2 * linux/fs/ext3/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@redhat.com), 1993
11 * Big-endian to little-endian byte-swapping/bitmaps by
12 * David S. Miller (davem@caip.rutgers.edu), 1995
15 #include <linux/time.h>
17 #include <linux/jbd.h>
18 #include <linux/ext3_fs.h>
19 #include <linux/ext3_jbd.h>
20 #include <linux/stat.h>
21 #include <linux/string.h>
22 #include <linux/quotaops.h>
23 #include <linux/buffer_head.h>
24 #include <linux/random.h>
25 #include <linux/vs_dlimit.h>
26 #include <linux/bitops.h>
27 #include <linux/vs_dlimit.h>
29 #include <asm/byteorder.h>
35 * ialloc.c contains the inodes allocation and deallocation routines
39 * The free inodes are managed by bitmaps. A file system contains several
40 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
41 * block for inodes, N blocks for the inode table and data blocks.
43 * The file system contains group descriptors which are located after the
44 * super block. Each descriptor contains the number of the bitmap block and
45 * the free blocks count in the block.
50 * Read the inode allocation bitmap for a given block_group, reading
51 * into the specified slot in the superblock's bitmap cache.
53 * Return buffer_head of bitmap on success or NULL.
55 static struct buffer_head *
56 read_inode_bitmap(struct super_block * sb, unsigned long block_group)
58 struct ext3_group_desc *desc;
59 struct buffer_head *bh = NULL;
61 desc = ext3_get_group_desc(sb, block_group, NULL);
65 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
67 ext3_error(sb, "read_inode_bitmap",
68 "Cannot read inode bitmap - "
69 "block_group = %lu, inode_bitmap = %u",
70 block_group, le32_to_cpu(desc->bg_inode_bitmap));
76 * NOTE! When we get the inode, we're the only people
77 * that have access to it, and as such there are no
78 * race conditions we have to worry about. The inode
79 * is not on the hash-lists, and it cannot be reached
80 * through the filesystem because the directory entry
81 * has been deleted earlier.
83 * HOWEVER: we must make sure that we get no aliases,
84 * which means that we have to call "clear_inode()"
85 * _before_ we mark the inode not in use in the inode
86 * bitmaps. Otherwise a newly created file might use
87 * the same inode number (not actually the same pointer
88 * though), and then we'd have two inodes sharing the
89 * same inode number and space on the harddisk.
91 void ext3_free_inode (handle_t *handle, struct inode * inode)
93 struct super_block * sb = inode->i_sb;
96 struct buffer_head *bitmap_bh = NULL;
97 struct buffer_head *bh2;
98 unsigned long block_group;
100 struct ext3_group_desc * gdp;
101 struct ext3_super_block * es;
102 struct ext3_sb_info *sbi;
105 if (atomic_read(&inode->i_count) > 1) {
106 printk ("ext3_free_inode: inode has count=%d\n",
107 atomic_read(&inode->i_count));
110 if (inode->i_nlink) {
111 printk ("ext3_free_inode: inode has nlink=%d\n",
116 printk("ext3_free_inode: inode on nonexistent device\n");
122 ext3_debug ("freeing inode %lu\n", ino);
125 * Note: we must free any quota before locking the superblock,
126 * as writing the quota to disk may need the lock as well.
129 ext3_xattr_delete_inode(handle, inode);
130 DLIMIT_FREE_INODE(sb, inode->i_xid);
131 DQUOT_FREE_INODE(inode);
134 is_directory = S_ISDIR(inode->i_mode);
136 /* Do this BEFORE marking the inode not in use or returning an error */
139 es = EXT3_SB(sb)->s_es;
140 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
141 ext3_error (sb, "ext3_free_inode",
142 "reserved or nonexistent inode %lu", ino);
145 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
146 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
147 bitmap_bh = read_inode_bitmap(sb, block_group);
151 BUFFER_TRACE(bitmap_bh, "get_write_access");
152 fatal = ext3_journal_get_write_access(handle, bitmap_bh);
156 /* Ok, now we can actually update the inode bitmaps.. */
157 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
158 bit, bitmap_bh->b_data))
159 ext3_error (sb, "ext3_free_inode",
160 "bit already cleared for inode %lu", ino);
162 gdp = ext3_get_group_desc (sb, block_group, &bh2);
164 BUFFER_TRACE(bh2, "get_write_access");
165 fatal = ext3_journal_get_write_access(handle, bh2);
166 if (fatal) goto error_return;
169 spin_lock(sb_bgl_lock(sbi, block_group));
170 gdp->bg_free_inodes_count = cpu_to_le16(
171 le16_to_cpu(gdp->bg_free_inodes_count) + 1);
173 gdp->bg_used_dirs_count = cpu_to_le16(
174 le16_to_cpu(gdp->bg_used_dirs_count) - 1);
175 spin_unlock(sb_bgl_lock(sbi, block_group));
176 percpu_counter_inc(&sbi->s_freeinodes_counter);
178 percpu_counter_dec(&sbi->s_dirs_counter);
181 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
182 err = ext3_journal_dirty_metadata(handle, bh2);
183 if (!fatal) fatal = err;
185 BUFFER_TRACE(bitmap_bh, "call ext3_journal_dirty_metadata");
186 err = ext3_journal_dirty_metadata(handle, bitmap_bh);
192 ext3_std_error(sb, fatal);
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.
202 * For other inodes, search forward from the parent directory\'s block
203 * group to find a free inode.
205 static int find_group_dir(struct super_block *sb, struct inode *parent)
207 int ngroups = EXT3_SB(sb)->s_groups_count;
209 struct ext3_group_desc *desc, *best_desc = NULL;
210 struct buffer_head *bh;
211 int group, best_group = -1;
213 freei = percpu_counter_read_positive(&EXT3_SB(sb)->s_freeinodes_counter);
214 avefreei = freei / ngroups;
216 for (group = 0; group < ngroups; group++) {
217 desc = ext3_get_group_desc (sb, group, &bh);
218 if (!desc || !desc->bg_free_inodes_count)
220 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
223 (le16_to_cpu(desc->bg_free_blocks_count) >
224 le16_to_cpu(best_desc->bg_free_blocks_count))) {
233 * Orlov's allocator for directories.
235 * We always try to spread first-level directories.
237 * If there are blockgroups with both free inodes and free blocks counts
238 * not worse than average we return one with smallest directory count.
239 * Otherwise we simply return a random group.
241 * For the rest rules look so:
243 * It's OK to put directory into a group unless
244 * it has too many directories already (max_dirs) or
245 * it has too few free inodes left (min_inodes) or
246 * it has too few free blocks left (min_blocks) or
247 * it's already running too large debt (max_debt).
248 * Parent's group is prefered, if it doesn't satisfy these
249 * conditions we search cyclically through the rest. If none
250 * of the groups look good we just look for a group with more
251 * free inodes than average (starting at parent's group).
253 * Debt is incremented each time we allocate a directory and decremented
254 * when we allocate an inode, within 0--255.
257 #define INODE_COST 64
258 #define BLOCK_COST 256
260 static int find_group_orlov(struct super_block *sb, struct inode *parent)
262 int parent_group = EXT3_I(parent)->i_block_group;
263 struct ext3_sb_info *sbi = EXT3_SB(sb);
264 struct ext3_super_block *es = sbi->s_es;
265 int ngroups = sbi->s_groups_count;
266 int inodes_per_group = EXT3_INODES_PER_GROUP(sb);
269 int blocks_per_dir, ndirs;
270 int max_debt, max_dirs, min_blocks, min_inodes;
272 struct ext3_group_desc *desc;
273 struct buffer_head *bh;
275 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
276 avefreei = freei / ngroups;
277 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
278 avefreeb = freeb / ngroups;
279 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
281 if ((parent == sb->s_root->d_inode) ||
282 (EXT3_I(parent)->i_flags & EXT3_TOPDIR_FL)) {
283 int best_ndir = inodes_per_group;
286 get_random_bytes(&group, sizeof(group));
287 parent_group = (unsigned)group % ngroups;
288 for (i = 0; i < ngroups; i++) {
289 group = (parent_group + i) % ngroups;
290 desc = ext3_get_group_desc (sb, group, &bh);
291 if (!desc || !desc->bg_free_inodes_count)
293 if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
295 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
297 if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
300 best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
307 blocks_per_dir = (le32_to_cpu(es->s_blocks_count) - freeb) / ndirs;
309 max_dirs = ndirs / ngroups + inodes_per_group / 16;
310 min_inodes = avefreei - inodes_per_group / 4;
311 min_blocks = avefreeb - EXT3_BLOCKS_PER_GROUP(sb) / 4;
313 max_debt = EXT3_BLOCKS_PER_GROUP(sb) / max(blocks_per_dir, BLOCK_COST);
314 if (max_debt * INODE_COST > inodes_per_group)
315 max_debt = inodes_per_group / INODE_COST;
321 for (i = 0; i < ngroups; i++) {
322 group = (parent_group + i) % ngroups;
323 desc = ext3_get_group_desc (sb, group, &bh);
324 if (!desc || !desc->bg_free_inodes_count)
326 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
328 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
330 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
336 for (i = 0; i < ngroups; i++) {
337 group = (parent_group + i) % ngroups;
338 desc = ext3_get_group_desc (sb, group, &bh);
339 if (!desc || !desc->bg_free_inodes_count)
341 if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
347 * The free-inodes counter is approximate, and for really small
348 * filesystems the above test can fail to find any blockgroups
357 static int find_group_other(struct super_block *sb, struct inode *parent)
359 int parent_group = EXT3_I(parent)->i_block_group;
360 int ngroups = EXT3_SB(sb)->s_groups_count;
361 struct ext3_group_desc *desc;
362 struct buffer_head *bh;
366 * Try to place the inode in its parent directory
368 group = parent_group;
369 desc = ext3_get_group_desc (sb, group, &bh);
370 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
371 le16_to_cpu(desc->bg_free_blocks_count))
375 * We're going to place this inode in a different blockgroup from its
376 * parent. We want to cause files in a common directory to all land in
377 * the same blockgroup. But we want files which are in a different
378 * directory which shares a blockgroup with our parent to land in a
379 * different blockgroup.
381 * So add our directory's i_ino into the starting point for the hash.
383 group = (group + parent->i_ino) % ngroups;
386 * Use a quadratic hash to find a group with a free inode and some free
389 for (i = 1; i < ngroups; i <<= 1) {
391 if (group >= ngroups)
393 desc = ext3_get_group_desc (sb, group, &bh);
394 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
395 le16_to_cpu(desc->bg_free_blocks_count))
400 * That failed: try linear search for a free inode, even if that group
401 * has no free blocks.
403 group = parent_group;
404 for (i = 0; i < ngroups; i++) {
405 if (++group >= ngroups)
407 desc = ext3_get_group_desc (sb, group, &bh);
408 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
416 * There are two policies for allocating an inode. If the new inode is
417 * a directory, then a forward search is made for a block group with both
418 * free space and a low directory-to-inode ratio; if that fails, then of
419 * the groups with above-average free space, that group with the fewest
420 * directories already is chosen.
422 * For other inodes, search forward from the parent directory's block
423 * group to find a free inode.
425 struct inode *ext3_new_inode(handle_t *handle, struct inode * dir, int mode)
427 struct super_block *sb;
428 struct buffer_head *bitmap_bh = NULL;
429 struct buffer_head *bh2;
431 unsigned long ino = 0;
432 struct inode * inode;
433 struct ext3_group_desc * gdp = NULL;
434 struct ext3_super_block * es;
435 struct ext3_inode_info *ei;
436 struct ext3_sb_info *sbi;
441 /* Cannot create files in a deleted directory */
442 if (!dir || !dir->i_nlink)
443 return ERR_PTR(-EPERM);
446 inode = new_inode(sb);
448 return ERR_PTR(-ENOMEM);
450 if (sb->s_flags & MS_TAGXID)
451 inode->i_xid = vx_current_xid();
455 if (DLIMIT_ALLOC_INODE(sb, inode->i_xid)) {
464 if (test_opt (sb, OLDALLOC))
465 group = find_group_dir(sb, dir);
467 group = find_group_orlov(sb, dir);
469 group = find_group_other(sb, dir);
475 for (i = 0; i < sbi->s_groups_count; i++) {
476 gdp = ext3_get_group_desc(sb, group, &bh2);
480 bitmap_bh = read_inode_bitmap(sb, group);
486 repeat_in_this_group:
487 ino = ext3_find_next_zero_bit((unsigned long *)
488 bitmap_bh->b_data, EXT3_INODES_PER_GROUP(sb), ino);
489 if (ino < EXT3_INODES_PER_GROUP(sb)) {
492 BUFFER_TRACE(bitmap_bh, "get_write_access");
493 err = ext3_journal_get_write_access_credits(handle,
494 bitmap_bh, &credits);
498 if (!ext3_set_bit_atomic(sb_bgl_lock(sbi, group),
499 ino, bitmap_bh->b_data)) {
501 BUFFER_TRACE(bitmap_bh,
502 "call ext3_journal_dirty_metadata");
503 err = ext3_journal_dirty_metadata(handle,
510 journal_release_buffer(handle, bitmap_bh, credits);
512 if (++ino < EXT3_INODES_PER_GROUP(sb))
513 goto repeat_in_this_group;
517 * This case is possible in concurrent environment. It is very
518 * rare. We cannot repeat the find_group_xxx() call because
519 * that will simply return the same blockgroup, because the
520 * group descriptor metadata has not yet been updated.
521 * So we just go onto the next blockgroup.
523 if (++group == sbi->s_groups_count)
530 ino += group * EXT3_INODES_PER_GROUP(sb) + 1;
531 if (ino < EXT3_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
532 ext3_error (sb, "ext3_new_inode",
533 "reserved inode or inode > inodes count - "
534 "block_group = %d, inode=%lu", group, ino);
539 BUFFER_TRACE(bh2, "get_write_access");
540 err = ext3_journal_get_write_access(handle, bh2);
542 spin_lock(sb_bgl_lock(sbi, group));
543 gdp->bg_free_inodes_count =
544 cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
546 gdp->bg_used_dirs_count =
547 cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
549 spin_unlock(sb_bgl_lock(sbi, group));
550 BUFFER_TRACE(bh2, "call ext3_journal_dirty_metadata");
551 err = ext3_journal_dirty_metadata(handle, bh2);
554 percpu_counter_dec(&sbi->s_freeinodes_counter);
556 percpu_counter_inc(&sbi->s_dirs_counter);
559 inode->i_uid = current->fsuid;
560 if (test_opt (sb, GRPID))
561 inode->i_gid = dir->i_gid;
562 else if (dir->i_mode & S_ISGID) {
563 inode->i_gid = dir->i_gid;
567 inode->i_gid = current->fsgid;
568 inode->i_mode = mode;
571 /* This is the optimal IO size (for stat), not the fs block size */
572 inode->i_blksize = PAGE_SIZE;
574 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
576 memset(ei->i_data, 0, sizeof(ei->i_data));
577 ei->i_next_alloc_block = 0;
578 ei->i_next_alloc_goal = 0;
579 ei->i_dir_start_lookup = 0;
582 ei->i_flags = EXT3_I(dir)->i_flags &
583 ~(EXT3_INDEX_FL|EXT3_IUNLINK_FL|EXT3_BARRIER_FL);
585 ei->i_flags &= ~(EXT3_IMMUTABLE_FL|EXT3_APPEND_FL);
586 /* dirsync only applies to directories */
588 ei->i_flags &= ~EXT3_DIRSYNC_FL;
589 #ifdef EXT3_FRAGMENTS
597 ei->i_rsv_window.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
598 ei->i_rsv_window.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
599 atomic_set(&ei->i_rsv_window.rsv_goal_size, EXT3_DEFAULT_RESERVE_BLOCKS);
600 atomic_set(&ei->i_rsv_window.rsv_alloc_hit, 0);
601 seqlock_init(&ei->i_rsv_window.rsv_seqlock);
602 ei->i_block_group = group;
604 ext3_set_inode_flags(inode);
605 if (IS_DIRSYNC(inode))
607 insert_inode_hash(inode);
608 spin_lock(&sbi->s_next_gen_lock);
609 inode->i_generation = sbi->s_next_generation++;
610 spin_unlock(&sbi->s_next_gen_lock);
612 ei->i_state = EXT3_STATE_NEW;
615 if(DQUOT_ALLOC_INODE(inode)) {
620 err = ext3_init_acl(handle, inode, dir);
622 DQUOT_FREE_INODE(inode);
625 err = ext3_mark_inode_dirty(handle, inode);
627 ext3_std_error(sb, err);
628 DQUOT_FREE_INODE(inode);
632 ext3_debug("allocating inode %lu\n", inode->i_ino);
635 DLIMIT_FREE_INODE(sb, inode->i_xid);
636 ext3_std_error(sb, err);
645 DLIMIT_FREE_INODE(sb, inode->i_xid);
646 inode->i_flags |= S_NOQUOTA;
653 /* Verify that we are loading a valid orphan from disk */
654 struct inode *ext3_orphan_get(struct super_block *sb, unsigned long ino)
656 unsigned long max_ino = le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count);
657 unsigned long block_group;
659 struct buffer_head *bitmap_bh = NULL;
660 struct inode *inode = NULL;
662 /* Error cases - e2fsck has already cleaned up for us */
664 ext3_warning(sb, __FUNCTION__,
665 "bad orphan ino %lu! e2fsck was run?\n", ino);
669 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
670 bit = (ino - 1) % EXT3_INODES_PER_GROUP(sb);
671 bitmap_bh = read_inode_bitmap(sb, block_group);
673 ext3_warning(sb, __FUNCTION__,
674 "inode bitmap error for orphan %lu\n", ino);
678 /* Having the inode bit set should be a 100% indicator that this
679 * is a valid orphan (no e2fsck run on fs). Orphans also include
680 * inodes that were being truncated, so we can't check i_nlink==0.
682 if (!ext3_test_bit(bit, bitmap_bh->b_data) ||
683 !(inode = iget(sb, ino)) || is_bad_inode(inode) ||
684 NEXT_ORPHAN(inode) > max_ino) {
685 ext3_warning(sb, __FUNCTION__,
686 "bad orphan inode %lu! e2fsck was run?\n", ino);
687 printk(KERN_NOTICE "ext3_test_bit(bit=%d, block=%llu) = %d\n",
688 bit, (unsigned long long)bitmap_bh->b_blocknr,
689 ext3_test_bit(bit, bitmap_bh->b_data));
690 printk(KERN_NOTICE "inode=%p\n", inode);
692 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
693 is_bad_inode(inode));
694 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
696 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
698 /* Avoid freeing blocks if we got a bad deleted inode */
699 if (inode && inode->i_nlink == 0)
709 unsigned long ext3_count_free_inodes (struct super_block * sb)
711 unsigned long desc_count;
712 struct ext3_group_desc *gdp;
715 struct ext3_super_block *es;
716 unsigned long bitmap_count, x;
717 struct buffer_head *bitmap_bh = NULL;
720 es = EXT3_SB(sb)->s_es;
724 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
725 gdp = ext3_get_group_desc (sb, i, NULL);
728 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
730 bitmap_bh = read_inode_bitmap(sb, i);
734 x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8);
735 printk("group %d: stored = %d, counted = %lu\n",
736 i, le16_to_cpu(gdp->bg_free_inodes_count), x);
740 printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n",
741 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
746 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
747 gdp = ext3_get_group_desc (sb, i, NULL);
750 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
756 /* Called at mount-time, super-block is locked */
757 unsigned long ext3_count_dirs (struct super_block * sb)
759 unsigned long count = 0;
762 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
763 struct ext3_group_desc *gdp = ext3_get_group_desc (sb, i, NULL);
766 count += le16_to_cpu(gdp->bg_used_dirs_count);
771 #ifdef CONFIG_EXT3_CHECK
772 /* Called at mount-time, super-block is locked */
773 void ext3_check_inodes_bitmap (struct super_block * sb)
775 struct ext3_super_block * es;
776 unsigned long desc_count, bitmap_count, x;
777 struct buffer_head *bitmap_bh = NULL;
778 struct ext3_group_desc * gdp;
781 es = EXT3_SB(sb)->s_es;
785 for (i = 0; i < EXT3_SB(sb)->s_groups_count; i++) {
786 gdp = ext3_get_group_desc (sb, i, NULL);
789 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
791 bitmap_bh = read_inode_bitmap(sb, i);
795 x = ext3_count_free(bitmap_bh, EXT3_INODES_PER_GROUP(sb) / 8);
796 if (le16_to_cpu(gdp->bg_free_inodes_count) != x)
797 ext3_error (sb, "ext3_check_inodes_bitmap",
798 "Wrong free inodes count in group %d, "
799 "stored = %d, counted = %lu", i,
800 le16_to_cpu(gdp->bg_free_inodes_count), x);
804 if (le32_to_cpu(es->s_free_inodes_count) != bitmap_count)
805 ext3_error (sb, "ext3_check_inodes_bitmap",
806 "Wrong free inodes count in super block, "
807 "stored = %lu, counted = %lu",
808 (unsigned long)le32_to_cpu(es->s_free_inodes_count),