made for a block group with both | |
651 | * free space and a low directory-to-inode ratio; if that fails, then of | |
652 | * the groups with above-average free space, that group with the fewest | |
653 | * directories already is chosen. | |
654 | * | |
655 | * For other inodes, search forward from the parent directory's block | |
656 | * group to find a free inode. | |
657 | */ | |
af5bc92d | 658 | struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode) |
ac27a0ec DK |
659 | { |
660 | struct super_block *sb; | |
3300beda AK |
661 | struct buffer_head *inode_bitmap_bh = NULL; |
662 | struct buffer_head *group_desc_bh; | |
2aa9fc4c | 663 | ext4_group_t group = 0; |
ac27a0ec | 664 | unsigned long ino = 0; |
af5bc92d TT |
665 | struct inode *inode; |
666 | struct ext4_group_desc *gdp = NULL; | |
667 | struct ext4_super_block *es; | |
617ba13b MC |
668 | struct ext4_inode_info *ei; |
669 | struct ext4_sb_info *sbi; | |
39341867 | 670 | int ret2, err = 0; |
ac27a0ec | 671 | struct inode *ret; |
2aa9fc4c AM |
672 | ext4_group_t i; |
673 | int free = 0; | |
772cb7c8 | 674 | ext4_group_t flex_group; |
ac27a0ec DK |
675 | |
676 | /* Cannot create files in a deleted directory */ | |
677 | if (!dir || !dir->i_nlink) | |
678 | return ERR_PTR(-EPERM); | |
679 | ||
680 | sb = dir->i_sb; | |
681 | inode = new_inode(sb); | |
682 | if (!inode) | |
683 | return ERR_PTR(-ENOMEM); | |
617ba13b | 684 | ei = EXT4_I(inode); |
ac27a0ec | 685 | |
617ba13b | 686 | sbi = EXT4_SB(sb); |
ac27a0ec | 687 | es = sbi->s_es; |
772cb7c8 JS |
688 | |
689 | if (sbi->s_log_groups_per_flex) { | |
690 | ret2 = find_group_flex(sb, dir, &group); | |
691 | goto got_group; | |
692 | } | |
693 | ||
ac27a0ec | 694 | if (S_ISDIR(mode)) { |
af5bc92d | 695 | if (test_opt(sb, OLDALLOC)) |
2aa9fc4c | 696 | ret2 = find_group_dir(sb, dir, &group); |
ac27a0ec | 697 | else |
2aa9fc4c | 698 | ret2 = find_group_orlov(sb, dir, &group); |
ac27a0ec | 699 | } else |
2aa9fc4c | 700 | ret2 = find_group_other(sb, dir, &group); |
ac27a0ec | 701 | |
772cb7c8 | 702 | got_group: |
ac27a0ec | 703 | err = -ENOSPC; |
2aa9fc4c | 704 | if (ret2 == -1) |
ac27a0ec DK |
705 | goto out; |
706 | ||
707 | for (i = 0; i < sbi->s_groups_count; i++) { | |
708 | err = -EIO; | |
709 | ||
3300beda | 710 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec DK |
711 | if (!gdp) |
712 | goto fail; | |
713 | ||
3300beda AK |
714 | brelse(inode_bitmap_bh); |
715 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
716 | if (!inode_bitmap_bh) | |
ac27a0ec DK |
717 | goto fail; |
718 | ||
719 | ino = 0; | |
720 | ||
721 | repeat_in_this_group: | |
617ba13b | 722 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
723 | inode_bitmap_bh->b_data, |
724 | EXT4_INODES_PER_GROUP(sb), ino); | |
725 | ||
617ba13b | 726 | if (ino < EXT4_INODES_PER_GROUP(sb)) { |
ac27a0ec | 727 | |
3300beda AK |
728 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
729 | err = ext4_journal_get_write_access(handle, | |
730 | inode_bitmap_bh); | |
ac27a0ec DK |
731 | if (err) |
732 | goto fail; | |
733 | ||
39341867 AK |
734 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
735 | err = ext4_journal_get_write_access(handle, | |
736 | group_desc_bh); | |
737 | if (err) | |
738 | goto fail; | |
739 | if (!ext4_claim_inode(sb, inode_bitmap_bh, | |
740 | ino, group, mode)) { | |
ac27a0ec | 741 | /* we won it */ |
3300beda | 742 | BUFFER_TRACE(inode_bitmap_bh, |
0390131b FM |
743 | "call ext4_handle_dirty_metadata"); |
744 | err = ext4_handle_dirty_metadata(handle, | |
3300beda AK |
745 | inode, |
746 | inode_bitmap_bh); | |
ac27a0ec DK |
747 | if (err) |
748 | goto fail; | |
39341867 AK |
749 | /* zero bit is inode number 1*/ |
750 | ino++; | |
ac27a0ec DK |
751 | goto got; |
752 | } | |
753 | /* we lost it */ | |
3300beda | 754 | ext4_handle_release_buffer(handle, inode_bitmap_bh); |
39341867 | 755 | ext4_handle_release_buffer(handle, group_desc_bh); |
ac27a0ec | 756 | |
617ba13b | 757 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
758 | goto repeat_in_this_group; |
759 | } | |
760 | ||
761 | /* | |
762 | * This case is possible in concurrent environment. It is very | |
763 | * rare. We cannot repeat the find_group_xxx() call because | |
764 | * that will simply return the same blockgroup, because the | |
765 | * group descriptor metadata has not yet been updated. | |
766 | * So we just go onto the next blockgroup. | |
767 | */ | |
768 | if (++group == sbi->s_groups_count) | |
769 | group = 0; | |
770 | } | |
771 | err = -ENOSPC; | |
772 | goto out; | |
773 | ||
774 | got: | |
717d50e4 AD |
775 | /* We may have to initialize the block bitmap if it isn't already */ |
776 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && | |
777 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
3300beda | 778 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 779 | |
3300beda AK |
780 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
781 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); | |
782 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 783 | if (err) { |
3300beda | 784 | brelse(block_bitmap_bh); |
717d50e4 AD |
785 | goto fail; |
786 | } | |
787 | ||
788 | free = 0; | |
789 | spin_lock(sb_bgl_lock(sbi, group)); | |
790 | /* recheck and clear flag under lock if we still need to */ | |
791 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
717d50e4 | 792 | free = ext4_free_blocks_after_init(sb, group, gdp); |
3300beda | 793 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
560671a0 | 794 | ext4_free_blks_set(sb, gdp, free); |
23712a9c FB |
795 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, |
796 | gdp); | |
717d50e4 AD |
797 | } |
798 | spin_unlock(sb_bgl_lock(sbi, group)); | |
799 | ||
800 | /* Don't need to dirty bitmap block if we didn't change it */ | |
801 | if (free) { | |
3300beda | 802 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
0390131b | 803 | err = ext4_handle_dirty_metadata(handle, |
3300beda | 804 | NULL, block_bitmap_bh); |
717d50e4 AD |
805 | } |
806 | ||
3300beda | 807 | brelse(block_bitmap_bh); |
717d50e4 AD |
808 | if (err) |
809 | goto fail; | |
810 | } | |
3300beda AK |
811 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
812 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
39341867 AK |
813 | if (err) |
814 | goto fail; | |
ac27a0ec DK |
815 | |
816 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
817 | if (S_ISDIR(mode)) | |
818 | percpu_counter_inc(&sbi->s_dirs_counter); | |
819 | sb->s_dirt = 1; | |
820 | ||
772cb7c8 JS |
821 | if (sbi->s_log_groups_per_flex) { |
822 | flex_group = ext4_flex_group(sbi, group); | |
823 | spin_lock(sb_bgl_lock(sbi, flex_group)); | |
824 | sbi->s_flex_groups[flex_group].free_inodes--; | |
825 | spin_unlock(sb_bgl_lock(sbi, flex_group)); | |
826 | } | |
827 | ||
4c9c544e | 828 | inode->i_uid = current_fsuid(); |
af5bc92d | 829 | if (test_opt(sb, GRPID)) |
ac27a0ec DK |
830 | inode->i_gid = dir->i_gid; |
831 | else if (dir->i_mode & S_ISGID) { | |
832 | inode->i_gid = dir->i_gid; | |
833 | if (S_ISDIR(mode)) | |
834 | mode |= S_ISGID; | |
835 | } else | |
4c9c544e | 836 | inode->i_gid = current_fsgid(); |
ac27a0ec DK |
837 | inode->i_mode = mode; |
838 | ||
717d50e4 | 839 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
840 | /* This is the optimal IO size (for stat), not the fs block size */ |
841 | inode->i_blocks = 0; | |
ef7f3835 KS |
842 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
843 | ext4_current_time(inode); | |
ac27a0ec DK |
844 | |
845 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
846 | ei->i_dir_start_lookup = 0; | |
847 | ei->i_disksize = 0; | |
848 | ||
42bf0383 AK |
849 | /* |
850 | * Don't inherit extent flag from directory. We set extent flag on | |
851 | * newly created directory and file only if -o extent mount option is | |
852 | * specified | |
853 | */ | |
854 | ei->i_flags = EXT4_I(dir)->i_flags & ~(EXT4_INDEX_FL|EXT4_EXTENTS_FL); | |
ac27a0ec | 855 | if (S_ISLNK(mode)) |
617ba13b | 856 | ei->i_flags &= ~(EXT4_IMMUTABLE_FL|EXT4_APPEND_FL); |
ac27a0ec DK |
857 | /* dirsync only applies to directories */ |
858 | if (!S_ISDIR(mode)) | |
617ba13b | 859 | ei->i_flags &= ~EXT4_DIRSYNC_FL; |
ac27a0ec | 860 | ei->i_file_acl = 0; |
ac27a0ec | 861 | ei->i_dtime = 0; |
ac27a0ec DK |
862 | ei->i_block_group = group; |
863 | ||
617ba13b | 864 | ext4_set_inode_flags(inode); |
ac27a0ec | 865 | if (IS_DIRSYNC(inode)) |
0390131b | 866 | ext4_handle_sync(handle); |
6b38e842 AV |
867 | if (insert_inode_locked(inode) < 0) { |
868 | err = -EINVAL; | |
869 | goto fail_drop; | |
870 | } | |
ac27a0ec DK |
871 | spin_lock(&sbi->s_next_gen_lock); |
872 | inode->i_generation = sbi->s_next_generation++; | |
873 | spin_unlock(&sbi->s_next_gen_lock); | |
874 | ||
617ba13b | 875 | ei->i_state = EXT4_STATE_NEW; |
ef7f3835 KS |
876 | |
877 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
878 | |
879 | ret = inode; | |
af5bc92d | 880 | if (DQUOT_ALLOC_INODE(inode)) { |
ac27a0ec DK |
881 | err = -EDQUOT; |
882 | goto fail_drop; | |
883 | } | |
884 | ||
617ba13b | 885 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
886 | if (err) |
887 | goto fail_free_drop; | |
888 | ||
af5bc92d | 889 | err = ext4_init_security(handle, inode, dir); |
ac27a0ec DK |
890 | if (err) |
891 | goto fail_free_drop; | |
892 | ||
a86c6181 | 893 | if (test_opt(sb, EXTENTS)) { |
e4079a11 | 894 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 895 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
42bf0383 AK |
896 | EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL; |
897 | ext4_ext_tree_init(handle, inode); | |
42bf0383 | 898 | } |
a86c6181 | 899 | } |
ac27a0ec | 900 | |
8753e88f AK |
901 | err = ext4_mark_inode_dirty(handle, inode); |
902 | if (err) { | |
903 | ext4_std_error(sb, err); | |
904 | goto fail_free_drop; | |
905 | } | |
906 | ||
617ba13b | 907 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
ac27a0ec DK |
908 | goto really_out; |
909 | fail: | |
617ba13b | 910 | ext4_std_error(sb, err); |
ac27a0ec DK |
911 | out: |
912 | iput(inode); | |
913 | ret = ERR_PTR(err); | |
914 | really_out: | |
3300beda | 915 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
916 | return ret; |
917 | ||
918 | fail_free_drop: | |
919 | DQUOT_FREE_INODE(inode); | |
920 | ||
921 | fail_drop: | |
922 | DQUOT_DROP(inode); | |
923 | inode->i_flags |= S_NOQUOTA; | |
924 | inode->i_nlink = 0; | |
6b38e842 | 925 | unlock_new_inode(inode); |
ac27a0ec | 926 | iput(inode); |
3300beda | 927 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
928 | return ERR_PTR(err); |
929 | } | |
930 | ||
931 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 932 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 933 | { |
617ba13b | 934 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 935 | ext4_group_t block_group; |
ac27a0ec | 936 | int bit; |
1d1fe1ee | 937 | struct buffer_head *bitmap_bh; |
ac27a0ec | 938 | struct inode *inode = NULL; |
1d1fe1ee | 939 | long err = -EIO; |
ac27a0ec DK |
940 | |
941 | /* Error cases - e2fsck has already cleaned up for us */ | |
942 | if (ino > max_ino) { | |
46e665e9 | 943 | ext4_warning(sb, __func__, |
ac27a0ec | 944 | "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 945 | goto error; |
ac27a0ec DK |
946 | } |
947 | ||
617ba13b MC |
948 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
949 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 950 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 951 | if (!bitmap_bh) { |
46e665e9 | 952 | ext4_warning(sb, __func__, |
ac27a0ec | 953 | "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 954 | goto error; |
ac27a0ec DK |
955 | } |
956 | ||
957 | /* Having the inode bit set should be a 100% indicator that this | |
958 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
959 | * inodes that were being truncated, so we can't check i_nlink==0. | |
960 | */ | |
1d1fe1ee DH |
961 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
962 | goto bad_orphan; | |
963 | ||
964 | inode = ext4_iget(sb, ino); | |
965 | if (IS_ERR(inode)) | |
966 | goto iget_failed; | |
967 | ||
91ef4caf DG |
968 | /* |
969 | * If the orphans has i_nlinks > 0 then it should be able to be | |
970 | * truncated, otherwise it won't be removed from the orphan list | |
971 | * during processing and an infinite loop will result. | |
972 | */ | |
973 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
974 | goto bad_orphan; | |
975 | ||
1d1fe1ee DH |
976 | if (NEXT_ORPHAN(inode) > max_ino) |
977 | goto bad_orphan; | |
978 | brelse(bitmap_bh); | |
979 | return inode; | |
980 | ||
981 | iget_failed: | |
982 | err = PTR_ERR(inode); | |
983 | inode = NULL; | |
984 | bad_orphan: | |
46e665e9 | 985 | ext4_warning(sb, __func__, |
1d1fe1ee DH |
986 | "bad orphan inode %lu! e2fsck was run?", ino); |
987 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", | |
988 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
989 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
990 | printk(KERN_NOTICE "inode=%p\n", inode); | |
991 | if (inode) { | |
992 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
993 | is_bad_inode(inode)); | |
994 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
995 | NEXT_ORPHAN(inode)); | |
996 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
91ef4caf | 997 | printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink); |
ac27a0ec | 998 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 999 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1000 | inode->i_blocks = 0; |
1001 | iput(inode); | |
ac27a0ec | 1002 | } |
ac27a0ec | 1003 | brelse(bitmap_bh); |
1d1fe1ee DH |
1004 | error: |
1005 | return ERR_PTR(err); | |
ac27a0ec DK |
1006 | } |
1007 | ||
af5bc92d | 1008 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1009 | { |
1010 | unsigned long desc_count; | |
617ba13b | 1011 | struct ext4_group_desc *gdp; |
fd2d4291 | 1012 | ext4_group_t i; |
617ba13b MC |
1013 | #ifdef EXT4FS_DEBUG |
1014 | struct ext4_super_block *es; | |
ac27a0ec DK |
1015 | unsigned long bitmap_count, x; |
1016 | struct buffer_head *bitmap_bh = NULL; | |
1017 | ||
617ba13b | 1018 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1019 | desc_count = 0; |
1020 | bitmap_count = 0; | |
1021 | gdp = NULL; | |
617ba13b | 1022 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
af5bc92d | 1023 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1024 | if (!gdp) |
1025 | continue; | |
560671a0 | 1026 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1027 | brelse(bitmap_bh); |
e29d1cde | 1028 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
1029 | if (!bitmap_bh) |
1030 | continue; | |
1031 | ||
617ba13b | 1032 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
c549a95d | 1033 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
560671a0 | 1034 | i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1035 | bitmap_count += x; |
1036 | } | |
1037 | brelse(bitmap_bh); | |
4776004f TT |
1038 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1039 | "stored = %u, computed = %lu, %lu\n", | |
1040 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1041 | return desc_count; |
1042 | #else | |
1043 | desc_count = 0; | |
617ba13b | 1044 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
af5bc92d | 1045 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1046 | if (!gdp) |
1047 | continue; | |
560671a0 | 1048 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1049 | cond_resched(); |
1050 | } | |
1051 | return desc_count; | |
1052 | #endif | |
1053 | } | |
1054 | ||
1055 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1056 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1057 | { |
1058 | unsigned long count = 0; | |
fd2d4291 | 1059 | ext4_group_t i; |
ac27a0ec | 1060 | |
617ba13b | 1061 | for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) { |
af5bc92d | 1062 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1063 | if (!gdp) |
1064 | continue; | |
560671a0 | 1065 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1066 | } |
1067 | return count; | |
1068 | } |