]>
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 | 109 | return NULL; |
bfff6873 | 110 | |
e29d1cde ES |
111 | bitmap_blk = ext4_inode_bitmap(sb, desc); |
112 | bh = sb_getblk(sb, bitmap_blk); | |
113 | if (unlikely(!bh)) { | |
12062ddd | 114 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 115 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
116 | block_group, bitmap_blk); |
117 | return NULL; | |
118 | } | |
2ccb5fb9 | 119 | if (bitmap_uptodate(bh)) |
e29d1cde ES |
120 | return bh; |
121 | ||
c806e68f | 122 | lock_buffer(bh); |
2ccb5fb9 AK |
123 | if (bitmap_uptodate(bh)) { |
124 | unlock_buffer(bh); | |
125 | return bh; | |
126 | } | |
bfff6873 | 127 | |
955ce5f5 | 128 | ext4_lock_group(sb, block_group); |
717d50e4 | 129 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
e29d1cde | 130 | ext4_init_inode_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 131 | set_bitmap_uptodate(bh); |
e29d1cde | 132 | set_buffer_uptodate(bh); |
955ce5f5 | 133 | ext4_unlock_group(sb, block_group); |
3300beda | 134 | unlock_buffer(bh); |
e29d1cde | 135 | return bh; |
717d50e4 | 136 | } |
955ce5f5 | 137 | ext4_unlock_group(sb, block_group); |
bfff6873 | 138 | |
2ccb5fb9 AK |
139 | if (buffer_uptodate(bh)) { |
140 | /* | |
141 | * if not uninit if bh is uptodate, | |
142 | * bitmap is also uptodate | |
143 | */ | |
144 | set_bitmap_uptodate(bh); | |
145 | unlock_buffer(bh); | |
146 | return bh; | |
147 | } | |
148 | /* | |
149 | * submit the buffer_head for read. We can | |
150 | * safely mark the bitmap as uptodate now. | |
151 | * We do it here so the bitmap uptodate bit | |
152 | * get set with buffer lock held. | |
153 | */ | |
154 | set_bitmap_uptodate(bh); | |
e29d1cde ES |
155 | if (bh_submit_read(bh) < 0) { |
156 | put_bh(bh); | |
12062ddd | 157 | ext4_error(sb, "Cannot read inode bitmap - " |
a9df9a49 | 158 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
159 | block_group, bitmap_blk); |
160 | return NULL; | |
161 | } | |
ac27a0ec DK |
162 | return bh; |
163 | } | |
164 | ||
165 | /* | |
166 | * NOTE! When we get the inode, we're the only people | |
167 | * that have access to it, and as such there are no | |
168 | * race conditions we have to worry about. The inode | |
169 | * is not on the hash-lists, and it cannot be reached | |
170 | * through the filesystem because the directory entry | |
171 | * has been deleted earlier. | |
172 | * | |
173 | * HOWEVER: we must make sure that we get no aliases, | |
174 | * which means that we have to call "clear_inode()" | |
175 | * _before_ we mark the inode not in use in the inode | |
176 | * bitmaps. Otherwise a newly created file might use | |
177 | * the same inode number (not actually the same pointer | |
178 | * though), and then we'd have two inodes sharing the | |
179 | * same inode number and space on the harddisk. | |
180 | */ | |
af5bc92d | 181 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 182 | { |
af5bc92d | 183 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
184 | int is_directory; |
185 | unsigned long ino; | |
186 | struct buffer_head *bitmap_bh = NULL; | |
187 | struct buffer_head *bh2; | |
fd2d4291 | 188 | ext4_group_t block_group; |
ac27a0ec | 189 | unsigned long bit; |
af5bc92d TT |
190 | struct ext4_group_desc *gdp; |
191 | struct ext4_super_block *es; | |
617ba13b | 192 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 193 | int fatal = 0, err, count, cleared; |
ac27a0ec DK |
194 | |
195 | if (atomic_read(&inode->i_count) > 1) { | |
4776004f TT |
196 | printk(KERN_ERR "ext4_free_inode: inode has count=%d\n", |
197 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
198 | return; |
199 | } | |
200 | if (inode->i_nlink) { | |
4776004f TT |
201 | printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n", |
202 | inode->i_nlink); | |
ac27a0ec DK |
203 | return; |
204 | } | |
205 | if (!sb) { | |
4776004f TT |
206 | printk(KERN_ERR "ext4_free_inode: inode on " |
207 | "nonexistent device\n"); | |
ac27a0ec DK |
208 | return; |
209 | } | |
617ba13b | 210 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
211 | |
212 | ino = inode->i_ino; | |
af5bc92d | 213 | ext4_debug("freeing inode %lu\n", ino); |
9bffad1e | 214 | trace_ext4_free_inode(inode); |
ac27a0ec DK |
215 | |
216 | /* | |
217 | * Note: we must free any quota before locking the superblock, | |
218 | * as writing the quota to disk may need the lock as well. | |
219 | */ | |
871a2931 | 220 | dquot_initialize(inode); |
617ba13b | 221 | ext4_xattr_delete_inode(handle, inode); |
63936dda | 222 | dquot_free_inode(inode); |
9f754758 | 223 | dquot_drop(inode); |
ac27a0ec DK |
224 | |
225 | is_directory = S_ISDIR(inode->i_mode); | |
226 | ||
227 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
0930fcc1 | 228 | ext4_clear_inode(inode); |
ac27a0ec | 229 | |
617ba13b MC |
230 | es = EXT4_SB(sb)->s_es; |
231 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
12062ddd | 232 | ext4_error(sb, "reserved or nonexistent inode %lu", ino); |
ac27a0ec DK |
233 | goto error_return; |
234 | } | |
617ba13b MC |
235 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
236 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 237 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec DK |
238 | if (!bitmap_bh) |
239 | goto error_return; | |
240 | ||
241 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 242 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
243 | if (fatal) |
244 | goto error_return; | |
245 | ||
d17413c0 DM |
246 | fatal = -ESRCH; |
247 | gdp = ext4_get_group_desc(sb, block_group, &bh2); | |
248 | if (gdp) { | |
ac27a0ec | 249 | BUFFER_TRACE(bh2, "get_write_access"); |
617ba13b | 250 | fatal = ext4_journal_get_write_access(handle, bh2); |
d17413c0 DM |
251 | } |
252 | ext4_lock_group(sb, block_group); | |
253 | cleared = ext4_clear_bit(bit, bitmap_bh->b_data); | |
254 | if (fatal || !cleared) { | |
255 | ext4_unlock_group(sb, block_group); | |
256 | goto out; | |
257 | } | |
7d39db14 | 258 | |
d17413c0 DM |
259 | count = ext4_free_inodes_count(sb, gdp) + 1; |
260 | ext4_free_inodes_set(sb, gdp, count); | |
261 | if (is_directory) { | |
262 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
263 | ext4_used_dirs_set(sb, gdp, count); | |
264 | percpu_counter_dec(&sbi->s_dirs_counter); | |
ac27a0ec | 265 | } |
d17413c0 DM |
266 | gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp); |
267 | ext4_unlock_group(sb, block_group); | |
ac27a0ec | 268 | |
d17413c0 DM |
269 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
270 | if (sbi->s_log_groups_per_flex) { | |
271 | ext4_group_t f = ext4_flex_group(sbi, block_group); | |
9f24e420 | 272 | |
d17413c0 DM |
273 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); |
274 | if (is_directory) | |
275 | atomic_dec(&sbi->s_flex_groups[f].used_dirs); | |
ac27a0ec | 276 | } |
d17413c0 DM |
277 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
278 | fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
279 | out: | |
280 | if (cleared) { | |
281 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); | |
282 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
283 | if (!fatal) | |
284 | fatal = err; | |
a0375156 | 285 | ext4_mark_super_dirty(sb); |
d17413c0 DM |
286 | } else |
287 | ext4_error(sb, "bit already cleared for inode %lu", ino); | |
288 | ||
ac27a0ec DK |
289 | error_return: |
290 | brelse(bitmap_bh); | |
617ba13b | 291 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
292 | } |
293 | ||
294 | /* | |
295 | * There are two policies for allocating an inode. If the new inode is | |
296 | * a directory, then a forward search is made for a block group with both | |
297 | * free space and a low directory-to-inode ratio; if that fails, then of | |
298 | * the groups with above-average free space, that group with the fewest | |
299 | * directories already is chosen. | |
300 | * | |
301 | * For other inodes, search forward from the parent directory\'s block | |
302 | * group to find a free inode. | |
303 | */ | |
2aa9fc4c AM |
304 | static int find_group_dir(struct super_block *sb, struct inode *parent, |
305 | ext4_group_t *best_group) | |
ac27a0ec | 306 | { |
8df9675f | 307 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 308 | unsigned int freei, avefreei; |
617ba13b | 309 | struct ext4_group_desc *desc, *best_desc = NULL; |
2aa9fc4c AM |
310 | ext4_group_t group; |
311 | int ret = -1; | |
ac27a0ec | 312 | |
617ba13b | 313 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
314 | avefreei = freei / ngroups; |
315 | ||
316 | for (group = 0; group < ngroups; group++) { | |
af5bc92d | 317 | desc = ext4_get_group_desc(sb, group, NULL); |
560671a0 | 318 | if (!desc || !ext4_free_inodes_count(sb, desc)) |
ac27a0ec | 319 | continue; |
560671a0 | 320 | if (ext4_free_inodes_count(sb, desc) < avefreei) |
ac27a0ec DK |
321 | continue; |
322 | if (!best_desc || | |
560671a0 AK |
323 | (ext4_free_blks_count(sb, desc) > |
324 | ext4_free_blks_count(sb, best_desc))) { | |
2aa9fc4c | 325 | *best_group = group; |
ac27a0ec | 326 | best_desc = desc; |
2aa9fc4c | 327 | ret = 0; |
ac27a0ec DK |
328 | } |
329 | } | |
2aa9fc4c | 330 | return ret; |
ac27a0ec DK |
331 | } |
332 | ||
772cb7c8 JS |
333 | #define free_block_ratio 10 |
334 | ||
335 | static int find_group_flex(struct super_block *sb, struct inode *parent, | |
336 | ext4_group_t *best_group) | |
337 | { | |
338 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
339 | struct ext4_group_desc *desc; | |
772cb7c8 JS |
340 | struct flex_groups *flex_group = sbi->s_flex_groups; |
341 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; | |
342 | ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group); | |
8df9675f | 343 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
772cb7c8 JS |
344 | int flex_size = ext4_flex_bg_size(sbi); |
345 | ext4_group_t best_flex = parent_fbg_group; | |
346 | int blocks_per_flex = sbi->s_blocks_per_group * flex_size; | |
347 | int flexbg_free_blocks; | |
348 | int flex_freeb_ratio; | |
349 | ext4_group_t n_fbg_groups; | |
350 | ext4_group_t i; | |
351 | ||
8df9675f | 352 | n_fbg_groups = (ngroups + flex_size - 1) >> |
772cb7c8 JS |
353 | sbi->s_log_groups_per_flex; |
354 | ||
355 | find_close_to_parent: | |
9f24e420 | 356 | flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks); |
772cb7c8 | 357 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; |
9f24e420 | 358 | if (atomic_read(&flex_group[best_flex].free_inodes) && |
772cb7c8 JS |
359 | flex_freeb_ratio > free_block_ratio) |
360 | goto found_flexbg; | |
361 | ||
362 | if (best_flex && best_flex == parent_fbg_group) { | |
363 | best_flex--; | |
364 | goto find_close_to_parent; | |
365 | } | |
366 | ||
367 | for (i = 0; i < n_fbg_groups; i++) { | |
368 | if (i == parent_fbg_group || i == parent_fbg_group - 1) | |
369 | continue; | |
370 | ||
9f24e420 | 371 | flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks); |
772cb7c8 JS |
372 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; |
373 | ||
374 | if (flex_freeb_ratio > free_block_ratio && | |
9f24e420 | 375 | (atomic_read(&flex_group[i].free_inodes))) { |
772cb7c8 JS |
376 | best_flex = i; |
377 | goto found_flexbg; | |
378 | } | |
379 | ||
9f24e420 TT |
380 | if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) || |
381 | ((atomic_read(&flex_group[i].free_blocks) > | |
382 | atomic_read(&flex_group[best_flex].free_blocks)) && | |
383 | atomic_read(&flex_group[i].free_inodes))) | |
772cb7c8 JS |
384 | best_flex = i; |
385 | } | |
386 | ||
9f24e420 TT |
387 | if (!atomic_read(&flex_group[best_flex].free_inodes) || |
388 | !atomic_read(&flex_group[best_flex].free_blocks)) | |
772cb7c8 JS |
389 | return -1; |
390 | ||
391 | found_flexbg: | |
392 | for (i = best_flex * flex_size; i < ngroups && | |
393 | i < (best_flex + 1) * flex_size; i++) { | |
88b6edd1 | 394 | desc = ext4_get_group_desc(sb, i, NULL); |
560671a0 | 395 | if (ext4_free_inodes_count(sb, desc)) { |
772cb7c8 JS |
396 | *best_group = i; |
397 | goto out; | |
398 | } | |
399 | } | |
400 | ||
401 | return -1; | |
402 | out: | |
403 | return 0; | |
404 | } | |
405 | ||
a4912123 TT |
406 | struct orlov_stats { |
407 | __u32 free_inodes; | |
408 | __u32 free_blocks; | |
409 | __u32 used_dirs; | |
410 | }; | |
411 | ||
412 | /* | |
413 | * Helper function for Orlov's allocator; returns critical information | |
414 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
415 | * is a block group number; otherwise it is flex_bg number. | |
416 | */ | |
417 | void get_orlov_stats(struct super_block *sb, ext4_group_t g, | |
418 | int flex_size, struct orlov_stats *stats) | |
419 | { | |
420 | struct ext4_group_desc *desc; | |
7d39db14 | 421 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 422 | |
7d39db14 TT |
423 | if (flex_size > 1) { |
424 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
425 | stats->free_blocks = atomic_read(&flex_group[g].free_blocks); | |
426 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); | |
427 | return; | |
428 | } | |
a4912123 | 429 | |
7d39db14 TT |
430 | desc = ext4_get_group_desc(sb, g, NULL); |
431 | if (desc) { | |
432 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
433 | stats->free_blocks = ext4_free_blks_count(sb, desc); | |
434 | stats->used_dirs = ext4_used_dirs_count(sb, desc); | |
435 | } else { | |
436 | stats->free_inodes = 0; | |
437 | stats->free_blocks = 0; | |
438 | stats->used_dirs = 0; | |
a4912123 TT |
439 | } |
440 | } | |
441 | ||
ac27a0ec DK |
442 | /* |
443 | * Orlov's allocator for directories. | |
444 | * | |
445 | * We always try to spread first-level directories. | |
446 | * | |
447 | * If there are blockgroups with both free inodes and free blocks counts | |
448 | * not worse than average we return one with smallest directory count. | |
449 | * Otherwise we simply return a random group. | |
450 | * | |
451 | * For the rest rules look so: | |
452 | * | |
453 | * It's OK to put directory into a group unless | |
454 | * it has too many directories already (max_dirs) or | |
455 | * it has too few free inodes left (min_inodes) or | |
456 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 457 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
458 | * conditions we search cyclically through the rest. If none |
459 | * of the groups look good we just look for a group with more | |
460 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
461 | */ |
462 | ||
2aa9fc4c | 463 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
f157a4aa TT |
464 | ext4_group_t *group, int mode, |
465 | const struct qstr *qstr) | |
ac27a0ec | 466 | { |
fd2d4291 | 467 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 468 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 469 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 470 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 471 | unsigned int freei, avefreei; |
617ba13b | 472 | ext4_fsblk_t freeb, avefreeb; |
ac27a0ec | 473 | unsigned int ndirs; |
a4912123 | 474 | int max_dirs, min_inodes; |
617ba13b | 475 | ext4_grpblk_t min_blocks; |
8df9675f | 476 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 477 | struct ext4_group_desc *desc; |
a4912123 TT |
478 | struct orlov_stats stats; |
479 | int flex_size = ext4_flex_bg_size(sbi); | |
f157a4aa | 480 | struct dx_hash_info hinfo; |
a4912123 | 481 | |
8df9675f | 482 | ngroups = real_ngroups; |
a4912123 | 483 | if (flex_size > 1) { |
8df9675f | 484 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
485 | sbi->s_log_groups_per_flex; |
486 | parent_group >>= sbi->s_log_groups_per_flex; | |
487 | } | |
ac27a0ec DK |
488 | |
489 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
490 | avefreei = freei / ngroups; | |
491 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3a5b2ecd | 492 | avefreeb = freeb; |
f4e5bc24 | 493 | do_div(avefreeb, ngroups); |
ac27a0ec DK |
494 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
495 | ||
a4912123 TT |
496 | if (S_ISDIR(mode) && |
497 | ((parent == sb->s_root->d_inode) || | |
12e9b892 | 498 | (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { |
ac27a0ec | 499 | int best_ndir = inodes_per_group; |
2aa9fc4c | 500 | int ret = -1; |
ac27a0ec | 501 | |
f157a4aa TT |
502 | if (qstr) { |
503 | hinfo.hash_version = DX_HASH_HALF_MD4; | |
504 | hinfo.seed = sbi->s_hash_seed; | |
505 | ext4fs_dirhash(qstr->name, qstr->len, &hinfo); | |
506 | grp = hinfo.hash; | |
507 | } else | |
508 | get_random_bytes(&grp, sizeof(grp)); | |
2aa9fc4c | 509 | parent_group = (unsigned)grp % ngroups; |
ac27a0ec | 510 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
511 | g = (parent_group + i) % ngroups; |
512 | get_orlov_stats(sb, g, flex_size, &stats); | |
513 | if (!stats.free_inodes) | |
ac27a0ec | 514 | continue; |
a4912123 | 515 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 516 | continue; |
a4912123 | 517 | if (stats.free_inodes < avefreei) |
ac27a0ec | 518 | continue; |
a4912123 | 519 | if (stats.free_blocks < avefreeb) |
ac27a0ec | 520 | continue; |
a4912123 | 521 | grp = g; |
2aa9fc4c | 522 | ret = 0; |
a4912123 TT |
523 | best_ndir = stats.used_dirs; |
524 | } | |
525 | if (ret) | |
526 | goto fallback; | |
527 | found_flex_bg: | |
528 | if (flex_size == 1) { | |
529 | *group = grp; | |
530 | return 0; | |
531 | } | |
532 | ||
533 | /* | |
534 | * We pack inodes at the beginning of the flexgroup's | |
535 | * inode tables. Block allocation decisions will do | |
536 | * something similar, although regular files will | |
537 | * start at 2nd block group of the flexgroup. See | |
538 | * ext4_ext_find_goal() and ext4_find_near(). | |
539 | */ | |
540 | grp *= flex_size; | |
541 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 542 | if (grp+i >= real_ngroups) |
a4912123 TT |
543 | break; |
544 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
545 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
546 | *group = grp+i; | |
547 | return 0; | |
548 | } | |
ac27a0ec | 549 | } |
ac27a0ec DK |
550 | goto fallback; |
551 | } | |
552 | ||
ac27a0ec | 553 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
554 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
555 | if (min_inodes < 1) | |
556 | min_inodes = 1; | |
557 | min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4; | |
558 | ||
559 | /* | |
560 | * Start looking in the flex group where we last allocated an | |
561 | * inode for this parent directory | |
562 | */ | |
563 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
564 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
565 | if (flex_size > 1) | |
566 | parent_group >>= sbi->s_log_groups_per_flex; | |
567 | } | |
ac27a0ec DK |
568 | |
569 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
570 | grp = (parent_group + i) % ngroups; |
571 | get_orlov_stats(sb, grp, flex_size, &stats); | |
572 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 573 | continue; |
a4912123 | 574 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 575 | continue; |
a4912123 | 576 | if (stats.free_blocks < min_blocks) |
ac27a0ec | 577 | continue; |
a4912123 | 578 | goto found_flex_bg; |
ac27a0ec DK |
579 | } |
580 | ||
581 | fallback: | |
8df9675f | 582 | ngroups = real_ngroups; |
a4912123 | 583 | avefreei = freei / ngroups; |
b5451f7b | 584 | fallback_retry: |
a4912123 | 585 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 586 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
587 | grp = (parent_group + i) % ngroups; |
588 | desc = ext4_get_group_desc(sb, grp, NULL); | |
560671a0 | 589 | if (desc && ext4_free_inodes_count(sb, desc) && |
a4912123 TT |
590 | ext4_free_inodes_count(sb, desc) >= avefreei) { |
591 | *group = grp; | |
2aa9fc4c | 592 | return 0; |
a4912123 | 593 | } |
ac27a0ec DK |
594 | } |
595 | ||
596 | if (avefreei) { | |
597 | /* | |
598 | * The free-inodes counter is approximate, and for really small | |
599 | * filesystems the above test can fail to find any blockgroups | |
600 | */ | |
601 | avefreei = 0; | |
b5451f7b | 602 | goto fallback_retry; |
ac27a0ec DK |
603 | } |
604 | ||
605 | return -1; | |
606 | } | |
607 | ||
2aa9fc4c | 608 | static int find_group_other(struct super_block *sb, struct inode *parent, |
a4912123 | 609 | ext4_group_t *group, int mode) |
ac27a0ec | 610 | { |
fd2d4291 | 611 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 612 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 613 | struct ext4_group_desc *desc; |
a4912123 TT |
614 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
615 | ||
616 | /* | |
617 | * Try to place the inode is the same flex group as its | |
618 | * parent. If we can't find space, use the Orlov algorithm to | |
619 | * find another flex group, and store that information in the | |
620 | * parent directory's inode information so that use that flex | |
621 | * group for future allocations. | |
622 | */ | |
623 | if (flex_size > 1) { | |
624 | int retry = 0; | |
625 | ||
626 | try_again: | |
627 | parent_group &= ~(flex_size-1); | |
628 | last = parent_group + flex_size; | |
629 | if (last > ngroups) | |
630 | last = ngroups; | |
631 | for (i = parent_group; i < last; i++) { | |
632 | desc = ext4_get_group_desc(sb, i, NULL); | |
633 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
634 | *group = i; | |
635 | return 0; | |
636 | } | |
637 | } | |
638 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
639 | retry = 1; | |
640 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
641 | goto try_again; | |
642 | } | |
643 | /* | |
644 | * If this didn't work, use the Orlov search algorithm | |
645 | * to find a new flex group; we pass in the mode to | |
646 | * avoid the topdir algorithms. | |
647 | */ | |
648 | *group = parent_group + flex_size; | |
649 | if (*group > ngroups) | |
650 | *group = 0; | |
f157a4aa | 651 | return find_group_orlov(sb, parent, group, mode, 0); |
a4912123 | 652 | } |
ac27a0ec DK |
653 | |
654 | /* | |
655 | * Try to place the inode in its parent directory | |
656 | */ | |
2aa9fc4c AM |
657 | *group = parent_group; |
658 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
659 | if (desc && ext4_free_inodes_count(sb, desc) && |
660 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 661 | return 0; |
ac27a0ec DK |
662 | |
663 | /* | |
664 | * We're going to place this inode in a different blockgroup from its | |
665 | * parent. We want to cause files in a common directory to all land in | |
666 | * the same blockgroup. But we want files which are in a different | |
667 | * directory which shares a blockgroup with our parent to land in a | |
668 | * different blockgroup. | |
669 | * | |
670 | * So add our directory's i_ino into the starting point for the hash. | |
671 | */ | |
2aa9fc4c | 672 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
673 | |
674 | /* | |
675 | * Use a quadratic hash to find a group with a free inode and some free | |
676 | * blocks. | |
677 | */ | |
678 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
679 | *group += i; |
680 | if (*group >= ngroups) | |
681 | *group -= ngroups; | |
682 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
683 | if (desc && ext4_free_inodes_count(sb, desc) && |
684 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 685 | return 0; |
ac27a0ec DK |
686 | } |
687 | ||
688 | /* | |
689 | * That failed: try linear search for a free inode, even if that group | |
690 | * has no free blocks. | |
691 | */ | |
2aa9fc4c | 692 | *group = parent_group; |
ac27a0ec | 693 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
694 | if (++*group >= ngroups) |
695 | *group = 0; | |
696 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 697 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 698 | return 0; |
ac27a0ec DK |
699 | } |
700 | ||
701 | return -1; | |
702 | } | |
703 | ||
39341867 AK |
704 | /* |
705 | * claim the inode from the inode bitmap. If the group | |
955ce5f5 | 706 | * is uninit we need to take the groups's ext4_group_lock |
39341867 AK |
707 | * and clear the uninit flag. The inode bitmap update |
708 | * and group desc uninit flag clear should be done | |
955ce5f5 | 709 | * after holding ext4_group_lock so that ext4_read_inode_bitmap |
39341867 AK |
710 | * doesn't race with the ext4_claim_inode |
711 | */ | |
712 | static int ext4_claim_inode(struct super_block *sb, | |
713 | struct buffer_head *inode_bitmap_bh, | |
714 | unsigned long ino, ext4_group_t group, int mode) | |
715 | { | |
716 | int free = 0, retval = 0, count; | |
717 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
bfff6873 | 718 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); |
39341867 AK |
719 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); |
720 | ||
bfff6873 LC |
721 | /* |
722 | * We have to be sure that new inode allocation does not race with | |
723 | * inode table initialization, because otherwise we may end up | |
724 | * allocating and writing new inode right before sb_issue_zeroout | |
725 | * takes place and overwriting our new inode with zeroes. So we | |
726 | * take alloc_sem to prevent it. | |
727 | */ | |
728 | down_read(&grp->alloc_sem); | |
955ce5f5 | 729 | ext4_lock_group(sb, group); |
39341867 AK |
730 | if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) { |
731 | /* not a free inode */ | |
732 | retval = 1; | |
733 | goto err_ret; | |
734 | } | |
735 | ino++; | |
736 | if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || | |
737 | ino > EXT4_INODES_PER_GROUP(sb)) { | |
955ce5f5 | 738 | ext4_unlock_group(sb, group); |
bfff6873 | 739 | up_read(&grp->alloc_sem); |
12062ddd | 740 | ext4_error(sb, "reserved inode or inode > inodes count - " |
39341867 AK |
741 | "block_group = %u, inode=%lu", group, |
742 | ino + group * EXT4_INODES_PER_GROUP(sb)); | |
743 | return 1; | |
744 | } | |
745 | /* If we didn't allocate from within the initialized part of the inode | |
746 | * table then we need to initialize up to this inode. */ | |
747 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { | |
748 | ||
749 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
750 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
751 | /* When marking the block group with | |
752 | * ~EXT4_BG_INODE_UNINIT we don't want to depend | |
753 | * on the value of bg_itable_unused even though | |
754 | * mke2fs could have initialized the same for us. | |
755 | * Instead we calculated the value below | |
756 | */ | |
757 | ||
758 | free = 0; | |
759 | } else { | |
760 | free = EXT4_INODES_PER_GROUP(sb) - | |
761 | ext4_itable_unused_count(sb, gdp); | |
762 | } | |
763 | ||
764 | /* | |
765 | * Check the relative inode number against the last used | |
766 | * relative inode number in this group. if it is greater | |
767 | * we need to update the bg_itable_unused count | |
768 | * | |
769 | */ | |
770 | if (ino > free) | |
771 | ext4_itable_unused_set(sb, gdp, | |
772 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
773 | } | |
774 | count = ext4_free_inodes_count(sb, gdp) - 1; | |
775 | ext4_free_inodes_set(sb, gdp, count); | |
776 | if (S_ISDIR(mode)) { | |
777 | count = ext4_used_dirs_count(sb, gdp) + 1; | |
778 | ext4_used_dirs_set(sb, gdp, count); | |
7d39db14 TT |
779 | if (sbi->s_log_groups_per_flex) { |
780 | ext4_group_t f = ext4_flex_group(sbi, group); | |
781 | ||
c4caae25 | 782 | atomic_inc(&sbi->s_flex_groups[f].used_dirs); |
7d39db14 | 783 | } |
39341867 AK |
784 | } |
785 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); | |
786 | err_ret: | |
955ce5f5 | 787 | ext4_unlock_group(sb, group); |
bfff6873 | 788 | up_read(&grp->alloc_sem); |
39341867 AK |
789 | return retval; |
790 | } | |
791 | ||
ac27a0ec DK |
792 | /* |
793 | * There are two policies for allocating an inode. If the new inode is | |
794 | * a directory, then a forward search is made for a block group with both | |
795 | * free space and a low directory-to-inode ratio; if that fails, then of | |
796 | * the groups with above-average free space, that group with the fewest | |
797 | * directories already is chosen. | |
798 | * | |
799 | * For other inodes, search forward from the parent directory's block | |
800 | * group to find a free inode. | |
801 | */ | |
f157a4aa | 802 | struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode, |
11013911 | 803 | const struct qstr *qstr, __u32 goal) |
ac27a0ec DK |
804 | { |
805 | struct super_block *sb; | |
3300beda AK |
806 | struct buffer_head *inode_bitmap_bh = NULL; |
807 | struct buffer_head *group_desc_bh; | |
8df9675f | 808 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 809 | unsigned long ino = 0; |
af5bc92d TT |
810 | struct inode *inode; |
811 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
812 | struct ext4_inode_info *ei; |
813 | struct ext4_sb_info *sbi; | |
39341867 | 814 | int ret2, err = 0; |
ac27a0ec | 815 | struct inode *ret; |
2aa9fc4c AM |
816 | ext4_group_t i; |
817 | int free = 0; | |
2842c3b5 | 818 | static int once = 1; |
772cb7c8 | 819 | ext4_group_t flex_group; |
ac27a0ec DK |
820 | |
821 | /* Cannot create files in a deleted directory */ | |
822 | if (!dir || !dir->i_nlink) | |
823 | return ERR_PTR(-EPERM); | |
824 | ||
825 | sb = dir->i_sb; | |
8df9675f | 826 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 827 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
828 | inode = new_inode(sb); |
829 | if (!inode) | |
830 | return ERR_PTR(-ENOMEM); | |
617ba13b | 831 | ei = EXT4_I(inode); |
617ba13b | 832 | sbi = EXT4_SB(sb); |
772cb7c8 | 833 | |
11013911 AD |
834 | if (!goal) |
835 | goal = sbi->s_inode_goal; | |
836 | ||
e6462869 | 837 | if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { |
11013911 AD |
838 | group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); |
839 | ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); | |
840 | ret2 = 0; | |
841 | goto got_group; | |
842 | } | |
843 | ||
a4912123 | 844 | if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) { |
772cb7c8 | 845 | ret2 = find_group_flex(sb, dir, &group); |
05bf9e83 | 846 | if (ret2 == -1) { |
a4912123 | 847 | ret2 = find_group_other(sb, dir, &group, mode); |
6b82f3cb | 848 | if (ret2 == 0 && once) { |
2842c3b5 | 849 | once = 0; |
05bf9e83 TT |
850 | printk(KERN_NOTICE "ext4: find_group_flex " |
851 | "failed, fallback succeeded dir %lu\n", | |
852 | dir->i_ino); | |
6b82f3cb | 853 | } |
05bf9e83 | 854 | } |
772cb7c8 JS |
855 | goto got_group; |
856 | } | |
857 | ||
ac27a0ec | 858 | if (S_ISDIR(mode)) { |
af5bc92d | 859 | if (test_opt(sb, OLDALLOC)) |
2aa9fc4c | 860 | ret2 = find_group_dir(sb, dir, &group); |
ac27a0ec | 861 | else |
f157a4aa | 862 | ret2 = find_group_orlov(sb, dir, &group, mode, qstr); |
ac27a0ec | 863 | } else |
a4912123 | 864 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 865 | |
772cb7c8 | 866 | got_group: |
a4912123 | 867 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 868 | err = -ENOSPC; |
2aa9fc4c | 869 | if (ret2 == -1) |
ac27a0ec DK |
870 | goto out; |
871 | ||
11013911 | 872 | for (i = 0; i < ngroups; i++, ino = 0) { |
ac27a0ec DK |
873 | err = -EIO; |
874 | ||
3300beda | 875 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec DK |
876 | if (!gdp) |
877 | goto fail; | |
878 | ||
3300beda AK |
879 | brelse(inode_bitmap_bh); |
880 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
881 | if (!inode_bitmap_bh) | |
ac27a0ec DK |
882 | goto fail; |
883 | ||
ac27a0ec | 884 | repeat_in_this_group: |
617ba13b | 885 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
886 | inode_bitmap_bh->b_data, |
887 | EXT4_INODES_PER_GROUP(sb), ino); | |
888 | ||
617ba13b | 889 | if (ino < EXT4_INODES_PER_GROUP(sb)) { |
ac27a0ec | 890 | |
3300beda AK |
891 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
892 | err = ext4_journal_get_write_access(handle, | |
893 | inode_bitmap_bh); | |
ac27a0ec DK |
894 | if (err) |
895 | goto fail; | |
896 | ||
39341867 AK |
897 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
898 | err = ext4_journal_get_write_access(handle, | |
899 | group_desc_bh); | |
900 | if (err) | |
901 | goto fail; | |
902 | if (!ext4_claim_inode(sb, inode_bitmap_bh, | |
903 | ino, group, mode)) { | |
ac27a0ec | 904 | /* we won it */ |
3300beda | 905 | BUFFER_TRACE(inode_bitmap_bh, |
0390131b FM |
906 | "call ext4_handle_dirty_metadata"); |
907 | err = ext4_handle_dirty_metadata(handle, | |
73b50c1c | 908 | NULL, |
3300beda | 909 | inode_bitmap_bh); |
ac27a0ec DK |
910 | if (err) |
911 | goto fail; | |
39341867 AK |
912 | /* zero bit is inode number 1*/ |
913 | ino++; | |
ac27a0ec DK |
914 | goto got; |
915 | } | |
916 | /* we lost it */ | |
3300beda | 917 | ext4_handle_release_buffer(handle, inode_bitmap_bh); |
39341867 | 918 | ext4_handle_release_buffer(handle, group_desc_bh); |
ac27a0ec | 919 | |
617ba13b | 920 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
921 | goto repeat_in_this_group; |
922 | } | |
923 | ||
924 | /* | |
925 | * This case is possible in concurrent environment. It is very | |
926 | * rare. We cannot repeat the find_group_xxx() call because | |
927 | * that will simply return the same blockgroup, because the | |
928 | * group descriptor metadata has not yet been updated. | |
929 | * So we just go onto the next blockgroup. | |
930 | */ | |
8df9675f | 931 | if (++group == ngroups) |
ac27a0ec DK |
932 | group = 0; |
933 | } | |
934 | err = -ENOSPC; | |
935 | goto out; | |
936 | ||
937 | got: | |
717d50e4 AD |
938 | /* We may have to initialize the block bitmap if it isn't already */ |
939 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && | |
940 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
3300beda | 941 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 942 | |
3300beda AK |
943 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
944 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); | |
945 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 946 | if (err) { |
3300beda | 947 | brelse(block_bitmap_bh); |
717d50e4 AD |
948 | goto fail; |
949 | } | |
950 | ||
951 | free = 0; | |
955ce5f5 | 952 | ext4_lock_group(sb, group); |
717d50e4 AD |
953 | /* recheck and clear flag under lock if we still need to */ |
954 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
717d50e4 | 955 | free = ext4_free_blocks_after_init(sb, group, gdp); |
3300beda | 956 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
560671a0 | 957 | ext4_free_blks_set(sb, gdp, free); |
23712a9c FB |
958 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, |
959 | gdp); | |
717d50e4 | 960 | } |
955ce5f5 | 961 | ext4_unlock_group(sb, group); |
717d50e4 AD |
962 | |
963 | /* Don't need to dirty bitmap block if we didn't change it */ | |
964 | if (free) { | |
3300beda | 965 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
0390131b | 966 | err = ext4_handle_dirty_metadata(handle, |
3300beda | 967 | NULL, block_bitmap_bh); |
717d50e4 AD |
968 | } |
969 | ||
3300beda | 970 | brelse(block_bitmap_bh); |
717d50e4 AD |
971 | if (err) |
972 | goto fail; | |
973 | } | |
3300beda AK |
974 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
975 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
39341867 AK |
976 | if (err) |
977 | goto fail; | |
ac27a0ec DK |
978 | |
979 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
980 | if (S_ISDIR(mode)) | |
981 | percpu_counter_inc(&sbi->s_dirs_counter); | |
a0375156 | 982 | ext4_mark_super_dirty(sb); |
ac27a0ec | 983 | |
772cb7c8 JS |
984 | if (sbi->s_log_groups_per_flex) { |
985 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 986 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 JS |
987 | } |
988 | ||
b10b8520 DM |
989 | if (test_opt(sb, GRPID)) { |
990 | inode->i_mode = mode; | |
991 | inode->i_uid = current_fsuid(); | |
ac27a0ec | 992 | inode->i_gid = dir->i_gid; |
ac27a0ec | 993 | } else |
b10b8520 | 994 | inode_init_owner(inode, dir, mode); |
ac27a0ec | 995 | |
717d50e4 | 996 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
997 | /* This is the optimal IO size (for stat), not the fs block size */ |
998 | inode->i_blocks = 0; | |
ef7f3835 KS |
999 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
1000 | ext4_current_time(inode); | |
ac27a0ec DK |
1001 | |
1002 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
1003 | ei->i_dir_start_lookup = 0; | |
1004 | ei->i_disksize = 0; | |
1005 | ||
42bf0383 | 1006 | /* |
2dc6b0d4 DG |
1007 | * Don't inherit extent flag from directory, amongst others. We set |
1008 | * extent flag on newly created directory and file only if -o extent | |
1009 | * mount option is specified | |
42bf0383 | 1010 | */ |
2dc6b0d4 DG |
1011 | ei->i_flags = |
1012 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 1013 | ei->i_file_acl = 0; |
ac27a0ec | 1014 | ei->i_dtime = 0; |
ac27a0ec | 1015 | ei->i_block_group = group; |
a4912123 | 1016 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1017 | |
617ba13b | 1018 | ext4_set_inode_flags(inode); |
ac27a0ec | 1019 | if (IS_DIRSYNC(inode)) |
0390131b | 1020 | ext4_handle_sync(handle); |
6b38e842 AV |
1021 | if (insert_inode_locked(inode) < 0) { |
1022 | err = -EINVAL; | |
1023 | goto fail_drop; | |
1024 | } | |
ac27a0ec DK |
1025 | spin_lock(&sbi->s_next_gen_lock); |
1026 | inode->i_generation = sbi->s_next_generation++; | |
1027 | spin_unlock(&sbi->s_next_gen_lock); | |
1028 | ||
19f5fb7a TT |
1029 | ei->i_state_flags = 0; |
1030 | ext4_set_inode_state(inode, EXT4_STATE_NEW); | |
ef7f3835 KS |
1031 | |
1032 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
1033 | |
1034 | ret = inode; | |
871a2931 | 1035 | dquot_initialize(inode); |
63936dda CH |
1036 | err = dquot_alloc_inode(inode); |
1037 | if (err) | |
ac27a0ec | 1038 | goto fail_drop; |
ac27a0ec | 1039 | |
617ba13b | 1040 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
1041 | if (err) |
1042 | goto fail_free_drop; | |
1043 | ||
af5bc92d | 1044 | err = ext4_init_security(handle, inode, dir); |
ac27a0ec DK |
1045 | if (err) |
1046 | goto fail_free_drop; | |
1047 | ||
83982b6f | 1048 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
e4079a11 | 1049 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1050 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
12e9b892 | 1051 | ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); |
42bf0383 | 1052 | ext4_ext_tree_init(handle, inode); |
42bf0383 | 1053 | } |
a86c6181 | 1054 | } |
ac27a0ec | 1055 | |
8753e88f AK |
1056 | err = ext4_mark_inode_dirty(handle, inode); |
1057 | if (err) { | |
1058 | ext4_std_error(sb, err); | |
1059 | goto fail_free_drop; | |
1060 | } | |
1061 | ||
617ba13b | 1062 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1063 | trace_ext4_allocate_inode(inode, dir, mode); |
ac27a0ec DK |
1064 | goto really_out; |
1065 | fail: | |
617ba13b | 1066 | ext4_std_error(sb, err); |
ac27a0ec DK |
1067 | out: |
1068 | iput(inode); | |
1069 | ret = ERR_PTR(err); | |
1070 | really_out: | |
3300beda | 1071 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1072 | return ret; |
1073 | ||
1074 | fail_free_drop: | |
63936dda | 1075 | dquot_free_inode(inode); |
ac27a0ec DK |
1076 | |
1077 | fail_drop: | |
9f754758 | 1078 | dquot_drop(inode); |
ac27a0ec DK |
1079 | inode->i_flags |= S_NOQUOTA; |
1080 | inode->i_nlink = 0; | |
6b38e842 | 1081 | unlock_new_inode(inode); |
ac27a0ec | 1082 | iput(inode); |
3300beda | 1083 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1084 | return ERR_PTR(err); |
1085 | } | |
1086 | ||
1087 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1088 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1089 | { |
617ba13b | 1090 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1091 | ext4_group_t block_group; |
ac27a0ec | 1092 | int bit; |
1d1fe1ee | 1093 | struct buffer_head *bitmap_bh; |
ac27a0ec | 1094 | struct inode *inode = NULL; |
1d1fe1ee | 1095 | long err = -EIO; |
ac27a0ec DK |
1096 | |
1097 | /* Error cases - e2fsck has already cleaned up for us */ | |
1098 | if (ino > max_ino) { | |
12062ddd | 1099 | ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 1100 | goto error; |
ac27a0ec DK |
1101 | } |
1102 | ||
617ba13b MC |
1103 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1104 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1105 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 1106 | if (!bitmap_bh) { |
12062ddd | 1107 | ext4_warning(sb, "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 1108 | goto error; |
ac27a0ec DK |
1109 | } |
1110 | ||
1111 | /* Having the inode bit set should be a 100% indicator that this | |
1112 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1113 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1114 | */ | |
1d1fe1ee DH |
1115 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1116 | goto bad_orphan; | |
1117 | ||
1118 | inode = ext4_iget(sb, ino); | |
1119 | if (IS_ERR(inode)) | |
1120 | goto iget_failed; | |
1121 | ||
91ef4caf DG |
1122 | /* |
1123 | * If the orphans has i_nlinks > 0 then it should be able to be | |
1124 | * truncated, otherwise it won't be removed from the orphan list | |
1125 | * during processing and an infinite loop will result. | |
1126 | */ | |
1127 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
1128 | goto bad_orphan; | |
1129 | ||
1d1fe1ee DH |
1130 | if (NEXT_ORPHAN(inode) > max_ino) |
1131 | goto bad_orphan; | |
1132 | brelse(bitmap_bh); | |
1133 | return inode; | |
1134 | ||
1135 | iget_failed: | |
1136 | err = PTR_ERR(inode); | |
1137 | inode = NULL; | |
1138 | bad_orphan: | |
12062ddd | 1139 | ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino); |
1d1fe1ee DH |
1140 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", |
1141 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
1142 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
1143 | printk(KERN_NOTICE "inode=%p\n", inode); | |
1144 | if (inode) { | |
1145 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
1146 | is_bad_inode(inode)); | |
1147 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
1148 | NEXT_ORPHAN(inode)); | |
1149 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
91ef4caf | 1150 | printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink); |
ac27a0ec | 1151 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1152 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1153 | inode->i_blocks = 0; |
1154 | iput(inode); | |
ac27a0ec | 1155 | } |
ac27a0ec | 1156 | brelse(bitmap_bh); |
1d1fe1ee DH |
1157 | error: |
1158 | return ERR_PTR(err); | |
ac27a0ec DK |
1159 | } |
1160 | ||
af5bc92d | 1161 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1162 | { |
1163 | unsigned long desc_count; | |
617ba13b | 1164 | struct ext4_group_desc *gdp; |
8df9675f | 1165 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1166 | #ifdef EXT4FS_DEBUG |
1167 | struct ext4_super_block *es; | |
ac27a0ec DK |
1168 | unsigned long bitmap_count, x; |
1169 | struct buffer_head *bitmap_bh = NULL; | |
1170 | ||
617ba13b | 1171 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1172 | desc_count = 0; |
1173 | bitmap_count = 0; | |
1174 | gdp = NULL; | |
8df9675f | 1175 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1176 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1177 | if (!gdp) |
1178 | continue; | |
560671a0 | 1179 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1180 | brelse(bitmap_bh); |
e29d1cde | 1181 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
1182 | if (!bitmap_bh) |
1183 | continue; | |
1184 | ||
617ba13b | 1185 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
c549a95d | 1186 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
785b4b3a | 1187 | (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1188 | bitmap_count += x; |
1189 | } | |
1190 | brelse(bitmap_bh); | |
4776004f TT |
1191 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1192 | "stored = %u, computed = %lu, %lu\n", | |
1193 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1194 | return desc_count; |
1195 | #else | |
1196 | desc_count = 0; | |
8df9675f | 1197 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1198 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1199 | if (!gdp) |
1200 | continue; | |
560671a0 | 1201 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1202 | cond_resched(); |
1203 | } | |
1204 | return desc_count; | |
1205 | #endif | |
1206 | } | |
1207 | ||
1208 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1209 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1210 | { |
1211 | unsigned long count = 0; | |
8df9675f | 1212 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1213 | |
8df9675f | 1214 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1215 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1216 | if (!gdp) |
1217 | continue; | |
560671a0 | 1218 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1219 | } |
1220 | return count; | |
1221 | } | |
bfff6873 LC |
1222 | |
1223 | /* | |
1224 | * Zeroes not yet zeroed inode table - just write zeroes through the whole | |
1225 | * inode table. Must be called without any spinlock held. The only place | |
1226 | * where it is called from on active part of filesystem is ext4lazyinit | |
1227 | * thread, so we do not need any special locks, however we have to prevent | |
1228 | * inode allocation from the current group, so we take alloc_sem lock, to | |
1229 | * block ext4_claim_inode until we are finished. | |
1230 | */ | |
1231 | extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, | |
1232 | int barrier) | |
1233 | { | |
1234 | struct ext4_group_info *grp = ext4_get_group_info(sb, group); | |
1235 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
1236 | struct ext4_group_desc *gdp = NULL; | |
1237 | struct buffer_head *group_desc_bh; | |
1238 | handle_t *handle; | |
1239 | ext4_fsblk_t blk; | |
1240 | int num, ret = 0, used_blks = 0; | |
1241 | unsigned long flags = BLKDEV_IFL_WAIT; | |
1242 | ||
1243 | /* This should not happen, but just to be sure check this */ | |
1244 | if (sb->s_flags & MS_RDONLY) { | |
1245 | ret = 1; | |
1246 | goto out; | |
1247 | } | |
1248 | ||
1249 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); | |
1250 | if (!gdp) | |
1251 | goto out; | |
1252 | ||
1253 | /* | |
1254 | * We do not need to lock this, because we are the only one | |
1255 | * handling this flag. | |
1256 | */ | |
1257 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) | |
1258 | goto out; | |
1259 | ||
1260 | handle = ext4_journal_start_sb(sb, 1); | |
1261 | if (IS_ERR(handle)) { | |
1262 | ret = PTR_ERR(handle); | |
1263 | goto out; | |
1264 | } | |
1265 | ||
1266 | down_write(&grp->alloc_sem); | |
1267 | /* | |
1268 | * If inode bitmap was already initialized there may be some | |
1269 | * used inodes so we need to skip blocks with used inodes in | |
1270 | * inode table. | |
1271 | */ | |
1272 | if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) | |
1273 | used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - | |
1274 | ext4_itable_unused_count(sb, gdp)), | |
1275 | sbi->s_inodes_per_block); | |
1276 | ||
1277 | blk = ext4_inode_table(sb, gdp) + used_blks; | |
1278 | num = sbi->s_itb_per_group - used_blks; | |
1279 | ||
1280 | BUFFER_TRACE(group_desc_bh, "get_write_access"); | |
1281 | ret = ext4_journal_get_write_access(handle, | |
1282 | group_desc_bh); | |
1283 | if (ret) | |
1284 | goto err_out; | |
1285 | ||
1286 | if (unlikely(num > EXT4_INODES_PER_GROUP(sb))) { | |
1287 | ext4_error(sb, "Something is wrong with group %u\n" | |
1288 | "Used itable blocks: %d" | |
1289 | "Itable blocks per group: %lu\n", | |
1290 | group, used_blks, sbi->s_itb_per_group); | |
1291 | ret = 1; | |
1292 | goto err_out; | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * Skip zeroout if the inode table is full. But we set the ZEROED | |
1297 | * flag anyway, because obviously, when it is full it does not need | |
1298 | * further zeroing. | |
1299 | */ | |
1300 | if (unlikely(num == 0)) | |
1301 | goto skip_zeroout; | |
1302 | ||
1303 | ext4_debug("going to zero out inode table in group %d\n", | |
1304 | group); | |
1305 | if (barrier) | |
1306 | flags |= BLKDEV_IFL_BARRIER; | |
1307 | ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS, flags); | |
1308 | if (ret < 0) | |
1309 | goto err_out; | |
1310 | ||
1311 | skip_zeroout: | |
1312 | ext4_lock_group(sb, group); | |
1313 | gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); | |
1314 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); | |
1315 | ext4_unlock_group(sb, group); | |
1316 | ||
1317 | BUFFER_TRACE(group_desc_bh, | |
1318 | "call ext4_handle_dirty_metadata"); | |
1319 | ret = ext4_handle_dirty_metadata(handle, NULL, | |
1320 | group_desc_bh); | |
1321 | ||
1322 | err_out: | |
1323 | up_write(&grp->alloc_sem); | |
1324 | ext4_journal_stop(handle); | |
1325 | out: | |
1326 | return ret; | |
1327 | } |