Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/ialloc.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * BSD ufs-inspired inode and directory allocation by | |
10 | * Stephen Tweedie (sct@redhat.com), 1993 | |
11 | * Big-endian to little-endian byte-swapping/bitmaps by | |
12 | * David S. Miller (davem@caip.rutgers.edu), 1995 | |
13 | */ | |
14 | ||
15 | #include <linux/time.h> | |
16 | #include <linux/fs.h> | |
dab291af | 17 | #include <linux/jbd2.h> |
ac27a0ec DK |
18 | #include <linux/stat.h> |
19 | #include <linux/string.h> | |
20 | #include <linux/quotaops.h> | |
21 | #include <linux/buffer_head.h> | |
22 | #include <linux/random.h> | |
23 | #include <linux/bitops.h> | |
3a5b2ecd | 24 | #include <linux/blkdev.h> |
ac27a0ec | 25 | #include <asm/byteorder.h> |
3dcf5451 CH |
26 | #include "ext4.h" |
27 | #include "ext4_jbd2.h" | |
ac27a0ec DK |
28 | #include "xattr.h" |
29 | #include "acl.h" | |
30 | ||
31 | /* | |
32 | * ialloc.c contains the inodes allocation and deallocation routines | |
33 | */ | |
34 | ||
35 | /* | |
36 | * The free inodes are managed by bitmaps. A file system contains several | |
37 | * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap | |
38 | * block for inodes, N blocks for the inode table and data blocks. | |
39 | * | |
40 | * The file system contains group descriptors which are located after the | |
41 | * super block. Each descriptor contains the number of the bitmap block and | |
42 | * the free blocks count in the block. | |
43 | */ | |
44 | ||
717d50e4 AD |
45 | /* |
46 | * To avoid calling the atomic setbit hundreds or thousands of times, we only | |
47 | * need to use it within a single byte (to ensure we get endianness right). | |
48 | * We can use memset for the rest of the bitmap as there are no other users. | |
49 | */ | |
50 | void mark_bitmap_end(int start_bit, int end_bit, char *bitmap) | |
51 | { | |
52 | int i; | |
53 | ||
54 | if (start_bit >= end_bit) | |
55 | return; | |
56 | ||
57 | ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); | |
58 | for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) | |
59 | ext4_set_bit(i, bitmap); | |
60 | if (i < end_bit) | |
61 | memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); | |
62 | } | |
63 | ||
64 | /* Initializes an uninitialized inode bitmap */ | |
fd2d4291 AM |
65 | unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh, |
66 | ext4_group_t block_group, | |
717d50e4 AD |
67 | struct ext4_group_desc *gdp) |
68 | { | |
69 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
70 | ||
71 | J_ASSERT_BH(bh, buffer_locked(bh)); | |
72 | ||
73 | /* If checksum is bad mark all blocks and inodes use to prevent | |
74 | * allocation, essentially implementing a per-group read-only flag. */ | |
75 | if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) { | |
a9df9a49 | 76 | ext4_error(sb, __func__, "Checksum bad for group %u", |
717d50e4 | 77 | block_group); |
560671a0 AK |
78 | ext4_free_blks_set(sb, gdp, 0); |
79 | ext4_free_inodes_set(sb, gdp, 0); | |
80 | ext4_itable_unused_set(sb, gdp, 0); | |
717d50e4 AD |
81 | memset(bh->b_data, 0xff, sb->s_blocksize); |
82 | return 0; | |
83 | } | |
84 | ||
85 | memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); | |
648f5879 | 86 | mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8, |
717d50e4 AD |
87 | bh->b_data); |
88 | ||
89 | return EXT4_INODES_PER_GROUP(sb); | |
90 | } | |
ac27a0ec DK |
91 | |
92 | /* | |
93 | * Read the inode allocation bitmap for a given block_group, reading | |
94 | * into the specified slot in the superblock's bitmap cache. | |
95 | * | |
96 | * Return buffer_head of bitmap on success or NULL. | |
97 | */ | |
98 | static struct buffer_head * | |
e29d1cde | 99 | ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) |
ac27a0ec | 100 | { |
617ba13b | 101 | struct ext4_group_desc *desc; |
ac27a0ec | 102 | struct buffer_head *bh = NULL; |
e29d1cde | 103 | ext4_fsblk_t bitmap_blk; |
ac27a0ec | 104 | |
617ba13b | 105 | desc = ext4_get_group_desc(sb, block_group, NULL); |
ac27a0ec | 106 | if (!desc) |
e29d1cde ES |
107 | return NULL; |
108 | bitmap_blk = ext4_inode_bitmap(sb, desc); | |
109 | bh = sb_getblk(sb, bitmap_blk); | |
110 | if (unlikely(!bh)) { | |
111 | ext4_error(sb, __func__, | |
112 | "Cannot read inode bitmap - " | |
a9df9a49 | 113 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
114 | block_group, bitmap_blk); |
115 | return NULL; | |
116 | } | |
2ccb5fb9 | 117 | if (bitmap_uptodate(bh)) |
e29d1cde ES |
118 | return bh; |
119 | ||
c806e68f | 120 | lock_buffer(bh); |
2ccb5fb9 AK |
121 | if (bitmap_uptodate(bh)) { |
122 | unlock_buffer(bh); | |
123 | return bh; | |
124 | } | |
955ce5f5 | 125 | ext4_lock_group(sb, block_group); |
717d50e4 | 126 | if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { |
e29d1cde | 127 | ext4_init_inode_bitmap(sb, bh, block_group, desc); |
2ccb5fb9 | 128 | set_bitmap_uptodate(bh); |
e29d1cde | 129 | set_buffer_uptodate(bh); |
955ce5f5 | 130 | ext4_unlock_group(sb, block_group); |
3300beda | 131 | unlock_buffer(bh); |
e29d1cde | 132 | return bh; |
717d50e4 | 133 | } |
955ce5f5 | 134 | ext4_unlock_group(sb, block_group); |
2ccb5fb9 AK |
135 | if (buffer_uptodate(bh)) { |
136 | /* | |
137 | * if not uninit if bh is uptodate, | |
138 | * bitmap is also uptodate | |
139 | */ | |
140 | set_bitmap_uptodate(bh); | |
141 | unlock_buffer(bh); | |
142 | return bh; | |
143 | } | |
144 | /* | |
145 | * submit the buffer_head for read. We can | |
146 | * safely mark the bitmap as uptodate now. | |
147 | * We do it here so the bitmap uptodate bit | |
148 | * get set with buffer lock held. | |
149 | */ | |
150 | set_bitmap_uptodate(bh); | |
e29d1cde ES |
151 | if (bh_submit_read(bh) < 0) { |
152 | put_bh(bh); | |
153 | ext4_error(sb, __func__, | |
ac27a0ec | 154 | "Cannot read inode bitmap - " |
a9df9a49 | 155 | "block_group = %u, inode_bitmap = %llu", |
e29d1cde ES |
156 | block_group, bitmap_blk); |
157 | return NULL; | |
158 | } | |
ac27a0ec DK |
159 | return bh; |
160 | } | |
161 | ||
162 | /* | |
163 | * NOTE! When we get the inode, we're the only people | |
164 | * that have access to it, and as such there are no | |
165 | * race conditions we have to worry about. The inode | |
166 | * is not on the hash-lists, and it cannot be reached | |
167 | * through the filesystem because the directory entry | |
168 | * has been deleted earlier. | |
169 | * | |
170 | * HOWEVER: we must make sure that we get no aliases, | |
171 | * which means that we have to call "clear_inode()" | |
172 | * _before_ we mark the inode not in use in the inode | |
173 | * bitmaps. Otherwise a newly created file might use | |
174 | * the same inode number (not actually the same pointer | |
175 | * though), and then we'd have two inodes sharing the | |
176 | * same inode number and space on the harddisk. | |
177 | */ | |
af5bc92d | 178 | void ext4_free_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 179 | { |
af5bc92d | 180 | struct super_block *sb = inode->i_sb; |
ac27a0ec DK |
181 | int is_directory; |
182 | unsigned long ino; | |
183 | struct buffer_head *bitmap_bh = NULL; | |
184 | struct buffer_head *bh2; | |
fd2d4291 | 185 | ext4_group_t block_group; |
ac27a0ec | 186 | unsigned long bit; |
af5bc92d TT |
187 | struct ext4_group_desc *gdp; |
188 | struct ext4_super_block *es; | |
617ba13b | 189 | struct ext4_sb_info *sbi; |
7ce9d5d1 | 190 | int fatal = 0, err, count, cleared; |
ac27a0ec DK |
191 | |
192 | if (atomic_read(&inode->i_count) > 1) { | |
4776004f TT |
193 | printk(KERN_ERR "ext4_free_inode: inode has count=%d\n", |
194 | atomic_read(&inode->i_count)); | |
ac27a0ec DK |
195 | return; |
196 | } | |
197 | if (inode->i_nlink) { | |
4776004f TT |
198 | printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n", |
199 | inode->i_nlink); | |
ac27a0ec DK |
200 | return; |
201 | } | |
202 | if (!sb) { | |
4776004f TT |
203 | printk(KERN_ERR "ext4_free_inode: inode on " |
204 | "nonexistent device\n"); | |
ac27a0ec DK |
205 | return; |
206 | } | |
617ba13b | 207 | sbi = EXT4_SB(sb); |
ac27a0ec DK |
208 | |
209 | ino = inode->i_ino; | |
af5bc92d | 210 | ext4_debug("freeing inode %lu\n", ino); |
ba80b101 TT |
211 | trace_mark(ext4_free_inode, |
212 | "dev %s ino %lu mode %d uid %lu gid %lu bocks %llu", | |
213 | sb->s_id, inode->i_ino, inode->i_mode, | |
214 | (unsigned long) inode->i_uid, (unsigned long) inode->i_gid, | |
215 | (unsigned long long) inode->i_blocks); | |
ac27a0ec DK |
216 | |
217 | /* | |
218 | * Note: we must free any quota before locking the superblock, | |
219 | * as writing the quota to disk may need the lock as well. | |
220 | */ | |
a269eb18 | 221 | vfs_dq_init(inode); |
617ba13b | 222 | ext4_xattr_delete_inode(handle, inode); |
a269eb18 JK |
223 | vfs_dq_free_inode(inode); |
224 | vfs_dq_drop(inode); | |
ac27a0ec DK |
225 | |
226 | is_directory = S_ISDIR(inode->i_mode); | |
227 | ||
228 | /* Do this BEFORE marking the inode not in use or returning an error */ | |
af5bc92d | 229 | clear_inode(inode); |
ac27a0ec | 230 | |
617ba13b MC |
231 | es = EXT4_SB(sb)->s_es; |
232 | if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { | |
af5bc92d TT |
233 | ext4_error(sb, "ext4_free_inode", |
234 | "reserved or nonexistent inode %lu", ino); | |
ac27a0ec DK |
235 | goto error_return; |
236 | } | |
617ba13b MC |
237 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
238 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 239 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec DK |
240 | if (!bitmap_bh) |
241 | goto error_return; | |
242 | ||
243 | BUFFER_TRACE(bitmap_bh, "get_write_access"); | |
617ba13b | 244 | fatal = ext4_journal_get_write_access(handle, bitmap_bh); |
ac27a0ec DK |
245 | if (fatal) |
246 | goto error_return; | |
247 | ||
248 | /* Ok, now we can actually update the inode bitmaps.. */ | |
955ce5f5 AK |
249 | cleared = ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group), |
250 | bit, bitmap_bh->b_data); | |
7ce9d5d1 | 251 | if (!cleared) |
af5bc92d TT |
252 | ext4_error(sb, "ext4_free_inode", |
253 | "bit already cleared for inode %lu", ino); | |
ac27a0ec | 254 | else { |
af5bc92d | 255 | gdp = ext4_get_group_desc(sb, block_group, &bh2); |
ac27a0ec DK |
256 | |
257 | BUFFER_TRACE(bh2, "get_write_access"); | |
617ba13b | 258 | fatal = ext4_journal_get_write_access(handle, bh2); |
ac27a0ec DK |
259 | if (fatal) goto error_return; |
260 | ||
261 | if (gdp) { | |
955ce5f5 | 262 | ext4_lock_group(sb, block_group); |
560671a0 AK |
263 | count = ext4_free_inodes_count(sb, gdp) + 1; |
264 | ext4_free_inodes_set(sb, gdp, count); | |
265 | if (is_directory) { | |
266 | count = ext4_used_dirs_count(sb, gdp) - 1; | |
267 | ext4_used_dirs_set(sb, gdp, count); | |
7d39db14 TT |
268 | if (sbi->s_log_groups_per_flex) { |
269 | ext4_group_t f; | |
270 | ||
271 | f = ext4_flex_group(sbi, block_group); | |
272 | atomic_dec(&sbi->s_flex_groups[f].free_inodes); | |
273 | } | |
274 | ||
560671a0 | 275 | } |
717d50e4 AD |
276 | gdp->bg_checksum = ext4_group_desc_csum(sbi, |
277 | block_group, gdp); | |
955ce5f5 | 278 | ext4_unlock_group(sb, block_group); |
ac27a0ec DK |
279 | percpu_counter_inc(&sbi->s_freeinodes_counter); |
280 | if (is_directory) | |
281 | percpu_counter_dec(&sbi->s_dirs_counter); | |
282 | ||
772cb7c8 | 283 | if (sbi->s_log_groups_per_flex) { |
9f24e420 TT |
284 | ext4_group_t f; |
285 | ||
286 | f = ext4_flex_group(sbi, block_group); | |
287 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); | |
772cb7c8 | 288 | } |
ac27a0ec | 289 | } |
0390131b FM |
290 | BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); |
291 | err = ext4_handle_dirty_metadata(handle, NULL, bh2); | |
ac27a0ec DK |
292 | if (!fatal) fatal = err; |
293 | } | |
0390131b FM |
294 | BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); |
295 | err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); | |
ac27a0ec DK |
296 | if (!fatal) |
297 | fatal = err; | |
298 | sb->s_dirt = 1; | |
299 | error_return: | |
300 | brelse(bitmap_bh); | |
617ba13b | 301 | ext4_std_error(sb, fatal); |
ac27a0ec DK |
302 | } |
303 | ||
304 | /* | |
305 | * There are two policies for allocating an inode. If the new inode is | |
306 | * a directory, then a forward search is made for a block group with both | |
307 | * free space and a low directory-to-inode ratio; if that fails, then of | |
308 | * the groups with above-average free space, that group with the fewest | |
309 | * directories already is chosen. | |
310 | * | |
311 | * For other inodes, search forward from the parent directory\'s block | |
312 | * group to find a free inode. | |
313 | */ | |
2aa9fc4c AM |
314 | static int find_group_dir(struct super_block *sb, struct inode *parent, |
315 | ext4_group_t *best_group) | |
ac27a0ec | 316 | { |
8df9675f | 317 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 318 | unsigned int freei, avefreei; |
617ba13b | 319 | struct ext4_group_desc *desc, *best_desc = NULL; |
2aa9fc4c AM |
320 | ext4_group_t group; |
321 | int ret = -1; | |
ac27a0ec | 322 | |
617ba13b | 323 | freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter); |
ac27a0ec DK |
324 | avefreei = freei / ngroups; |
325 | ||
326 | for (group = 0; group < ngroups; group++) { | |
af5bc92d | 327 | desc = ext4_get_group_desc(sb, group, NULL); |
560671a0 | 328 | if (!desc || !ext4_free_inodes_count(sb, desc)) |
ac27a0ec | 329 | continue; |
560671a0 | 330 | if (ext4_free_inodes_count(sb, desc) < avefreei) |
ac27a0ec DK |
331 | continue; |
332 | if (!best_desc || | |
560671a0 AK |
333 | (ext4_free_blks_count(sb, desc) > |
334 | ext4_free_blks_count(sb, best_desc))) { | |
2aa9fc4c | 335 | *best_group = group; |
ac27a0ec | 336 | best_desc = desc; |
2aa9fc4c | 337 | ret = 0; |
ac27a0ec DK |
338 | } |
339 | } | |
2aa9fc4c | 340 | return ret; |
ac27a0ec DK |
341 | } |
342 | ||
772cb7c8 JS |
343 | #define free_block_ratio 10 |
344 | ||
345 | static int find_group_flex(struct super_block *sb, struct inode *parent, | |
346 | ext4_group_t *best_group) | |
347 | { | |
348 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
349 | struct ext4_group_desc *desc; | |
772cb7c8 JS |
350 | struct flex_groups *flex_group = sbi->s_flex_groups; |
351 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; | |
352 | ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group); | |
8df9675f | 353 | ext4_group_t ngroups = ext4_get_groups_count(sb); |
772cb7c8 JS |
354 | int flex_size = ext4_flex_bg_size(sbi); |
355 | ext4_group_t best_flex = parent_fbg_group; | |
356 | int blocks_per_flex = sbi->s_blocks_per_group * flex_size; | |
357 | int flexbg_free_blocks; | |
358 | int flex_freeb_ratio; | |
359 | ext4_group_t n_fbg_groups; | |
360 | ext4_group_t i; | |
361 | ||
8df9675f | 362 | n_fbg_groups = (ngroups + flex_size - 1) >> |
772cb7c8 JS |
363 | sbi->s_log_groups_per_flex; |
364 | ||
365 | find_close_to_parent: | |
9f24e420 | 366 | flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks); |
772cb7c8 | 367 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; |
9f24e420 | 368 | if (atomic_read(&flex_group[best_flex].free_inodes) && |
772cb7c8 JS |
369 | flex_freeb_ratio > free_block_ratio) |
370 | goto found_flexbg; | |
371 | ||
372 | if (best_flex && best_flex == parent_fbg_group) { | |
373 | best_flex--; | |
374 | goto find_close_to_parent; | |
375 | } | |
376 | ||
377 | for (i = 0; i < n_fbg_groups; i++) { | |
378 | if (i == parent_fbg_group || i == parent_fbg_group - 1) | |
379 | continue; | |
380 | ||
9f24e420 | 381 | flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks); |
772cb7c8 JS |
382 | flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex; |
383 | ||
384 | if (flex_freeb_ratio > free_block_ratio && | |
9f24e420 | 385 | (atomic_read(&flex_group[i].free_inodes))) { |
772cb7c8 JS |
386 | best_flex = i; |
387 | goto found_flexbg; | |
388 | } | |
389 | ||
9f24e420 TT |
390 | if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) || |
391 | ((atomic_read(&flex_group[i].free_blocks) > | |
392 | atomic_read(&flex_group[best_flex].free_blocks)) && | |
393 | atomic_read(&flex_group[i].free_inodes))) | |
772cb7c8 JS |
394 | best_flex = i; |
395 | } | |
396 | ||
9f24e420 TT |
397 | if (!atomic_read(&flex_group[best_flex].free_inodes) || |
398 | !atomic_read(&flex_group[best_flex].free_blocks)) | |
772cb7c8 JS |
399 | return -1; |
400 | ||
401 | found_flexbg: | |
402 | for (i = best_flex * flex_size; i < ngroups && | |
403 | i < (best_flex + 1) * flex_size; i++) { | |
88b6edd1 | 404 | desc = ext4_get_group_desc(sb, i, NULL); |
560671a0 | 405 | if (ext4_free_inodes_count(sb, desc)) { |
772cb7c8 JS |
406 | *best_group = i; |
407 | goto out; | |
408 | } | |
409 | } | |
410 | ||
411 | return -1; | |
412 | out: | |
413 | return 0; | |
414 | } | |
415 | ||
a4912123 TT |
416 | struct orlov_stats { |
417 | __u32 free_inodes; | |
418 | __u32 free_blocks; | |
419 | __u32 used_dirs; | |
420 | }; | |
421 | ||
422 | /* | |
423 | * Helper function for Orlov's allocator; returns critical information | |
424 | * for a particular block group or flex_bg. If flex_size is 1, then g | |
425 | * is a block group number; otherwise it is flex_bg number. | |
426 | */ | |
427 | void get_orlov_stats(struct super_block *sb, ext4_group_t g, | |
428 | int flex_size, struct orlov_stats *stats) | |
429 | { | |
430 | struct ext4_group_desc *desc; | |
7d39db14 | 431 | struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; |
a4912123 | 432 | |
7d39db14 TT |
433 | if (flex_size > 1) { |
434 | stats->free_inodes = atomic_read(&flex_group[g].free_inodes); | |
435 | stats->free_blocks = atomic_read(&flex_group[g].free_blocks); | |
436 | stats->used_dirs = atomic_read(&flex_group[g].used_dirs); | |
437 | return; | |
438 | } | |
a4912123 | 439 | |
7d39db14 TT |
440 | desc = ext4_get_group_desc(sb, g, NULL); |
441 | if (desc) { | |
442 | stats->free_inodes = ext4_free_inodes_count(sb, desc); | |
443 | stats->free_blocks = ext4_free_blks_count(sb, desc); | |
444 | stats->used_dirs = ext4_used_dirs_count(sb, desc); | |
445 | } else { | |
446 | stats->free_inodes = 0; | |
447 | stats->free_blocks = 0; | |
448 | stats->used_dirs = 0; | |
a4912123 TT |
449 | } |
450 | } | |
451 | ||
ac27a0ec DK |
452 | /* |
453 | * Orlov's allocator for directories. | |
454 | * | |
455 | * We always try to spread first-level directories. | |
456 | * | |
457 | * If there are blockgroups with both free inodes and free blocks counts | |
458 | * not worse than average we return one with smallest directory count. | |
459 | * Otherwise we simply return a random group. | |
460 | * | |
461 | * For the rest rules look so: | |
462 | * | |
463 | * It's OK to put directory into a group unless | |
464 | * it has too many directories already (max_dirs) or | |
465 | * it has too few free inodes left (min_inodes) or | |
466 | * it has too few free blocks left (min_blocks) or | |
1cc8dcf5 | 467 | * Parent's group is preferred, if it doesn't satisfy these |
ac27a0ec DK |
468 | * conditions we search cyclically through the rest. If none |
469 | * of the groups look good we just look for a group with more | |
470 | * free inodes than average (starting at parent's group). | |
ac27a0ec DK |
471 | */ |
472 | ||
2aa9fc4c | 473 | static int find_group_orlov(struct super_block *sb, struct inode *parent, |
a4912123 | 474 | ext4_group_t *group, int mode) |
ac27a0ec | 475 | { |
fd2d4291 | 476 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
617ba13b | 477 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
8df9675f | 478 | ext4_group_t real_ngroups = ext4_get_groups_count(sb); |
617ba13b | 479 | int inodes_per_group = EXT4_INODES_PER_GROUP(sb); |
ac27a0ec | 480 | unsigned int freei, avefreei; |
617ba13b | 481 | ext4_fsblk_t freeb, avefreeb; |
ac27a0ec | 482 | unsigned int ndirs; |
a4912123 | 483 | int max_dirs, min_inodes; |
617ba13b | 484 | ext4_grpblk_t min_blocks; |
8df9675f | 485 | ext4_group_t i, grp, g, ngroups; |
617ba13b | 486 | struct ext4_group_desc *desc; |
a4912123 TT |
487 | struct orlov_stats stats; |
488 | int flex_size = ext4_flex_bg_size(sbi); | |
489 | ||
8df9675f | 490 | ngroups = real_ngroups; |
a4912123 | 491 | if (flex_size > 1) { |
8df9675f | 492 | ngroups = (real_ngroups + flex_size - 1) >> |
a4912123 TT |
493 | sbi->s_log_groups_per_flex; |
494 | parent_group >>= sbi->s_log_groups_per_flex; | |
495 | } | |
ac27a0ec DK |
496 | |
497 | freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); | |
498 | avefreei = freei / ngroups; | |
499 | freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter); | |
3a5b2ecd | 500 | avefreeb = freeb; |
f4e5bc24 | 501 | do_div(avefreeb, ngroups); |
ac27a0ec DK |
502 | ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); |
503 | ||
a4912123 TT |
504 | if (S_ISDIR(mode) && |
505 | ((parent == sb->s_root->d_inode) || | |
506 | (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) { | |
ac27a0ec | 507 | int best_ndir = inodes_per_group; |
2aa9fc4c | 508 | int ret = -1; |
ac27a0ec | 509 | |
2aa9fc4c AM |
510 | get_random_bytes(&grp, sizeof(grp)); |
511 | parent_group = (unsigned)grp % ngroups; | |
ac27a0ec | 512 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
513 | g = (parent_group + i) % ngroups; |
514 | get_orlov_stats(sb, g, flex_size, &stats); | |
515 | if (!stats.free_inodes) | |
ac27a0ec | 516 | continue; |
a4912123 | 517 | if (stats.used_dirs >= best_ndir) |
ac27a0ec | 518 | continue; |
a4912123 | 519 | if (stats.free_inodes < avefreei) |
ac27a0ec | 520 | continue; |
a4912123 | 521 | if (stats.free_blocks < avefreeb) |
ac27a0ec | 522 | continue; |
a4912123 | 523 | grp = g; |
2aa9fc4c | 524 | ret = 0; |
a4912123 TT |
525 | best_ndir = stats.used_dirs; |
526 | } | |
527 | if (ret) | |
528 | goto fallback; | |
529 | found_flex_bg: | |
530 | if (flex_size == 1) { | |
531 | *group = grp; | |
532 | return 0; | |
533 | } | |
534 | ||
535 | /* | |
536 | * We pack inodes at the beginning of the flexgroup's | |
537 | * inode tables. Block allocation decisions will do | |
538 | * something similar, although regular files will | |
539 | * start at 2nd block group of the flexgroup. See | |
540 | * ext4_ext_find_goal() and ext4_find_near(). | |
541 | */ | |
542 | grp *= flex_size; | |
543 | for (i = 0; i < flex_size; i++) { | |
8df9675f | 544 | if (grp+i >= real_ngroups) |
a4912123 TT |
545 | break; |
546 | desc = ext4_get_group_desc(sb, grp+i, NULL); | |
547 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
548 | *group = grp+i; | |
549 | return 0; | |
550 | } | |
ac27a0ec | 551 | } |
ac27a0ec DK |
552 | goto fallback; |
553 | } | |
554 | ||
ac27a0ec | 555 | max_dirs = ndirs / ngroups + inodes_per_group / 16; |
a4912123 TT |
556 | min_inodes = avefreei - inodes_per_group*flex_size / 4; |
557 | if (min_inodes < 1) | |
558 | min_inodes = 1; | |
559 | min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4; | |
560 | ||
561 | /* | |
562 | * Start looking in the flex group where we last allocated an | |
563 | * inode for this parent directory | |
564 | */ | |
565 | if (EXT4_I(parent)->i_last_alloc_group != ~0) { | |
566 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
567 | if (flex_size > 1) | |
568 | parent_group >>= sbi->s_log_groups_per_flex; | |
569 | } | |
ac27a0ec DK |
570 | |
571 | for (i = 0; i < ngroups; i++) { | |
a4912123 TT |
572 | grp = (parent_group + i) % ngroups; |
573 | get_orlov_stats(sb, grp, flex_size, &stats); | |
574 | if (stats.used_dirs >= max_dirs) | |
ac27a0ec | 575 | continue; |
a4912123 | 576 | if (stats.free_inodes < min_inodes) |
ac27a0ec | 577 | continue; |
a4912123 | 578 | if (stats.free_blocks < min_blocks) |
ac27a0ec | 579 | continue; |
a4912123 | 580 | goto found_flex_bg; |
ac27a0ec DK |
581 | } |
582 | ||
583 | fallback: | |
8df9675f | 584 | ngroups = real_ngroups; |
a4912123 | 585 | avefreei = freei / ngroups; |
b5451f7b | 586 | fallback_retry: |
a4912123 | 587 | parent_group = EXT4_I(parent)->i_block_group; |
ac27a0ec | 588 | for (i = 0; i < ngroups; i++) { |
a4912123 TT |
589 | grp = (parent_group + i) % ngroups; |
590 | desc = ext4_get_group_desc(sb, grp, NULL); | |
560671a0 | 591 | if (desc && ext4_free_inodes_count(sb, desc) && |
a4912123 TT |
592 | ext4_free_inodes_count(sb, desc) >= avefreei) { |
593 | *group = grp; | |
2aa9fc4c | 594 | return 0; |
a4912123 | 595 | } |
ac27a0ec DK |
596 | } |
597 | ||
598 | if (avefreei) { | |
599 | /* | |
600 | * The free-inodes counter is approximate, and for really small | |
601 | * filesystems the above test can fail to find any blockgroups | |
602 | */ | |
603 | avefreei = 0; | |
b5451f7b | 604 | goto fallback_retry; |
ac27a0ec DK |
605 | } |
606 | ||
607 | return -1; | |
608 | } | |
609 | ||
2aa9fc4c | 610 | static int find_group_other(struct super_block *sb, struct inode *parent, |
a4912123 | 611 | ext4_group_t *group, int mode) |
ac27a0ec | 612 | { |
fd2d4291 | 613 | ext4_group_t parent_group = EXT4_I(parent)->i_block_group; |
8df9675f | 614 | ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); |
617ba13b | 615 | struct ext4_group_desc *desc; |
a4912123 TT |
616 | int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); |
617 | ||
618 | /* | |
619 | * Try to place the inode is the same flex group as its | |
620 | * parent. If we can't find space, use the Orlov algorithm to | |
621 | * find another flex group, and store that information in the | |
622 | * parent directory's inode information so that use that flex | |
623 | * group for future allocations. | |
624 | */ | |
625 | if (flex_size > 1) { | |
626 | int retry = 0; | |
627 | ||
628 | try_again: | |
629 | parent_group &= ~(flex_size-1); | |
630 | last = parent_group + flex_size; | |
631 | if (last > ngroups) | |
632 | last = ngroups; | |
633 | for (i = parent_group; i < last; i++) { | |
634 | desc = ext4_get_group_desc(sb, i, NULL); | |
635 | if (desc && ext4_free_inodes_count(sb, desc)) { | |
636 | *group = i; | |
637 | return 0; | |
638 | } | |
639 | } | |
640 | if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { | |
641 | retry = 1; | |
642 | parent_group = EXT4_I(parent)->i_last_alloc_group; | |
643 | goto try_again; | |
644 | } | |
645 | /* | |
646 | * If this didn't work, use the Orlov search algorithm | |
647 | * to find a new flex group; we pass in the mode to | |
648 | * avoid the topdir algorithms. | |
649 | */ | |
650 | *group = parent_group + flex_size; | |
651 | if (*group > ngroups) | |
652 | *group = 0; | |
653 | return find_group_orlov(sb, parent, group, mode); | |
654 | } | |
ac27a0ec DK |
655 | |
656 | /* | |
657 | * Try to place the inode in its parent directory | |
658 | */ | |
2aa9fc4c AM |
659 | *group = parent_group; |
660 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
661 | if (desc && ext4_free_inodes_count(sb, desc) && |
662 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 663 | return 0; |
ac27a0ec DK |
664 | |
665 | /* | |
666 | * We're going to place this inode in a different blockgroup from its | |
667 | * parent. We want to cause files in a common directory to all land in | |
668 | * the same blockgroup. But we want files which are in a different | |
669 | * directory which shares a blockgroup with our parent to land in a | |
670 | * different blockgroup. | |
671 | * | |
672 | * So add our directory's i_ino into the starting point for the hash. | |
673 | */ | |
2aa9fc4c | 674 | *group = (*group + parent->i_ino) % ngroups; |
ac27a0ec DK |
675 | |
676 | /* | |
677 | * Use a quadratic hash to find a group with a free inode and some free | |
678 | * blocks. | |
679 | */ | |
680 | for (i = 1; i < ngroups; i <<= 1) { | |
2aa9fc4c AM |
681 | *group += i; |
682 | if (*group >= ngroups) | |
683 | *group -= ngroups; | |
684 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 AK |
685 | if (desc && ext4_free_inodes_count(sb, desc) && |
686 | ext4_free_blks_count(sb, desc)) | |
2aa9fc4c | 687 | return 0; |
ac27a0ec DK |
688 | } |
689 | ||
690 | /* | |
691 | * That failed: try linear search for a free inode, even if that group | |
692 | * has no free blocks. | |
693 | */ | |
2aa9fc4c | 694 | *group = parent_group; |
ac27a0ec | 695 | for (i = 0; i < ngroups; i++) { |
2aa9fc4c AM |
696 | if (++*group >= ngroups) |
697 | *group = 0; | |
698 | desc = ext4_get_group_desc(sb, *group, NULL); | |
560671a0 | 699 | if (desc && ext4_free_inodes_count(sb, desc)) |
2aa9fc4c | 700 | return 0; |
ac27a0ec DK |
701 | } |
702 | ||
703 | return -1; | |
704 | } | |
705 | ||
39341867 AK |
706 | /* |
707 | * claim the inode from the inode bitmap. If the group | |
955ce5f5 | 708 | * is uninit we need to take the groups's ext4_group_lock |
39341867 AK |
709 | * and clear the uninit flag. The inode bitmap update |
710 | * and group desc uninit flag clear should be done | |
955ce5f5 | 711 | * after holding ext4_group_lock so that ext4_read_inode_bitmap |
39341867 AK |
712 | * doesn't race with the ext4_claim_inode |
713 | */ | |
714 | static int ext4_claim_inode(struct super_block *sb, | |
715 | struct buffer_head *inode_bitmap_bh, | |
716 | unsigned long ino, ext4_group_t group, int mode) | |
717 | { | |
718 | int free = 0, retval = 0, count; | |
719 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
720 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL); | |
721 | ||
955ce5f5 | 722 | ext4_lock_group(sb, group); |
39341867 AK |
723 | if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) { |
724 | /* not a free inode */ | |
725 | retval = 1; | |
726 | goto err_ret; | |
727 | } | |
728 | ino++; | |
729 | if ((group == 0 && ino < EXT4_FIRST_INO(sb)) || | |
730 | ino > EXT4_INODES_PER_GROUP(sb)) { | |
955ce5f5 | 731 | ext4_unlock_group(sb, group); |
39341867 AK |
732 | ext4_error(sb, __func__, |
733 | "reserved inode or inode > inodes count - " | |
734 | "block_group = %u, inode=%lu", group, | |
735 | ino + group * EXT4_INODES_PER_GROUP(sb)); | |
736 | return 1; | |
737 | } | |
738 | /* If we didn't allocate from within the initialized part of the inode | |
739 | * table then we need to initialize up to this inode. */ | |
740 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) { | |
741 | ||
742 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { | |
743 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); | |
744 | /* When marking the block group with | |
745 | * ~EXT4_BG_INODE_UNINIT we don't want to depend | |
746 | * on the value of bg_itable_unused even though | |
747 | * mke2fs could have initialized the same for us. | |
748 | * Instead we calculated the value below | |
749 | */ | |
750 | ||
751 | free = 0; | |
752 | } else { | |
753 | free = EXT4_INODES_PER_GROUP(sb) - | |
754 | ext4_itable_unused_count(sb, gdp); | |
755 | } | |
756 | ||
757 | /* | |
758 | * Check the relative inode number against the last used | |
759 | * relative inode number in this group. if it is greater | |
760 | * we need to update the bg_itable_unused count | |
761 | * | |
762 | */ | |
763 | if (ino > free) | |
764 | ext4_itable_unused_set(sb, gdp, | |
765 | (EXT4_INODES_PER_GROUP(sb) - ino)); | |
766 | } | |
767 | count = ext4_free_inodes_count(sb, gdp) - 1; | |
768 | ext4_free_inodes_set(sb, gdp, count); | |
769 | if (S_ISDIR(mode)) { | |
770 | count = ext4_used_dirs_count(sb, gdp) + 1; | |
771 | ext4_used_dirs_set(sb, gdp, count); | |
7d39db14 TT |
772 | if (sbi->s_log_groups_per_flex) { |
773 | ext4_group_t f = ext4_flex_group(sbi, group); | |
774 | ||
775 | atomic_inc(&sbi->s_flex_groups[f].free_inodes); | |
776 | } | |
39341867 AK |
777 | } |
778 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp); | |
779 | err_ret: | |
955ce5f5 | 780 | ext4_unlock_group(sb, group); |
39341867 AK |
781 | return retval; |
782 | } | |
783 | ||
ac27a0ec DK |
784 | /* |
785 | * There are two policies for allocating an inode. If the new inode is | |
786 | * a directory, then a forward search is made for a block group with both | |
787 | * free space and a low directory-to-inode ratio; if that fails, then of | |
788 | * the groups with above-average free space, that group with the fewest | |
789 | * directories already is chosen. | |
790 | * | |
791 | * For other inodes, search forward from the parent directory's block | |
792 | * group to find a free inode. | |
793 | */ | |
af5bc92d | 794 | struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode) |
ac27a0ec DK |
795 | { |
796 | struct super_block *sb; | |
3300beda AK |
797 | struct buffer_head *inode_bitmap_bh = NULL; |
798 | struct buffer_head *group_desc_bh; | |
8df9675f | 799 | ext4_group_t ngroups, group = 0; |
ac27a0ec | 800 | unsigned long ino = 0; |
af5bc92d TT |
801 | struct inode *inode; |
802 | struct ext4_group_desc *gdp = NULL; | |
617ba13b MC |
803 | struct ext4_inode_info *ei; |
804 | struct ext4_sb_info *sbi; | |
39341867 | 805 | int ret2, err = 0; |
ac27a0ec | 806 | struct inode *ret; |
2aa9fc4c AM |
807 | ext4_group_t i; |
808 | int free = 0; | |
2842c3b5 | 809 | static int once = 1; |
772cb7c8 | 810 | ext4_group_t flex_group; |
ac27a0ec DK |
811 | |
812 | /* Cannot create files in a deleted directory */ | |
813 | if (!dir || !dir->i_nlink) | |
814 | return ERR_PTR(-EPERM); | |
815 | ||
816 | sb = dir->i_sb; | |
8df9675f | 817 | ngroups = ext4_get_groups_count(sb); |
ba80b101 TT |
818 | trace_mark(ext4_request_inode, "dev %s dir %lu mode %d", sb->s_id, |
819 | dir->i_ino, mode); | |
ac27a0ec DK |
820 | inode = new_inode(sb); |
821 | if (!inode) | |
822 | return ERR_PTR(-ENOMEM); | |
617ba13b | 823 | ei = EXT4_I(inode); |
617ba13b | 824 | sbi = EXT4_SB(sb); |
772cb7c8 | 825 | |
a4912123 | 826 | if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) { |
772cb7c8 | 827 | ret2 = find_group_flex(sb, dir, &group); |
05bf9e83 | 828 | if (ret2 == -1) { |
a4912123 | 829 | ret2 = find_group_other(sb, dir, &group, mode); |
6b82f3cb | 830 | if (ret2 == 0 && once) { |
2842c3b5 | 831 | once = 0; |
05bf9e83 TT |
832 | printk(KERN_NOTICE "ext4: find_group_flex " |
833 | "failed, fallback succeeded dir %lu\n", | |
834 | dir->i_ino); | |
6b82f3cb | 835 | } |
05bf9e83 | 836 | } |
772cb7c8 JS |
837 | goto got_group; |
838 | } | |
839 | ||
ac27a0ec | 840 | if (S_ISDIR(mode)) { |
af5bc92d | 841 | if (test_opt(sb, OLDALLOC)) |
2aa9fc4c | 842 | ret2 = find_group_dir(sb, dir, &group); |
ac27a0ec | 843 | else |
a4912123 | 844 | ret2 = find_group_orlov(sb, dir, &group, mode); |
ac27a0ec | 845 | } else |
a4912123 | 846 | ret2 = find_group_other(sb, dir, &group, mode); |
ac27a0ec | 847 | |
772cb7c8 | 848 | got_group: |
a4912123 | 849 | EXT4_I(dir)->i_last_alloc_group = group; |
ac27a0ec | 850 | err = -ENOSPC; |
2aa9fc4c | 851 | if (ret2 == -1) |
ac27a0ec DK |
852 | goto out; |
853 | ||
8df9675f | 854 | for (i = 0; i < ngroups; i++) { |
ac27a0ec DK |
855 | err = -EIO; |
856 | ||
3300beda | 857 | gdp = ext4_get_group_desc(sb, group, &group_desc_bh); |
ac27a0ec DK |
858 | if (!gdp) |
859 | goto fail; | |
860 | ||
3300beda AK |
861 | brelse(inode_bitmap_bh); |
862 | inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); | |
863 | if (!inode_bitmap_bh) | |
ac27a0ec DK |
864 | goto fail; |
865 | ||
866 | ino = 0; | |
867 | ||
868 | repeat_in_this_group: | |
617ba13b | 869 | ino = ext4_find_next_zero_bit((unsigned long *) |
3300beda AK |
870 | inode_bitmap_bh->b_data, |
871 | EXT4_INODES_PER_GROUP(sb), ino); | |
872 | ||
617ba13b | 873 | if (ino < EXT4_INODES_PER_GROUP(sb)) { |
ac27a0ec | 874 | |
3300beda AK |
875 | BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); |
876 | err = ext4_journal_get_write_access(handle, | |
877 | inode_bitmap_bh); | |
ac27a0ec DK |
878 | if (err) |
879 | goto fail; | |
880 | ||
39341867 AK |
881 | BUFFER_TRACE(group_desc_bh, "get_write_access"); |
882 | err = ext4_journal_get_write_access(handle, | |
883 | group_desc_bh); | |
884 | if (err) | |
885 | goto fail; | |
886 | if (!ext4_claim_inode(sb, inode_bitmap_bh, | |
887 | ino, group, mode)) { | |
ac27a0ec | 888 | /* we won it */ |
3300beda | 889 | BUFFER_TRACE(inode_bitmap_bh, |
0390131b FM |
890 | "call ext4_handle_dirty_metadata"); |
891 | err = ext4_handle_dirty_metadata(handle, | |
3300beda AK |
892 | inode, |
893 | inode_bitmap_bh); | |
ac27a0ec DK |
894 | if (err) |
895 | goto fail; | |
39341867 AK |
896 | /* zero bit is inode number 1*/ |
897 | ino++; | |
ac27a0ec DK |
898 | goto got; |
899 | } | |
900 | /* we lost it */ | |
3300beda | 901 | ext4_handle_release_buffer(handle, inode_bitmap_bh); |
39341867 | 902 | ext4_handle_release_buffer(handle, group_desc_bh); |
ac27a0ec | 903 | |
617ba13b | 904 | if (++ino < EXT4_INODES_PER_GROUP(sb)) |
ac27a0ec DK |
905 | goto repeat_in_this_group; |
906 | } | |
907 | ||
908 | /* | |
909 | * This case is possible in concurrent environment. It is very | |
910 | * rare. We cannot repeat the find_group_xxx() call because | |
911 | * that will simply return the same blockgroup, because the | |
912 | * group descriptor metadata has not yet been updated. | |
913 | * So we just go onto the next blockgroup. | |
914 | */ | |
8df9675f | 915 | if (++group == ngroups) |
ac27a0ec DK |
916 | group = 0; |
917 | } | |
918 | err = -ENOSPC; | |
919 | goto out; | |
920 | ||
921 | got: | |
717d50e4 AD |
922 | /* We may have to initialize the block bitmap if it isn't already */ |
923 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) && | |
924 | gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
3300beda | 925 | struct buffer_head *block_bitmap_bh; |
717d50e4 | 926 | |
3300beda AK |
927 | block_bitmap_bh = ext4_read_block_bitmap(sb, group); |
928 | BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); | |
929 | err = ext4_journal_get_write_access(handle, block_bitmap_bh); | |
717d50e4 | 930 | if (err) { |
3300beda | 931 | brelse(block_bitmap_bh); |
717d50e4 AD |
932 | goto fail; |
933 | } | |
934 | ||
935 | free = 0; | |
955ce5f5 | 936 | ext4_lock_group(sb, group); |
717d50e4 AD |
937 | /* recheck and clear flag under lock if we still need to */ |
938 | if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { | |
717d50e4 | 939 | free = ext4_free_blocks_after_init(sb, group, gdp); |
3300beda | 940 | gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); |
560671a0 | 941 | ext4_free_blks_set(sb, gdp, free); |
23712a9c FB |
942 | gdp->bg_checksum = ext4_group_desc_csum(sbi, group, |
943 | gdp); | |
717d50e4 | 944 | } |
955ce5f5 | 945 | ext4_unlock_group(sb, group); |
717d50e4 AD |
946 | |
947 | /* Don't need to dirty bitmap block if we didn't change it */ | |
948 | if (free) { | |
3300beda | 949 | BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); |
0390131b | 950 | err = ext4_handle_dirty_metadata(handle, |
3300beda | 951 | NULL, block_bitmap_bh); |
717d50e4 AD |
952 | } |
953 | ||
3300beda | 954 | brelse(block_bitmap_bh); |
717d50e4 AD |
955 | if (err) |
956 | goto fail; | |
957 | } | |
3300beda AK |
958 | BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); |
959 | err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); | |
39341867 AK |
960 | if (err) |
961 | goto fail; | |
ac27a0ec DK |
962 | |
963 | percpu_counter_dec(&sbi->s_freeinodes_counter); | |
964 | if (S_ISDIR(mode)) | |
965 | percpu_counter_inc(&sbi->s_dirs_counter); | |
966 | sb->s_dirt = 1; | |
967 | ||
772cb7c8 JS |
968 | if (sbi->s_log_groups_per_flex) { |
969 | flex_group = ext4_flex_group(sbi, group); | |
9f24e420 | 970 | atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); |
772cb7c8 JS |
971 | } |
972 | ||
4c9c544e | 973 | inode->i_uid = current_fsuid(); |
af5bc92d | 974 | if (test_opt(sb, GRPID)) |
ac27a0ec DK |
975 | inode->i_gid = dir->i_gid; |
976 | else if (dir->i_mode & S_ISGID) { | |
977 | inode->i_gid = dir->i_gid; | |
978 | if (S_ISDIR(mode)) | |
979 | mode |= S_ISGID; | |
980 | } else | |
4c9c544e | 981 | inode->i_gid = current_fsgid(); |
ac27a0ec DK |
982 | inode->i_mode = mode; |
983 | ||
717d50e4 | 984 | inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); |
ac27a0ec DK |
985 | /* This is the optimal IO size (for stat), not the fs block size */ |
986 | inode->i_blocks = 0; | |
ef7f3835 KS |
987 | inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = |
988 | ext4_current_time(inode); | |
ac27a0ec DK |
989 | |
990 | memset(ei->i_data, 0, sizeof(ei->i_data)); | |
991 | ei->i_dir_start_lookup = 0; | |
992 | ei->i_disksize = 0; | |
993 | ||
42bf0383 | 994 | /* |
2dc6b0d4 DG |
995 | * Don't inherit extent flag from directory, amongst others. We set |
996 | * extent flag on newly created directory and file only if -o extent | |
997 | * mount option is specified | |
42bf0383 | 998 | */ |
2dc6b0d4 DG |
999 | ei->i_flags = |
1000 | ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); | |
ac27a0ec | 1001 | ei->i_file_acl = 0; |
ac27a0ec | 1002 | ei->i_dtime = 0; |
ac27a0ec | 1003 | ei->i_block_group = group; |
a4912123 | 1004 | ei->i_last_alloc_group = ~0; |
ac27a0ec | 1005 | |
617ba13b | 1006 | ext4_set_inode_flags(inode); |
ac27a0ec | 1007 | if (IS_DIRSYNC(inode)) |
0390131b | 1008 | ext4_handle_sync(handle); |
6b38e842 AV |
1009 | if (insert_inode_locked(inode) < 0) { |
1010 | err = -EINVAL; | |
1011 | goto fail_drop; | |
1012 | } | |
ac27a0ec DK |
1013 | spin_lock(&sbi->s_next_gen_lock); |
1014 | inode->i_generation = sbi->s_next_generation++; | |
1015 | spin_unlock(&sbi->s_next_gen_lock); | |
1016 | ||
617ba13b | 1017 | ei->i_state = EXT4_STATE_NEW; |
ef7f3835 KS |
1018 | |
1019 | ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize; | |
ac27a0ec DK |
1020 | |
1021 | ret = inode; | |
a269eb18 | 1022 | if (vfs_dq_alloc_inode(inode)) { |
ac27a0ec DK |
1023 | err = -EDQUOT; |
1024 | goto fail_drop; | |
1025 | } | |
1026 | ||
617ba13b | 1027 | err = ext4_init_acl(handle, inode, dir); |
ac27a0ec DK |
1028 | if (err) |
1029 | goto fail_free_drop; | |
1030 | ||
af5bc92d | 1031 | err = ext4_init_security(handle, inode, dir); |
ac27a0ec DK |
1032 | if (err) |
1033 | goto fail_free_drop; | |
1034 | ||
83982b6f | 1035 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) { |
e4079a11 | 1036 | /* set extent flag only for directory, file and normal symlink*/ |
e65187e6 | 1037 | if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { |
42bf0383 AK |
1038 | EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL; |
1039 | ext4_ext_tree_init(handle, inode); | |
42bf0383 | 1040 | } |
a86c6181 | 1041 | } |
ac27a0ec | 1042 | |
8753e88f AK |
1043 | err = ext4_mark_inode_dirty(handle, inode); |
1044 | if (err) { | |
1045 | ext4_std_error(sb, err); | |
1046 | goto fail_free_drop; | |
1047 | } | |
1048 | ||
617ba13b | 1049 | ext4_debug("allocating inode %lu\n", inode->i_ino); |
ba80b101 TT |
1050 | trace_mark(ext4_allocate_inode, "dev %s ino %lu dir %lu mode %d", |
1051 | sb->s_id, inode->i_ino, dir->i_ino, mode); | |
ac27a0ec DK |
1052 | goto really_out; |
1053 | fail: | |
617ba13b | 1054 | ext4_std_error(sb, err); |
ac27a0ec DK |
1055 | out: |
1056 | iput(inode); | |
1057 | ret = ERR_PTR(err); | |
1058 | really_out: | |
3300beda | 1059 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1060 | return ret; |
1061 | ||
1062 | fail_free_drop: | |
a269eb18 | 1063 | vfs_dq_free_inode(inode); |
ac27a0ec DK |
1064 | |
1065 | fail_drop: | |
a269eb18 | 1066 | vfs_dq_drop(inode); |
ac27a0ec DK |
1067 | inode->i_flags |= S_NOQUOTA; |
1068 | inode->i_nlink = 0; | |
6b38e842 | 1069 | unlock_new_inode(inode); |
ac27a0ec | 1070 | iput(inode); |
3300beda | 1071 | brelse(inode_bitmap_bh); |
ac27a0ec DK |
1072 | return ERR_PTR(err); |
1073 | } | |
1074 | ||
1075 | /* Verify that we are loading a valid orphan from disk */ | |
617ba13b | 1076 | struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) |
ac27a0ec | 1077 | { |
617ba13b | 1078 | unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); |
fd2d4291 | 1079 | ext4_group_t block_group; |
ac27a0ec | 1080 | int bit; |
1d1fe1ee | 1081 | struct buffer_head *bitmap_bh; |
ac27a0ec | 1082 | struct inode *inode = NULL; |
1d1fe1ee | 1083 | long err = -EIO; |
ac27a0ec DK |
1084 | |
1085 | /* Error cases - e2fsck has already cleaned up for us */ | |
1086 | if (ino > max_ino) { | |
46e665e9 | 1087 | ext4_warning(sb, __func__, |
ac27a0ec | 1088 | "bad orphan ino %lu! e2fsck was run?", ino); |
1d1fe1ee | 1089 | goto error; |
ac27a0ec DK |
1090 | } |
1091 | ||
617ba13b MC |
1092 | block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
1093 | bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); | |
e29d1cde | 1094 | bitmap_bh = ext4_read_inode_bitmap(sb, block_group); |
ac27a0ec | 1095 | if (!bitmap_bh) { |
46e665e9 | 1096 | ext4_warning(sb, __func__, |
ac27a0ec | 1097 | "inode bitmap error for orphan %lu", ino); |
1d1fe1ee | 1098 | goto error; |
ac27a0ec DK |
1099 | } |
1100 | ||
1101 | /* Having the inode bit set should be a 100% indicator that this | |
1102 | * is a valid orphan (no e2fsck run on fs). Orphans also include | |
1103 | * inodes that were being truncated, so we can't check i_nlink==0. | |
1104 | */ | |
1d1fe1ee DH |
1105 | if (!ext4_test_bit(bit, bitmap_bh->b_data)) |
1106 | goto bad_orphan; | |
1107 | ||
1108 | inode = ext4_iget(sb, ino); | |
1109 | if (IS_ERR(inode)) | |
1110 | goto iget_failed; | |
1111 | ||
91ef4caf DG |
1112 | /* |
1113 | * If the orphans has i_nlinks > 0 then it should be able to be | |
1114 | * truncated, otherwise it won't be removed from the orphan list | |
1115 | * during processing and an infinite loop will result. | |
1116 | */ | |
1117 | if (inode->i_nlink && !ext4_can_truncate(inode)) | |
1118 | goto bad_orphan; | |
1119 | ||
1d1fe1ee DH |
1120 | if (NEXT_ORPHAN(inode) > max_ino) |
1121 | goto bad_orphan; | |
1122 | brelse(bitmap_bh); | |
1123 | return inode; | |
1124 | ||
1125 | iget_failed: | |
1126 | err = PTR_ERR(inode); | |
1127 | inode = NULL; | |
1128 | bad_orphan: | |
46e665e9 | 1129 | ext4_warning(sb, __func__, |
1d1fe1ee DH |
1130 | "bad orphan inode %lu! e2fsck was run?", ino); |
1131 | printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n", | |
1132 | bit, (unsigned long long)bitmap_bh->b_blocknr, | |
1133 | ext4_test_bit(bit, bitmap_bh->b_data)); | |
1134 | printk(KERN_NOTICE "inode=%p\n", inode); | |
1135 | if (inode) { | |
1136 | printk(KERN_NOTICE "is_bad_inode(inode)=%d\n", | |
1137 | is_bad_inode(inode)); | |
1138 | printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n", | |
1139 | NEXT_ORPHAN(inode)); | |
1140 | printk(KERN_NOTICE "max_ino=%lu\n", max_ino); | |
91ef4caf | 1141 | printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink); |
ac27a0ec | 1142 | /* Avoid freeing blocks if we got a bad deleted inode */ |
1d1fe1ee | 1143 | if (inode->i_nlink == 0) |
ac27a0ec DK |
1144 | inode->i_blocks = 0; |
1145 | iput(inode); | |
ac27a0ec | 1146 | } |
ac27a0ec | 1147 | brelse(bitmap_bh); |
1d1fe1ee DH |
1148 | error: |
1149 | return ERR_PTR(err); | |
ac27a0ec DK |
1150 | } |
1151 | ||
af5bc92d | 1152 | unsigned long ext4_count_free_inodes(struct super_block *sb) |
ac27a0ec DK |
1153 | { |
1154 | unsigned long desc_count; | |
617ba13b | 1155 | struct ext4_group_desc *gdp; |
8df9675f | 1156 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
617ba13b MC |
1157 | #ifdef EXT4FS_DEBUG |
1158 | struct ext4_super_block *es; | |
ac27a0ec DK |
1159 | unsigned long bitmap_count, x; |
1160 | struct buffer_head *bitmap_bh = NULL; | |
1161 | ||
617ba13b | 1162 | es = EXT4_SB(sb)->s_es; |
ac27a0ec DK |
1163 | desc_count = 0; |
1164 | bitmap_count = 0; | |
1165 | gdp = NULL; | |
8df9675f | 1166 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1167 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1168 | if (!gdp) |
1169 | continue; | |
560671a0 | 1170 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec | 1171 | brelse(bitmap_bh); |
e29d1cde | 1172 | bitmap_bh = ext4_read_inode_bitmap(sb, i); |
ac27a0ec DK |
1173 | if (!bitmap_bh) |
1174 | continue; | |
1175 | ||
617ba13b | 1176 | x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8); |
c549a95d | 1177 | printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", |
560671a0 | 1178 | i, ext4_free_inodes_count(sb, gdp), x); |
ac27a0ec DK |
1179 | bitmap_count += x; |
1180 | } | |
1181 | brelse(bitmap_bh); | |
4776004f TT |
1182 | printk(KERN_DEBUG "ext4_count_free_inodes: " |
1183 | "stored = %u, computed = %lu, %lu\n", | |
1184 | le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); | |
ac27a0ec DK |
1185 | return desc_count; |
1186 | #else | |
1187 | desc_count = 0; | |
8df9675f | 1188 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1189 | gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1190 | if (!gdp) |
1191 | continue; | |
560671a0 | 1192 | desc_count += ext4_free_inodes_count(sb, gdp); |
ac27a0ec DK |
1193 | cond_resched(); |
1194 | } | |
1195 | return desc_count; | |
1196 | #endif | |
1197 | } | |
1198 | ||
1199 | /* Called at mount-time, super-block is locked */ | |
af5bc92d | 1200 | unsigned long ext4_count_dirs(struct super_block * sb) |
ac27a0ec DK |
1201 | { |
1202 | unsigned long count = 0; | |
8df9675f | 1203 | ext4_group_t i, ngroups = ext4_get_groups_count(sb); |
ac27a0ec | 1204 | |
8df9675f | 1205 | for (i = 0; i < ngroups; i++) { |
af5bc92d | 1206 | struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); |
ac27a0ec DK |
1207 | if (!gdp) |
1208 | continue; | |
560671a0 | 1209 | count += ext4_used_dirs_count(sb, gdp); |
ac27a0ec DK |
1210 | } |
1211 | return count; | |
1212 | } |