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 | */ | |
61d08673 | 53 | void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap) |
717d50e4 AD |
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 */ | |
1f109d5a TT |
68 | static unsigned ext4_init_inode_bitmap(struct super_block *sb, |
69 | struct buffer_head *bh, | |
70 | ext4_group_t block_group, | |
71 | struct ext4_group_desc *gdp) | |
717d50e4 AD |
72 | { |
73 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
74 | ||
75 | J_ASSERT_BH(bh, buffer_locked(bh)); | |
76 | ||
77 | /* If checksum is bad mark all blocks and inodes use to prevent | |
78 | * allocation, essentially implementing a per-group read-only flag. */ | |
79 | if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) { | |
12062ddd | 80 | ext4_error(sb, "Checksum bad for group %u", block_group); |
021b65bb | 81 | ext4_free_group_clusters_set(sb, gdp, 0); |
560671a0 AK |
82 | ext4_free_inodes_set(sb, gdp, 0); |
83 | ext4_itable_unused_set(sb, gdp, 0); | |
717d50e4 AD |
84 | memset(bh->b_data, 0xff, sb->s_blocksize); |
85 | return 0; | |
86 | } | |
87 | ||
88 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
61d08673 | 89 | ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
717d50e4 AD |
90 | bh->b_data); |
91 | ||
92 | return EXT4_INODES_PER_GROUP(sb); | |
93 | } | |
ac27a0ec DK |
94 | |
95 | /* | |
96 | * Read the inode allocation bitmap for a given block_group, reading | |
97 | * into the specified slot in the superblock's bitmap cache. | |
98 | * | |
99 | * Return buffer_head of bitmap on success or NULL. | |
100 | */ | |
101 | static struct buffer_head * | |
e29d1cde | 102 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 103 | { |
617ba13b | 104 | struct ext4_group_desc *desc; |
ac27a0ec | 105 | struct buffer_head *bh = NULL; |
e29d1cde | 106 | ext4_fsblk_t bitmap_blk; |
ac27a0ec | 107 | |
617ba13b | 108 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 109 | if (!desc) |
e29d1cde | 110 | return NULL; |
bfff6873 | 111 | |
e29d1cde ES |
112 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
113 | bh = sb_getblk(sb, bitmap_blk); | |
114 | if (unlikely(!bh)) { | |
12062ddd | 115 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 116 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
117 | block_group, bitmap_blk); |
118 | return NULL; | |
119 | } | |
2ccb5fb9 | 120 | if (bitmap_uptodate(bh)) |
e29d1cde ES |
121 | return bh; |
122 | ||
c806e68f | 123 | lock_buffer(bh); |
2ccb5fb9 AK |
124 | if (bitmap_uptodate(bh)) { |
125 | unlock_buffer(bh); | |
126 | return bh; | |
127 | } | |
bfff6873 | 128 | |
955ce5f5 | 129 | ext4_lock_group(sb, block_group); |
717d50e4 | 130 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
e29d1cde | 131 | ext4_init_inode_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 132 | set_bitmap_uptodate(bh); |
e29d1cde | 133 | set_buffer_uptodate(bh); |
955ce5f5 | 134 | ext4_unlock_group(sb, block_group); |
3300beda | 135 | unlock_buffer(bh); |
e29d1cde | 136 | return bh; |
717d50e4 | 137 | } |
955ce5f5 | 138 | ext4_unlock_group(sb, block_group); |
bfff6873 | 139 | |
2ccb5fb9 AK |
140 | if (buffer_uptodate(bh)) { |
141 | /* | |
142 | * if not uninit if bh is uptodate, | |
143 | * bitmap is also uptodate | |
144 | */ | |
145 | set_bitmap_uptodate(bh); | |
146 | unlock_buffer(bh); | |
147 | return bh; | |
148 | } | |
149 | /* | |
150 | * submit the buffer_head for read. We can | |
151 | * safely mark the bitmap as uptodate now. | |
152 | * We do it here so the bitmap uptodate bit | |
153 | * get set with buffer lock held. | |
154 | */ | |
0562e0ba | 155 | trace_ext4_load_inode_bitmap(sb, block_group); |
2ccb5fb9 | 156 | set_bitmap_uptodate(bh); |
e29d1cde ES |
157 | if (bh_submit_read(bh) < 0) { |
158 | put_bh(bh); | |
12062ddd | 159 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 160 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
161 | block_group, bitmap_blk); |
162 | return NULL; | |
163 | } | |
ac27a0ec DK |
164 | return bh; |
165 | } | |
166 | ||
167 | /* | |
168 | * NOTE! When we get the inode, we're the only people | |
169 | * that have access to it, and as such there are no | |
170 | * race conditions we have to worry about. The inode | |
171 | * is not on the hash-lists, and it cannot be reached | |
172 | * through the filesystem because the directory entry | |
173 | * has been deleted earlier. | |
174 | * | |
175 | * HOWEVER: we must make sure that we get no aliases, | |
176 | * which means that we have to call "clear_inode()" | |
177 | * _before_ we mark the inode not in use in the inode | |
178 | * bitmaps. Otherwise a newly created file might use | |
179 | * the same inode number (not actually the same pointer | |
180 | * though), and then we'd have two inodes sharing the | |
181 | * same inode number and space on the harddisk. | |
182 | */ | |
af5bc92d | 183 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 184 | { |
af5bc92d | 185 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
186 | int is_directory; |
187 | unsigned long ino; | |
188 | struct buffer_head *bitmap_bh = NULL; | |
189 | struct buffer_head *bh2; | |
fd2d4291 | 190 | ext4_group_t block_group; |
ac27a0ec | 191 | unsigned long bit; |
af5bc92d TT |
192 | struct ext4_group_desc *gdp; |
193 | struct ext4_super_block *es; | |
617ba13b | 194 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 195 | int fatal = 0, err, count, cleared; |
ac27a0ec DK |
196 | |
197 | if (atomic_read(&inode->i_count) > 1) { | |
4776004f TT |
198 | printk(KERN_ERR "ext4_free_inode: inode has count=%d\n", |
199 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
200 | return; |
201 | } | |
202 | if (inode->i_nlink) { | |
4776004f TT |
203 | printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n", |
204 | inode->i_nlink); | |
ac27a0ec DK |
205 | return; |
206 | } | |
207 | if (!sb) { | |
4776004f TT |
208 | printk(KERN_ERR "ext4_free_inode: inode on " |
209 | "nonexistent device\n"); | |
ac27a0ec DK |
210 | return; |
211 | } | |
617ba13b | 212 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
213 | |
214 | ino = inode->i_ino; | |
af5bc92d | 215 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 216 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
217 | |
218 | /* | |
219 | * Note: we must free any quota before locking the superblock, | |
220 | * as writing the quota to disk may need the lock as well. | |
221 | */ | |
871a2931 | 222 | dquot_initialize(inode); |
617ba13b | 223 | ext4_xattr_delete_inode(handle, inode); |
63936dda | 224 | dquot_free_inode(inode); |
9f754758 | 225 | dquot_drop(inode); |
ac27a0ec DK |
226 | |
227 | is_directory = S_ISDIR(inode->i_mode); | |
228 | ||
229 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 230 | ext4_clear_inode(inode); |
ac27a0ec | 231 | |
617ba13b MC |
232 | es = EXT4_SB(sb)->s_es; |
233 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 234 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
235 | goto error_return; |
236 | } | |
617ba13b MC |
237 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
238 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 239 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec DK |
240 | if (!bitmap_bh) |
241 | goto error_return; | |
242 | ||
243 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 244 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
245 | if (fatal) |
246 | goto error_return; | |
247 | ||
d17413c0 DM |
248 | fatal = -ESRCH; |
249 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
250 | if (gdp) { | |
ac27a0ec | 251 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 252 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
253 | } |
254 | ext4_lock_group(sb, block_group); | |
255 | cleared = ext4_clear_bit(bit, bitmap_bh->b_data); | |
256 | if (fatal || !cleared) { | |
257 | ext4_unlock_group(sb, block_group); | |
258 | goto out; | |
259 | } | |
7d39db14 | 260 | |
d17413c0 DM |
261 | count = ext4_free_inodes_count(sb, gdp) + 1; |
262 | ext4_free_inodes_set(sb, gdp, count); | |
263 | if (is_directory) { | |
264 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
265 | ext4_used_dirs_set(sb, gdp, count); | |
266 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 267 | } |
d17413c0 DM |
268 | gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp); |
269 | ext4_unlock_group(sb, block_group); | |
ac27a0ec | 270 | |
d17413c0 DM |
271 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
272 | if (sbi->s_log_groups_per_flex) { | |
273 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 274 | |
d17413c0 DM |
275 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
276 | if (is_directory) | |
277 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 278 | } |
d17413c0 DM |
279 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
280 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
281 | out: | |
282 | if (cleared) { | |
283 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
284 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
285 | if (!fatal) | |
286 | fatal = err; | |
a0375156 | 287 | ext4_mark_super_dirty(sb); |
d17413c0 DM |
288 | } else |
289 | ext4_error(sb, "bit already cleared for inode %lu", ino); | |
290 | ||
ac27a0ec DK |
291 | error_return: |
292 | brelse(bitmap_bh); | |
617ba13b | 293 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
294 | } |
295 | ||
296 | /* | |
297 | * There are two policies for allocating an inode. If the new inode is | |
298 | * a directory, then a forward search is made for a block group with both | |
299 | * free space and a low directory-to-inode ratio; if that fails, then of | |
300 | * the groups with above-average free space, that group with the fewest | |
301 | * directories already is chosen. | |
302 | * | |
303 | * For other inodes, search forward from the parent directory\'s block | |
304 | * group to find a free inode. | |
305 | */ | |
2aa9fc4c AM |
306 | static int find_group_dir(struct super_block *sb, struct inode *parent, |
307 | ext4_group_t *best_group) | |
ac27a0ec | 308 | { |
8df9675f | 309 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 310 | unsigned int freei, avefreei; |
617ba13b | 311 | struct ext4_group_desc *desc, *best_desc = NULL; |
2aa9fc4c AM |
312 | ext4_group_t group; |
313 | int ret = -1; | |
ac27a0ec | 314 | |
617ba13b | 315 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
316 | avefreei = freei / ngroups; |
317 | ||
318 | for (group = 0; group < ngroups; group++) { | |
af5bc92d | 319 | desc = ext4_get_group_desc(sb, group, NULL); |
560671a0 | 320 | if (!desc || !ext4_free_inodes_count(sb, desc)) |
ac27a0ec | 321 | continue; |
560671a0 | 322 | if (ext4_free_inodes_count(sb, desc) < avefreei) |
ac27a0ec DK |
323 | continue; |
324 | if (!best_desc || | |
021b65bb TT |
325 | (ext4_free_group_clusters(sb, desc) > |
326 | ext4_free_group_clusters(sb, best_desc))) { | |
2aa9fc4c | 327 | *best_group = group; |
ac27a0ec | 328 | best_desc = desc; |
2aa9fc4c | 329 | ret = 0; |
ac27a0ec DK |
330 | } |
331 | } | |
2aa9fc4c | 332 | return ret; |
ac27a0ec DK |
333 | } |
334 | ||
772cb7c8 JS |
335 | #define free_block_ratio 10 |
336 | ||
337 | static int find_group_flex(struct super_block *sb, struct inode *parent, | |
338 | ext4_group_t *best_group) | |
339 | { | |
340 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
341 | struct ext4_group_desc *desc; | |
772cb7c8 JS |
342 | struct flex_groups *flex_group = sbi->s_flex_groups; |
343 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; | |
344 | ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group); | |
8df9675f | 345 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
772cb7c8 JS |
346 | int flex_size = ext4_flex_bg_size(sbi); |
347 | ext4_group_t best_flex = parent_fbg_group; | |
348 | int blocks_per_flex = sbi->s_blocks_per_group * flex_size; | |
24aaa8ef | 349 | int flexbg_free_clusters; |
772cb7c8 JS |
350 | int flex_freeb_ratio; |
351 | ext4_group_t n_fbg_groups; | |
352 | ext4_group_t i; | |
353 | ||
8df9675f | 354 | n_fbg_groups = (ngroups + flex_size - 1) >> |
772cb7c8 JS |
355 | sbi->s_log_groups_per_flex; |
356 | ||
357 | find_close_to_parent: | |
24aaa8ef TT |
358 | flexbg_free_clusters = atomic_read(&flex_group[best_flex].free_clusters); |
359 | flex_freeb_ratio = EXT4_C2B(sbi, flexbg_free_clusters) * 100 / | |
360 | blocks_per_flex; | |
9f24e420 | 361 | if (atomic_read(&flex_group[best_flex].free_inodes) && |
772cb7c8 JS |
362 | flex_freeb_ratio > free_block_ratio) |
363 | goto found_flexbg; | |
364 | ||
365 | if (best_flex && best_flex == parent_fbg_group) { | |
366 | best_flex--; | |
367 | goto find_close_to_parent; | |
368 | } | |
369 | ||
370 | for (i = 0; i < n_fbg_groups; i++) { | |
371 | if (i == parent_fbg_group || i == parent_fbg_group - 1) | |
372 | continue; | |
373 | ||
24aaa8ef TT |
374 | flexbg_free_clusters = atomic_read(&flex_group[i].free_clusters); |
375 | flex_freeb_ratio = EXT4_C2B(sbi, flexbg_free_clusters) * 100 / | |
376 | blocks_per_flex; | |
772cb7c8 JS |
377 | |
378 | if (flex_freeb_ratio > free_block_ratio && | |
9f24e420 | 379 | (atomic_read(&flex_group[i].free_inodes))) { |
772cb7c8 JS |
380 | best_flex = i; |
381 | goto found_flexbg; | |
382 | } | |
383 | ||
9f24e420 | 384 | if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) || |
24aaa8ef TT |
385 | ((atomic_read(&flex_group[i].free_clusters) > |
386 | atomic_read(&flex_group[best_flex].free_clusters)) && | |
9f24e420 | 387 | atomic_read(&flex_group[i].free_inodes))) |
772cb7c8 JS |
388 | best_flex = i; |
389 | } | |
390 | ||
9f24e420 | 391 | if (!atomic_read(&flex_group[best_flex].free_inodes) || |
24aaa8ef | 392 | !atomic_read(&flex_group[best_flex].free_clusters)) |
772cb7c8 JS |
393 | return -1; |
394 | ||
395 | found_flexbg: | |
396 | for (i = best_flex * flex_size; i < ngroups && | |
397 | i < (best_flex + 1) * flex_size; i++) { | |
88b6edd1 | 398 | desc = ext4_get_group_desc(sb, i, NULL); |
560671a0 | 399 | if (ext4_free_inodes_count(sb, desc)) { |
772cb7c8 JS |
400 | *best_group = i; |
401 | goto out; | |
402 | } | |
403 | } | |
404 | ||
405 | return -1; | |
406 | out: | |
407 | return 0; | |
408 | } | |
409 | ||
a4912123 TT |
410 | struct orlov_stats { |
411 | __u32 free_inodes; | |
24aaa8ef | 412 | __u32 free_clusters; |
a4912123 TT |
413 | __u32 used_dirs; |
414 | }; | |
415 | ||
416 | /* | |
417 | * Helper function for Orlov's allocator; returns critical information | |
418 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
419 | * is a block group number; otherwise it is flex_bg number. | |
420 | */ | |
1f109d5a TT |
421 | static void get_orlov_stats(struct super_block *sb, ext4_group_t g, |
422 | int flex_size, struct orlov_stats *stats) | |
a4912123 TT |
423 | { |
424 | struct ext4_group_desc *desc; | |
7d39db14 | 425 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 426 | |
7d39db14 TT |
427 | if (flex_size > 1) { |
428 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
24aaa8ef | 429 | stats->free_clusters = atomic_read(&flex_group[g].free_clusters); |
7d39db14 TT |
430 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); |
431 | return; | |
432 | } | |
a4912123 | 433 | |
7d39db14 TT |
434 | desc = ext4_get_group_desc(sb, g, NULL); |
435 | if (desc) { | |
436 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
021b65bb | 437 | stats->free_clusters = ext4_free_group_clusters(sb, desc); |
7d39db14 TT |
438 | stats->used_dirs = ext4_used_dirs_count(sb, desc); |
439 | } else { | |
440 | stats->free_inodes = 0; | |
24aaa8ef | 441 | stats->free_clusters = 0; |
7d39db14 | 442 | stats->used_dirs = 0; |
a4912123 TT |
443 | } |
444 | } | |
445 | ||
ac27a0ec DK |
446 | /* |
447 | * Orlov's allocator for directories. | |
448 | * | |
449 | * We always try to spread first-level directories. | |
450 | * | |
451 | * If there are blockgroups with both free inodes and free blocks counts | |
452 | * not worse than average we return one with smallest directory count. | |
453 | * Otherwise we simply return a random group. | |
454 | * | |
455 | * For the rest rules look so: | |
456 | * | |
457 | * It's OK to put directory into a group unless | |
458 | * it has too many directories already (max_dirs) or | |
459 | * it has too few free inodes left (min_inodes) or | |
460 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 461 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
462 | * conditions we search cyclically through the rest. If none |
463 | * of the groups look good we just look for a group with more | |
464 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
465 | */ |
466 | ||
2aa9fc4c | 467 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
f157a4aa TT |
468 | ext4_group_t *group, int mode, |
469 | const struct qstr *qstr) | |
ac27a0ec | 470 | { |
fd2d4291 | 471 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 472 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 473 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 474 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 475 | unsigned int freei, avefreei; |
24aaa8ef | 476 | ext4_fsblk_t freeb, avefreec; |
ac27a0ec | 477 | unsigned int ndirs; |
a4912123 | 478 | int max_dirs, min_inodes; |
24aaa8ef | 479 | ext4_grpblk_t min_clusters; |
8df9675f | 480 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 481 | struct ext4_group_desc *desc; |
a4912123 TT |
482 | struct orlov_stats stats; |
483 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 484 | struct dx_hash_info hinfo; |
a4912123 | 485 | |
8df9675f | 486 | ngroups = real_ngroups; |
a4912123 | 487 | if (flex_size > 1) { |
8df9675f | 488 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
489 | sbi->s_log_groups_per_flex; |
490 | parent_group >>= sbi->s_log_groups_per_flex; | |
491 | } | |
ac27a0ec DK |
492 | |
493 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
494 | avefreei = freei / ngroups; | |
57042651 TT |
495 | freeb = EXT4_C2B(sbi, |
496 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
24aaa8ef TT |
497 | avefreec = freeb; |
498 | do_div(avefreec, ngroups); | |
ac27a0ec DK |
499 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
500 | ||
a4912123 TT |
501 | if (S_ISDIR(mode) && |
502 | ((parent == sb->s_root->d_inode) || | |
12e9b892 | 503 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 504 | int best_ndir = inodes_per_group; |
2aa9fc4c | 505 | int ret = -1; |
ac27a0ec | 506 | |
f157a4aa TT |
507 | if (qstr) { |
508 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
509 | hinfo.seed = sbi->s_hash_seed; | |
510 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
511 | grp = hinfo.hash; | |
512 | } else | |
513 | get_random_bytes(&grp, sizeof(grp)); | |
2aa9fc4c | 514 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 515 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
516 | g = (parent_group + i) % ngroups; |
517 | get_orlov_stats(sb, g, flex_size, &stats); | |
518 | if (!stats.free_inodes) | |
ac27a0ec | 519 | continue; |
a4912123 | 520 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 521 | continue; |
a4912123 | 522 | if (stats.free_inodes < avefreei) |
ac27a0ec | 523 | continue; |
24aaa8ef | 524 | if (stats.free_clusters < avefreec) |
ac27a0ec | 525 | continue; |
a4912123 | 526 | grp = g; |
2aa9fc4c | 527 | ret = 0; |
a4912123 TT |
528 | best_ndir = stats.used_dirs; |
529 | } | |
530 | if (ret) | |
531 | goto fallback; | |
532 | found_flex_bg: | |
533 | if (flex_size == 1) { | |
534 | *group = grp; | |
535 | return 0; | |
536 | } | |
537 | ||
538 | /* | |
539 | * We pack inodes at the beginning of the flexgroup's | |
540 | * inode tables. Block allocation decisions will do | |
541 | * something similar, although regular files will | |
542 | * start at 2nd block group of the flexgroup. See | |
543 | * ext4_ext_find_goal() and ext4_find_near(). | |
544 | */ | |
545 | grp *= flex_size; | |
546 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 547 | if (grp+i >= real_ngroups) |
a4912123 TT |
548 | break; |
549 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
550 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
551 | *group = grp+i; | |
552 | return 0; | |
553 | } | |
ac27a0ec | 554 | } |
ac27a0ec DK |
555 | goto fallback; |
556 | } | |
557 | ||
ac27a0ec | 558 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
559 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
560 | if (min_inodes < 1) | |
561 | min_inodes = 1; | |
24aaa8ef | 562 | min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; |
a4912123 TT |
563 | |
564 | /* | |
565 | * Start looking in the flex group where we last allocated an | |
566 | * inode for this parent directory | |
567 | */ | |
568 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
569 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
570 | if (flex_size > 1) | |
571 | parent_group >>= sbi->s_log_groups_per_flex; | |
572 | } | |
ac27a0ec DK |
573 | |
574 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
575 | grp = (parent_group + i) % ngroups; |
576 | get_orlov_stats(sb, grp, flex_size, &stats); | |
577 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 578 | continue; |
a4912123 | 579 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 580 | continue; |
24aaa8ef | 581 | if (stats.free_clusters < min_clusters) |
ac27a0ec | 582 | continue; |
a4912123 | 583 | goto found_flex_bg; |
ac27a0ec DK |
584 | } |
585 | ||
586 | fallback: | |
8df9675f | 587 | ngroups = real_ngroups; |
a4912123 | 588 | avefreei = freei / ngroups; |
b5451f7b | 589 | fallback_retry: |
a4912123 | 590 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 591 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
592 | grp = (parent_group + i) % ngroups; |
593 | desc = ext4_get_group_desc(sb, grp, NULL); | |
560671a0 | 594 | if (desc && ext4_free_inodes_count(sb, desc) && |
a4912123 TT |
595 | ext4_free_inodes_count(sb, desc) >= avefreei) { |
596 | *group = grp; | |
2aa9fc4c | 597 | return 0; |
a4912123 | 598 | } |
ac27a0ec DK |
599 | } |
600 | ||
601 | if (avefreei) { | |
602 | /* | |
603 | * The free-inodes counter is approximate, and for really small | |
604 | * filesystems the above test can fail to find any blockgroups | |
605 | */ | |
606 | avefreei = 0; | |
b5451f7b | 607 | goto fallback_retry; |
ac27a0ec DK |
608 | } |
609 | ||
610 | return -1; | |
611 | } | |
612 | ||
2aa9fc4c | 613 | static int find_group_other(struct super_block *sb, struct inode *parent, |
a4912123 | 614 | ext4_group_t *group, int mode) |
ac27a0ec | 615 | { |
fd2d4291 | 616 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 617 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 618 | struct ext4_group_desc *desc; |
a4912123 TT |
619 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
620 | ||
621 | /* | |
622 | * Try to place the inode is the same flex group as its | |
623 | * parent. If we can't find space, use the Orlov algorithm to | |
624 | * find another flex group, and store that information in the | |
625 | * parent directory's inode information so that use that flex | |
626 | * group for future allocations. | |
627 | */ | |
628 | if (flex_size > 1) { | |
629 | int retry = 0; | |
630 | ||
631 | try_again: | |
632 | parent_group &= ~(flex_size-1); | |
633 | last = parent_group + flex_size; | |
634 | if (last > ngroups) | |
635 | last = ngroups; | |
636 | for (i = parent_group; i < last; i++) { | |
637 | desc = ext4_get_group_desc(sb, i, NULL); | |
638 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
639 | *group = i; | |
640 | return 0; | |
641 | } | |
642 | } | |
643 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
644 | retry = 1; | |
645 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
646 | goto try_again; | |
647 | } | |
648 | /* | |
649 | * If this didn't work, use the Orlov search algorithm | |
650 | * to find a new flex group; we pass in the mode to | |
651 | * avoid the topdir algorithms. | |
652 | */ | |
653 | *group = parent_group + flex_size; | |
654 | if (*group > ngroups) | |
655 | *group = 0; | |
7dc57615 | 656 | return find_group_orlov(sb, parent, group, mode, NULL); |
a4912123 | 657 | } |
ac27a0ec DK |
658 | |
659 | /* | |
660 | * Try to place the inode in its parent directory | |
661 | */ | |
2aa9fc4c AM |
662 | *group = parent_group; |
663 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 664 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 665 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 666 | return 0; |
ac27a0ec DK |
667 | |
668 | /* | |
669 | * We're going to place this inode in a different blockgroup from its | |
670 | * parent. We want to cause files in a common directory to all land in | |
671 | * the same blockgroup. But we want files which are in a different | |
672 | * directory which shares a blockgroup with our parent to land in a | |
673 | * different blockgroup. | |
674 | * | |
675 | * So add our directory's i_ino into the starting point for the hash. | |
676 | */ | |
2aa9fc4c | 677 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
678 | |
679 | /* | |
680 | * Use a quadratic hash to find a group with a free inode and some free | |
681 | * blocks. | |
682 | */ | |
683 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
684 | *group += i; |
685 | if (*group >= ngroups) | |
686 | *group -= ngroups; | |
687 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 688 | if (desc && ext4_free_inodes_count(sb, desc) && |
021b65bb | 689 | ext4_free_group_clusters(sb, desc)) |
2aa9fc4c | 690 | return 0; |
ac27a0ec DK |
691 | } |
692 | ||
693 | /* | |
694 | * That failed: try linear search for a free inode, even if that group | |
695 | * has no free blocks. | |
696 | */ | |
2aa9fc4c | 697 | *group = parent_group; |
ac27a0ec | 698 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
699 | if (++*group >= ngroups) |
700 | *group = 0; | |
701 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 702 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 703 | return 0; |
ac27a0ec DK |
704 | } |
705 | ||
706 | return -1; | |
707 | } | |
708 | ||
39341867 AK |
709 | /* |
710 | * claim the inode from the inode bitmap. If the group | |
955ce5f5 | 711 | * is uninit we need to take the groups's ext4_group_lock |
39341867 AK |
712 | * and clear the uninit flag. The inode bitmap update |
713 | * and group desc uninit flag clear should be done | |
955ce5f5 | 714 | * after holding ext4_group_lock so that ext4_read_inode_bitmap |
39341867 AK |
715 | * doesn't race with the ext4_claim_inode |
716 | */ | |
717 | static int ext4_claim_inode(struct super_block *sb, | |
718 | struct buffer_head *inode_bitmap_bh, | |
719 | unsigned long ino, ext4_group_t group, int mode) | |
720 | { | |
721 | int free = 0, retval = 0, count; | |
722 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
bfff6873 | 723 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); |
39341867 AK |
724 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); |
725 | ||
bfff6873 LC |
726 | /* |
727 | * We have to be sure that new inode allocation does not race with | |
728 | * inode table initialization, because otherwise we may end up | |
729 | * allocating and writing new inode right before sb_issue_zeroout | |
730 | * takes place and overwriting our new inode with zeroes. So we | |
731 | * take alloc_sem to prevent it. | |
732 | */ | |
733 | down_read(&grp->alloc_sem); | |
955ce5f5 | 734 | ext4_lock_group(sb, group); |
39341867 AK |
735 | if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) { |
736 | /* not a free inode */ | |
737 | retval = 1; | |
738 | goto err_ret; | |
739 | } | |
740 | ino++; | |
741 | if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || | |
742 | ino > EXT4_INODES_PER_GROUP(sb)) { | |
955ce5f5 | 743 | ext4_unlock_group(sb, group); |
bfff6873 | 744 | up_read(&grp->alloc_sem); |
12062ddd | 745 | ext4_error(sb, "reserved inode or inode > inodes count - " |
39341867 AK |
746 | "block_group = %u, inode=%lu", group, |
747 | ino + group * EXT4_INODES_PER_GROUP(sb)); | |
748 | return 1; | |
749 | } | |
750 | /* If we didn't allocate from within the initialized part of the inode | |
751 | * table then we need to initialize up to this inode. */ | |
752 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { | |
753 | ||
754 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
755 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
756 | /* When marking the block group with | |
757 | * ~EXT4_BG_INODE_UNINIT we don't want to depend | |
758 | * on the value of bg_itable_unused even though | |
759 | * mke2fs could have initialized the same for us. | |
760 | * Instead we calculated the value below | |
761 | */ | |
762 | ||
763 | free = 0; | |
764 | } else { | |
765 | free = EXT4_INODES_PER_GROUP(sb) - | |
766 | ext4_itable_unused_count(sb, gdp); | |
767 | } | |
768 | ||
769 | /* | |
770 | * Check the relative inode number against the last used | |
771 | * relative inode number in this group. if it is greater | |
772 | * we need to update the bg_itable_unused count | |
773 | * | |
774 | */ | |
775 | if (ino > free) | |
776 | ext4_itable_unused_set(sb, gdp, | |
777 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
778 | } | |
779 | count = ext4_free_inodes_count(sb, gdp) - 1; | |
780 | ext4_free_inodes_set(sb, gdp, count); | |
781 | if (S_ISDIR(mode)) { | |
782 | count = ext4_used_dirs_count(sb, gdp) + 1; | |
783 | ext4_used_dirs_set(sb, gdp, count); | |
7d39db14 TT |
784 | if (sbi->s_log_groups_per_flex) { |
785 | ext4_group_t f = ext4_flex_group(sbi, group); | |
786 | ||
c4caae25 | 787 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); |
7d39db14 | 788 | } |
39341867 AK |
789 | } |
790 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); | |
791 | err_ret: | |
955ce5f5 | 792 | ext4_unlock_group(sb, group); |
bfff6873 | 793 | up_read(&grp->alloc_sem); |
39341867 AK |
794 | return retval; |
795 | } | |
796 | ||
ac27a0ec DK |
797 | /* |
798 | * There are two policies for allocating an inode. If the new inode is | |
799 | * a directory, then a forward search is made for a block group with both | |
800 | * free space and a low directory-to-inode ratio; if that fails, then of | |
801 | * the groups with above-average free space, that group with the fewest | |
802 | * directories already is chosen. | |
803 | * | |
804 | * For other inodes, search forward from the parent directory's block | |
805 | * group to find a free inode. | |
806 | */ | |
f157a4aa | 807 | struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode, |
11013911 | 808 | const struct qstr *qstr, __u32 goal) |
ac27a0ec DK |
809 | { |
810 | struct super_block *sb; | |
3300beda AK |
811 | struct buffer_head *inode_bitmap_bh = NULL; |
812 | struct buffer_head *group_desc_bh; | |
8df9675f | 813 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 814 | unsigned long ino = 0; |
af5bc92d TT |
815 | struct inode *inode; |
816 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
817 | struct ext4_inode_info *ei; |
818 | struct ext4_sb_info *sbi; | |
39341867 | 819 | int ret2, err = 0; |
ac27a0ec | 820 | struct inode *ret; |
2aa9fc4c | 821 | ext4_group_t i; |
2842c3b5 | 822 | static int once = 1; |
772cb7c8 | 823 | ext4_group_t flex_group; |
ac27a0ec DK |
824 | |
825 | /* Cannot create files in a deleted directory */ | |
826 | if (!dir || !dir->i_nlink) | |
827 | return ERR_PTR(-EPERM); | |
828 | ||
829 | sb = dir->i_sb; | |
8df9675f | 830 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 831 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
832 | inode = new_inode(sb); |
833 | if (!inode) | |
834 | return ERR_PTR(-ENOMEM); | |
617ba13b | 835 | ei = EXT4_I(inode); |
617ba13b | 836 | sbi = EXT4_SB(sb); |
772cb7c8 | 837 | |
11013911 AD |
838 | if (!goal) |
839 | goal = sbi->s_inode_goal; | |
840 | ||
e6462869 | 841 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
842 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
843 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
844 | ret2 = 0; | |
845 | goto got_group; | |
846 | } | |
847 | ||
a4912123 | 848 | if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) { |
772cb7c8 | 849 | ret2 = find_group_flex(sb, dir, &group); |
05bf9e83 | 850 | if (ret2 == -1) { |
a4912123 | 851 | ret2 = find_group_other(sb, dir, &group, mode); |
6b82f3cb | 852 | if (ret2 == 0 && once) { |
2842c3b5 | 853 | once = 0; |
05bf9e83 TT |
854 | printk(KERN_NOTICE "ext4: find_group_flex " |
855 | "failed, fallback succeeded dir %lu\n", | |
856 | dir->i_ino); | |
6b82f3cb | 857 | } |
05bf9e83 | 858 | } |
772cb7c8 JS |
859 | goto got_group; |
860 | } | |
861 | ||
ac27a0ec | 862 | if (S_ISDIR(mode)) { |
af5bc92d | 863 | if (test_opt(sb, OLDALLOC)) |
2aa9fc4c | 864 | ret2 = find_group_dir(sb, dir, &group); |
ac27a0ec | 865 | else |
f157a4aa | 866 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); |
ac27a0ec | 867 | } else |
a4912123 | 868 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 869 | |
772cb7c8 | 870 | got_group: |
a4912123 | 871 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 872 | err = -ENOSPC; |
2aa9fc4c | 873 | if (ret2 == -1) |
ac27a0ec DK |
874 | goto out; |
875 | ||
11013911 | 876 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
877 | err = -EIO; |
878 | ||
3300beda | 879 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec DK |
880 | if (!gdp) |
881 | goto fail; | |
882 | ||
3300beda AK |
883 | brelse(inode_bitmap_bh); |
884 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
885 | if (!inode_bitmap_bh) | |
ac27a0ec DK |
886 | goto fail; |
887 | ||
ac27a0ec | 888 | repeat_in_this_group: |
617ba13b | 889 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
890 | inode_bitmap_bh->b_data, |
891 | EXT4_INODES_PER_GROUP(sb), ino); | |
892 | ||
617ba13b | 893 | if (ino < EXT4_INODES_PER_GROUP(sb)) { |
ac27a0ec | 894 | |
3300beda AK |
895 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
896 | err = ext4_journal_get_write_access(handle, | |
897 | inode_bitmap_bh); | |
ac27a0ec DK |
898 | if (err) |
899 | goto fail; | |
900 | ||
39341867 AK |
901 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
902 | err = ext4_journal_get_write_access(handle, | |
903 | group_desc_bh); | |
904 | if (err) | |
905 | goto fail; | |
906 | if (!ext4_claim_inode(sb, inode_bitmap_bh, | |
907 | ino, group, mode)) { | |
ac27a0ec | 908 | /* we won it */ |
3300beda | 909 | BUFFER_TRACE(inode_bitmap_bh, |
0390131b FM |
910 | "call ext4_handle_dirty_metadata"); |
911 | err = ext4_handle_dirty_metadata(handle, | |
73b50c1c | 912 | NULL, |
3300beda | 913 | inode_bitmap_bh); |
ac27a0ec DK |
914 | if (err) |
915 | goto fail; | |
39341867 AK |
916 | /* zero bit is inode number 1*/ |
917 | ino++; | |
ac27a0ec DK |
918 | goto got; |
919 | } | |
920 | /* we lost it */ | |
3300beda | 921 | ext4_handle_release_buffer(handle, inode_bitmap_bh); |
39341867 | 922 | ext4_handle_release_buffer(handle, group_desc_bh); |
ac27a0ec | 923 | |
617ba13b | 924 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
925 | goto repeat_in_this_group; |
926 | } | |
927 | ||
928 | /* | |
929 | * This case is possible in concurrent environment. It is very | |
930 | * rare. We cannot repeat the find_group_xxx() call because | |
931 | * that will simply return the same blockgroup, because the | |
932 | * group descriptor metadata has not yet been updated. | |
933 | * So we just go onto the next blockgroup. | |
934 | */ | |
8df9675f | 935 | if (++group == ngroups) |
ac27a0ec DK |
936 | group = 0; |
937 | } | |
938 | err = -ENOSPC; | |
939 | goto out; | |
940 | ||
941 | got: | |
717d50e4 AD |
942 | /* We may have to initialize the block bitmap if it isn't already */ |
943 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && | |
944 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
3300beda | 945 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 946 | |
3300beda AK |
947 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
948 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); | |
949 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 950 | if (err) { |
3300beda | 951 | brelse(block_bitmap_bh); |
717d50e4 AD |
952 | goto fail; |
953 | } | |
954 | ||
fd034a84 TT |
955 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
956 | err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); | |
957 | brelse(block_bitmap_bh); | |
958 | ||
717d50e4 | 959 | /* recheck and clear flag under lock if we still need to */ |
fd034a84 | 960 | ext4_lock_group(sb, group); |
717d50e4 | 961 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { |
3300beda | 962 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
021b65bb | 963 | ext4_free_group_clusters_set(sb, gdp, |
cff1dfd7 | 964 | ext4_free_clusters_after_init(sb, group, gdp)); |
23712a9c FB |
965 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, |
966 | gdp); | |
717d50e4 | 967 | } |
955ce5f5 | 968 | ext4_unlock_group(sb, group); |
717d50e4 | 969 | |
717d50e4 AD |
970 | if (err) |
971 | goto fail; | |
972 | } | |
3300beda AK |
973 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
974 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
39341867 AK |
975 | if (err) |
976 | goto fail; | |
ac27a0ec DK |
977 | |
978 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
979 | if (S_ISDIR(mode)) | |
980 | percpu_counter_inc(&sbi->s_dirs_counter); | |
a0375156 | 981 | ext4_mark_super_dirty(sb); |
ac27a0ec | 982 | |
772cb7c8 JS |
983 | if (sbi->s_log_groups_per_flex) { |
984 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 985 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 JS |
986 | } |
987 | ||
b10b8520 DM |
988 | if (test_opt(sb, GRPID)) { |
989 | inode->i_mode = mode; | |
990 | inode->i_uid = current_fsuid(); | |
ac27a0ec | 991 | inode->i_gid = dir->i_gid; |
ac27a0ec | 992 | } else |
b10b8520 | 993 | inode_init_owner(inode, dir, mode); |
ac27a0ec | 994 | |
717d50e4 | 995 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
996 | /* This is the optimal IO size (for stat), not the fs block size */ |
997 | inode->i_blocks = 0; | |
ef7f3835 KS |
998 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
999 | ext4_current_time(inode); | |
ac27a0ec DK |
1000 | |
1001 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
1002 | ei->i_dir_start_lookup = 0; | |
1003 | ei->i_disksize = 0; | |
1004 | ||
42bf0383 | 1005 | /* |
2dc6b0d4 DG |
1006 | * Don't inherit extent flag from directory, amongst others. We set |
1007 | * extent flag on newly created directory and file only if -o extent | |
1008 | * mount option is specified | |
42bf0383 | 1009 | */ |
2dc6b0d4 DG |
1010 | ei->i_flags = |
1011 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 1012 | ei->i_file_acl = 0; |
ac27a0ec | 1013 | ei->i_dtime = 0; |
ac27a0ec | 1014 | ei->i_block_group = group; |
a4912123 | 1015 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1016 | |
617ba13b | 1017 | ext4_set_inode_flags(inode); |
ac27a0ec | 1018 | if (IS_DIRSYNC(inode)) |
0390131b | 1019 | ext4_handle_sync(handle); |
6b38e842 AV |
1020 | if (insert_inode_locked(inode) < 0) { |
1021 | err = -EINVAL; | |
1022 | goto fail_drop; | |
1023 | } | |
ac27a0ec DK |
1024 | spin_lock(&sbi->s_next_gen_lock); |
1025 | inode->i_generation = sbi->s_next_generation++; | |
1026 | spin_unlock(&sbi->s_next_gen_lock); | |
1027 | ||
353eb83c | 1028 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
19f5fb7a | 1029 | ext4_set_inode_state(inode, EXT4_STATE_NEW); |
ef7f3835 KS |
1030 | |
1031 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
1032 | |
1033 | ret = inode; | |
871a2931 | 1034 | dquot_initialize(inode); |
63936dda CH |
1035 | err = dquot_alloc_inode(inode); |
1036 | if (err) | |
ac27a0ec | 1037 | goto fail_drop; |
ac27a0ec | 1038 | |
617ba13b | 1039 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
1040 | if (err) |
1041 | goto fail_free_drop; | |
1042 | ||
2a7dba39 | 1043 | err = ext4_init_security(handle, inode, dir, qstr); |
ac27a0ec DK |
1044 | if (err) |
1045 | goto fail_free_drop; | |
1046 | ||
83982b6f | 1047 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
e4079a11 | 1048 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1049 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1050 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1051 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1052 | } |
a86c6181 | 1053 | } |
ac27a0ec | 1054 | |
688f869c TT |
1055 | if (ext4_handle_valid(handle)) { |
1056 | ei->i_sync_tid = handle->h_transaction->t_tid; | |
1057 | ei->i_datasync_tid = handle->h_transaction->t_tid; | |
1058 | } | |
1059 | ||
8753e88f AK |
1060 | err = ext4_mark_inode_dirty(handle, inode); |
1061 | if (err) { | |
1062 | ext4_std_error(sb, err); | |
1063 | goto fail_free_drop; | |
1064 | } | |
1065 | ||
617ba13b | 1066 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1067 | trace_ext4_allocate_inode(inode, dir, mode); |
ac27a0ec DK |
1068 | goto really_out; |
1069 | fail: | |
617ba13b | 1070 | ext4_std_error(sb, err); |
ac27a0ec DK |
1071 | out: |
1072 | iput(inode); | |
1073 | ret = ERR_PTR(err); | |
1074 | really_out: | |
3300beda | 1075 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1076 | return ret; |
1077 | ||
1078 | fail_free_drop: | |
63936dda | 1079 | dquot_free_inode(inode); |
ac27a0ec DK |
1080 | |
1081 | fail_drop: | |
9f754758 | 1082 | dquot_drop(inode); |
ac27a0ec DK |
1083 | inode->i_flags |= S_NOQUOTA; |
1084 | inode->i_nlink = 0; | |
6b38e842 | 1085 | unlock_new_inode(inode); |
ac27a0ec | 1086 | iput(inode); |
3300beda | 1087 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1088 | return ERR_PTR(err); |
1089 | } | |
1090 | ||
1091 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1092 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1093 | { |
617ba13b | 1094 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1095 | ext4_group_t block_group; |
ac27a0ec | 1096 | int bit; |
1d1fe1ee | 1097 | struct buffer_head *bitmap_bh; |
ac27a0ec | 1098 | struct inode *inode = NULL; |
1d1fe1ee | 1099 | long err = -EIO; |
ac27a0ec DK |
1100 | |
1101 | /* Error cases - e2fsck has already cleaned up for us */ | |
1102 | if (ino > max_ino) { | |
12062ddd | 1103 | ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 1104 | goto error; |
ac27a0ec DK |
1105 | } |
1106 | ||
617ba13b MC |
1107 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1108 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1109 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 1110 | if (!bitmap_bh) { |
12062ddd | 1111 | ext4_warning(sb, "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 1112 | goto error; |
ac27a0ec DK |
1113 | } |
1114 | ||
1115 | /* Having the inode bit set should be a 100% indicator that this | |
1116 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1117 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1118 | */ | |
1d1fe1ee DH |
1119 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1120 | goto bad_orphan; | |
1121 | ||
1122 | inode = ext4_iget(sb, ino); | |
1123 | if (IS_ERR(inode)) | |
1124 | goto iget_failed; | |
1125 | ||
91ef4caf DG |
1126 | /* |
1127 | * If the orphans has i_nlinks > 0 then it should be able to be | |
1128 | * truncated, otherwise it won't be removed from the orphan list | |
1129 | * during processing and an infinite loop will result. | |
1130 | */ | |
1131 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
1132 | goto bad_orphan; | |
1133 | ||
1d1fe1ee DH |
1134 | if (NEXT_ORPHAN(inode) > max_ino) |
1135 | goto bad_orphan; | |
1136 | brelse(bitmap_bh); | |
1137 | return inode; | |
1138 | ||
1139 | iget_failed: | |
1140 | err = PTR_ERR(inode); | |
1141 | inode = NULL; | |
1142 | bad_orphan: | |
12062ddd | 1143 | ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino); |
1d1fe1ee DH |
1144 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", |
1145 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
1146 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
1147 | printk(KERN_NOTICE "inode=%p\n", inode); | |
1148 | if (inode) { | |
1149 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
1150 | is_bad_inode(inode)); | |
1151 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
1152 | NEXT_ORPHAN(inode)); | |
1153 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
91ef4caf | 1154 | printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink); |
ac27a0ec | 1155 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1156 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1157 | inode->i_blocks = 0; |
1158 | iput(inode); | |
ac27a0ec | 1159 | } |
ac27a0ec | 1160 | brelse(bitmap_bh); |
1d1fe1ee DH |
1161 | error: |
1162 | return ERR_PTR(err); | |
ac27a0ec DK |
1163 | } |
1164 | ||
af5bc92d | 1165 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1166 | { |
1167 | unsigned long desc_count; | |
617ba13b | 1168 | struct ext4_group_desc *gdp; |
8df9675f | 1169 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1170 | #ifdef EXT4FS_DEBUG |
1171 | struct ext4_super_block *es; | |
ac27a0ec DK |
1172 | unsigned long bitmap_count, x; |
1173 | struct buffer_head *bitmap_bh = NULL; | |
1174 | ||
617ba13b | 1175 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1176 | desc_count = 0; |
1177 | bitmap_count = 0; | |
1178 | gdp = NULL; | |
8df9675f | 1179 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1180 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1181 | if (!gdp) |
1182 | continue; | |
560671a0 | 1183 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1184 | brelse(bitmap_bh); |
e29d1cde | 1185 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
1186 | if (!bitmap_bh) |
1187 | continue; | |
1188 | ||
617ba13b | 1189 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
c549a95d | 1190 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1191 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1192 | bitmap_count += x; |
1193 | } | |
1194 | brelse(bitmap_bh); | |
4776004f TT |
1195 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1196 | "stored = %u, computed = %lu, %lu\n", | |
1197 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1198 | return desc_count; |
1199 | #else | |
1200 | desc_count = 0; | |
8df9675f | 1201 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1202 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1203 | if (!gdp) |
1204 | continue; | |
560671a0 | 1205 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1206 | cond_resched(); |
1207 | } | |
1208 | return desc_count; | |
1209 | #endif | |
1210 | } | |
1211 | ||
1212 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1213 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1214 | { |
1215 | unsigned long count = 0; | |
8df9675f | 1216 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1217 | |
8df9675f | 1218 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1219 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1220 | if (!gdp) |
1221 | continue; | |
560671a0 | 1222 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1223 | } |
1224 | return count; | |
1225 | } | |
bfff6873 LC |
1226 | |
1227 | /* | |
1228 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1229 | * inode table. Must be called without any spinlock held. The only place | |
1230 | * where it is called from on active part of filesystem is ext4lazyinit | |
1231 | * thread, so we do not need any special locks, however we have to prevent | |
1232 | * inode allocation from the current group, so we take alloc_sem lock, to | |
1233 | * block ext4_claim_inode until we are finished. | |
1234 | */ | |
1235 | extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, | |
1236 | int barrier) | |
1237 | { | |
1238 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1239 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1240 | struct ext4_group_desc *gdp = NULL; | |
1241 | struct buffer_head *group_desc_bh; | |
1242 | handle_t *handle; | |
1243 | ext4_fsblk_t blk; | |
1244 | int num, ret = 0, used_blks = 0; | |
bfff6873 LC |
1245 | |
1246 | /* This should not happen, but just to be sure check this */ | |
1247 | if (sb->s_flags & MS_RDONLY) { | |
1248 | ret = 1; | |
1249 | goto out; | |
1250 | } | |
1251 | ||
1252 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1253 | if (!gdp) | |
1254 | goto out; | |
1255 | ||
1256 | /* | |
1257 | * We do not need to lock this, because we are the only one | |
1258 | * handling this flag. | |
1259 | */ | |
1260 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1261 | goto out; | |
1262 | ||
1263 | handle = ext4_journal_start_sb(sb, 1); | |
1264 | if (IS_ERR(handle)) { | |
1265 | ret = PTR_ERR(handle); | |
1266 | goto out; | |
1267 | } | |
1268 | ||
1269 | down_write(&grp->alloc_sem); | |
1270 | /* | |
1271 | * If inode bitmap was already initialized there may be some | |
1272 | * used inodes so we need to skip blocks with used inodes in | |
1273 | * inode table. | |
1274 | */ | |
1275 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1276 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1277 | ext4_itable_unused_count(sb, gdp)), | |
1278 | sbi->s_inodes_per_block); | |
1279 | ||
857ac889 LC |
1280 | if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) { |
1281 | ext4_error(sb, "Something is wrong with group %u\n" | |
1282 | "Used itable blocks: %d" | |
1283 | "itable unused count: %u\n", | |
1284 | group, used_blks, | |
1285 | ext4_itable_unused_count(sb, gdp)); | |
1286 | ret = 1; | |
33853a0d | 1287 | goto err_out; |
857ac889 LC |
1288 | } |
1289 | ||
bfff6873 LC |
1290 | blk = ext4_inode_table(sb, gdp) + used_blks; |
1291 | num = sbi->s_itb_per_group - used_blks; | |
1292 | ||
1293 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1294 | ret = ext4_journal_get_write_access(handle, | |
1295 | group_desc_bh); | |
1296 | if (ret) | |
1297 | goto err_out; | |
1298 | ||
bfff6873 LC |
1299 | /* |
1300 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1301 | * flag anyway, because obviously, when it is full it does not need | |
1302 | * further zeroing. | |
1303 | */ | |
1304 | if (unlikely(num == 0)) | |
1305 | goto skip_zeroout; | |
1306 | ||
1307 | ext4_debug("going to zero out inode table in group %d\n", | |
1308 | group); | |
a107e5a3 | 1309 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); |
bfff6873 LC |
1310 | if (ret < 0) |
1311 | goto err_out; | |
a107e5a3 TT |
1312 | if (barrier) |
1313 | blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); | |
bfff6873 LC |
1314 | |
1315 | skip_zeroout: | |
1316 | ext4_lock_group(sb, group); | |
1317 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
1318 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); | |
1319 | ext4_unlock_group(sb, group); | |
1320 | ||
1321 | BUFFER_TRACE(group_desc_bh, | |
1322 | "call ext4_handle_dirty_metadata"); | |
1323 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1324 | group_desc_bh); | |
1325 | ||
1326 | err_out: | |
1327 | up_write(&grp->alloc_sem); | |
1328 | ext4_journal_stop(handle); | |
1329 | out: | |
1330 | return ret; | |
1331 | } |