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)) { | |
a9df9a49 | 79 | ext4_error(sb, __func__, "Checksum bad for group %u", |
717d50e4 | 80 | block_group); |
560671a0 AK |
81 | ext4_free_blks_set(sb, gdp, 0); |
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); | |
648f5879 | 89 | 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 ES |
110 | return NULL; |
111 | bitmap_blk = ext4_inode_bitmap(sb, desc); | |
112 | bh = sb_getblk(sb, bitmap_blk); | |
113 | if (unlikely(!bh)) { | |
114 | ext4_error(sb, __func__, | |
115 | "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 | } | |
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); |
2ccb5fb9 AK |
138 | if (buffer_uptodate(bh)) { |
139 | /* | |
140 | * if not uninit if bh is uptodate, | |
141 | * bitmap is also uptodate | |
142 | */ | |
143 | set_bitmap_uptodate(bh); | |
144 | unlock_buffer(bh); | |
145 | return bh; | |
146 | } | |
147 | /* | |
148 | * submit the buffer_head for read. We can | |
149 | * safely mark the bitmap as uptodate now. | |
150 | * We do it here so the bitmap uptodate bit | |
151 | * get set with buffer lock held. | |
152 | */ | |
153 | set_bitmap_uptodate(bh); | |
e29d1cde ES |
154 | if (bh_submit_read(bh) < 0) { |
155 | put_bh(bh); | |
156 | ext4_error(sb, __func__, | |
ac27a0ec | 157 | "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 | */ | |
a269eb18 | 220 | vfs_dq_init(inode); |
617ba13b | 221 | ext4_xattr_delete_inode(handle, inode); |
a269eb18 JK |
222 | vfs_dq_free_inode(inode); |
223 | vfs_dq_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 */ | |
af5bc92d | 228 | 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)) { | |
af5bc92d TT |
232 | ext4_error(sb, "ext4_free_inode", |
233 | "reserved or nonexistent inode %lu", ino); | |
ac27a0ec DK |
234 | goto error_return; |
235 | } | |
617ba13b MC |
236 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
237 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 238 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec DK |
239 | if (!bitmap_bh) |
240 | goto error_return; | |
241 | ||
242 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 243 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
244 | if (fatal) |
245 | goto error_return; | |
246 | ||
247 | /* Ok, now we can actually update the inode bitmaps.. */ | |
955ce5f5 AK |
248 | cleared = ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group), |
249 | bit, bitmap_bh->b_data); | |
7ce9d5d1 | 250 | if (!cleared) |
af5bc92d TT |
251 | ext4_error(sb, "ext4_free_inode", |
252 | "bit already cleared for inode %lu", ino); | |
ac27a0ec | 253 | else { |
af5bc92d | 254 | gdp = ext4_get_group_desc(sb, block_group, &bh2); |
ac27a0ec DK |
255 | |
256 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 257 | fatal = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec DK |
258 | if (fatal) goto error_return; |
259 | ||
260 | if (gdp) { | |
955ce5f5 | 261 | ext4_lock_group(sb, block_group); |
560671a0 AK |
262 | count = ext4_free_inodes_count(sb, gdp) + 1; |
263 | ext4_free_inodes_set(sb, gdp, count); | |
264 | if (is_directory) { | |
265 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
266 | ext4_used_dirs_set(sb, gdp, count); | |
7d39db14 TT |
267 | if (sbi->s_log_groups_per_flex) { |
268 | ext4_group_t f; | |
269 | ||
270 | f = ext4_flex_group(sbi, block_group); | |
271 | atomic_dec(&sbi->s_flex_groups[f].free_inodes); | |
272 | } | |
273 | ||
560671a0 | 274 | } |
717d50e4 AD |
275 | gdp->bg_checksum = ext4_group_desc_csum(sbi, |
276 | block_group, gdp); | |
955ce5f5 | 277 | ext4_unlock_group(sb, block_group); |
ac27a0ec DK |
278 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
279 | if (is_directory) | |
280 | percpu_counter_dec(&sbi->s_dirs_counter); | |
281 | ||
772cb7c8 | 282 | if (sbi->s_log_groups_per_flex) { |
9f24e420 TT |
283 | ext4_group_t f; |
284 | ||
285 | f = ext4_flex_group(sbi, block_group); | |
286 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); | |
772cb7c8 | 287 | } |
ac27a0ec | 288 | } |
0390131b FM |
289 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
290 | err = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
ac27a0ec DK |
291 | if (!fatal) fatal = err; |
292 | } | |
0390131b FM |
293 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); |
294 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
ac27a0ec DK |
295 | if (!fatal) |
296 | fatal = err; | |
297 | sb->s_dirt = 1; | |
298 | error_return: | |
299 | brelse(bitmap_bh); | |
617ba13b | 300 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
301 | } |
302 | ||
303 | /* | |
304 | * There are two policies for allocating an inode. If the new inode is | |
305 | * a directory, then a forward search is made for a block group with both | |
306 | * free space and a low directory-to-inode ratio; if that fails, then of | |
307 | * the groups with above-average free space, that group with the fewest | |
308 | * directories already is chosen. | |
309 | * | |
310 | * For other inodes, search forward from the parent directory\'s block | |
311 | * group to find a free inode. | |
312 | */ | |
2aa9fc4c AM |
313 | static int find_group_dir(struct super_block *sb, struct inode *parent, |
314 | ext4_group_t *best_group) | |
ac27a0ec | 315 | { |
8df9675f | 316 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 317 | unsigned int freei, avefreei; |
617ba13b | 318 | struct ext4_group_desc *desc, *best_desc = NULL; |
2aa9fc4c AM |
319 | ext4_group_t group; |
320 | int ret = -1; | |
ac27a0ec | 321 | |
617ba13b | 322 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
323 | avefreei = freei / ngroups; |
324 | ||
325 | for (group = 0; group < ngroups; group++) { | |
af5bc92d | 326 | desc = ext4_get_group_desc(sb, group, NULL); |
560671a0 | 327 | if (!desc || !ext4_free_inodes_count(sb, desc)) |
ac27a0ec | 328 | continue; |
560671a0 | 329 | if (ext4_free_inodes_count(sb, desc) < avefreei) |
ac27a0ec DK |
330 | continue; |
331 | if (!best_desc || | |
560671a0 AK |
332 | (ext4_free_blks_count(sb, desc) > |
333 | ext4_free_blks_count(sb, best_desc))) { | |
2aa9fc4c | 334 | *best_group = group; |
ac27a0ec | 335 | best_desc = desc; |
2aa9fc4c | 336 | ret = 0; |
ac27a0ec DK |
337 | } |
338 | } | |
2aa9fc4c | 339 | return ret; |
ac27a0ec DK |
340 | } |
341 | ||
772cb7c8 JS |
342 | #define free_block_ratio 10 |
343 | ||
344 | static int find_group_flex(struct super_block *sb, struct inode *parent, | |
345 | ext4_group_t *best_group) | |
346 | { | |
347 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
348 | struct ext4_group_desc *desc; | |
772cb7c8 JS |
349 | struct flex_groups *flex_group = sbi->s_flex_groups; |
350 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; | |
351 | ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group); | |
8df9675f | 352 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
772cb7c8 JS |
353 | int flex_size = ext4_flex_bg_size(sbi); |
354 | ext4_group_t best_flex = parent_fbg_group; | |
355 | int blocks_per_flex = sbi->s_blocks_per_group * flex_size; | |
356 | int flexbg_free_blocks; | |
357 | int flex_freeb_ratio; | |
358 | ext4_group_t n_fbg_groups; | |
359 | ext4_group_t i; | |
360 | ||
8df9675f | 361 | n_fbg_groups = (ngroups + flex_size - 1) >> |
772cb7c8 JS |
362 | sbi->s_log_groups_per_flex; |
363 | ||
364 | find_close_to_parent: | |
9f24e420 | 365 | flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks); |
772cb7c8 | 366 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; |
9f24e420 | 367 | if (atomic_read(&flex_group[best_flex].free_inodes) && |
772cb7c8 JS |
368 | flex_freeb_ratio > free_block_ratio) |
369 | goto found_flexbg; | |
370 | ||
371 | if (best_flex && best_flex == parent_fbg_group) { | |
372 | best_flex--; | |
373 | goto find_close_to_parent; | |
374 | } | |
375 | ||
376 | for (i = 0; i < n_fbg_groups; i++) { | |
377 | if (i == parent_fbg_group || i == parent_fbg_group - 1) | |
378 | continue; | |
379 | ||
9f24e420 | 380 | flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks); |
772cb7c8 JS |
381 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; |
382 | ||
383 | if (flex_freeb_ratio > free_block_ratio && | |
9f24e420 | 384 | (atomic_read(&flex_group[i].free_inodes))) { |
772cb7c8 JS |
385 | best_flex = i; |
386 | goto found_flexbg; | |
387 | } | |
388 | ||
9f24e420 TT |
389 | if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) || |
390 | ((atomic_read(&flex_group[i].free_blocks) > | |
391 | atomic_read(&flex_group[best_flex].free_blocks)) && | |
392 | atomic_read(&flex_group[i].free_inodes))) | |
772cb7c8 JS |
393 | best_flex = i; |
394 | } | |
395 | ||
9f24e420 TT |
396 | if (!atomic_read(&flex_group[best_flex].free_inodes) || |
397 | !atomic_read(&flex_group[best_flex].free_blocks)) | |
772cb7c8 JS |
398 | return -1; |
399 | ||
400 | found_flexbg: | |
401 | for (i = best_flex * flex_size; i < ngroups && | |
402 | i < (best_flex + 1) * flex_size; i++) { | |
88b6edd1 | 403 | desc = ext4_get_group_desc(sb, i, NULL); |
560671a0 | 404 | if (ext4_free_inodes_count(sb, desc)) { |
772cb7c8 JS |
405 | *best_group = i; |
406 | goto out; | |
407 | } | |
408 | } | |
409 | ||
410 | return -1; | |
411 | out: | |
412 | return 0; | |
413 | } | |
414 | ||
a4912123 TT |
415 | struct orlov_stats { |
416 | __u32 free_inodes; | |
417 | __u32 free_blocks; | |
418 | __u32 used_dirs; | |
419 | }; | |
420 | ||
421 | /* | |
422 | * Helper function for Orlov's allocator; returns critical information | |
423 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
424 | * is a block group number; otherwise it is flex_bg number. | |
425 | */ | |
426 | void get_orlov_stats(struct super_block *sb, ext4_group_t g, | |
427 | int flex_size, struct orlov_stats *stats) | |
428 | { | |
429 | struct ext4_group_desc *desc; | |
7d39db14 | 430 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 431 | |
7d39db14 TT |
432 | if (flex_size > 1) { |
433 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
434 | stats->free_blocks = atomic_read(&flex_group[g].free_blocks); | |
435 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); | |
436 | return; | |
437 | } | |
a4912123 | 438 | |
7d39db14 TT |
439 | desc = ext4_get_group_desc(sb, g, NULL); |
440 | if (desc) { | |
441 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
442 | stats->free_blocks = ext4_free_blks_count(sb, desc); | |
443 | stats->used_dirs = ext4_used_dirs_count(sb, desc); | |
444 | } else { | |
445 | stats->free_inodes = 0; | |
446 | stats->free_blocks = 0; | |
447 | stats->used_dirs = 0; | |
a4912123 TT |
448 | } |
449 | } | |
450 | ||
ac27a0ec DK |
451 | /* |
452 | * Orlov's allocator for directories. | |
453 | * | |
454 | * We always try to spread first-level directories. | |
455 | * | |
456 | * If there are blockgroups with both free inodes and free blocks counts | |
457 | * not worse than average we return one with smallest directory count. | |
458 | * Otherwise we simply return a random group. | |
459 | * | |
460 | * For the rest rules look so: | |
461 | * | |
462 | * It's OK to put directory into a group unless | |
463 | * it has too many directories already (max_dirs) or | |
464 | * it has too few free inodes left (min_inodes) or | |
465 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 466 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
467 | * conditions we search cyclically through the rest. If none |
468 | * of the groups look good we just look for a group with more | |
469 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
470 | */ |
471 | ||
2aa9fc4c | 472 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
a4912123 | 473 | ext4_group_t *group, int mode) |
ac27a0ec | 474 | { |
fd2d4291 | 475 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 476 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 477 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 478 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 479 | unsigned int freei, avefreei; |
617ba13b | 480 | ext4_fsblk_t freeb, avefreeb; |
ac27a0ec | 481 | unsigned int ndirs; |
a4912123 | 482 | int max_dirs, min_inodes; |
617ba13b | 483 | ext4_grpblk_t min_blocks; |
8df9675f | 484 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 485 | struct ext4_group_desc *desc; |
a4912123 TT |
486 | struct orlov_stats stats; |
487 | int flex_size = ext4_flex_bg_size(sbi); | |
488 | ||
8df9675f | 489 | ngroups = real_ngroups; |
a4912123 | 490 | if (flex_size > 1) { |
8df9675f | 491 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
492 | sbi->s_log_groups_per_flex; |
493 | parent_group >>= sbi->s_log_groups_per_flex; | |
494 | } | |
ac27a0ec DK |
495 | |
496 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
497 | avefreei = freei / ngroups; | |
498 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3a5b2ecd | 499 | avefreeb = freeb; |
f4e5bc24 | 500 | do_div(avefreeb, ngroups); |
ac27a0ec DK |
501 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
502 | ||
a4912123 TT |
503 | if (S_ISDIR(mode) && |
504 | ((parent == sb->s_root->d_inode) || | |
505 | (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) { | |
ac27a0ec | 506 | int best_ndir = inodes_per_group; |
2aa9fc4c | 507 | int ret = -1; |
ac27a0ec | 508 | |
2aa9fc4c AM |
509 | get_random_bytes(&grp, sizeof(grp)); |
510 | parent_group = (unsigned)grp % ngroups; | |
ac27a0ec | 511 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
512 | g = (parent_group + i) % ngroups; |
513 | get_orlov_stats(sb, g, flex_size, &stats); | |
514 | if (!stats.free_inodes) | |
ac27a0ec | 515 | continue; |
a4912123 | 516 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 517 | continue; |
a4912123 | 518 | if (stats.free_inodes < avefreei) |
ac27a0ec | 519 | continue; |
a4912123 | 520 | if (stats.free_blocks < avefreeb) |
ac27a0ec | 521 | continue; |
a4912123 | 522 | grp = g; |
2aa9fc4c | 523 | ret = 0; |
a4912123 TT |
524 | best_ndir = stats.used_dirs; |
525 | } | |
526 | if (ret) | |
527 | goto fallback; | |
528 | found_flex_bg: | |
529 | if (flex_size == 1) { | |
530 | *group = grp; | |
531 | return 0; | |
532 | } | |
533 | ||
534 | /* | |
535 | * We pack inodes at the beginning of the flexgroup's | |
536 | * inode tables. Block allocation decisions will do | |
537 | * something similar, although regular files will | |
538 | * start at 2nd block group of the flexgroup. See | |
539 | * ext4_ext_find_goal() and ext4_find_near(). | |
540 | */ | |
541 | grp *= flex_size; | |
542 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 543 | if (grp+i >= real_ngroups) |
a4912123 TT |
544 | break; |
545 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
546 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
547 | *group = grp+i; | |
548 | return 0; | |
549 | } | |
ac27a0ec | 550 | } |
ac27a0ec DK |
551 | goto fallback; |
552 | } | |
553 | ||
ac27a0ec | 554 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
555 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
556 | if (min_inodes < 1) | |
557 | min_inodes = 1; | |
558 | min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4; | |
559 | ||
560 | /* | |
561 | * Start looking in the flex group where we last allocated an | |
562 | * inode for this parent directory | |
563 | */ | |
564 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
565 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
566 | if (flex_size > 1) | |
567 | parent_group >>= sbi->s_log_groups_per_flex; | |
568 | } | |
ac27a0ec DK |
569 | |
570 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
571 | grp = (parent_group + i) % ngroups; |
572 | get_orlov_stats(sb, grp, flex_size, &stats); | |
573 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 574 | continue; |
a4912123 | 575 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 576 | continue; |
a4912123 | 577 | if (stats.free_blocks < min_blocks) |
ac27a0ec | 578 | continue; |
a4912123 | 579 | goto found_flex_bg; |
ac27a0ec DK |
580 | } |
581 | ||
582 | fallback: | |
8df9675f | 583 | ngroups = real_ngroups; |
a4912123 | 584 | avefreei = freei / ngroups; |
b5451f7b | 585 | fallback_retry: |
a4912123 | 586 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 587 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
588 | grp = (parent_group + i) % ngroups; |
589 | desc = ext4_get_group_desc(sb, grp, NULL); | |
560671a0 | 590 | if (desc && ext4_free_inodes_count(sb, desc) && |
a4912123 TT |
591 | ext4_free_inodes_count(sb, desc) >= avefreei) { |
592 | *group = grp; | |
2aa9fc4c | 593 | return 0; |
a4912123 | 594 | } |
ac27a0ec DK |
595 | } |
596 | ||
597 | if (avefreei) { | |
598 | /* | |
599 | * The free-inodes counter is approximate, and for really small | |
600 | * filesystems the above test can fail to find any blockgroups | |
601 | */ | |
602 | avefreei = 0; | |
b5451f7b | 603 | goto fallback_retry; |
ac27a0ec DK |
604 | } |
605 | ||
606 | return -1; | |
607 | } | |
608 | ||
2aa9fc4c | 609 | static int find_group_other(struct super_block *sb, struct inode *parent, |
a4912123 | 610 | ext4_group_t *group, int mode) |
ac27a0ec | 611 | { |
fd2d4291 | 612 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 613 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 614 | struct ext4_group_desc *desc; |
a4912123 TT |
615 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
616 | ||
617 | /* | |
618 | * Try to place the inode is the same flex group as its | |
619 | * parent. If we can't find space, use the Orlov algorithm to | |
620 | * find another flex group, and store that information in the | |
621 | * parent directory's inode information so that use that flex | |
622 | * group for future allocations. | |
623 | */ | |
624 | if (flex_size > 1) { | |
625 | int retry = 0; | |
626 | ||
627 | try_again: | |
628 | parent_group &= ~(flex_size-1); | |
629 | last = parent_group + flex_size; | |
630 | if (last > ngroups) | |
631 | last = ngroups; | |
632 | for (i = parent_group; i < last; i++) { | |
633 | desc = ext4_get_group_desc(sb, i, NULL); | |
634 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
635 | *group = i; | |
636 | return 0; | |
637 | } | |
638 | } | |
639 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
640 | retry = 1; | |
641 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
642 | goto try_again; | |
643 | } | |
644 | /* | |
645 | * If this didn't work, use the Orlov search algorithm | |
646 | * to find a new flex group; we pass in the mode to | |
647 | * avoid the topdir algorithms. | |
648 | */ | |
649 | *group = parent_group + flex_size; | |
650 | if (*group > ngroups) | |
651 | *group = 0; | |
652 | return find_group_orlov(sb, parent, group, mode); | |
653 | } | |
ac27a0ec DK |
654 | |
655 | /* | |
656 | * Try to place the inode in its parent directory | |
657 | */ | |
2aa9fc4c AM |
658 | *group = parent_group; |
659 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
660 | if (desc && ext4_free_inodes_count(sb, desc) && |
661 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 662 | return 0; |
ac27a0ec DK |
663 | |
664 | /* | |
665 | * We're going to place this inode in a different blockgroup from its | |
666 | * parent. We want to cause files in a common directory to all land in | |
667 | * the same blockgroup. But we want files which are in a different | |
668 | * directory which shares a blockgroup with our parent to land in a | |
669 | * different blockgroup. | |
670 | * | |
671 | * So add our directory's i_ino into the starting point for the hash. | |
672 | */ | |
2aa9fc4c | 673 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
674 | |
675 | /* | |
676 | * Use a quadratic hash to find a group with a free inode and some free | |
677 | * blocks. | |
678 | */ | |
679 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
680 | *group += i; |
681 | if (*group >= ngroups) | |
682 | *group -= ngroups; | |
683 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
684 | if (desc && ext4_free_inodes_count(sb, desc) && |
685 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 686 | return 0; |
ac27a0ec DK |
687 | } |
688 | ||
689 | /* | |
690 | * That failed: try linear search for a free inode, even if that group | |
691 | * has no free blocks. | |
692 | */ | |
2aa9fc4c | 693 | *group = parent_group; |
ac27a0ec | 694 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
695 | if (++*group >= ngroups) |
696 | *group = 0; | |
697 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 698 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 699 | return 0; |
ac27a0ec DK |
700 | } |
701 | ||
702 | return -1; | |
703 | } | |
704 | ||
39341867 AK |
705 | /* |
706 | * claim the inode from the inode bitmap. If the group | |
955ce5f5 | 707 | * is uninit we need to take the groups's ext4_group_lock |
39341867 AK |
708 | * and clear the uninit flag. The inode bitmap update |
709 | * and group desc uninit flag clear should be done | |
955ce5f5 | 710 | * after holding ext4_group_lock so that ext4_read_inode_bitmap |
39341867 AK |
711 | * doesn't race with the ext4_claim_inode |
712 | */ | |
713 | static int ext4_claim_inode(struct super_block *sb, | |
714 | struct buffer_head *inode_bitmap_bh, | |
715 | unsigned long ino, ext4_group_t group, int mode) | |
716 | { | |
717 | int free = 0, retval = 0, count; | |
718 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
719 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); | |
720 | ||
955ce5f5 | 721 | ext4_lock_group(sb, group); |
39341867 AK |
722 | if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) { |
723 | /* not a free inode */ | |
724 | retval = 1; | |
725 | goto err_ret; | |
726 | } | |
727 | ino++; | |
728 | if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || | |
729 | ino > EXT4_INODES_PER_GROUP(sb)) { | |
955ce5f5 | 730 | ext4_unlock_group(sb, group); |
39341867 AK |
731 | ext4_error(sb, __func__, |
732 | "reserved inode or inode > inodes count - " | |
733 | "block_group = %u, inode=%lu", group, | |
734 | ino + group * EXT4_INODES_PER_GROUP(sb)); | |
735 | return 1; | |
736 | } | |
737 | /* If we didn't allocate from within the initialized part of the inode | |
738 | * table then we need to initialize up to this inode. */ | |
739 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { | |
740 | ||
741 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
742 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
743 | /* When marking the block group with | |
744 | * ~EXT4_BG_INODE_UNINIT we don't want to depend | |
745 | * on the value of bg_itable_unused even though | |
746 | * mke2fs could have initialized the same for us. | |
747 | * Instead we calculated the value below | |
748 | */ | |
749 | ||
750 | free = 0; | |
751 | } else { | |
752 | free = EXT4_INODES_PER_GROUP(sb) - | |
753 | ext4_itable_unused_count(sb, gdp); | |
754 | } | |
755 | ||
756 | /* | |
757 | * Check the relative inode number against the last used | |
758 | * relative inode number in this group. if it is greater | |
759 | * we need to update the bg_itable_unused count | |
760 | * | |
761 | */ | |
762 | if (ino > free) | |
763 | ext4_itable_unused_set(sb, gdp, | |
764 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
765 | } | |
766 | count = ext4_free_inodes_count(sb, gdp) - 1; | |
767 | ext4_free_inodes_set(sb, gdp, count); | |
768 | if (S_ISDIR(mode)) { | |
769 | count = ext4_used_dirs_count(sb, gdp) + 1; | |
770 | ext4_used_dirs_set(sb, gdp, count); | |
7d39db14 TT |
771 | if (sbi->s_log_groups_per_flex) { |
772 | ext4_group_t f = ext4_flex_group(sbi, group); | |
773 | ||
774 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); | |
775 | } | |
39341867 AK |
776 | } |
777 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); | |
778 | err_ret: | |
955ce5f5 | 779 | ext4_unlock_group(sb, group); |
39341867 AK |
780 | return retval; |
781 | } | |
782 | ||
ac27a0ec DK |
783 | /* |
784 | * There are two policies for allocating an inode. If the new inode is | |
785 | * a directory, then a forward search is made for a block group with both | |
786 | * free space and a low directory-to-inode ratio; if that fails, then of | |
787 | * the groups with above-average free space, that group with the fewest | |
788 | * directories already is chosen. | |
789 | * | |
790 | * For other inodes, search forward from the parent directory's block | |
791 | * group to find a free inode. | |
792 | */ | |
af5bc92d | 793 | struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode) |
ac27a0ec DK |
794 | { |
795 | struct super_block *sb; | |
3300beda AK |
796 | struct buffer_head *inode_bitmap_bh = NULL; |
797 | struct buffer_head *group_desc_bh; | |
8df9675f | 798 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 799 | unsigned long ino = 0; |
af5bc92d TT |
800 | struct inode *inode; |
801 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
802 | struct ext4_inode_info *ei; |
803 | struct ext4_sb_info *sbi; | |
39341867 | 804 | int ret2, err = 0; |
ac27a0ec | 805 | struct inode *ret; |
2aa9fc4c AM |
806 | ext4_group_t i; |
807 | int free = 0; | |
2842c3b5 | 808 | static int once = 1; |
772cb7c8 | 809 | ext4_group_t flex_group; |
ac27a0ec DK |
810 | |
811 | /* Cannot create files in a deleted directory */ | |
812 | if (!dir || !dir->i_nlink) | |
813 | return ERR_PTR(-EPERM); | |
814 | ||
815 | sb = dir->i_sb; | |
8df9675f | 816 | ngroups = ext4_get_groups_count(sb); |
9bffad1e | 817 | trace_ext4_request_inode(dir, mode); |
ac27a0ec DK |
818 | inode = new_inode(sb); |
819 | if (!inode) | |
820 | return ERR_PTR(-ENOMEM); | |
617ba13b | 821 | ei = EXT4_I(inode); |
617ba13b | 822 | sbi = EXT4_SB(sb); |
772cb7c8 | 823 | |
a4912123 | 824 | if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) { |
772cb7c8 | 825 | ret2 = find_group_flex(sb, dir, &group); |
05bf9e83 | 826 | if (ret2 == -1) { |
a4912123 | 827 | ret2 = find_group_other(sb, dir, &group, mode); |
6b82f3cb | 828 | if (ret2 == 0 && once) { |
2842c3b5 | 829 | once = 0; |
05bf9e83 TT |
830 | printk(KERN_NOTICE "ext4: find_group_flex " |
831 | "failed, fallback succeeded dir %lu\n", | |
832 | dir->i_ino); | |
6b82f3cb | 833 | } |
05bf9e83 | 834 | } |
772cb7c8 JS |
835 | goto got_group; |
836 | } | |
837 | ||
ac27a0ec | 838 | if (S_ISDIR(mode)) { |
af5bc92d | 839 | if (test_opt(sb, OLDALLOC)) |
2aa9fc4c | 840 | ret2 = find_group_dir(sb, dir, &group); |
ac27a0ec | 841 | else |
a4912123 | 842 | ret2 = find_group_orlov(sb, dir, &group, mode); |
ac27a0ec | 843 | } else |
a4912123 | 844 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 845 | |
772cb7c8 | 846 | got_group: |
a4912123 | 847 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 848 | err = -ENOSPC; |
2aa9fc4c | 849 | if (ret2 == -1) |
ac27a0ec DK |
850 | goto out; |
851 | ||
8df9675f | 852 | for (i = 0; i < ngroups; i++) { |
ac27a0ec DK |
853 | err = -EIO; |
854 | ||
3300beda | 855 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec DK |
856 | if (!gdp) |
857 | goto fail; | |
858 | ||
3300beda AK |
859 | brelse(inode_bitmap_bh); |
860 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
861 | if (!inode_bitmap_bh) | |
ac27a0ec DK |
862 | goto fail; |
863 | ||
864 | ino = 0; | |
865 | ||
866 | repeat_in_this_group: | |
617ba13b | 867 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
868 | inode_bitmap_bh->b_data, |
869 | EXT4_INODES_PER_GROUP(sb), ino); | |
870 | ||
617ba13b | 871 | if (ino < EXT4_INODES_PER_GROUP(sb)) { |
ac27a0ec | 872 | |
3300beda AK |
873 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
874 | err = ext4_journal_get_write_access(handle, | |
875 | inode_bitmap_bh); | |
ac27a0ec DK |
876 | if (err) |
877 | goto fail; | |
878 | ||
39341867 AK |
879 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
880 | err = ext4_journal_get_write_access(handle, | |
881 | group_desc_bh); | |
882 | if (err) | |
883 | goto fail; | |
884 | if (!ext4_claim_inode(sb, inode_bitmap_bh, | |
885 | ino, group, mode)) { | |
ac27a0ec | 886 | /* we won it */ |
3300beda | 887 | BUFFER_TRACE(inode_bitmap_bh, |
0390131b FM |
888 | "call ext4_handle_dirty_metadata"); |
889 | err = ext4_handle_dirty_metadata(handle, | |
3300beda AK |
890 | inode, |
891 | inode_bitmap_bh); | |
ac27a0ec DK |
892 | if (err) |
893 | goto fail; | |
39341867 AK |
894 | /* zero bit is inode number 1*/ |
895 | ino++; | |
ac27a0ec DK |
896 | goto got; |
897 | } | |
898 | /* we lost it */ | |
3300beda | 899 | ext4_handle_release_buffer(handle, inode_bitmap_bh); |
39341867 | 900 | ext4_handle_release_buffer(handle, group_desc_bh); |
ac27a0ec | 901 | |
617ba13b | 902 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
903 | goto repeat_in_this_group; |
904 | } | |
905 | ||
906 | /* | |
907 | * This case is possible in concurrent environment. It is very | |
908 | * rare. We cannot repeat the find_group_xxx() call because | |
909 | * that will simply return the same blockgroup, because the | |
910 | * group descriptor metadata has not yet been updated. | |
911 | * So we just go onto the next blockgroup. | |
912 | */ | |
8df9675f | 913 | if (++group == ngroups) |
ac27a0ec DK |
914 | group = 0; |
915 | } | |
916 | err = -ENOSPC; | |
917 | goto out; | |
918 | ||
919 | got: | |
717d50e4 AD |
920 | /* We may have to initialize the block bitmap if it isn't already */ |
921 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && | |
922 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
3300beda | 923 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 924 | |
3300beda AK |
925 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
926 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); | |
927 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 928 | if (err) { |
3300beda | 929 | brelse(block_bitmap_bh); |
717d50e4 AD |
930 | goto fail; |
931 | } | |
932 | ||
933 | free = 0; | |
955ce5f5 | 934 | ext4_lock_group(sb, group); |
717d50e4 AD |
935 | /* recheck and clear flag under lock if we still need to */ |
936 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
717d50e4 | 937 | free = ext4_free_blocks_after_init(sb, group, gdp); |
3300beda | 938 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
560671a0 | 939 | ext4_free_blks_set(sb, gdp, free); |
23712a9c FB |
940 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, |
941 | gdp); | |
717d50e4 | 942 | } |
955ce5f5 | 943 | ext4_unlock_group(sb, group); |
717d50e4 AD |
944 | |
945 | /* Don't need to dirty bitmap block if we didn't change it */ | |
946 | if (free) { | |
3300beda | 947 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
0390131b | 948 | err = ext4_handle_dirty_metadata(handle, |
3300beda | 949 | NULL, block_bitmap_bh); |
717d50e4 AD |
950 | } |
951 | ||
3300beda | 952 | brelse(block_bitmap_bh); |
717d50e4 AD |
953 | if (err) |
954 | goto fail; | |
955 | } | |
3300beda AK |
956 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
957 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
39341867 AK |
958 | if (err) |
959 | goto fail; | |
ac27a0ec DK |
960 | |
961 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
962 | if (S_ISDIR(mode)) | |
963 | percpu_counter_inc(&sbi->s_dirs_counter); | |
964 | sb->s_dirt = 1; | |
965 | ||
772cb7c8 JS |
966 | if (sbi->s_log_groups_per_flex) { |
967 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 968 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 JS |
969 | } |
970 | ||
4c9c544e | 971 | inode->i_uid = current_fsuid(); |
af5bc92d | 972 | if (test_opt(sb, GRPID)) |
ac27a0ec DK |
973 | inode->i_gid = dir->i_gid; |
974 | else if (dir->i_mode & S_ISGID) { | |
975 | inode->i_gid = dir->i_gid; | |
976 | if (S_ISDIR(mode)) | |
977 | mode |= S_ISGID; | |
978 | } else | |
4c9c544e | 979 | inode->i_gid = current_fsgid(); |
ac27a0ec DK |
980 | inode->i_mode = mode; |
981 | ||
717d50e4 | 982 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
983 | /* This is the optimal IO size (for stat), not the fs block size */ |
984 | inode->i_blocks = 0; | |
ef7f3835 KS |
985 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
986 | ext4_current_time(inode); | |
ac27a0ec DK |
987 | |
988 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
989 | ei->i_dir_start_lookup = 0; | |
990 | ei->i_disksize = 0; | |
991 | ||
42bf0383 | 992 | /* |
2dc6b0d4 DG |
993 | * Don't inherit extent flag from directory, amongst others. We set |
994 | * extent flag on newly created directory and file only if -o extent | |
995 | * mount option is specified | |
42bf0383 | 996 | */ |
2dc6b0d4 DG |
997 | ei->i_flags = |
998 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 999 | ei->i_file_acl = 0; |
ac27a0ec | 1000 | ei->i_dtime = 0; |
ac27a0ec | 1001 | ei->i_block_group = group; |
a4912123 | 1002 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1003 | |
617ba13b | 1004 | ext4_set_inode_flags(inode); |
ac27a0ec | 1005 | if (IS_DIRSYNC(inode)) |
0390131b | 1006 | ext4_handle_sync(handle); |
6b38e842 AV |
1007 | if (insert_inode_locked(inode) < 0) { |
1008 | err = -EINVAL; | |
1009 | goto fail_drop; | |
1010 | } | |
ac27a0ec DK |
1011 | spin_lock(&sbi->s_next_gen_lock); |
1012 | inode->i_generation = sbi->s_next_generation++; | |
1013 | spin_unlock(&sbi->s_next_gen_lock); | |
1014 | ||
617ba13b | 1015 | ei->i_state = EXT4_STATE_NEW; |
ef7f3835 KS |
1016 | |
1017 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
1018 | |
1019 | ret = inode; | |
a269eb18 | 1020 | if (vfs_dq_alloc_inode(inode)) { |
ac27a0ec DK |
1021 | err = -EDQUOT; |
1022 | goto fail_drop; | |
1023 | } | |
1024 | ||
617ba13b | 1025 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
1026 | if (err) |
1027 | goto fail_free_drop; | |
1028 | ||
af5bc92d | 1029 | err = ext4_init_security(handle, inode, dir); |
ac27a0ec DK |
1030 | if (err) |
1031 | goto fail_free_drop; | |
1032 | ||
83982b6f | 1033 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
e4079a11 | 1034 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1035 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
42bf0383 AK |
1036 | EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL; |
1037 | ext4_ext_tree_init(handle, inode); | |
42bf0383 | 1038 | } |
a86c6181 | 1039 | } |
ac27a0ec | 1040 | |
8753e88f AK |
1041 | err = ext4_mark_inode_dirty(handle, inode); |
1042 | if (err) { | |
1043 | ext4_std_error(sb, err); | |
1044 | goto fail_free_drop; | |
1045 | } | |
1046 | ||
617ba13b | 1047 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
9bffad1e | 1048 | trace_ext4_allocate_inode(inode, dir, mode); |
ac27a0ec DK |
1049 | goto really_out; |
1050 | fail: | |
617ba13b | 1051 | ext4_std_error(sb, err); |
ac27a0ec DK |
1052 | out: |
1053 | iput(inode); | |
1054 | ret = ERR_PTR(err); | |
1055 | really_out: | |
3300beda | 1056 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1057 | return ret; |
1058 | ||
1059 | fail_free_drop: | |
a269eb18 | 1060 | vfs_dq_free_inode(inode); |
ac27a0ec DK |
1061 | |
1062 | fail_drop: | |
a269eb18 | 1063 | vfs_dq_drop(inode); |
ac27a0ec DK |
1064 | inode->i_flags |= S_NOQUOTA; |
1065 | inode->i_nlink = 0; | |
6b38e842 | 1066 | unlock_new_inode(inode); |
ac27a0ec | 1067 | iput(inode); |
3300beda | 1068 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1069 | return ERR_PTR(err); |
1070 | } | |
1071 | ||
1072 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1073 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1074 | { |
617ba13b | 1075 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1076 | ext4_group_t block_group; |
ac27a0ec | 1077 | int bit; |
1d1fe1ee | 1078 | struct buffer_head *bitmap_bh; |
ac27a0ec | 1079 | struct inode *inode = NULL; |
1d1fe1ee | 1080 | long err = -EIO; |
ac27a0ec DK |
1081 | |
1082 | /* Error cases - e2fsck has already cleaned up for us */ | |
1083 | if (ino > max_ino) { | |
46e665e9 | 1084 | ext4_warning(sb, __func__, |
ac27a0ec | 1085 | "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 1086 | goto error; |
ac27a0ec DK |
1087 | } |
1088 | ||
617ba13b MC |
1089 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1090 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1091 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 1092 | if (!bitmap_bh) { |
46e665e9 | 1093 | ext4_warning(sb, __func__, |
ac27a0ec | 1094 | "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 1095 | goto error; |
ac27a0ec DK |
1096 | } |
1097 | ||
1098 | /* Having the inode bit set should be a 100% indicator that this | |
1099 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1100 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1101 | */ | |
1d1fe1ee DH |
1102 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1103 | goto bad_orphan; | |
1104 | ||
1105 | inode = ext4_iget(sb, ino); | |
1106 | if (IS_ERR(inode)) | |
1107 | goto iget_failed; | |
1108 | ||
91ef4caf DG |
1109 | /* |
1110 | * If the orphans has i_nlinks > 0 then it should be able to be | |
1111 | * truncated, otherwise it won't be removed from the orphan list | |
1112 | * during processing and an infinite loop will result. | |
1113 | */ | |
1114 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
1115 | goto bad_orphan; | |
1116 | ||
1d1fe1ee DH |
1117 | if (NEXT_ORPHAN(inode) > max_ino) |
1118 | goto bad_orphan; | |
1119 | brelse(bitmap_bh); | |
1120 | return inode; | |
1121 | ||
1122 | iget_failed: | |
1123 | err = PTR_ERR(inode); | |
1124 | inode = NULL; | |
1125 | bad_orphan: | |
46e665e9 | 1126 | ext4_warning(sb, __func__, |
1d1fe1ee DH |
1127 | "bad orphan inode %lu! e2fsck was run?", ino); |
1128 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", | |
1129 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
1130 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
1131 | printk(KERN_NOTICE "inode=%p\n", inode); | |
1132 | if (inode) { | |
1133 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
1134 | is_bad_inode(inode)); | |
1135 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
1136 | NEXT_ORPHAN(inode)); | |
1137 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
91ef4caf | 1138 | printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink); |
ac27a0ec | 1139 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1140 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1141 | inode->i_blocks = 0; |
1142 | iput(inode); | |
ac27a0ec | 1143 | } |
ac27a0ec | 1144 | brelse(bitmap_bh); |
1d1fe1ee DH |
1145 | error: |
1146 | return ERR_PTR(err); | |
ac27a0ec DK |
1147 | } |
1148 | ||
af5bc92d | 1149 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1150 | { |
1151 | unsigned long desc_count; | |
617ba13b | 1152 | struct ext4_group_desc *gdp; |
8df9675f | 1153 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1154 | #ifdef EXT4FS_DEBUG |
1155 | struct ext4_super_block *es; | |
ac27a0ec DK |
1156 | unsigned long bitmap_count, x; |
1157 | struct buffer_head *bitmap_bh = NULL; | |
1158 | ||
617ba13b | 1159 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1160 | desc_count = 0; |
1161 | bitmap_count = 0; | |
1162 | gdp = NULL; | |
8df9675f | 1163 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1164 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1165 | if (!gdp) |
1166 | continue; | |
560671a0 | 1167 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1168 | brelse(bitmap_bh); |
e29d1cde | 1169 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
1170 | if (!bitmap_bh) |
1171 | continue; | |
1172 | ||
617ba13b | 1173 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
c549a95d | 1174 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
560671a0 | 1175 | i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1176 | bitmap_count += x; |
1177 | } | |
1178 | brelse(bitmap_bh); | |
4776004f TT |
1179 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1180 | "stored = %u, computed = %lu, %lu\n", | |
1181 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1182 | return desc_count; |
1183 | #else | |
1184 | desc_count = 0; | |
8df9675f | 1185 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1186 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1187 | if (!gdp) |
1188 | continue; | |
560671a0 | 1189 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1190 | cond_resched(); |
1191 | } | |
1192 | return desc_count; | |
1193 | #endif | |
1194 | } | |
1195 | ||
1196 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1197 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1198 | { |
1199 | unsigned long count = 0; | |
8df9675f | 1200 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1201 | |
8df9675f | 1202 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1203 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1204 | if (!gdp) |
1205 | continue; | |
560671a0 | 1206 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1207 | } |
1208 | return count; | |
1209 | } |