sched: latencytop support
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ext2 / ialloc.c
blob5deb8b74e649ed56c9455780b9ad258d2b974c75
1 /*
2 * linux/fs/ext2/ialloc.c
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)
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
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"
25 * ialloc.c contains the inodes allocation and deallocation routines
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.
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.
40 * Read the inode allocation bitmap for a given block_group, reading
41 * into the specified slot in the superblock's bitmap cache.
43 * Return buffer_head of bitmap on success or NULL.
45 static struct buffer_head *
46 read_inode_bitmap(struct super_block * sb, unsigned long block_group)
48 struct ext2_group_desc *desc;
49 struct buffer_head *bh = NULL;
51 desc = ext2_get_group_desc(sb, block_group, NULL);
52 if (!desc)
53 goto error_out;
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));
61 error_out:
62 return bh;
65 static void ext2_release_inode(struct super_block *sb, int group, int dir)
67 struct ext2_group_desc * desc;
68 struct buffer_head *bh;
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;
77 spin_lock(sb_bgl_lock(EXT2_SB(sb), group));
78 desc->bg_free_inodes_count =
79 cpu_to_le16(le16_to_cpu(desc->bg_free_inodes_count) + 1);
80 if (dir)
81 desc->bg_used_dirs_count =
82 cpu_to_le16(le16_to_cpu(desc->bg_used_dirs_count) - 1);
83 spin_unlock(sb_bgl_lock(EXT2_SB(sb), group));
84 if (dir)
85 percpu_counter_dec(&EXT2_SB(sb)->s_dirs_counter);
86 sb->s_dirt = 1;
87 mark_buffer_dirty(bh);
91 * NOTE! When we get the inode, we're the only people
92 * that have access to it, and as such there are no
93 * race conditions we have to worry about. The inode
94 * is not on the hash-lists, and it cannot be reached
95 * through the filesystem because the directory entry
96 * has been deleted earlier.
98 * HOWEVER: we must make sure that we get no aliases,
99 * which means that we have to call "clear_inode()"
100 * _before_ we mark the inode not in use in the inode
101 * bitmaps. Otherwise a newly created file might use
102 * the same inode number (not actually the same pointer
103 * though), and then we'd have two inodes sharing the
104 * same inode number and space on the harddisk.
106 void ext2_free_inode (struct inode * inode)
108 struct super_block * sb = inode->i_sb;
109 int is_directory;
110 unsigned long ino;
111 struct buffer_head *bitmap_bh = NULL;
112 unsigned long block_group;
113 unsigned long bit;
114 struct ext2_super_block * es;
116 ino = inode->i_ino;
117 ext2_debug ("freeing inode %lu\n", ino);
120 * Note: we must free any quota before locking the superblock,
121 * as writing the quota to disk may need the lock as well.
123 if (!is_bad_inode(inode)) {
124 /* Quota is already initialized in iput() */
125 ext2_xattr_delete_inode(inode);
126 DQUOT_FREE_INODE(inode);
127 DQUOT_DROP(inode);
130 es = EXT2_SB(sb)->s_es;
131 is_directory = S_ISDIR(inode->i_mode);
133 /* Do this BEFORE marking the inode not in use or returning an error */
134 clear_inode (inode);
136 if (ino < EXT2_FIRST_INO(sb) ||
137 ino > le32_to_cpu(es->s_inodes_count)) {
138 ext2_error (sb, "ext2_free_inode",
139 "reserved or nonexistent inode %lu", ino);
140 goto error_return;
142 block_group = (ino - 1) / EXT2_INODES_PER_GROUP(sb);
143 bit = (ino - 1) % EXT2_INODES_PER_GROUP(sb);
144 brelse(bitmap_bh);
145 bitmap_bh = read_inode_bitmap(sb, block_group);
146 if (!bitmap_bh)
147 goto error_return;
149 /* Ok, now we can actually update the inode bitmaps.. */
150 if (!ext2_clear_bit_atomic(sb_bgl_lock(EXT2_SB(sb), block_group),
151 bit, (void *) bitmap_bh->b_data))
152 ext2_error (sb, "ext2_free_inode",
153 "bit already cleared for inode %lu", ino);
154 else
155 ext2_release_inode(sb, block_group, is_directory);
156 mark_buffer_dirty(bitmap_bh);
157 if (sb->s_flags & MS_SYNCHRONOUS)
158 sync_dirty_buffer(bitmap_bh);
159 error_return:
160 brelse(bitmap_bh);
164 * We perform asynchronous prereading of the new inode's inode block when
165 * we create the inode, in the expectation that the inode will be written
166 * back soon. There are two reasons:
168 * - When creating a large number of files, the async prereads will be
169 * nicely merged into large reads
170 * - When writing out a large number of inodes, we don't need to keep on
171 * stalling the writes while we read the inode block.
173 * FIXME: ext2_get_group_desc() needs to be simplified.
175 static void ext2_preread_inode(struct inode *inode)
177 unsigned long block_group;
178 unsigned long offset;
179 unsigned long block;
180 struct ext2_group_desc * gdp;
181 struct backing_dev_info *bdi;
183 bdi = inode->i_mapping->backing_dev_info;
184 if (bdi_read_congested(bdi))
185 return;
186 if (bdi_write_congested(bdi))
187 return;
189 block_group = (inode->i_ino - 1) / EXT2_INODES_PER_GROUP(inode->i_sb);
190 gdp = ext2_get_group_desc(inode->i_sb, block_group, NULL);
191 if (gdp == NULL)
192 return;
195 * Figure out the offset within the block group inode table
197 offset = ((inode->i_ino - 1) % EXT2_INODES_PER_GROUP(inode->i_sb)) *
198 EXT2_INODE_SIZE(inode->i_sb);
199 block = le32_to_cpu(gdp->bg_inode_table) +
200 (offset >> EXT2_BLOCK_SIZE_BITS(inode->i_sb));
201 sb_breadahead(inode->i_sb, block);
205 * There are two policies for allocating an inode. If the new inode is
206 * a directory, then a forward search is made for a block group with both
207 * free space and a low directory-to-inode ratio; if that fails, then of
208 * the groups with above-average free space, that group with the fewest
209 * directories already is chosen.
211 * For other inodes, search forward from the parent directory\'s block
212 * group to find a free inode.
214 static int find_group_dir(struct super_block *sb, struct inode *parent)
216 int ngroups = EXT2_SB(sb)->s_groups_count;
217 int avefreei = ext2_count_free_inodes(sb) / ngroups;
218 struct ext2_group_desc *desc, *best_desc = NULL;
219 int group, best_group = -1;
221 for (group = 0; group < ngroups; group++) {
222 desc = ext2_get_group_desc (sb, group, NULL);
223 if (!desc || !desc->bg_free_inodes_count)
224 continue;
225 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
226 continue;
227 if (!best_desc ||
228 (le16_to_cpu(desc->bg_free_blocks_count) >
229 le16_to_cpu(best_desc->bg_free_blocks_count))) {
230 best_group = group;
231 best_desc = desc;
234 if (!best_desc)
235 return -1;
237 return best_group;
241 * Orlov's allocator for directories.
243 * We always try to spread first-level directories.
245 * If there are blockgroups with both free inodes and free blocks counts
246 * not worse than average we return one with smallest directory count.
247 * Otherwise we simply return a random group.
249 * For the rest rules look so:
251 * It's OK to put directory into a group unless
252 * it has too many directories already (max_dirs) or
253 * it has too few free inodes left (min_inodes) or
254 * it has too few free blocks left (min_blocks) or
255 * it's already running too large debt (max_debt).
256 * Parent's group is prefered, if it doesn't satisfy these
257 * conditions we search cyclically through the rest. If none
258 * of the groups look good we just look for a group with more
259 * free inodes than average (starting at parent's group).
261 * Debt is incremented each time we allocate a directory and decremented
262 * when we allocate an inode, within 0--255.
265 #define INODE_COST 64
266 #define BLOCK_COST 256
268 static int find_group_orlov(struct super_block *sb, struct inode *parent)
270 int parent_group = EXT2_I(parent)->i_block_group;
271 struct ext2_sb_info *sbi = EXT2_SB(sb);
272 struct ext2_super_block *es = sbi->s_es;
273 int ngroups = sbi->s_groups_count;
274 int inodes_per_group = EXT2_INODES_PER_GROUP(sb);
275 int freei;
276 int avefreei;
277 int free_blocks;
278 int avefreeb;
279 int blocks_per_dir;
280 int ndirs;
281 int max_debt, max_dirs, min_blocks, min_inodes;
282 int group = -1, i;
283 struct ext2_group_desc *desc;
285 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
286 avefreei = freei / ngroups;
287 free_blocks = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
288 avefreeb = free_blocks / ngroups;
289 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
291 if ((parent == sb->s_root->d_inode) ||
292 (EXT2_I(parent)->i_flags & EXT2_TOPDIR_FL)) {
293 struct ext2_group_desc *best_desc = NULL;
294 int best_ndir = inodes_per_group;
295 int best_group = -1;
297 get_random_bytes(&group, sizeof(group));
298 parent_group = (unsigned)group % ngroups;
299 for (i = 0; i < ngroups; i++) {
300 group = (parent_group + i) % ngroups;
301 desc = ext2_get_group_desc (sb, group, NULL);
302 if (!desc || !desc->bg_free_inodes_count)
303 continue;
304 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
305 continue;
306 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
307 continue;
308 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
309 continue;
310 best_group = group;
311 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
312 best_desc = desc;
314 if (best_group >= 0) {
315 desc = best_desc;
316 group = best_group;
317 goto found;
319 goto fallback;
322 if (ndirs == 0)
323 ndirs = 1; /* percpu_counters are approximate... */
325 blocks_per_dir = (le32_to_cpu(es->s_blocks_count)-free_blocks) / ndirs;
327 max_dirs = ndirs / ngroups + inodes_per_group / 16;
328 min_inodes = avefreei - inodes_per_group / 4;
329 min_blocks = avefreeb - EXT2_BLOCKS_PER_GROUP(sb) / 4;
331 max_debt = EXT2_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST);
332 if (max_debt * INODE_COST > inodes_per_group)
333 max_debt = inodes_per_group / INODE_COST;
334 if (max_debt > 255)
335 max_debt = 255;
336 if (max_debt == 0)
337 max_debt = 1;
339 for (i = 0; i < ngroups; i++) {
340 group = (parent_group + i) % ngroups;
341 desc = ext2_get_group_desc (sb, group, NULL);
342 if (!desc || !desc->bg_free_inodes_count)
343 continue;
344 if (sbi->s_debts[group] >= max_debt)
345 continue;
346 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
347 continue;
348 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
349 continue;
350 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
351 continue;
352 goto found;
355 fallback:
356 for (i = 0; i < ngroups; i++) {
357 group = (parent_group + i) % ngroups;
358 desc = ext2_get_group_desc (sb, group, NULL);
359 if (!desc || !desc->bg_free_inodes_count)
360 continue;
361 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
362 goto found;
365 if (avefreei) {
367 * The free-inodes counter is approximate, and for really small
368 * filesystems the above test can fail to find any blockgroups
370 avefreei = 0;
371 goto fallback;
374 return -1;
376 found:
377 return group;
380 static int find_group_other(struct super_block *sb, struct inode *parent)
382 int parent_group = EXT2_I(parent)->i_block_group;
383 int ngroups = EXT2_SB(sb)->s_groups_count;
384 struct ext2_group_desc *desc;
385 int group, i;
388 * Try to place the inode in its parent directory
390 group = parent_group;
391 desc = ext2_get_group_desc (sb, group, NULL);
392 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
393 le16_to_cpu(desc->bg_free_blocks_count))
394 goto found;
397 * We're going to place this inode in a different blockgroup from its
398 * parent. We want to cause files in a common directory to all land in
399 * the same blockgroup. But we want files which are in a different
400 * directory which shares a blockgroup with our parent to land in a
401 * different blockgroup.
403 * So add our directory's i_ino into the starting point for the hash.
405 group = (group + parent->i_ino) % ngroups;
408 * Use a quadratic hash to find a group with a free inode and some
409 * free blocks.
411 for (i = 1; i < ngroups; i <<= 1) {
412 group += i;
413 if (group >= ngroups)
414 group -= ngroups;
415 desc = ext2_get_group_desc (sb, group, NULL);
416 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
417 le16_to_cpu(desc->bg_free_blocks_count))
418 goto found;
422 * That failed: try linear search for a free inode, even if that group
423 * has no free blocks.
425 group = parent_group;
426 for (i = 0; i < ngroups; i++) {
427 if (++group >= ngroups)
428 group = 0;
429 desc = ext2_get_group_desc (sb, group, NULL);
430 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
431 goto found;
434 return -1;
436 found:
437 return group;
440 struct inode *ext2_new_inode(struct inode *dir, int mode)
442 struct super_block *sb;
443 struct buffer_head *bitmap_bh = NULL;
444 struct buffer_head *bh2;
445 int group, i;
446 ino_t ino = 0;
447 struct inode * inode;
448 struct ext2_group_desc *gdp;
449 struct ext2_super_block *es;
450 struct ext2_inode_info *ei;
451 struct ext2_sb_info *sbi;
452 int err;
454 sb = dir->i_sb;
455 inode = new_inode(sb);
456 if (!inode)
457 return ERR_PTR(-ENOMEM);
459 ei = EXT2_I(inode);
460 sbi = EXT2_SB(sb);
461 es = sbi->s_es;
462 if (S_ISDIR(mode)) {
463 if (test_opt(sb, OLDALLOC))
464 group = find_group_dir(sb, dir);
465 else
466 group = find_group_orlov(sb, dir);
467 } else
468 group = find_group_other(sb, dir);
470 if (group == -1) {
471 err = -ENOSPC;
472 goto fail;
475 for (i = 0; i < sbi->s_groups_count; i++) {
476 gdp = ext2_get_group_desc(sb, group, &bh2);
477 brelse(bitmap_bh);
478 bitmap_bh = read_inode_bitmap(sb, group);
479 if (!bitmap_bh) {
480 err = -EIO;
481 goto fail;
483 ino = 0;
485 repeat_in_this_group:
486 ino = ext2_find_next_zero_bit((unsigned long *)bitmap_bh->b_data,
487 EXT2_INODES_PER_GROUP(sb), ino);
488 if (ino >= EXT2_INODES_PER_GROUP(sb)) {
490 * Rare race: find_group_xx() decided that there were
491 * free inodes in this group, but by the time we tried
492 * to allocate one, they're all gone. This can also
493 * occur because the counters which find_group_orlov()
494 * uses are approximate. So just go and search the
495 * next block group.
497 if (++group == sbi->s_groups_count)
498 group = 0;
499 continue;
501 if (ext2_set_bit_atomic(sb_bgl_lock(sbi, group),
502 ino, bitmap_bh->b_data)) {
503 /* we lost this inode */
504 if (++ino >= EXT2_INODES_PER_GROUP(sb)) {
505 /* this group is exhausted, try next group */
506 if (++group == sbi->s_groups_count)
507 group = 0;
508 continue;
510 /* try to find free inode in the same group */
511 goto repeat_in_this_group;
513 goto got;
517 * Scanned all blockgroups.
519 err = -ENOSPC;
520 goto fail;
521 got:
522 mark_buffer_dirty(bitmap_bh);
523 if (sb->s_flags & MS_SYNCHRONOUS)
524 sync_dirty_buffer(bitmap_bh);
525 brelse(bitmap_bh);
527 ino += group * EXT2_INODES_PER_GROUP(sb) + 1;
528 if (ino < EXT2_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
529 ext2_error (sb, "ext2_new_inode",
530 "reserved inode or inode > inodes count - "
531 "block_group = %d,inode=%lu", group,
532 (unsigned long) ino);
533 err = -EIO;
534 goto fail;
537 percpu_counter_add(&sbi->s_freeinodes_counter, -1);
538 if (S_ISDIR(mode))
539 percpu_counter_inc(&sbi->s_dirs_counter);
541 spin_lock(sb_bgl_lock(sbi, group));
542 gdp->bg_free_inodes_count =
543 cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
544 if (S_ISDIR(mode)) {
545 if (sbi->s_debts[group] < 255)
546 sbi->s_debts[group]++;
547 gdp->bg_used_dirs_count =
548 cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
549 } else {
550 if (sbi->s_debts[group])
551 sbi->s_debts[group]--;
553 spin_unlock(sb_bgl_lock(sbi, group));
555 sb->s_dirt = 1;
556 mark_buffer_dirty(bh2);
557 inode->i_uid = current->fsuid;
558 if (test_opt (sb, GRPID))
559 inode->i_gid = dir->i_gid;
560 else if (dir->i_mode & S_ISGID) {
561 inode->i_gid = dir->i_gid;
562 if (S_ISDIR(mode))
563 mode |= S_ISGID;
564 } else
565 inode->i_gid = current->fsgid;
566 inode->i_mode = mode;
568 inode->i_ino = ino;
569 inode->i_blocks = 0;
570 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
571 memset(ei->i_data, 0, sizeof(ei->i_data));
572 ei->i_flags = EXT2_I(dir)->i_flags & ~EXT2_BTREE_FL;
573 if (S_ISLNK(mode))
574 ei->i_flags &= ~(EXT2_IMMUTABLE_FL|EXT2_APPEND_FL);
575 /* dirsync is only applied to directories */
576 if (!S_ISDIR(mode))
577 ei->i_flags &= ~EXT2_DIRSYNC_FL;
578 ei->i_faddr = 0;
579 ei->i_frag_no = 0;
580 ei->i_frag_size = 0;
581 ei->i_file_acl = 0;
582 ei->i_dir_acl = 0;
583 ei->i_dtime = 0;
584 ei->i_block_alloc_info = NULL;
585 ei->i_block_group = group;
586 ei->i_dir_start_lookup = 0;
587 ei->i_state = EXT2_STATE_NEW;
588 ext2_set_inode_flags(inode);
589 spin_lock(&sbi->s_next_gen_lock);
590 inode->i_generation = sbi->s_next_generation++;
591 spin_unlock(&sbi->s_next_gen_lock);
592 insert_inode_hash(inode);
594 if (DQUOT_ALLOC_INODE(inode)) {
595 err = -EDQUOT;
596 goto fail_drop;
599 err = ext2_init_acl(inode, dir);
600 if (err)
601 goto fail_free_drop;
603 err = ext2_init_security(inode,dir);
604 if (err)
605 goto fail_free_drop;
607 mark_inode_dirty(inode);
608 ext2_debug("allocating inode %lu\n", inode->i_ino);
609 ext2_preread_inode(inode);
610 return inode;
612 fail_free_drop:
613 DQUOT_FREE_INODE(inode);
615 fail_drop:
616 DQUOT_DROP(inode);
617 inode->i_flags |= S_NOQUOTA;
618 inode->i_nlink = 0;
619 iput(inode);
620 return ERR_PTR(err);
622 fail:
623 make_bad_inode(inode);
624 iput(inode);
625 return ERR_PTR(err);
628 unsigned long ext2_count_free_inodes (struct super_block * sb)
630 struct ext2_group_desc *desc;
631 unsigned long desc_count = 0;
632 int i;
634 #ifdef EXT2FS_DEBUG
635 struct ext2_super_block *es;
636 unsigned long bitmap_count = 0;
637 struct buffer_head *bitmap_bh = NULL;
639 es = EXT2_SB(sb)->s_es;
640 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
641 unsigned x;
643 desc = ext2_get_group_desc (sb, i, NULL);
644 if (!desc)
645 continue;
646 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
647 brelse(bitmap_bh);
648 bitmap_bh = read_inode_bitmap(sb, i);
649 if (!bitmap_bh)
650 continue;
652 x = ext2_count_free(bitmap_bh, EXT2_INODES_PER_GROUP(sb) / 8);
653 printk("group %d: stored = %d, counted = %u\n",
654 i, le16_to_cpu(desc->bg_free_inodes_count), x);
655 bitmap_count += x;
657 brelse(bitmap_bh);
658 printk("ext2_count_free_inodes: stored = %lu, computed = %lu, %lu\n",
659 percpu_counter_read(&EXT2_SB(sb)->s_freeinodes_counter),
660 desc_count, bitmap_count);
661 return desc_count;
662 #else
663 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
664 desc = ext2_get_group_desc (sb, i, NULL);
665 if (!desc)
666 continue;
667 desc_count += le16_to_cpu(desc->bg_free_inodes_count);
669 return desc_count;
670 #endif
673 /* Called at mount-time, super-block is locked */
674 unsigned long ext2_count_dirs (struct super_block * sb)
676 unsigned long count = 0;
677 int i;
679 for (i = 0; i < EXT2_SB(sb)->s_groups_count; i++) {
680 struct ext2_group_desc *gdp = ext2_get_group_desc (sb, i, NULL);
681 if (!gdp)
682 continue;
683 count += le16_to_cpu(gdp->bg_used_dirs_count);
685 return count;