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> |
9bffad1e | 26 | |
3dcf5451 CH |
27 | #include "ext4.h" |
28 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
29 | #include "xattr.h" |
30 | #include "acl.h" | |
31 | ||
9bffad1e TT |
32 | #include <trace/events/ext4.h> |
33 | ||
ac27a0ec DK |
34 | /* |
35 | * ialloc.c contains the inodes allocation and deallocation routines | |
36 | */ | |
37 | ||
38 | /* | |
39 | * The free inodes are managed by bitmaps. A file system contains several | |
40 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
41 | * block for inodes, N blocks for the inode table and data blocks. | |
42 | * | |
43 | * The file system contains group descriptors which are located after the | |
44 | * super block. Each descriptor contains the number of the bitmap block and | |
45 | * the free blocks count in the block. | |
46 | */ | |
47 | ||
717d50e4 AD |
48 | /* |
49 | * To avoid calling the atomic setbit hundreds or thousands of times, we only | |
50 | * need to use it within a single byte (to ensure we get endianness right). | |
51 | * We can use memset for the rest of the bitmap as there are no other users. | |
52 | */ | |
53 | void mark_bitmap_end(int start_bit, int end_bit, char *bitmap) | |
54 | { | |
55 | int i; | |
56 | ||
57 | if (start_bit >= end_bit) | |
58 | return; | |
59 | ||
60 | ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); | |
61 | for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) | |
62 | ext4_set_bit(i, bitmap); | |
63 | if (i < end_bit) | |
64 | memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); | |
65 | } | |
66 | ||
67 | /* Initializes an uninitialized inode bitmap */ | |
fd2d4291 AM |
68 | unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh, |
69 | ext4_group_t block_group, | |
717d50e4 AD |
70 | struct ext4_group_desc *gdp) |
71 | { | |
72 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
73 | ||
74 | J_ASSERT_BH(bh, buffer_locked(bh)); | |
75 | ||
76 | /* If checksum is bad mark all blocks and inodes use to prevent | |
77 | * allocation, essentially implementing a per-group read-only flag. */ | |
78 | if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) { | |
12062ddd | 79 | ext4_error(sb, "Checksum bad for group %u", block_group); |
560671a0 AK |
80 | ext4_free_blks_set(sb, gdp, 0); |
81 | ext4_free_inodes_set(sb, gdp, 0); | |
82 | ext4_itable_unused_set(sb, gdp, 0); | |
717d50e4 AD |
83 | memset(bh->b_data, 0xff, sb->s_blocksize); |
84 | return 0; | |
85 | } | |
86 | ||
87 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
648f5879 | 88 | mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
717d50e4 AD |
89 | bh->b_data); |
90 | ||
91 | return EXT4_INODES_PER_GROUP(sb); | |
92 | } | |
ac27a0ec DK |
93 | |
94 | /* | |
95 | * Read the inode allocation bitmap for a given block_group, reading | |
96 | * into the specified slot in the superblock's bitmap cache. | |
97 | * | |
98 | * Return buffer_head of bitmap on success or NULL. | |
99 | */ | |
100 | static struct buffer_head * | |
e29d1cde | 101 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 102 | { |
617ba13b | 103 | struct ext4_group_desc *desc; |
ac27a0ec | 104 | struct buffer_head *bh = NULL; |
e29d1cde | 105 | ext4_fsblk_t bitmap_blk; |
ac27a0ec | 106 | |
617ba13b | 107 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 108 | if (!desc) |
e29d1cde ES |
109 | return NULL; |
110 | bitmap_blk = ext4_inode_bitmap(sb, desc); | |
111 | bh = sb_getblk(sb, bitmap_blk); | |
112 | if (unlikely(!bh)) { | |
12062ddd | 113 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 114 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
115 | block_group, bitmap_blk); |
116 | return NULL; | |
117 | } | |
2ccb5fb9 | 118 | if (bitmap_uptodate(bh)) |
e29d1cde ES |
119 | return bh; |
120 | ||
c806e68f | 121 | lock_buffer(bh); |
2ccb5fb9 AK |
122 | if (bitmap_uptodate(bh)) { |
123 | unlock_buffer(bh); | |
124 | return bh; | |
125 | } | |
955ce5f5 | 126 | ext4_lock_group(sb, block_group); |
717d50e4 | 127 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
e29d1cde | 128 | ext4_init_inode_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 129 | set_bitmap_uptodate(bh); |
e29d1cde | 130 | set_buffer_uptodate(bh); |
955ce5f5 | 131 | ext4_unlock_group(sb, block_group); |
3300beda | 132 | unlock_buffer(bh); |
e29d1cde | 133 | return bh; |
717d50e4 | 134 | } |
955ce5f5 | 135 | ext4_unlock_group(sb, block_group); |
2ccb5fb9 AK |
136 | if (buffer_uptodate(bh)) { |
137 | /* | |
138 | * if not uninit if bh is uptodate, | |
139 | * bitmap is also uptodate | |
140 | */ | |
141 | set_bitmap_uptodate(bh); | |
142 | unlock_buffer(bh); | |
143 | return bh; | |
144 | } | |
145 | /* | |
146 | * submit the buffer_head for read. We can | |
147 | * safely mark the bitmap as uptodate now. | |
148 | * We do it here so the bitmap uptodate bit | |
149 | * get set with buffer lock held. | |
150 | */ | |
151 | set_bitmap_uptodate(bh); | |
e29d1cde ES |
152 | if (bh_submit_read(bh) < 0) { |
153 | put_bh(bh); | |
12062ddd | 154 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 155 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
156 | block_group, bitmap_blk); |
157 | return NULL; | |
158 | } | |
ac27a0ec DK |
159 | return bh; |
160 | } | |
161 | ||
162 | /* | |
163 | * NOTE! When we get the inode, we're the only people | |
164 | * that have access to it, and as such there are no | |
165 | * race conditions we have to worry about. The inode | |
166 | * is not on the hash-lists, and it cannot be reached | |
167 | * through the filesystem because the directory entry | |
168 | * has been deleted earlier. | |
169 | * | |
170 | * HOWEVER: we must make sure that we get no aliases, | |
171 | * which means that we have to call "clear_inode()" | |
172 | * _before_ we mark the inode not in use in the inode | |
173 | * bitmaps. Otherwise a newly created file might use | |
174 | * the same inode number (not actually the same pointer | |
175 | * though), and then we'd have two inodes sharing the | |
176 | * same inode number and space on the harddisk. | |
177 | */ | |
af5bc92d | 178 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 179 | { |
af5bc92d | 180 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
181 | int is_directory; |
182 | unsigned long ino; | |
183 | struct buffer_head *bitmap_bh = NULL; | |
184 | struct buffer_head *bh2; | |
fd2d4291 | 185 | ext4_group_t block_group; |
ac27a0ec | 186 | unsigned long bit; |
af5bc92d TT |
187 | struct ext4_group_desc *gdp; |
188 | struct ext4_super_block *es; | |
617ba13b | 189 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 190 | int fatal = 0, err, count, cleared; |
ac27a0ec DK |
191 | |
192 | if (atomic_read(&inode->i_count) > 1) { | |
4776004f TT |
193 | printk(KERN_ERR "ext4_free_inode: inode has count=%d\n", |
194 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
195 | return; |
196 | } | |
197 | if (inode->i_nlink) { | |
4776004f TT |
198 | printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n", |
199 | inode->i_nlink); | |
ac27a0ec DK |
200 | return; |
201 | } | |
202 | if (!sb) { | |
4776004f TT |
203 | printk(KERN_ERR "ext4_free_inode: inode on " |
204 | "nonexistent device\n"); | |
ac27a0ec DK |
205 | return; |
206 | } | |
617ba13b | 207 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
208 | |
209 | ino = inode->i_ino; | |
af5bc92d | 210 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 211 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
212 | |
213 | /* | |
214 | * Note: we must free any quota before locking the superblock, | |
215 | * as writing the quota to disk may need the lock as well. | |
216 | */ | |
871a2931 | 217 | dquot_initialize(inode); |
617ba13b | 218 | ext4_xattr_delete_inode(handle, inode); |
63936dda | 219 | dquot_free_inode(inode); |
9f754758 | 220 | dquot_drop(inode); |
ac27a0ec DK |
221 | |
222 | is_directory = S_ISDIR(inode->i_mode); | |
223 | ||
224 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
af5bc92d | 225 | clear_inode(inode); |
ac27a0ec | 226 | |
617ba13b MC |
227 | es = EXT4_SB(sb)->s_es; |
228 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 229 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
230 | goto error_return; |
231 | } | |
617ba13b MC |
232 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
233 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 234 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec DK |
235 | if (!bitmap_bh) |
236 | goto error_return; | |
237 | ||
238 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 239 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
240 | if (fatal) |
241 | goto error_return; | |
242 | ||
d17413c0 DM |
243 | fatal = -ESRCH; |
244 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
245 | if (gdp) { | |
ac27a0ec | 246 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 247 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
248 | } |
249 | ext4_lock_group(sb, block_group); | |
250 | cleared = ext4_clear_bit(bit, bitmap_bh->b_data); | |
251 | if (fatal || !cleared) { | |
252 | ext4_unlock_group(sb, block_group); | |
253 | goto out; | |
254 | } | |
7d39db14 | 255 | |
d17413c0 DM |
256 | count = ext4_free_inodes_count(sb, gdp) + 1; |
257 | ext4_free_inodes_set(sb, gdp, count); | |
258 | if (is_directory) { | |
259 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
260 | ext4_used_dirs_set(sb, gdp, count); | |
261 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 262 | } |
d17413c0 DM |
263 | gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp); |
264 | ext4_unlock_group(sb, block_group); | |
ac27a0ec | 265 | |
d17413c0 DM |
266 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
267 | if (sbi->s_log_groups_per_flex) { | |
268 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 269 | |
d17413c0 DM |
270 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
271 | if (is_directory) | |
272 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 273 | } |
d17413c0 DM |
274 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
275 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
276 | out: | |
277 | if (cleared) { | |
278 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
279 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
280 | if (!fatal) | |
281 | fatal = err; | |
282 | sb->s_dirt = 1; | |
283 | } else | |
284 | ext4_error(sb, "bit already cleared for inode %lu", ino); | |
285 | ||
ac27a0ec DK |
286 | error_return: |
287 | brelse(bitmap_bh); | |
617ba13b | 288 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
289 | } |
290 | ||
291 | /* | |
292 | * There are two policies for allocating an inode. If the new inode is | |
293 | * a directory, then a forward search is made for a block group with both | |
294 | * free space and a low directory-to-inode ratio; if that fails, then of | |
295 | * the groups with above-average free space, that group with the fewest | |
296 | * directories already is chosen. | |
297 | * | |
298 | * For other inodes, search forward from the parent directory\'s block | |
299 | * group to find a free inode. | |
300 | */ | |
2aa9fc4c AM |
301 | static int find_group_dir(struct super_block *sb, struct inode *parent, |
302 | ext4_group_t *best_group) | |
ac27a0ec | 303 | { |
8df9675f | 304 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 305 | unsigned int freei, avefreei; |
617ba13b | 306 | struct ext4_group_desc *desc, *best_desc = NULL; |
2aa9fc4c AM |
307 | ext4_group_t group; |
308 | int ret = -1; | |
ac27a0ec | 309 | |
617ba13b | 310 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
311 | avefreei = freei / ngroups; |
312 | ||
313 | for (group = 0; group < ngroups; group++) { | |
af5bc92d | 314 | desc = ext4_get_group_desc(sb, group, NULL); |
560671a0 | 315 | if (!desc || !ext4_free_inodes_count(sb, desc)) |
ac27a0ec | 316 | continue; |
560671a0 | 317 | if (ext4_free_inodes_count(sb, desc) < avefreei) |
ac27a0ec DK |
318 | continue; |
319 | if (!best_desc || | |
560671a0 AK |
320 | (ext4_free_blks_count(sb, desc) > |
321 | ext4_free_blks_count(sb, best_desc))) { | |
2aa9fc4c | 322 | *best_group = group; |
ac27a0ec | 323 | best_desc = desc; |
2aa9fc4c | 324 | ret = 0; |
ac27a0ec DK |
325 | } |
326 | } | |
2aa9fc4c | 327 | return ret; |
ac27a0ec DK |
328 | } |
329 | ||
772cb7c8 JS |
330 | #define free_block_ratio 10 |
331 | ||
332 | static int find_group_flex(struct super_block *sb, struct inode *parent, | |
333 | ext4_group_t *best_group) | |
334 | { | |
335 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
336 | struct ext4_group_desc *desc; | |
772cb7c8 JS |
337 | struct flex_groups *flex_group = sbi->s_flex_groups; |
338 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; | |
339 | ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group); | |
8df9675f | 340 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
772cb7c8 JS |
341 | int flex_size = ext4_flex_bg_size(sbi); |
342 | ext4_group_t best_flex = parent_fbg_group; | |
343 | int blocks_per_flex = sbi->s_blocks_per_group * flex_size; | |
344 | int flexbg_free_blocks; | |
345 | int flex_freeb_ratio; | |
346 | ext4_group_t n_fbg_groups; | |
347 | ext4_group_t i; | |
348 | ||
8df9675f | 349 | n_fbg_groups = (ngroups + flex_size - 1) >> |
772cb7c8 JS |
350 | sbi->s_log_groups_per_flex; |
351 | ||
352 | find_close_to_parent: | |
9f24e420 | 353 | flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks); |
772cb7c8 | 354 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; |
9f24e420 | 355 | if (atomic_read(&flex_group[best_flex].free_inodes) && |
772cb7c8 JS |
356 | flex_freeb_ratio > free_block_ratio) |
357 | goto found_flexbg; | |
358 | ||
359 | if (best_flex && best_flex == parent_fbg_group) { | |
360 | best_flex--; | |
361 | goto find_close_to_parent; | |
362 | } | |
363 | ||
364 | for (i = 0; i < n_fbg_groups; i++) { | |
365 | if (i == parent_fbg_group || i == parent_fbg_group - 1) | |
366 | continue; | |
367 | ||
9f24e420 | 368 | flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks); |
772cb7c8 JS |
369 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; |
370 | ||
371 | if (flex_freeb_ratio > free_block_ratio && | |
9f24e420 | 372 | (atomic_read(&flex_group[i].free_inodes))) { |
772cb7c8 JS |
373 | best_flex = i; |
374 | goto found_flexbg; | |
375 | } | |
376 | ||
9f24e420 TT |
377 | if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) || |
378 | ((atomic_read(&flex_group[i].free_blocks) > | |
379 | atomic_read(&flex_group[best_flex].free_blocks)) && | |
380 | atomic_read(&flex_group[i].free_inodes))) | |
772cb7c8 JS |
381 | best_flex = i; |
382 | } | |
383 | ||
9f24e420 TT |
384 | if (!atomic_read(&flex_group[best_flex].free_inodes) || |
385 | !atomic_read(&flex_group[best_flex].free_blocks)) | |
772cb7c8 JS |
386 | return -1; |
387 | ||
388 | found_flexbg: | |
389 | for (i = best_flex * flex_size; i < ngroups && | |
390 | i < (best_flex + 1) * flex_size; i++) { | |
88b6edd1 | 391 | desc = ext4_get_group_desc(sb, i, NULL); |
560671a0 | 392 | if (ext4_free_inodes_count(sb, desc)) { |
772cb7c8 JS |
393 | *best_group = i; |
394 | goto out; | |
395 | } | |
396 | } | |
397 | ||
398 | return -1; | |
399 | out: | |
400 | return 0; | |
401 | } | |
402 | ||
a4912123 TT |
403 | struct orlov_stats { |
404 | __u32 free_inodes; | |
405 | __u32 free_blocks; | |
406 | __u32 used_dirs; | |
407 | }; | |
408 | ||
409 | /* | |
410 | * Helper function for Orlov's allocator; returns critical information | |
411 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
412 | * is a block group number; otherwise it is flex_bg number. | |
413 | */ | |
414 | void get_orlov_stats(struct super_block *sb, ext4_group_t g, | |
415 | int flex_size, struct orlov_stats *stats) | |
416 | { | |
417 | struct ext4_group_desc *desc; | |
7d39db14 | 418 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 419 | |
7d39db14 TT |
420 | if (flex_size > 1) { |
421 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
422 | stats->free_blocks = atomic_read(&flex_group[g].free_blocks); | |
423 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); | |
424 | return; | |
425 | } | |
a4912123 | 426 | |
7d39db14 TT |
427 | desc = ext4_get_group_desc(sb, g, NULL); |
428 | if (desc) { | |
429 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
430 | stats->free_blocks = ext4_free_blks_count(sb, desc); | |
431 | stats->used_dirs = ext4_used_dirs_count(sb, desc); | |
432 | } else { | |
433 | stats->free_inodes = 0; | |
434 | stats->free_blocks = 0; | |
435 | stats->used_dirs = 0; | |
a4912123 TT |
436 | } |
437 | } | |
438 | ||
ac27a0ec DK |
439 | /* |
440 | * Orlov's allocator for directories. | |
441 | * | |
442 | * We always try to spread first-level directories. | |
443 | * | |
444 | * If there are blockgroups with both free inodes and free blocks counts | |
445 | * not worse than average we return one with smallest directory count. | |
446 | * Otherwise we simply return a random group. | |
447 | * | |
448 | * For the rest rules look so: | |
449 | * | |
450 | * It's OK to put directory into a group unless | |
451 | * it has too many directories already (max_dirs) or | |
452 | * it has too few free inodes left (min_inodes) or | |
453 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 454 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
455 | * conditions we search cyclically through the rest. If none |
456 | * of the groups look good we just look for a group with more | |
457 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
458 | */ |
459 | ||
2aa9fc4c | 460 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
f157a4aa TT |
461 | ext4_group_t *group, int mode, |
462 | const struct qstr *qstr) | |
ac27a0ec | 463 | { |
fd2d4291 | 464 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 465 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 466 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 467 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 468 | unsigned int freei, avefreei; |
617ba13b | 469 | ext4_fsblk_t freeb, avefreeb; |
ac27a0ec | 470 | unsigned int ndirs; |
a4912123 | 471 | int max_dirs, min_inodes; |
617ba13b | 472 | ext4_grpblk_t min_blocks; |
8df9675f | 473 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 474 | struct ext4_group_desc *desc; |
a4912123 TT |
475 | struct orlov_stats stats; |
476 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 477 | struct dx_hash_info hinfo; |
a4912123 | 478 | |
8df9675f | 479 | ngroups = real_ngroups; |
a4912123 | 480 | if (flex_size > 1) { |
8df9675f | 481 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
482 | sbi->s_log_groups_per_flex; |
483 | parent_group >>= sbi->s_log_groups_per_flex; | |
484 | } | |
ac27a0ec DK |
485 | |
486 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
487 | avefreei = freei / ngroups; | |
488 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3a5b2ecd | 489 | avefreeb = freeb; |
f4e5bc24 | 490 | do_div(avefreeb, ngroups); |
ac27a0ec DK |
491 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
492 | ||
a4912123 TT |
493 | if (S_ISDIR(mode) && |
494 | ((parent == sb->s_root->d_inode) || | |
12e9b892 | 495 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 496 | int best_ndir = inodes_per_group; |
2aa9fc4c | 497 | int ret = -1; |
ac27a0ec | 498 | |
f157a4aa TT |
499 | if (qstr) { |
500 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
501 | hinfo.seed = sbi->s_hash_seed; | |
502 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
503 | grp = hinfo.hash; | |
504 | } else | |
505 | get_random_bytes(&grp, sizeof(grp)); | |
2aa9fc4c | 506 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 507 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
508 | g = (parent_group + i) % ngroups; |
509 | get_orlov_stats(sb, g, flex_size, &stats); | |
510 | if (!stats.free_inodes) | |
ac27a0ec | 511 | continue; |
a4912123 | 512 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 513 | continue; |
a4912123 | 514 | if (stats.free_inodes < avefreei) |
ac27a0ec | 515 | continue; |
a4912123 | 516 | if (stats.free_blocks < avefreeb) |
ac27a0ec | 517 | continue; |
a4912123 | 518 | grp = g; |
2aa9fc4c | 519 | ret = 0; |
a4912123 TT |
520 | best_ndir = stats.used_dirs; |
521 | } | |
522 | if (ret) | |
523 | goto fallback; | |
524 | found_flex_bg: | |
525 | if (flex_size == 1) { | |
526 | *group = grp; | |
527 | return 0; | |
528 | } | |
529 | ||
530 | /* | |
531 | * We pack inodes at the beginning of the flexgroup's | |
532 | * inode tables. Block allocation decisions will do | |
533 | * something similar, although regular files will | |
534 | * start at 2nd block group of the flexgroup. See | |
535 | * ext4_ext_find_goal() and ext4_find_near(). | |
536 | */ | |
537 | grp *= flex_size; | |
538 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 539 | if (grp+i >= real_ngroups) |
a4912123 TT |
540 | break; |
541 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
542 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
543 | *group = grp+i; | |
544 | return 0; | |
545 | } | |
ac27a0ec | 546 | } |
ac27a0ec DK |
547 | goto fallback; |
548 | } | |
549 | ||
ac27a0ec | 550 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
551 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
552 | if (min_inodes < 1) | |
553 | min_inodes = 1; | |
554 | min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4; | |
555 | ||
556 | /* | |
557 | * Start looking in the flex group where we last allocated an | |
558 | * inode for this parent directory | |
559 | */ | |
560 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
561 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
562 | if (flex_size > 1) | |
563 | parent_group >>= sbi->s_log_groups_per_flex; | |
564 | } | |
ac27a0ec DK |
565 | |
566 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
567 | grp = (parent_group + i) % ngroups; |
568 | get_orlov_stats(sb, grp, flex_size, &stats); | |
569 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 570 | continue; |
a4912123 | 571 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 572 | continue; |
a4912123 | 573 | if (stats.free_blocks < min_blocks) |
ac27a0ec | 574 | continue; |
a4912123 | 575 | goto found_flex_bg; |
ac27a0ec DK |
576 | } |
577 | ||
578 | fallback: | |
8df9675f | 579 | ngroups = real_ngroups; |
a4912123 | 580 | avefreei = freei / ngroups; |
b5451f7b | 581 | fallback_retry: |
a4912123 | 582 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 583 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
584 | grp = (parent_group + i) % ngroups; |
585 | desc = ext4_get_group_desc(sb, grp, NULL); | |
560671a0 | 586 | if (desc && ext4_free_inodes_count(sb, desc) && |
a4912123 TT |
587 | ext4_free_inodes_count(sb, desc) >= avefreei) { |
588 | *group = grp; | |
2aa9fc4c | 589 | return 0; |
a4912123 | 590 | } |
ac27a0ec DK |
591 | } |
592 | ||
593 | if (avefreei) { | |
594 | /* | |
595 | * The free-inodes counter is approximate, and for really small | |
596 | * filesystems the above test can fail to find any blockgroups | |
597 | */ | |
598 | avefreei = 0; | |
b5451f7b | 599 | goto fallback_retry; |
ac27a0ec DK |
600 | } |
601 | ||
602 | return -1; | |
603 | } | |
604 | ||
2aa9fc4c | 605 | static int find_group_other(struct super_block *sb, struct inode *parent, |
a4912123 | 606 | ext4_group_t *group, int mode) |
ac27a0ec | 607 | { |
fd2d4291 | 608 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 609 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 610 | struct ext4_group_desc *desc; |
a4912123 TT |
611 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
612 | ||
613 | /* | |
614 | * Try to place the inode is the same flex group as its | |
615 | * parent. If we can't find space, use the Orlov algorithm to | |
616 | * find another flex group, and store that information in the | |
617 | * parent directory's inode information so that use that flex | |
618 | * group for future allocations. | |
619 | */ | |
620 | if (flex_size > 1) { | |
621 | int retry = 0; | |
622 | ||
623 | try_again: | |
624 | parent_group &= ~(flex_size-1); | |
625 | last = parent_group + flex_size; | |
626 | if (last > ngroups) | |
627 | last = ngroups; | |
628 | for (i = parent_group; i < last; i++) { | |
629 | desc = ext4_get_group_desc(sb, i, NULL); | |
630 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
631 | *group = i; | |
632 | return 0; | |
633 | } | |
634 | } | |
635 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
636 | retry = 1; | |
637 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
638 | goto try_again; | |
639 | } | |
640 | /* | |
641 | * If this didn't work, use the Orlov search algorithm | |
642 | * to find a new flex group; we pass in the mode to | |
643 | * avoid the topdir algorithms. | |
644 | */ | |
645 | *group = parent_group + flex_size; | |
646 | if (*group > ngroups) | |
647 | *group = 0; | |
f157a4aa | 648 | return find_group_orlov(sb, parent, group, mode, 0); |
a4912123 | 649 | } |
ac27a0ec DK |
650 | |
651 | /* | |
652 | * Try to place the inode in its parent directory | |
653 | */ | |
2aa9fc4c AM |
654 | *group = parent_group; |
655 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
656 | if (desc && ext4_free_inodes_count(sb, desc) && |
657 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 658 | return 0; |
ac27a0ec DK |
659 | |
660 | /* | |
661 | * We're going to place this inode in a different blockgroup from its | |
662 | * parent. We want to cause files in a common directory to all land in | |
663 | * the same blockgroup. But we want files which are in a different | |
664 | * directory which shares a blockgroup with our parent to land in a | |
665 | * different blockgroup. | |
666 | * | |
667 | * So add our directory's i_ino into the starting point for the hash. | |
668 | */ | |
2aa9fc4c | 669 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
670 | |
671 | /* | |
672 | * Use a quadratic hash to find a group with a free inode and some free | |
673 | * blocks. | |
674 | */ | |
675 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
676 | *group += i; |
677 | if (*group >= ngroups) | |
678 | *group -= ngroups; | |
679 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
680 | if (desc && ext4_free_inodes_count(sb, desc) && |
681 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 682 | return 0; |
ac27a0ec DK |
683 | } |
684 | ||
685 | /* | |
686 | * That failed: try linear search for a free inode, even if that group | |
687 | * has no free blocks. | |
688 | */ | |
2aa9fc4c | 689 | *group = parent_group; |
ac27a0ec | 690 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
691 | if (++*group >= ngroups) |
692 | *group = 0; | |
693 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 694 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 695 | return 0; |
ac27a0ec DK |
696 | } |
697 | ||
698 | return -1; | |
699 | } | |
700 | ||
39341867 AK |
701 | /* |
702 | * claim the inode from the inode bitmap. If the group | |
955ce5f5 | 703 | * is uninit we need to take the groups's ext4_group_lock |
39341867 AK |
704 | * and clear the uninit flag. The inode bitmap update |
705 | * and group desc uninit flag clear should be done | |
955ce5f5 | 706 | * after holding ext4_group_lock so that ext4_read_inode_bitmap |
39341867 AK |
707 | * doesn't race with the ext4_claim_inode |
708 | */ | |
709 | static int ext4_claim_inode(struct super_block *sb, | |
710 | struct buffer_head *inode_bitmap_bh, | |
711 | unsigned long ino, ext4_group_t group, int mode) | |
712 | { | |
713 | int free = 0, retval = 0, count; | |
714 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
715 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); | |
716 | ||
955ce5f5 | 717 | ext4_lock_group(sb, group); |
39341867 AK |
718 | if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) { |
719 | /* not a free inode */ | |
720 | retval = 1; | |
721 | goto err_ret; | |
722 | } | |
723 | ino++; | |
724 | if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || | |
725 | ino > EXT4_INODES_PER_GROUP(sb)) { | |
955ce5f5 | 726 | ext4_unlock_group(sb, group); |
12062ddd | 727 | ext4_error(sb, "reserved inode or inode > inodes count - " |
39341867 AK |
728 | "block_group = %u, inode=%lu", group, |
729 | ino + group * EXT4_INODES_PER_GROUP(sb)); | |
730 | return 1; | |
731 | } | |
732 | /* If we didn't allocate from within the initialized part of the inode | |
733 | * table then we need to initialize up to this inode. */ | |
734 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { | |
735 | ||
736 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
737 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
738 | /* When marking the block group with | |
739 | * ~EXT4_BG_INODE_UNINIT we don't want to depend | |
740 | * on the value of bg_itable_unused even though | |
741 | * mke2fs could have initialized the same for us. | |
742 | * Instead we calculated the value below | |
743 | */ | |
744 | ||
745 | free = 0; | |
746 | } else { | |
747 | free = EXT4_INODES_PER_GROUP(sb) - | |
748 | ext4_itable_unused_count(sb, gdp); | |
749 | } | |
750 | ||
751 | /* | |
752 | * Check the relative inode number against the last used | |
753 | * relative inode number in this group. if it is greater | |
754 | * we need to update the bg_itable_unused count | |
755 | * | |
756 | */ | |
757 | if (ino > free) | |
758 | ext4_itable_unused_set(sb, gdp, | |
759 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
760 | } | |
761 | count = ext4_free_inodes_count(sb, gdp) - 1; | |
762 | ext4_free_inodes_set(sb, gdp, count); | |
763 | if (S_ISDIR(mode)) { | |
764 | count = ext4_used_dirs_count(sb, gdp) + 1; | |
765 | ext4_used_dirs_set(sb, gdp, count); | |
7d39db14 TT |
766 | if (sbi->s_log_groups_per_flex) { |
767 | ext4_group_t f = ext4_flex_group(sbi, group); | |
768 | ||
c4caae25 | 769 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); |
7d39db14 | 770 | } |
39341867 AK |
771 | } |
772 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); | |
773 | err_ret: | |
955ce5f5 | 774 | ext4_unlock_group(sb, group); |
39341867 AK |
775 | return retval; |
776 | } | |
777 | ||
ac27a0ec DK |
778 | /* |
779 | * There are two policies for allocating an inode. If the new inode is | |
780 | * a directory, then a forward search is made for a block group with both | |
781 | * free space and a low directory-to-inode ratio; if that fails, then of | |
782 | * the groups with above-average free space, that group with the fewest | |
783 | * directories already is chosen. | |
784 | * | |
785 | * For other inodes, search forward from the parent directory's block | |
786 | * group to find a free inode. | |
787 | */ | |
f157a4aa | 788 | struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode, |
11013911 | 789 | const struct qstr *qstr, __u32 goal) |
ac27a0ec DK |
790 | { |
791 | struct super_block *sb; | |
3300beda AK |
792 | struct buffer_head *inode_bitmap_bh = NULL; |
793 | struct buffer_head *group_desc_bh; | |
8df9675f | 794 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 795 | unsigned long ino = 0; |
af5bc92d TT |
796 | struct inode *inode; |
797 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
798 | struct ext4_inode_info *ei; |
799 | struct ext4_sb_info *sbi; | |
39341867 | 800 | int ret2, err = 0; |
ac27a0ec | 801 | struct inode *ret; |
2aa9fc4c AM |
802 | ext4_group_t i; |
803 | int free = 0; | |
2842c3b5 | 804 | static int once = 1; |
772cb7c8 | 805 | ext4_group_t flex_group; |
ac27a0ec DK |
806 | |
807 | /* Cannot create files in a deleted directory */ | |
808 | if (!dir || !dir->i_nlink) | |
809 | return ERR_PTR(-EPERM); | |
810 | ||
811 | sb = dir->i_sb; | |
8df9675f | 812 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 813 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
814 | inode = new_inode(sb); |
815 | if (!inode) | |
816 | return ERR_PTR(-ENOMEM); | |
617ba13b | 817 | ei = EXT4_I(inode); |
617ba13b | 818 | sbi = EXT4_SB(sb); |
772cb7c8 | 819 | |
11013911 AD |
820 | if (!goal) |
821 | goal = sbi->s_inode_goal; | |
822 | ||
e6462869 | 823 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
824 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
825 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
826 | ret2 = 0; | |
827 | goto got_group; | |
828 | } | |
829 | ||
a4912123 | 830 | if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) { |
772cb7c8 | 831 | ret2 = find_group_flex(sb, dir, &group); |
05bf9e83 | 832 | if (ret2 == -1) { |
a4912123 | 833 | ret2 = find_group_other(sb, dir, &group, mode); |
6b82f3cb | 834 | if (ret2 == 0 && once) { |
2842c3b5 | 835 | once = 0; |
05bf9e83 TT |
836 | printk(KERN_NOTICE "ext4: find_group_flex " |
837 | "failed, fallback succeeded dir %lu\n", | |
838 | dir->i_ino); | |
6b82f3cb | 839 | } |
05bf9e83 | 840 | } |
772cb7c8 JS |
841 | goto got_group; |
842 | } | |
843 | ||
ac27a0ec | 844 | if (S_ISDIR(mode)) { |
af5bc92d | 845 | if (test_opt(sb, OLDALLOC)) |
2aa9fc4c | 846 | ret2 = find_group_dir(sb, dir, &group); |
ac27a0ec | 847 | else |
f157a4aa | 848 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); |
ac27a0ec | 849 | } else |
a4912123 | 850 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 851 | |
772cb7c8 | 852 | got_group: |
a4912123 | 853 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 854 | err = -ENOSPC; |
2aa9fc4c | 855 | if (ret2 == -1) |
ac27a0ec DK |
856 | goto out; |
857 | ||
11013911 | 858 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
859 | err = -EIO; |
860 | ||
3300beda | 861 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec DK |
862 | if (!gdp) |
863 | goto fail; | |
864 | ||
3300beda AK |
865 | brelse(inode_bitmap_bh); |
866 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
867 | if (!inode_bitmap_bh) | |
ac27a0ec DK |
868 | goto fail; |
869 | ||
ac27a0ec | 870 | repeat_in_this_group: |
617ba13b | 871 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
872 | inode_bitmap_bh->b_data, |
873 | EXT4_INODES_PER_GROUP(sb), ino); | |
874 | ||
617ba13b | 875 | if (ino < EXT4_INODES_PER_GROUP(sb)) { |
ac27a0ec | 876 | |
3300beda AK |
877 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
878 | err = ext4_journal_get_write_access(handle, | |
879 | inode_bitmap_bh); | |
ac27a0ec DK |
880 | if (err) |
881 | goto fail; | |
882 | ||
39341867 AK |
883 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
884 | err = ext4_journal_get_write_access(handle, | |
885 | group_desc_bh); | |
886 | if (err) | |
887 | goto fail; | |
888 | if (!ext4_claim_inode(sb, inode_bitmap_bh, | |
889 | ino, group, mode)) { | |
ac27a0ec | 890 | /* we won it */ |
3300beda | 891 | BUFFER_TRACE(inode_bitmap_bh, |
0390131b FM |
892 | "call ext4_handle_dirty_metadata"); |
893 | err = ext4_handle_dirty_metadata(handle, | |
73b50c1c | 894 | NULL, |
3300beda | 895 | inode_bitmap_bh); |
ac27a0ec DK |
896 | if (err) |
897 | goto fail; | |
39341867 AK |
898 | /* zero bit is inode number 1*/ |
899 | ino++; | |
ac27a0ec DK |
900 | goto got; |
901 | } | |
902 | /* we lost it */ | |
3300beda | 903 | ext4_handle_release_buffer(handle, inode_bitmap_bh); |
39341867 | 904 | ext4_handle_release_buffer(handle, group_desc_bh); |
ac27a0ec | 905 | |
617ba13b | 906 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
907 | goto repeat_in_this_group; |
908 | } | |
909 | ||
910 | /* | |
911 | * This case is possible in concurrent environment. It is very | |
912 | * rare. We cannot repeat the find_group_xxx() call because | |
913 | * that will simply return the same blockgroup, because the | |
914 | * group descriptor metadata has not yet been updated. | |
915 | * So we just go onto the next blockgroup. | |
916 | */ | |
8df9675f | 917 | if (++group == ngroups) |
ac27a0ec DK |
918 | group = 0; |
919 | } | |
920 | err = -ENOSPC; | |
921 | goto out; | |
922 | ||
923 | got: | |
717d50e4 AD |
924 | /* We may have to initialize the block bitmap if it isn't already */ |
925 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && | |
926 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
3300beda | 927 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 928 | |
3300beda AK |
929 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
930 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); | |
931 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 932 | if (err) { |
3300beda | 933 | brelse(block_bitmap_bh); |
717d50e4 AD |
934 | goto fail; |
935 | } | |
936 | ||
937 | free = 0; | |
955ce5f5 | 938 | ext4_lock_group(sb, group); |
717d50e4 AD |
939 | /* recheck and clear flag under lock if we still need to */ |
940 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
717d50e4 | 941 | free = ext4_free_blocks_after_init(sb, group, gdp); |
3300beda | 942 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
560671a0 | 943 | ext4_free_blks_set(sb, gdp, free); |
23712a9c FB |
944 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, |
945 | gdp); | |
717d50e4 | 946 | } |
955ce5f5 | 947 | ext4_unlock_group(sb, group); |
717d50e4 AD |
948 | |
949 | /* Don't need to dirty bitmap block if we didn't change it */ | |
950 | if (free) { | |
3300beda | 951 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
0390131b | 952 | err = ext4_handle_dirty_metadata(handle, |
3300beda | 953 | NULL, block_bitmap_bh); |
717d50e4 AD |
954 | } |
955 | ||
3300beda | 956 | brelse(block_bitmap_bh); |
717d50e4 AD |
957 | if (err) |
958 | goto fail; | |
959 | } | |
3300beda AK |
960 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
961 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
39341867 AK |
962 | if (err) |
963 | goto fail; | |
ac27a0ec DK |
964 | |
965 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
966 | if (S_ISDIR(mode)) | |
967 | percpu_counter_inc(&sbi->s_dirs_counter); | |
968 | sb->s_dirt = 1; | |
969 | ||
772cb7c8 JS |
970 | if (sbi->s_log_groups_per_flex) { |
971 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 972 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 JS |
973 | } |
974 | ||
b10b8520 DM |
975 | if (test_opt(sb, GRPID)) { |
976 | inode->i_mode = mode; | |
977 | inode->i_uid = current_fsuid(); | |
ac27a0ec | 978 | inode->i_gid = dir->i_gid; |
ac27a0ec | 979 | } else |
b10b8520 | 980 | inode_init_owner(inode, dir, mode); |
ac27a0ec | 981 | |
717d50e4 | 982 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
983 | /* This is the optimal IO size (for stat), not the fs block size */ |
984 | inode->i_blocks = 0; | |
ef7f3835 KS |
985 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
986 | ext4_current_time(inode); | |
ac27a0ec DK |
987 | |
988 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
989 | ei->i_dir_start_lookup = 0; | |
990 | ei->i_disksize = 0; | |
991 | ||
42bf0383 | 992 | /* |
2dc6b0d4 DG |
993 | * Don't inherit extent flag from directory, amongst others. We set |
994 | * extent flag on newly created directory and file only if -o extent | |
995 | * mount option is specified | |
42bf0383 | 996 | */ |
2dc6b0d4 DG |
997 | ei->i_flags = |
998 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 999 | ei->i_file_acl = 0; |
ac27a0ec | 1000 | ei->i_dtime = 0; |
ac27a0ec | 1001 | ei->i_block_group = group; |
a4912123 | 1002 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1003 | |
617ba13b | 1004 | ext4_set_inode_flags(inode); |
ac27a0ec | 1005 | if (IS_DIRSYNC(inode)) |
0390131b | 1006 | ext4_handle_sync(handle); |
6b38e842 AV |
1007 | if (insert_inode_locked(inode) < 0) { |
1008 | err = -EINVAL; | |
1009 | goto fail_drop; | |
1010 | } | |
ac27a0ec DK |
1011 | spin_lock(&sbi->s_next_gen_lock); |
1012 | inode->i_generation = sbi->s_next_generation++; | |
1013 | spin_unlock(&sbi->s_next_gen_lock); | |
1014 | ||
19f5fb7a TT |
1015 | ei->i_state_flags = 0; |
1016 | ext4_set_inode_state(inode, EXT4_STATE_NEW); | |
ef7f3835 KS |
1017 | |
1018 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
1019 | |
1020 | ret = inode; | |
871a2931 | 1021 | dquot_initialize(inode); |
63936dda CH |
1022 | err = dquot_alloc_inode(inode); |
1023 | if (err) | |
ac27a0ec | 1024 | goto fail_drop; |
ac27a0ec | 1025 | |
617ba13b | 1026 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
1027 | if (err) |
1028 | goto fail_free_drop; | |
1029 | ||
af5bc92d | 1030 | err = ext4_init_security(handle, inode, dir); |
ac27a0ec DK |
1031 | if (err) |
1032 | goto fail_free_drop; | |
1033 | ||
83982b6f | 1034 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
e4079a11 | 1035 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1036 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1037 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1038 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1039 | } |
a86c6181 | 1040 | } |
ac27a0ec | 1041 | |
8753e88f AK |
1042 | err = ext4_mark_inode_dirty(handle, inode); |
1043 | if (err) { | |
1044 | ext4_std_error(sb, err); | |
1045 | goto fail_free_drop; | |
1046 | } | |
1047 | ||
617ba13b | 1048 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1049 | trace_ext4_allocate_inode(inode, dir, mode); |
ac27a0ec DK |
1050 | goto really_out; |
1051 | fail: | |
617ba13b | 1052 | ext4_std_error(sb, err); |
ac27a0ec DK |
1053 | out: |
1054 | iput(inode); | |
1055 | ret = ERR_PTR(err); | |
1056 | really_out: | |
3300beda | 1057 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1058 | return ret; |
1059 | ||
1060 | fail_free_drop: | |
63936dda | 1061 | dquot_free_inode(inode); |
ac27a0ec DK |
1062 | |
1063 | fail_drop: | |
9f754758 | 1064 | dquot_drop(inode); |
ac27a0ec DK |
1065 | inode->i_flags |= S_NOQUOTA; |
1066 | inode->i_nlink = 0; | |
6b38e842 | 1067 | unlock_new_inode(inode); |
ac27a0ec | 1068 | iput(inode); |
3300beda | 1069 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1070 | return ERR_PTR(err); |
1071 | } | |
1072 | ||
1073 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1074 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1075 | { |
617ba13b | 1076 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1077 | ext4_group_t block_group; |
ac27a0ec | 1078 | int bit; |
1d1fe1ee | 1079 | struct buffer_head *bitmap_bh; |
ac27a0ec | 1080 | struct inode *inode = NULL; |
1d1fe1ee | 1081 | long err = -EIO; |
ac27a0ec DK |
1082 | |
1083 | /* Error cases - e2fsck has already cleaned up for us */ | |
1084 | if (ino > max_ino) { | |
12062ddd | 1085 | ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 1086 | goto error; |
ac27a0ec DK |
1087 | } |
1088 | ||
617ba13b MC |
1089 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1090 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1091 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 1092 | if (!bitmap_bh) { |
12062ddd | 1093 | ext4_warning(sb, "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 1094 | goto error; |
ac27a0ec DK |
1095 | } |
1096 | ||
1097 | /* Having the inode bit set should be a 100% indicator that this | |
1098 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1099 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1100 | */ | |
1d1fe1ee DH |
1101 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1102 | goto bad_orphan; | |
1103 | ||
1104 | inode = ext4_iget(sb, ino); | |
1105 | if (IS_ERR(inode)) | |
1106 | goto iget_failed; | |
1107 | ||
91ef4caf DG |
1108 | /* |
1109 | * If the orphans has i_nlinks > 0 then it should be able to be | |
1110 | * truncated, otherwise it won't be removed from the orphan list | |
1111 | * during processing and an infinite loop will result. | |
1112 | */ | |
1113 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
1114 | goto bad_orphan; | |
1115 | ||
1d1fe1ee DH |
1116 | if (NEXT_ORPHAN(inode) > max_ino) |
1117 | goto bad_orphan; | |
1118 | brelse(bitmap_bh); | |
1119 | return inode; | |
1120 | ||
1121 | iget_failed: | |
1122 | err = PTR_ERR(inode); | |
1123 | inode = NULL; | |
1124 | bad_orphan: | |
12062ddd | 1125 | ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino); |
1d1fe1ee DH |
1126 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", |
1127 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
1128 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
1129 | printk(KERN_NOTICE "inode=%p\n", inode); | |
1130 | if (inode) { | |
1131 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
1132 | is_bad_inode(inode)); | |
1133 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
1134 | NEXT_ORPHAN(inode)); | |
1135 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
91ef4caf | 1136 | printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink); |
ac27a0ec | 1137 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1138 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1139 | inode->i_blocks = 0; |
1140 | iput(inode); | |
ac27a0ec | 1141 | } |
ac27a0ec | 1142 | brelse(bitmap_bh); |
1d1fe1ee DH |
1143 | error: |
1144 | return ERR_PTR(err); | |
ac27a0ec DK |
1145 | } |
1146 | ||
af5bc92d | 1147 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1148 | { |
1149 | unsigned long desc_count; | |
617ba13b | 1150 | struct ext4_group_desc *gdp; |
8df9675f | 1151 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1152 | #ifdef EXT4FS_DEBUG |
1153 | struct ext4_super_block *es; | |
ac27a0ec DK |
1154 | unsigned long bitmap_count, x; |
1155 | struct buffer_head *bitmap_bh = NULL; | |
1156 | ||
617ba13b | 1157 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1158 | desc_count = 0; |
1159 | bitmap_count = 0; | |
1160 | gdp = NULL; | |
8df9675f | 1161 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1162 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1163 | if (!gdp) |
1164 | continue; | |
560671a0 | 1165 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1166 | brelse(bitmap_bh); |
e29d1cde | 1167 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
1168 | if (!bitmap_bh) |
1169 | continue; | |
1170 | ||
617ba13b | 1171 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
c549a95d | 1172 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1173 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1174 | bitmap_count += x; |
1175 | } | |
1176 | brelse(bitmap_bh); | |
4776004f TT |
1177 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1178 | "stored = %u, computed = %lu, %lu\n", | |
1179 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1180 | return desc_count; |
1181 | #else | |
1182 | desc_count = 0; | |
8df9675f | 1183 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1184 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1185 | if (!gdp) |
1186 | continue; | |
560671a0 | 1187 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1188 | cond_resched(); |
1189 | } | |
1190 | return desc_count; | |
1191 | #endif | |
1192 | } | |
1193 | ||
1194 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1195 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1196 | { |
1197 | unsigned long count = 0; | |
8df9675f | 1198 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1199 | |
8df9675f | 1200 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1201 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1202 | if (!gdp) |
1203 | continue; | |
560671a0 | 1204 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1205 | } |
1206 | return count; | |
1207 | } |