1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext2/ialloc.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * BSD ufs-inspired inode and directory allocation by
11 * Stephen Tweedie (sct@dcs.ed.ac.uk), 1993
12 * Big-endian to little-endian byte-swapping/bitmaps by
13 * David S. Miller (davem@caip.rutgers.edu), 1995
14 */
15
16#include <linux/quotaops.h>
17#include <linux/sched.h>
18#include <linux/backing-dev.h>
19#include <linux/buffer_head.h>
20#include <linux/random.h>
21#include "ext2.h"
22#include "xattr.h"
23#include "acl.h"
24
25/*
26 * ialloc.c contains the inodes allocation and deallocation routines
27 */
28
29/*
30 * The free inodes are managed by bitmaps. A file system contains several
31 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
32 * block for inodes, N blocks for the inode table and data blocks.
33 *
34 * The file system contains group descriptors which are located after the
35 * super block. Each descriptor contains the number of the bitmap block and
36 * the free blocks count in the block.
37 */
38
39
40/*
41 * Read the inode allocation bitmap for a given block_group, reading
42 * into the specified slot in the superblock's bitmap cache.
43 *
44 * Return buffer_head of bitmap on success or NULL.
45 */
46static struct buffer_head *
47read_inode_bitmap(struct super_block * sb, unsigned long block_group)
48{
49 struct ext2_group_desc *desc;
50 struct buffer_head *bh = NULL;
51
52 desc = ext2_get_group_desc(sb, block_group, NULL);
53 if (!desc)
54 goto error_out;
55
56 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
57 if (!bh)
58 ext2_error(sb, "read_inode_bitmap",
59 "Cannot read inode bitmap - "
60 "block_group = %lu, inode_bitmap = %u",
61 block_group, le32_to_cpu(desc->bg_inode_bitmap));
62error_out:
63 return bh;
64}
65
66static void ext2_release_inode(struct super_block *sb, int group, int dir)
67{
68 struct ext2_group_desc * desc;
69 struct buffer_head *bh;
70
71 desc = ext2_get_group_desc(sb, block_group: group, bh: &bh);
72 if (!desc) {
73 ext2_error(sb, "ext2_release_inode",
74 "can't get descriptor for group %d", group);
75 return;
76 }
77
78 spin_lock(lock: sb_bgl_lock(sbi: EXT2_SB(sb), block_group: group));
79 le16_add_cpu(var: &desc->bg_free_inodes_count, val: 1);
80 if (dir)
81 le16_add_cpu(var: &desc->bg_used_dirs_count, val: -1);
82 spin_unlock(lock: sb_bgl_lock(sbi: EXT2_SB(sb), block_group: group));
83 percpu_counter_inc(fbc: &EXT2_SB(sb)->s_freeinodes_counter);
84 if (dir)
85 percpu_counter_dec(fbc: &EXT2_SB(sb)->s_dirs_counter);
86 mark_buffer_dirty(bh);
87}
88
89/*
90 * NOTE! When we get the inode, we're the only people
91 * that have access to it, and as such there are no
92 * race conditions we have to worry about. The inode
93 * is not on the hash-lists, and it cannot be reached
94 * through the filesystem because the directory entry
95 * has been deleted earlier.
96 *
97 * HOWEVER: we must make sure that we get no aliases,
98 * which means that we have to call "clear_inode()"
99 * _before_ we mark the inode not in use in the inode
100 * bitmaps. Otherwise a newly created file might use
101 * the same inode number (not actually the same pointer
102 * though), and then we'd have two inodes sharing the
103 * same inode number and space on the harddisk.
104 */
105void ext2_free_inode (struct inode * inode)
106{
107 struct super_block * sb = inode->i_sb;
108 int is_directory;
109 unsigned long ino;
110 struct buffer_head *bitmap_bh;
111 unsigned long block_group;
112 unsigned long bit;
113 struct ext2_super_block * es;
114
115 ino = inode->i_ino;
116 ext2_debug ("freeing inode %lu\n", ino);
117
118 /*
119 * Note: we must free any quota before locking the superblock,
120 * as writing the quota to disk may need the lock as well.
121 */
122 /* Quota is already initialized in iput() */
123 dquot_free_inode(inode);
124 dquot_drop(inode);
125
126 es = EXT2_SB(sb)->s_es;
127 is_directory = S_ISDIR(inode->i_mode);
128
129 if (ino < EXT2_FIRST_INO(sb) ||
130 ino > le32_to_cpu(es->s_inodes_count)) {
131 ext2_error (sb, "ext2_free_inode",
132 "reserved or nonexistent inode %lu", ino);
133 return;
134 }
135 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
136 bit = (ino - 1) % EXT2_INODES_PER_GROUP(sb);
137 bitmap_bh = read_inode_bitmap(sb, block_group);
138 if (!bitmap_bh)
139 return;
140
141 /* Ok, now we can actually update the inode bitmaps.. */
142 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb), block_group),
143 bit, (void *) bitmap_bh->b_data))
144 ext2_error (sb, "ext2_free_inode",
145 "bit already cleared for inode %lu", ino);
146 else
147 ext2_release_inode(sb, group: block_group, dir: is_directory);
148 mark_buffer_dirty(bh: bitmap_bh);
149 if (sb->s_flags & SB_SYNCHRONOUS)
150 sync_dirty_buffer(bh: bitmap_bh);
151
152 brelse(bh: bitmap_bh);
153}
154
155/*
156 * We perform asynchronous prereading of the new inode's inode block when
157 * we create the inode, in the expectation that the inode will be written
158 * back soon. There are two reasons:
159 *
160 * - When creating a large number of files, the async prereads will be
161 * nicely merged into large reads
162 * - When writing out a large number of inodes, we don't need to keep on
163 * stalling the writes while we read the inode block.
164 *
165 * FIXME: ext2_get_group_desc() needs to be simplified.
166 */
167static void ext2_preread_inode(struct inode *inode)
168{
169 unsigned long block_group;
170 unsigned long offset;
171 unsigned long block;
172 struct ext2_group_desc * gdp;
173
174 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
175 gdp = ext2_get_group_desc(sb: inode->i_sb, block_group, NULL);
176 if (gdp == NULL)
177 return;
178
179 /*
180 * Figure out the offset within the block group inode table
181 */
182 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) *
183 EXT2_INODE_SIZE(inode->i_sb);
184 block = le32_to_cpu(gdp->bg_inode_table) +
185 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb));
186 sb_breadahead(sb: inode->i_sb, block);
187}
188
189/*
190 * There are two policies for allocating an inode. If the new inode is
191 * a directory, then a forward search is made for a block group with both
192 * free space and a low directory-to-inode ratio; if that fails, then of
193 * the groups with above-average free space, that group with the fewest
194 * directories already is chosen.
195 *
196 * For other inodes, search forward from the parent directory\'s block
197 * group to find a free inode.
198 */
199static int find_group_dir(struct super_block *sb, struct inode *parent)
200{
201 int ngroups = EXT2_SB(sb)->s_groups_count;
202 int avefreei = ext2_count_free_inodes(sb) / ngroups;
203 struct ext2_group_desc *desc, *best_desc = NULL;
204 int group, best_group = -1;
205
206 for (group = 0; group < ngroups; group++) {
207 desc = ext2_get_group_desc (sb, block_group: group, NULL);
208 if (!desc || !desc->bg_free_inodes_count)
209 continue;
210 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
211 continue;
212 if (!best_desc ||
213 (le16_to_cpu(desc->bg_free_blocks_count) >
214 le16_to_cpu(best_desc->bg_free_blocks_count))) {
215 best_group = group;
216 best_desc = desc;
217 }
218 }
219
220 return best_group;
221}
222
223/*
224 * Orlov's allocator for directories.
225 *
226 * We always try to spread first-level directories.
227 *
228 * If there are blockgroups with both free inodes and free blocks counts
229 * not worse than average we return one with smallest directory count.
230 * Otherwise we simply return a random group.
231 *
232 * For the rest rules look so:
233 *
234 * It's OK to put directory into a group unless
235 * it has too many directories already (max_dirs) or
236 * it has too few free inodes left (min_inodes) or
237 * it has too few free blocks left (min_blocks) or
238 * it's already running too large debt (max_debt).
239 * Parent's group is preferred, if it doesn't satisfy these
240 * conditions we search cyclically through the rest. If none
241 * of the groups look good we just look for a group with more
242 * free inodes than average (starting at parent's group).
243 *
244 * Debt is incremented each time we allocate a directory and decremented
245 * when we allocate an inode, within 0--255.
246 */
247
248#define INODE_COST 64
249#define BLOCK_COST 256
250
251static int find_group_orlov(struct super_block *sb, struct inode *parent)
252{
253 int parent_group = EXT2_I(inode: parent)->i_block_group;
254 struct ext2_sb_info *sbi = EXT2_SB(sb);
255 struct ext2_super_block *es = sbi->s_es;
256 int ngroups = sbi->s_groups_count;
257 int inodes_per_group = EXT2_INODES_PER_GROUP(sb);
258 int freei;
259 int avefreei;
260 int free_blocks;
261 int avefreeb;
262 int blocks_per_dir;
263 int ndirs;
264 int max_debt, max_dirs, min_blocks, min_inodes;
265 int group = -1, i;
266 struct ext2_group_desc *desc;
267
268 freei = percpu_counter_read_positive(fbc: &sbi->s_freeinodes_counter);
269 avefreei = freei / ngroups;
270 free_blocks = percpu_counter_read_positive(fbc: &sbi->s_freeblocks_counter);
271 avefreeb = free_blocks / ngroups;
272 ndirs = percpu_counter_read_positive(fbc: &sbi->s_dirs_counter);
273
274 if ((parent == d_inode(dentry: sb->s_root)) ||
275 (EXT2_I(inode: parent)->i_flags & EXT2_TOPDIR_FL)) {
276 int best_ndir = inodes_per_group;
277 int best_group = -1;
278
279 parent_group = get_random_u32_below(ceil: ngroups);
280 for (i = 0; i < ngroups; i++) {
281 group = (parent_group + i) % ngroups;
282 desc = ext2_get_group_desc (sb, block_group: group, NULL);
283 if (!desc || !desc->bg_free_inodes_count)
284 continue;
285 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
286 continue;
287 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
288 continue;
289 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
290 continue;
291 best_group = group;
292 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
293 }
294 if (best_group >= 0) {
295 group = best_group;
296 goto found;
297 }
298 goto fallback;
299 }
300
301 if (ndirs == 0)
302 ndirs = 1; /* percpu_counters are approximate... */
303
304 blocks_per_dir = (le32_to_cpu(es->s_blocks_count)-free_blocks) / ndirs;
305
306 max_dirs = ndirs / ngroups + inodes_per_group / 16;
307 min_inodes = avefreei - inodes_per_group / 4;
308 min_blocks = avefreeb - EXT2_BLOCKS_PER_GROUP(sb) / 4;
309
310 max_debt = EXT2_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST);
311 if (max_debt * INODE_COST > inodes_per_group)
312 max_debt = inodes_per_group / INODE_COST;
313 if (max_debt > 255)
314 max_debt = 255;
315 if (max_debt == 0)
316 max_debt = 1;
317
318 for (i = 0; i < ngroups; i++) {
319 group = (parent_group + i) % ngroups;
320 desc = ext2_get_group_desc (sb, block_group: group, NULL);
321 if (!desc || !desc->bg_free_inodes_count)
322 continue;
323 if (sbi->s_debts[group] >= max_debt)
324 continue;
325 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
326 continue;
327 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
328 continue;
329 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
330 continue;
331 goto found;
332 }
333
334fallback:
335 for (i = 0; i < ngroups; i++) {
336 group = (parent_group + i) % ngroups;
337 desc = ext2_get_group_desc (sb, block_group: group, NULL);
338 if (!desc || !desc->bg_free_inodes_count)
339 continue;
340 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
341 goto found;
342 }
343
344 if (avefreei) {
345 /*
346 * The free-inodes counter is approximate, and for really small
347 * filesystems the above test can fail to find any blockgroups
348 */
349 avefreei = 0;
350 goto fallback;
351 }
352
353 return -1;
354
355found:
356 return group;
357}
358
359static int find_group_other(struct super_block *sb, struct inode *parent)
360{
361 int parent_group = EXT2_I(inode: parent)->i_block_group;
362 int ngroups = EXT2_SB(sb)->s_groups_count;
363 struct ext2_group_desc *desc;
364 int group, i;
365
366 /*
367 * Try to place the inode in its parent directory
368 */
369 group = parent_group;
370 desc = ext2_get_group_desc (sb, block_group: group, NULL);
371 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
372 le16_to_cpu(desc->bg_free_blocks_count))
373 goto found;
374
375 /*
376 * We're going to place this inode in a different blockgroup from its
377 * parent. We want to cause files in a common directory to all land in
378 * the same blockgroup. But we want files which are in a different
379 * directory which shares a blockgroup with our parent to land in a
380 * different blockgroup.
381 *
382 * So add our directory's i_ino into the starting point for the hash.
383 */
384 group = (group + parent->i_ino) % ngroups;
385
386 /*
387 * Use a quadratic hash to find a group with a free inode and some
388 * free blocks.
389 */
390 for (i = 1; i < ngroups; i <<= 1) {
391 group += i;
392 if (group >= ngroups)
393 group -= ngroups;
394 desc = ext2_get_group_desc (sb, block_group: group, NULL);
395 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
396 le16_to_cpu(desc->bg_free_blocks_count))
397 goto found;
398 }
399
400 /*
401 * That failed: try linear search for a free inode, even if that group
402 * has no free blocks.
403 */
404 group = parent_group;
405 for (i = 0; i < ngroups; i++) {
406 if (++group >= ngroups)
407 group = 0;
408 desc = ext2_get_group_desc (sb, block_group: group, NULL);
409 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
410 goto found;
411 }
412
413 return -1;
414
415found:
416 return group;
417}
418
419struct inode *ext2_new_inode(struct inode *dir, umode_t mode,
420 const struct qstr *qstr)
421{
422 struct super_block *sb;
423 struct buffer_head *bitmap_bh = NULL;
424 struct buffer_head *bh2;
425 int group, i;
426 ino_t ino = 0;
427 struct inode * inode;
428 struct ext2_group_desc *gdp;
429 struct ext2_super_block *es;
430 struct ext2_inode_info *ei;
431 struct ext2_sb_info *sbi;
432 int err;
433
434 sb = dir->i_sb;
435 inode = new_inode(sb);
436 if (!inode)
437 return ERR_PTR(error: -ENOMEM);
438
439 ei = EXT2_I(inode);
440 sbi = EXT2_SB(sb);
441 es = sbi->s_es;
442 if (S_ISDIR(mode)) {
443 if (test_opt(sb, OLDALLOC))
444 group = find_group_dir(sb, parent: dir);
445 else
446 group = find_group_orlov(sb, parent: dir);
447 } else
448 group = find_group_other(sb, parent: dir);
449
450 if (group == -1) {
451 err = -ENOSPC;
452 goto fail;
453 }
454
455 for (i = 0; i < sbi->s_groups_count; i++) {
456 gdp = ext2_get_group_desc(sb, block_group: group, bh: &bh2);
457 if (!gdp) {
458 if (++group == sbi->s_groups_count)
459 group = 0;
460 continue;
461 }
462 brelse(bh: bitmap_bh);
463 bitmap_bh = read_inode_bitmap(sb, block_group: group);
464 if (!bitmap_bh) {
465 err = -EIO;
466 goto fail;
467 }
468 ino = 0;
469
470repeat_in_this_group:
471 ino = ext2_find_next_zero_bit(addr: (unsigned long *)bitmap_bh->b_data,
472 EXT2_INODES_PER_GROUP(sb), offset: ino);
473 if (ino >= EXT2_INODES_PER_GROUP(sb)) {
474 /*
475 * Rare race: find_group_xx() decided that there were
476 * free inodes in this group, but by the time we tried
477 * to allocate one, they're all gone. This can also
478 * occur because the counters which find_group_orlov()
479 * uses are approximate. So just go and search the
480 * next block group.
481 */
482 if (++group == sbi->s_groups_count)
483 group = 0;
484 continue;
485 }
486 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group),
487 ino, bitmap_bh->b_data)) {
488 /* we lost this inode */
489 if (++ino >= EXT2_INODES_PER_GROUP(sb)) {
490 /* this group is exhausted, try next group */
491 if (++group == sbi->s_groups_count)
492 group = 0;
493 continue;
494 }
495 /* try to find free inode in the same group */
496 goto repeat_in_this_group;
497 }
498 goto got;
499 }
500
501 /*
502 * Scanned all blockgroups.
503 */
504 brelse(bh: bitmap_bh);
505 err = -ENOSPC;
506 goto fail;
507got:
508 mark_buffer_dirty(bh: bitmap_bh);
509 if (sb->s_flags & SB_SYNCHRONOUS)
510 sync_dirty_buffer(bh: bitmap_bh);
511 brelse(bh: bitmap_bh);
512
513 ino += group * EXT2_INODES_PER_GROUP(sb) + 1;
514 if (ino < EXT2_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
515 ext2_error (sb, "ext2_new_inode",
516 "reserved inode or inode > inodes count - "
517 "block_group = %d,inode=%lu", group,
518 (unsigned long) ino);
519 err = -EIO;
520 goto fail;
521 }
522
523 percpu_counter_dec(fbc: &sbi->s_freeinodes_counter);
524 if (S_ISDIR(mode))
525 percpu_counter_inc(fbc: &sbi->s_dirs_counter);
526
527 spin_lock(lock: sb_bgl_lock(sbi, block_group: group));
528 le16_add_cpu(var: &gdp->bg_free_inodes_count, val: -1);
529 if (S_ISDIR(mode)) {
530 if (sbi->s_debts[group] < 255)
531 sbi->s_debts[group]++;
532 le16_add_cpu(var: &gdp->bg_used_dirs_count, val: 1);
533 } else {
534 if (sbi->s_debts[group])
535 sbi->s_debts[group]--;
536 }
537 spin_unlock(lock: sb_bgl_lock(sbi, block_group: group));
538
539 mark_buffer_dirty(bh: bh2);
540 if (test_opt(sb, GRPID)) {
541 inode->i_mode = mode;
542 inode->i_uid = current_fsuid();
543 inode->i_gid = dir->i_gid;
544 } else
545 inode_init_owner(idmap: &nop_mnt_idmap, inode, dir, mode);
546
547 inode->i_ino = ino;
548 inode->i_blocks = 0;
549 simple_inode_init_ts(inode);
550 memset(ei->i_data, 0, sizeof(ei->i_data));
551 ei->i_flags =
552 ext2_mask_flags(mode, flags: EXT2_I(inode: dir)->i_flags & EXT2_FL_INHERITED);
553 ei->i_faddr = 0;
554 ei->i_frag_no = 0;
555 ei->i_frag_size = 0;
556 ei->i_file_acl = 0;
557 ei->i_dir_acl = 0;
558 ei->i_dtime = 0;
559 ei->i_block_alloc_info = NULL;
560 ei->i_block_group = group;
561 ei->i_dir_start_lookup = 0;
562 ei->i_state = EXT2_STATE_NEW;
563 ext2_set_inode_flags(inode);
564 spin_lock(lock: &sbi->s_next_gen_lock);
565 inode->i_generation = sbi->s_next_generation++;
566 spin_unlock(lock: &sbi->s_next_gen_lock);
567 if (insert_inode_locked(inode) < 0) {
568 ext2_error(sb, "ext2_new_inode",
569 "inode number already in use - inode=%lu",
570 (unsigned long) ino);
571 err = -EIO;
572 goto fail;
573 }
574
575 err = dquot_initialize(inode);
576 if (err)
577 goto fail_drop;
578
579 err = dquot_alloc_inode(inode);
580 if (err)
581 goto fail_drop;
582
583 err = ext2_init_acl(inode, dir);
584 if (err)
585 goto fail_free_drop;
586
587 err = ext2_init_security(inode, dir, qstr);
588 if (err)
589 goto fail_free_drop;
590
591 mark_inode_dirty(inode);
592 ext2_debug("allocating inode %lu\n", inode->i_ino);
593 ext2_preread_inode(inode);
594 return inode;
595
596fail_free_drop:
597 dquot_free_inode(inode);
598
599fail_drop:
600 dquot_drop(inode);
601 inode->i_flags |= S_NOQUOTA;
602 clear_nlink(inode);
603 discard_new_inode(inode);
604 return ERR_PTR(error: err);
605
606fail:
607 make_bad_inode(inode);
608 iput(inode);
609 return ERR_PTR(error: err);
610}
611
612unsigned long ext2_count_free_inodes (struct super_block * sb)
613{
614 struct ext2_group_desc *desc;
615 unsigned long desc_count = 0;
616 int i;
617
618#ifdef EXT2FS_DEBUG
619 struct ext2_super_block *es;
620 unsigned long bitmap_count = 0;
621 struct buffer_head *bitmap_bh = NULL;
622
623 es = EXT2_SB(sb)->s_es;
624 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
625 unsigned x;
626
627 desc = ext2_get_group_desc (sb, i, NULL);
628 if (!desc)
629 continue;
630 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
631 brelse(bitmap_bh);
632 bitmap_bh = read_inode_bitmap(sb, i);
633 if (!bitmap_bh)
634 continue;
635
636 x = ext2_count_free(bitmap_bh, EXT2_INODES_PER_GROUP(sb) / 8);
637 printk("group %d: stored = %d, counted = %u\n",
638 i, le16_to_cpu(desc->bg_free_inodes_count), x);
639 bitmap_count += x;
640 }
641 brelse(bitmap_bh);
642 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
643 (unsigned long)
644 percpu_counter_read(&EXT2_SB(sb)->s_freeinodes_counter),
645 desc_count, bitmap_count);
646 return desc_count;
647#else
648 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
649 desc = ext2_get_group_desc (sb, block_group: i, NULL);
650 if (!desc)
651 continue;
652 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
653 }
654 return desc_count;
655#endif
656}
657
658/* Called at mount-time, super-block is locked */
659unsigned long ext2_count_dirs (struct super_block * sb)
660{
661 unsigned long count = 0;
662 int i;
663
664 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
665 struct ext2_group_desc *gdp = ext2_get_group_desc (sb, block_group: i, NULL);
666 if (!gdp)
667 continue;
668 count += le16_to_cpu(gdp->bg_used_dirs_count);
669 }
670 return count;
671}
672
673

source code of linux/fs/ext2/ialloc.c