Loading...
1/*
2 * linux/fs/ext2/ialloc.c
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@dcs.ed.ac.uk), 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/quotaops.h>
16#include <linux/sched.h>
17#include <linux/backing-dev.h>
18#include <linux/buffer_head.h>
19#include <linux/random.h>
20#include "ext2.h"
21#include "xattr.h"
22#include "acl.h"
23
24/*
25 * ialloc.c contains the inodes allocation and deallocation routines
26 */
27
28/*
29 * The free inodes are managed by bitmaps. A file system contains several
30 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
31 * block for inodes, N blocks for the inode table and data blocks.
32 *
33 * The file system contains group descriptors which are located after the
34 * super block. Each descriptor contains the number of the bitmap block and
35 * the free blocks count in the block.
36 */
37
38
39/*
40 * Read the inode allocation bitmap for a given block_group, reading
41 * into the specified slot in the superblock's bitmap cache.
42 *
43 * Return buffer_head of bitmap on success or NULL.
44 */
45static struct buffer_head *
46read_inode_bitmap(struct super_block * sb, unsigned long block_group)
47{
48 struct ext2_group_desc *desc;
49 struct buffer_head *bh = NULL;
50
51 desc = ext2_get_group_desc(sb, block_group, NULL);
52 if (!desc)
53 goto error_out;
54
55 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
56 if (!bh)
57 ext2_error(sb, "read_inode_bitmap",
58 "Cannot read inode bitmap - "
59 "block_group = %lu, inode_bitmap = %u",
60 block_group, le32_to_cpu(desc->bg_inode_bitmap));
61error_out:
62 return bh;
63}
64
65static void ext2_release_inode(struct super_block *sb, int group, int dir)
66{
67 struct ext2_group_desc * desc;
68 struct buffer_head *bh;
69
70 desc = ext2_get_group_desc(sb, group, &bh);
71 if (!desc) {
72 ext2_error(sb, "ext2_release_inode",
73 "can't get descriptor for group %d", group);
74 return;
75 }
76
77 spin_lock(sb_bgl_lock(EXT2_SB(sb), group));
78 le16_add_cpu(&desc->bg_free_inodes_count, 1);
79 if (dir)
80 le16_add_cpu(&desc->bg_used_dirs_count, -1);
81 spin_unlock(sb_bgl_lock(EXT2_SB(sb), group));
82 if (dir)
83 percpu_counter_dec(&EXT2_SB(sb)->s_dirs_counter);
84 mark_buffer_dirty(bh);
85}
86
87/*
88 * NOTE! When we get the inode, we're the only people
89 * that have access to it, and as such there are no
90 * race conditions we have to worry about. The inode
91 * is not on the hash-lists, and it cannot be reached
92 * through the filesystem because the directory entry
93 * has been deleted earlier.
94 *
95 * HOWEVER: we must make sure that we get no aliases,
96 * which means that we have to call "clear_inode()"
97 * _before_ we mark the inode not in use in the inode
98 * bitmaps. Otherwise a newly created file might use
99 * the same inode number (not actually the same pointer
100 * though), and then we'd have two inodes sharing the
101 * same inode number and space on the harddisk.
102 */
103void ext2_free_inode (struct inode * inode)
104{
105 struct super_block * sb = inode->i_sb;
106 int is_directory;
107 unsigned long ino;
108 struct buffer_head *bitmap_bh;
109 unsigned long block_group;
110 unsigned long bit;
111 struct ext2_super_block * es;
112
113 ino = inode->i_ino;
114 ext2_debug ("freeing inode %lu\n", ino);
115
116 /*
117 * Note: we must free any quota before locking the superblock,
118 * as writing the quota to disk may need the lock as well.
119 */
120 /* Quota is already initialized in iput() */
121 dquot_free_inode(inode);
122 dquot_drop(inode);
123
124 es = EXT2_SB(sb)->s_es;
125 is_directory = S_ISDIR(inode->i_mode);
126
127 if (ino < EXT2_FIRST_INO(sb) ||
128 ino > le32_to_cpu(es->s_inodes_count)) {
129 ext2_error (sb, "ext2_free_inode",
130 "reserved or nonexistent inode %lu", ino);
131 return;
132 }
133 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
134 bit = (ino - 1) % EXT2_INODES_PER_GROUP(sb);
135 bitmap_bh = read_inode_bitmap(sb, block_group);
136 if (!bitmap_bh)
137 return;
138
139 /* Ok, now we can actually update the inode bitmaps.. */
140 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb), block_group),
141 bit, (void *) bitmap_bh->b_data))
142 ext2_error (sb, "ext2_free_inode",
143 "bit already cleared for inode %lu", ino);
144 else
145 ext2_release_inode(sb, block_group, is_directory);
146 mark_buffer_dirty(bitmap_bh);
147 if (sb->s_flags & MS_SYNCHRONOUS)
148 sync_dirty_buffer(bitmap_bh);
149
150 brelse(bitmap_bh);
151}
152
153/*
154 * We perform asynchronous prereading of the new inode's inode block when
155 * we create the inode, in the expectation that the inode will be written
156 * back soon. There are two reasons:
157 *
158 * - When creating a large number of files, the async prereads will be
159 * nicely merged into large reads
160 * - When writing out a large number of inodes, we don't need to keep on
161 * stalling the writes while we read the inode block.
162 *
163 * FIXME: ext2_get_group_desc() needs to be simplified.
164 */
165static void ext2_preread_inode(struct inode *inode)
166{
167 unsigned long block_group;
168 unsigned long offset;
169 unsigned long block;
170 struct ext2_group_desc * gdp;
171 struct backing_dev_info *bdi;
172
173 bdi = inode_to_bdi(inode);
174 if (bdi_read_congested(bdi))
175 return;
176 if (bdi_write_congested(bdi))
177 return;
178
179 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
180 gdp = ext2_get_group_desc(inode->i_sb, block_group, NULL);
181 if (gdp == NULL)
182 return;
183
184 /*
185 * Figure out the offset within the block group inode table
186 */
187 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) *
188 EXT2_INODE_SIZE(inode->i_sb);
189 block = le32_to_cpu(gdp->bg_inode_table) +
190 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb));
191 sb_breadahead(inode->i_sb, block);
192}
193
194/*
195 * There are two policies for allocating an inode. If the new inode is
196 * a directory, then a forward search is made for a block group with both
197 * free space and a low directory-to-inode ratio; if that fails, then of
198 * the groups with above-average free space, that group with the fewest
199 * directories already is chosen.
200 *
201 * For other inodes, search forward from the parent directory\'s block
202 * group to find a free inode.
203 */
204static int find_group_dir(struct super_block *sb, struct inode *parent)
205{
206 int ngroups = EXT2_SB(sb)->s_groups_count;
207 int avefreei = ext2_count_free_inodes(sb) / ngroups;
208 struct ext2_group_desc *desc, *best_desc = NULL;
209 int group, best_group = -1;
210
211 for (group = 0; group < ngroups; group++) {
212 desc = ext2_get_group_desc (sb, group, NULL);
213 if (!desc || !desc->bg_free_inodes_count)
214 continue;
215 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
216 continue;
217 if (!best_desc ||
218 (le16_to_cpu(desc->bg_free_blocks_count) >
219 le16_to_cpu(best_desc->bg_free_blocks_count))) {
220 best_group = group;
221 best_desc = desc;
222 }
223 }
224 if (!best_desc)
225 return -1;
226
227 return best_group;
228}
229
230/*
231 * Orlov's allocator for directories.
232 *
233 * We always try to spread first-level directories.
234 *
235 * If there are blockgroups with both free inodes and free blocks counts
236 * not worse than average we return one with smallest directory count.
237 * Otherwise we simply return a random group.
238 *
239 * For the rest rules look so:
240 *
241 * It's OK to put directory into a group unless
242 * it has too many directories already (max_dirs) or
243 * it has too few free inodes left (min_inodes) or
244 * it has too few free blocks left (min_blocks) or
245 * it's already running too large debt (max_debt).
246 * Parent's group is preferred, if it doesn't satisfy these
247 * conditions we search cyclically through the rest. If none
248 * of the groups look good we just look for a group with more
249 * free inodes than average (starting at parent's group).
250 *
251 * Debt is incremented each time we allocate a directory and decremented
252 * when we allocate an inode, within 0--255.
253 */
254
255#define INODE_COST 64
256#define BLOCK_COST 256
257
258static int find_group_orlov(struct super_block *sb, struct inode *parent)
259{
260 int parent_group = EXT2_I(parent)->i_block_group;
261 struct ext2_sb_info *sbi = EXT2_SB(sb);
262 struct ext2_super_block *es = sbi->s_es;
263 int ngroups = sbi->s_groups_count;
264 int inodes_per_group = EXT2_INODES_PER_GROUP(sb);
265 int freei;
266 int avefreei;
267 int free_blocks;
268 int avefreeb;
269 int blocks_per_dir;
270 int ndirs;
271 int max_debt, max_dirs, min_blocks, min_inodes;
272 int group = -1, i;
273 struct ext2_group_desc *desc;
274
275 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
276 avefreei = freei / ngroups;
277 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
278 avefreeb = free_blocks / ngroups;
279 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
280
281 if ((parent == d_inode(sb->s_root)) ||
282 (EXT2_I(parent)->i_flags & EXT2_TOPDIR_FL)) {
283 struct ext2_group_desc *best_desc = NULL;
284 int best_ndir = inodes_per_group;
285 int best_group = -1;
286
287 group = prandom_u32();
288 parent_group = (unsigned)group % ngroups;
289 for (i = 0; i < ngroups; i++) {
290 group = (parent_group + i) % ngroups;
291 desc = ext2_get_group_desc (sb, group, NULL);
292 if (!desc || !desc->bg_free_inodes_count)
293 continue;
294 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
295 continue;
296 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
297 continue;
298 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
299 continue;
300 best_group = group;
301 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
302 best_desc = desc;
303 }
304 if (best_group >= 0) {
305 desc = best_desc;
306 group = best_group;
307 goto found;
308 }
309 goto fallback;
310 }
311
312 if (ndirs == 0)
313 ndirs = 1; /* percpu_counters are approximate... */
314
315 blocks_per_dir = (le32_to_cpu(es->s_blocks_count)-free_blocks) / ndirs;
316
317 max_dirs = ndirs / ngroups + inodes_per_group / 16;
318 min_inodes = avefreei - inodes_per_group / 4;
319 min_blocks = avefreeb - EXT2_BLOCKS_PER_GROUP(sb) / 4;
320
321 max_debt = EXT2_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST);
322 if (max_debt * INODE_COST > inodes_per_group)
323 max_debt = inodes_per_group / INODE_COST;
324 if (max_debt > 255)
325 max_debt = 255;
326 if (max_debt == 0)
327 max_debt = 1;
328
329 for (i = 0; i < ngroups; i++) {
330 group = (parent_group + i) % ngroups;
331 desc = ext2_get_group_desc (sb, group, NULL);
332 if (!desc || !desc->bg_free_inodes_count)
333 continue;
334 if (sbi->s_debts[group] >= max_debt)
335 continue;
336 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
337 continue;
338 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
339 continue;
340 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
341 continue;
342 goto found;
343 }
344
345fallback:
346 for (i = 0; i < ngroups; i++) {
347 group = (parent_group + i) % ngroups;
348 desc = ext2_get_group_desc (sb, group, NULL);
349 if (!desc || !desc->bg_free_inodes_count)
350 continue;
351 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
352 goto found;
353 }
354
355 if (avefreei) {
356 /*
357 * The free-inodes counter is approximate, and for really small
358 * filesystems the above test can fail to find any blockgroups
359 */
360 avefreei = 0;
361 goto fallback;
362 }
363
364 return -1;
365
366found:
367 return group;
368}
369
370static int find_group_other(struct super_block *sb, struct inode *parent)
371{
372 int parent_group = EXT2_I(parent)->i_block_group;
373 int ngroups = EXT2_SB(sb)->s_groups_count;
374 struct ext2_group_desc *desc;
375 int group, i;
376
377 /*
378 * Try to place the inode in its parent directory
379 */
380 group = parent_group;
381 desc = ext2_get_group_desc (sb, group, NULL);
382 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
383 le16_to_cpu(desc->bg_free_blocks_count))
384 goto found;
385
386 /*
387 * We're going to place this inode in a different blockgroup from its
388 * parent. We want to cause files in a common directory to all land in
389 * the same blockgroup. But we want files which are in a different
390 * directory which shares a blockgroup with our parent to land in a
391 * different blockgroup.
392 *
393 * So add our directory's i_ino into the starting point for the hash.
394 */
395 group = (group + parent->i_ino) % ngroups;
396
397 /*
398 * Use a quadratic hash to find a group with a free inode and some
399 * free blocks.
400 */
401 for (i = 1; i < ngroups; i <<= 1) {
402 group += i;
403 if (group >= ngroups)
404 group -= ngroups;
405 desc = ext2_get_group_desc (sb, group, NULL);
406 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
407 le16_to_cpu(desc->bg_free_blocks_count))
408 goto found;
409 }
410
411 /*
412 * That failed: try linear search for a free inode, even if that group
413 * has no free blocks.
414 */
415 group = parent_group;
416 for (i = 0; i < ngroups; i++) {
417 if (++group >= ngroups)
418 group = 0;
419 desc = ext2_get_group_desc (sb, group, NULL);
420 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
421 goto found;
422 }
423
424 return -1;
425
426found:
427 return group;
428}
429
430struct inode *ext2_new_inode(struct inode *dir, umode_t mode,
431 const struct qstr *qstr)
432{
433 struct super_block *sb;
434 struct buffer_head *bitmap_bh = NULL;
435 struct buffer_head *bh2;
436 int group, i;
437 ino_t ino = 0;
438 struct inode * inode;
439 struct ext2_group_desc *gdp;
440 struct ext2_super_block *es;
441 struct ext2_inode_info *ei;
442 struct ext2_sb_info *sbi;
443 int err;
444
445 sb = dir->i_sb;
446 inode = new_inode(sb);
447 if (!inode)
448 return ERR_PTR(-ENOMEM);
449
450 ei = EXT2_I(inode);
451 sbi = EXT2_SB(sb);
452 es = sbi->s_es;
453 if (S_ISDIR(mode)) {
454 if (test_opt(sb, OLDALLOC))
455 group = find_group_dir(sb, dir);
456 else
457 group = find_group_orlov(sb, dir);
458 } else
459 group = find_group_other(sb, dir);
460
461 if (group == -1) {
462 err = -ENOSPC;
463 goto fail;
464 }
465
466 for (i = 0; i < sbi->s_groups_count; i++) {
467 gdp = ext2_get_group_desc(sb, group, &bh2);
468 brelse(bitmap_bh);
469 bitmap_bh = read_inode_bitmap(sb, group);
470 if (!bitmap_bh) {
471 err = -EIO;
472 goto fail;
473 }
474 ino = 0;
475
476repeat_in_this_group:
477 ino = ext2_find_next_zero_bit((unsigned long *)bitmap_bh->b_data,
478 EXT2_INODES_PER_GROUP(sb), ino);
479 if (ino >= EXT2_INODES_PER_GROUP(sb)) {
480 /*
481 * Rare race: find_group_xx() decided that there were
482 * free inodes in this group, but by the time we tried
483 * to allocate one, they're all gone. This can also
484 * occur because the counters which find_group_orlov()
485 * uses are approximate. So just go and search the
486 * next block group.
487 */
488 if (++group == sbi->s_groups_count)
489 group = 0;
490 continue;
491 }
492 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group),
493 ino, bitmap_bh->b_data)) {
494 /* we lost this inode */
495 if (++ino >= EXT2_INODES_PER_GROUP(sb)) {
496 /* this group is exhausted, try next group */
497 if (++group == sbi->s_groups_count)
498 group = 0;
499 continue;
500 }
501 /* try to find free inode in the same group */
502 goto repeat_in_this_group;
503 }
504 goto got;
505 }
506
507 /*
508 * Scanned all blockgroups.
509 */
510 err = -ENOSPC;
511 goto fail;
512got:
513 mark_buffer_dirty(bitmap_bh);
514 if (sb->s_flags & MS_SYNCHRONOUS)
515 sync_dirty_buffer(bitmap_bh);
516 brelse(bitmap_bh);
517
518 ino += group * EXT2_INODES_PER_GROUP(sb) + 1;
519 if (ino < EXT2_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
520 ext2_error (sb, "ext2_new_inode",
521 "reserved inode or inode > inodes count - "
522 "block_group = %d,inode=%lu", group,
523 (unsigned long) ino);
524 err = -EIO;
525 goto fail;
526 }
527
528 percpu_counter_add(&sbi->s_freeinodes_counter, -1);
529 if (S_ISDIR(mode))
530 percpu_counter_inc(&sbi->s_dirs_counter);
531
532 spin_lock(sb_bgl_lock(sbi, group));
533 le16_add_cpu(&gdp->bg_free_inodes_count, -1);
534 if (S_ISDIR(mode)) {
535 if (sbi->s_debts[group] < 255)
536 sbi->s_debts[group]++;
537 le16_add_cpu(&gdp->bg_used_dirs_count, 1);
538 } else {
539 if (sbi->s_debts[group])
540 sbi->s_debts[group]--;
541 }
542 spin_unlock(sb_bgl_lock(sbi, group));
543
544 mark_buffer_dirty(bh2);
545 if (test_opt(sb, GRPID)) {
546 inode->i_mode = mode;
547 inode->i_uid = current_fsuid();
548 inode->i_gid = dir->i_gid;
549 } else
550 inode_init_owner(inode, dir, mode);
551
552 inode->i_ino = ino;
553 inode->i_blocks = 0;
554 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
555 memset(ei->i_data, 0, sizeof(ei->i_data));
556 ei->i_flags =
557 ext2_mask_flags(mode, EXT2_I(dir)->i_flags & EXT2_FL_INHERITED);
558 ei->i_faddr = 0;
559 ei->i_frag_no = 0;
560 ei->i_frag_size = 0;
561 ei->i_file_acl = 0;
562 ei->i_dir_acl = 0;
563 ei->i_dtime = 0;
564 ei->i_block_alloc_info = NULL;
565 ei->i_block_group = group;
566 ei->i_dir_start_lookup = 0;
567 ei->i_state = EXT2_STATE_NEW;
568 ext2_set_inode_flags(inode);
569 spin_lock(&sbi->s_next_gen_lock);
570 inode->i_generation = sbi->s_next_generation++;
571 spin_unlock(&sbi->s_next_gen_lock);
572 if (insert_inode_locked(inode) < 0) {
573 ext2_error(sb, "ext2_new_inode",
574 "inode number already in use - inode=%lu",
575 (unsigned long) ino);
576 err = -EIO;
577 goto fail;
578 }
579
580 err = dquot_initialize(inode);
581 if (err)
582 goto fail_drop;
583
584 err = dquot_alloc_inode(inode);
585 if (err)
586 goto fail_drop;
587
588 err = ext2_init_acl(inode, dir);
589 if (err)
590 goto fail_free_drop;
591
592 err = ext2_init_security(inode, dir, qstr);
593 if (err)
594 goto fail_free_drop;
595
596 mark_inode_dirty(inode);
597 ext2_debug("allocating inode %lu\n", inode->i_ino);
598 ext2_preread_inode(inode);
599 return inode;
600
601fail_free_drop:
602 dquot_free_inode(inode);
603
604fail_drop:
605 dquot_drop(inode);
606 inode->i_flags |= S_NOQUOTA;
607 clear_nlink(inode);
608 unlock_new_inode(inode);
609 iput(inode);
610 return ERR_PTR(err);
611
612fail:
613 make_bad_inode(inode);
614 iput(inode);
615 return ERR_PTR(err);
616}
617
618unsigned long ext2_count_free_inodes (struct super_block * sb)
619{
620 struct ext2_group_desc *desc;
621 unsigned long desc_count = 0;
622 int i;
623
624#ifdef EXT2FS_DEBUG
625 struct ext2_super_block *es;
626 unsigned long bitmap_count = 0;
627 struct buffer_head *bitmap_bh = NULL;
628
629 es = EXT2_SB(sb)->s_es;
630 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
631 unsigned x;
632
633 desc = ext2_get_group_desc (sb, i, NULL);
634 if (!desc)
635 continue;
636 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
637 brelse(bitmap_bh);
638 bitmap_bh = read_inode_bitmap(sb, i);
639 if (!bitmap_bh)
640 continue;
641
642 x = ext2_count_free(bitmap_bh, EXT2_INODES_PER_GROUP(sb) / 8);
643 printk("group %d: stored = %d, counted = %u\n",
644 i, le16_to_cpu(desc->bg_free_inodes_count), x);
645 bitmap_count += x;
646 }
647 brelse(bitmap_bh);
648 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
649 (unsigned long)
650 percpu_counter_read(&EXT2_SB(sb)->s_freeinodes_counter),
651 desc_count, bitmap_count);
652 return desc_count;
653#else
654 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
655 desc = ext2_get_group_desc (sb, i, NULL);
656 if (!desc)
657 continue;
658 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
659 }
660 return desc_count;
661#endif
662}
663
664/* Called at mount-time, super-block is locked */
665unsigned long ext2_count_dirs (struct super_block * sb)
666{
667 unsigned long count = 0;
668 int i;
669
670 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
671 struct ext2_group_desc *gdp = ext2_get_group_desc (sb, i, NULL);
672 if (!gdp)
673 continue;
674 count += le16_to_cpu(gdp->bg_used_dirs_count);
675 }
676 return count;
677}
678
1/*
2 * linux/fs/ext2/ialloc.c
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@dcs.ed.ac.uk), 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/quotaops.h>
16#include <linux/sched.h>
17#include <linux/backing-dev.h>
18#include <linux/buffer_head.h>
19#include <linux/random.h>
20#include "ext2.h"
21#include "xattr.h"
22#include "acl.h"
23
24/*
25 * ialloc.c contains the inodes allocation and deallocation routines
26 */
27
28/*
29 * The free inodes are managed by bitmaps. A file system contains several
30 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
31 * block for inodes, N blocks for the inode table and data blocks.
32 *
33 * The file system contains group descriptors which are located after the
34 * super block. Each descriptor contains the number of the bitmap block and
35 * the free blocks count in the block.
36 */
37
38
39/*
40 * Read the inode allocation bitmap for a given block_group, reading
41 * into the specified slot in the superblock's bitmap cache.
42 *
43 * Return buffer_head of bitmap on success or NULL.
44 */
45static struct buffer_head *
46read_inode_bitmap(struct super_block * sb, unsigned long block_group)
47{
48 struct ext2_group_desc *desc;
49 struct buffer_head *bh = NULL;
50
51 desc = ext2_get_group_desc(sb, block_group, NULL);
52 if (!desc)
53 goto error_out;
54
55 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
56 if (!bh)
57 ext2_error(sb, "read_inode_bitmap",
58 "Cannot read inode bitmap - "
59 "block_group = %lu, inode_bitmap = %u",
60 block_group, le32_to_cpu(desc->bg_inode_bitmap));
61error_out:
62 return bh;
63}
64
65static void ext2_release_inode(struct super_block *sb, int group, int dir)
66{
67 struct ext2_group_desc * desc;
68 struct buffer_head *bh;
69
70 desc = ext2_get_group_desc(sb, group, &bh);
71 if (!desc) {
72 ext2_error(sb, "ext2_release_inode",
73 "can't get descriptor for group %d", group);
74 return;
75 }
76
77 spin_lock(sb_bgl_lock(EXT2_SB(sb), group));
78 le16_add_cpu(&desc->bg_free_inodes_count, 1);
79 if (dir)
80 le16_add_cpu(&desc->bg_used_dirs_count, -1);
81 spin_unlock(sb_bgl_lock(EXT2_SB(sb), group));
82 if (dir)
83 percpu_counter_dec(&EXT2_SB(sb)->s_dirs_counter);
84 mark_buffer_dirty(bh);
85}
86
87/*
88 * NOTE! When we get the inode, we're the only people
89 * that have access to it, and as such there are no
90 * race conditions we have to worry about. The inode
91 * is not on the hash-lists, and it cannot be reached
92 * through the filesystem because the directory entry
93 * has been deleted earlier.
94 *
95 * HOWEVER: we must make sure that we get no aliases,
96 * which means that we have to call "clear_inode()"
97 * _before_ we mark the inode not in use in the inode
98 * bitmaps. Otherwise a newly created file might use
99 * the same inode number (not actually the same pointer
100 * though), and then we'd have two inodes sharing the
101 * same inode number and space on the harddisk.
102 */
103void ext2_free_inode (struct inode * inode)
104{
105 struct super_block * sb = inode->i_sb;
106 int is_directory;
107 unsigned long ino;
108 struct buffer_head *bitmap_bh;
109 unsigned long block_group;
110 unsigned long bit;
111 struct ext2_super_block * es;
112
113 ino = inode->i_ino;
114 ext2_debug ("freeing inode %lu\n", ino);
115
116 /*
117 * Note: we must free any quota before locking the superblock,
118 * as writing the quota to disk may need the lock as well.
119 */
120 /* Quota is already initialized in iput() */
121 ext2_xattr_delete_inode(inode);
122 dquot_free_inode(inode);
123 dquot_drop(inode);
124
125 es = EXT2_SB(sb)->s_es;
126 is_directory = S_ISDIR(inode->i_mode);
127
128 if (ino < EXT2_FIRST_INO(sb) ||
129 ino > le32_to_cpu(es->s_inodes_count)) {
130 ext2_error (sb, "ext2_free_inode",
131 "reserved or nonexistent inode %lu", ino);
132 return;
133 }
134 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
135 bit = (ino - 1) % EXT2_INODES_PER_GROUP(sb);
136 bitmap_bh = read_inode_bitmap(sb, block_group);
137 if (!bitmap_bh)
138 return;
139
140 /* Ok, now we can actually update the inode bitmaps.. */
141 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb), block_group),
142 bit, (void *) bitmap_bh->b_data))
143 ext2_error (sb, "ext2_free_inode",
144 "bit already cleared for inode %lu", ino);
145 else
146 ext2_release_inode(sb, block_group, is_directory);
147 mark_buffer_dirty(bitmap_bh);
148 if (sb->s_flags & MS_SYNCHRONOUS)
149 sync_dirty_buffer(bitmap_bh);
150
151 brelse(bitmap_bh);
152}
153
154/*
155 * We perform asynchronous prereading of the new inode's inode block when
156 * we create the inode, in the expectation that the inode will be written
157 * back soon. There are two reasons:
158 *
159 * - When creating a large number of files, the async prereads will be
160 * nicely merged into large reads
161 * - When writing out a large number of inodes, we don't need to keep on
162 * stalling the writes while we read the inode block.
163 *
164 * FIXME: ext2_get_group_desc() needs to be simplified.
165 */
166static void ext2_preread_inode(struct inode *inode)
167{
168 unsigned long block_group;
169 unsigned long offset;
170 unsigned long block;
171 struct ext2_group_desc * gdp;
172 struct backing_dev_info *bdi;
173
174 bdi = inode->i_mapping->backing_dev_info;
175 if (bdi_read_congested(bdi))
176 return;
177 if (bdi_write_congested(bdi))
178 return;
179
180 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
181 gdp = ext2_get_group_desc(inode->i_sb, block_group, NULL);
182 if (gdp == NULL)
183 return;
184
185 /*
186 * Figure out the offset within the block group inode table
187 */
188 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) *
189 EXT2_INODE_SIZE(inode->i_sb);
190 block = le32_to_cpu(gdp->bg_inode_table) +
191 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb));
192 sb_breadahead(inode->i_sb, block);
193}
194
195/*
196 * There are two policies for allocating an inode. If the new inode is
197 * a directory, then a forward search is made for a block group with both
198 * free space and a low directory-to-inode ratio; if that fails, then of
199 * the groups with above-average free space, that group with the fewest
200 * directories already is chosen.
201 *
202 * For other inodes, search forward from the parent directory\'s block
203 * group to find a free inode.
204 */
205static int find_group_dir(struct super_block *sb, struct inode *parent)
206{
207 int ngroups = EXT2_SB(sb)->s_groups_count;
208 int avefreei = ext2_count_free_inodes(sb) / ngroups;
209 struct ext2_group_desc *desc, *best_desc = NULL;
210 int group, best_group = -1;
211
212 for (group = 0; group < ngroups; group++) {
213 desc = ext2_get_group_desc (sb, group, NULL);
214 if (!desc || !desc->bg_free_inodes_count)
215 continue;
216 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
217 continue;
218 if (!best_desc ||
219 (le16_to_cpu(desc->bg_free_blocks_count) >
220 le16_to_cpu(best_desc->bg_free_blocks_count))) {
221 best_group = group;
222 best_desc = desc;
223 }
224 }
225 if (!best_desc)
226 return -1;
227
228 return best_group;
229}
230
231/*
232 * Orlov's allocator for directories.
233 *
234 * We always try to spread first-level directories.
235 *
236 * If there are blockgroups with both free inodes and free blocks counts
237 * not worse than average we return one with smallest directory count.
238 * Otherwise we simply return a random group.
239 *
240 * For the rest rules look so:
241 *
242 * It's OK to put directory into a group unless
243 * it has too many directories already (max_dirs) or
244 * it has too few free inodes left (min_inodes) or
245 * it has too few free blocks left (min_blocks) or
246 * it's already running too large debt (max_debt).
247 * Parent's group is preferred, if it doesn't satisfy these
248 * conditions we search cyclically through the rest. If none
249 * of the groups look good we just look for a group with more
250 * free inodes than average (starting at parent's group).
251 *
252 * Debt is incremented each time we allocate a directory and decremented
253 * when we allocate an inode, within 0--255.
254 */
255
256#define INODE_COST 64
257#define BLOCK_COST 256
258
259static int find_group_orlov(struct super_block *sb, struct inode *parent)
260{
261 int parent_group = EXT2_I(parent)->i_block_group;
262 struct ext2_sb_info *sbi = EXT2_SB(sb);
263 struct ext2_super_block *es = sbi->s_es;
264 int ngroups = sbi->s_groups_count;
265 int inodes_per_group = EXT2_INODES_PER_GROUP(sb);
266 int freei;
267 int avefreei;
268 int free_blocks;
269 int avefreeb;
270 int blocks_per_dir;
271 int ndirs;
272 int max_debt, max_dirs, min_blocks, min_inodes;
273 int group = -1, i;
274 struct ext2_group_desc *desc;
275
276 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
277 avefreei = freei / ngroups;
278 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
279 avefreeb = free_blocks / ngroups;
280 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
281
282 if ((parent == sb->s_root->d_inode) ||
283 (EXT2_I(parent)->i_flags & EXT2_TOPDIR_FL)) {
284 struct ext2_group_desc *best_desc = NULL;
285 int best_ndir = inodes_per_group;
286 int best_group = -1;
287
288 get_random_bytes(&group, sizeof(group));
289 parent_group = (unsigned)group % ngroups;
290 for (i = 0; i < ngroups; i++) {
291 group = (parent_group + i) % ngroups;
292 desc = ext2_get_group_desc (sb, group, NULL);
293 if (!desc || !desc->bg_free_inodes_count)
294 continue;
295 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
296 continue;
297 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
298 continue;
299 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
300 continue;
301 best_group = group;
302 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
303 best_desc = desc;
304 }
305 if (best_group >= 0) {
306 desc = best_desc;
307 group = best_group;
308 goto found;
309 }
310 goto fallback;
311 }
312
313 if (ndirs == 0)
314 ndirs = 1; /* percpu_counters are approximate... */
315
316 blocks_per_dir = (le32_to_cpu(es->s_blocks_count)-free_blocks) / ndirs;
317
318 max_dirs = ndirs / ngroups + inodes_per_group / 16;
319 min_inodes = avefreei - inodes_per_group / 4;
320 min_blocks = avefreeb - EXT2_BLOCKS_PER_GROUP(sb) / 4;
321
322 max_debt = EXT2_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST);
323 if (max_debt * INODE_COST > inodes_per_group)
324 max_debt = inodes_per_group / INODE_COST;
325 if (max_debt > 255)
326 max_debt = 255;
327 if (max_debt == 0)
328 max_debt = 1;
329
330 for (i = 0; i < ngroups; i++) {
331 group = (parent_group + i) % ngroups;
332 desc = ext2_get_group_desc (sb, group, NULL);
333 if (!desc || !desc->bg_free_inodes_count)
334 continue;
335 if (sbi->s_debts[group] >= max_debt)
336 continue;
337 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
338 continue;
339 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
340 continue;
341 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
342 continue;
343 goto found;
344 }
345
346fallback:
347 for (i = 0; i < ngroups; i++) {
348 group = (parent_group + i) % ngroups;
349 desc = ext2_get_group_desc (sb, group, NULL);
350 if (!desc || !desc->bg_free_inodes_count)
351 continue;
352 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
353 goto found;
354 }
355
356 if (avefreei) {
357 /*
358 * The free-inodes counter is approximate, and for really small
359 * filesystems the above test can fail to find any blockgroups
360 */
361 avefreei = 0;
362 goto fallback;
363 }
364
365 return -1;
366
367found:
368 return group;
369}
370
371static int find_group_other(struct super_block *sb, struct inode *parent)
372{
373 int parent_group = EXT2_I(parent)->i_block_group;
374 int ngroups = EXT2_SB(sb)->s_groups_count;
375 struct ext2_group_desc *desc;
376 int group, i;
377
378 /*
379 * Try to place the inode in its parent directory
380 */
381 group = parent_group;
382 desc = ext2_get_group_desc (sb, group, NULL);
383 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
384 le16_to_cpu(desc->bg_free_blocks_count))
385 goto found;
386
387 /*
388 * We're going to place this inode in a different blockgroup from its
389 * parent. We want to cause files in a common directory to all land in
390 * the same blockgroup. But we want files which are in a different
391 * directory which shares a blockgroup with our parent to land in a
392 * different blockgroup.
393 *
394 * So add our directory's i_ino into the starting point for the hash.
395 */
396 group = (group + parent->i_ino) % ngroups;
397
398 /*
399 * Use a quadratic hash to find a group with a free inode and some
400 * free blocks.
401 */
402 for (i = 1; i < ngroups; i <<= 1) {
403 group += i;
404 if (group >= ngroups)
405 group -= ngroups;
406 desc = ext2_get_group_desc (sb, group, NULL);
407 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
408 le16_to_cpu(desc->bg_free_blocks_count))
409 goto found;
410 }
411
412 /*
413 * That failed: try linear search for a free inode, even if that group
414 * has no free blocks.
415 */
416 group = parent_group;
417 for (i = 0; i < ngroups; i++) {
418 if (++group >= ngroups)
419 group = 0;
420 desc = ext2_get_group_desc (sb, group, NULL);
421 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
422 goto found;
423 }
424
425 return -1;
426
427found:
428 return group;
429}
430
431struct inode *ext2_new_inode(struct inode *dir, umode_t mode,
432 const struct qstr *qstr)
433{
434 struct super_block *sb;
435 struct buffer_head *bitmap_bh = NULL;
436 struct buffer_head *bh2;
437 int group, i;
438 ino_t ino = 0;
439 struct inode * inode;
440 struct ext2_group_desc *gdp;
441 struct ext2_super_block *es;
442 struct ext2_inode_info *ei;
443 struct ext2_sb_info *sbi;
444 int err;
445
446 sb = dir->i_sb;
447 inode = new_inode(sb);
448 if (!inode)
449 return ERR_PTR(-ENOMEM);
450
451 ei = EXT2_I(inode);
452 sbi = EXT2_SB(sb);
453 es = sbi->s_es;
454 if (S_ISDIR(mode)) {
455 if (test_opt(sb, OLDALLOC))
456 group = find_group_dir(sb, dir);
457 else
458 group = find_group_orlov(sb, dir);
459 } else
460 group = find_group_other(sb, dir);
461
462 if (group == -1) {
463 err = -ENOSPC;
464 goto fail;
465 }
466
467 for (i = 0; i < sbi->s_groups_count; i++) {
468 gdp = ext2_get_group_desc(sb, group, &bh2);
469 brelse(bitmap_bh);
470 bitmap_bh = read_inode_bitmap(sb, group);
471 if (!bitmap_bh) {
472 err = -EIO;
473 goto fail;
474 }
475 ino = 0;
476
477repeat_in_this_group:
478 ino = ext2_find_next_zero_bit((unsigned long *)bitmap_bh->b_data,
479 EXT2_INODES_PER_GROUP(sb), ino);
480 if (ino >= EXT2_INODES_PER_GROUP(sb)) {
481 /*
482 * Rare race: find_group_xx() decided that there were
483 * free inodes in this group, but by the time we tried
484 * to allocate one, they're all gone. This can also
485 * occur because the counters which find_group_orlov()
486 * uses are approximate. So just go and search the
487 * next block group.
488 */
489 if (++group == sbi->s_groups_count)
490 group = 0;
491 continue;
492 }
493 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group),
494 ino, bitmap_bh->b_data)) {
495 /* we lost this inode */
496 if (++ino >= EXT2_INODES_PER_GROUP(sb)) {
497 /* this group is exhausted, try next group */
498 if (++group == sbi->s_groups_count)
499 group = 0;
500 continue;
501 }
502 /* try to find free inode in the same group */
503 goto repeat_in_this_group;
504 }
505 goto got;
506 }
507
508 /*
509 * Scanned all blockgroups.
510 */
511 err = -ENOSPC;
512 goto fail;
513got:
514 mark_buffer_dirty(bitmap_bh);
515 if (sb->s_flags & MS_SYNCHRONOUS)
516 sync_dirty_buffer(bitmap_bh);
517 brelse(bitmap_bh);
518
519 ino += group * EXT2_INODES_PER_GROUP(sb) + 1;
520 if (ino < EXT2_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
521 ext2_error (sb, "ext2_new_inode",
522 "reserved inode or inode > inodes count - "
523 "block_group = %d,inode=%lu", group,
524 (unsigned long) ino);
525 err = -EIO;
526 goto fail;
527 }
528
529 percpu_counter_add(&sbi->s_freeinodes_counter, -1);
530 if (S_ISDIR(mode))
531 percpu_counter_inc(&sbi->s_dirs_counter);
532
533 spin_lock(sb_bgl_lock(sbi, group));
534 le16_add_cpu(&gdp->bg_free_inodes_count, -1);
535 if (S_ISDIR(mode)) {
536 if (sbi->s_debts[group] < 255)
537 sbi->s_debts[group]++;
538 le16_add_cpu(&gdp->bg_used_dirs_count, 1);
539 } else {
540 if (sbi->s_debts[group])
541 sbi->s_debts[group]--;
542 }
543 spin_unlock(sb_bgl_lock(sbi, group));
544
545 mark_buffer_dirty(bh2);
546 if (test_opt(sb, GRPID)) {
547 inode->i_mode = mode;
548 inode->i_uid = current_fsuid();
549 inode->i_gid = dir->i_gid;
550 } else
551 inode_init_owner(inode, dir, mode);
552
553 inode->i_ino = ino;
554 inode->i_blocks = 0;
555 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
556 memset(ei->i_data, 0, sizeof(ei->i_data));
557 ei->i_flags =
558 ext2_mask_flags(mode, EXT2_I(dir)->i_flags & EXT2_FL_INHERITED);
559 ei->i_faddr = 0;
560 ei->i_frag_no = 0;
561 ei->i_frag_size = 0;
562 ei->i_file_acl = 0;
563 ei->i_dir_acl = 0;
564 ei->i_dtime = 0;
565 ei->i_block_alloc_info = NULL;
566 ei->i_block_group = group;
567 ei->i_dir_start_lookup = 0;
568 ei->i_state = EXT2_STATE_NEW;
569 ext2_set_inode_flags(inode);
570 spin_lock(&sbi->s_next_gen_lock);
571 inode->i_generation = sbi->s_next_generation++;
572 spin_unlock(&sbi->s_next_gen_lock);
573 if (insert_inode_locked(inode) < 0) {
574 ext2_error(sb, "ext2_new_inode",
575 "inode number already in use - inode=%lu",
576 (unsigned long) ino);
577 err = -EIO;
578 goto fail;
579 }
580
581 dquot_initialize(inode);
582 err = dquot_alloc_inode(inode);
583 if (err)
584 goto fail_drop;
585
586 err = ext2_init_acl(inode, dir);
587 if (err)
588 goto fail_free_drop;
589
590 err = ext2_init_security(inode, dir, qstr);
591 if (err)
592 goto fail_free_drop;
593
594 mark_inode_dirty(inode);
595 ext2_debug("allocating inode %lu\n", inode->i_ino);
596 ext2_preread_inode(inode);
597 return inode;
598
599fail_free_drop:
600 dquot_free_inode(inode);
601
602fail_drop:
603 dquot_drop(inode);
604 inode->i_flags |= S_NOQUOTA;
605 clear_nlink(inode);
606 unlock_new_inode(inode);
607 iput(inode);
608 return ERR_PTR(err);
609
610fail:
611 make_bad_inode(inode);
612 iput(inode);
613 return ERR_PTR(err);
614}
615
616unsigned long ext2_count_free_inodes (struct super_block * sb)
617{
618 struct ext2_group_desc *desc;
619 unsigned long desc_count = 0;
620 int i;
621
622#ifdef EXT2FS_DEBUG
623 struct ext2_super_block *es;
624 unsigned long bitmap_count = 0;
625 struct buffer_head *bitmap_bh = NULL;
626
627 es = EXT2_SB(sb)->s_es;
628 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
629 unsigned x;
630
631 desc = ext2_get_group_desc (sb, i, NULL);
632 if (!desc)
633 continue;
634 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
635 brelse(bitmap_bh);
636 bitmap_bh = read_inode_bitmap(sb, i);
637 if (!bitmap_bh)
638 continue;
639
640 x = ext2_count_free(bitmap_bh, EXT2_INODES_PER_GROUP(sb) / 8);
641 printk("group %d: stored = %d, counted = %u\n",
642 i, le16_to_cpu(desc->bg_free_inodes_count), x);
643 bitmap_count += x;
644 }
645 brelse(bitmap_bh);
646 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
647 percpu_counter_read(&EXT2_SB(sb)->s_freeinodes_counter),
648 desc_count, bitmap_count);
649 return desc_count;
650#else
651 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
652 desc = ext2_get_group_desc (sb, i, NULL);
653 if (!desc)
654 continue;
655 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
656 }
657 return desc_count;
658#endif
659}
660
661/* Called at mount-time, super-block is locked */
662unsigned long ext2_count_dirs (struct super_block * sb)
663{
664 unsigned long count = 0;
665 int i;
666
667 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
668 struct ext2_group_desc *gdp = ext2_get_group_desc (sb, i, NULL);
669 if (!gdp)
670 continue;
671 count += le16_to_cpu(gdp->bg_used_dirs_count);
672 }
673 return count;
674}
675