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/quotaops.h>
16 #include <linux/random.h>
23 * ialloc.c contains the inodes allocation and deallocation routines
27 * The free inodes are managed by bitmaps. A file system contains several
28 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
29 * block for inodes, N blocks for the inode table and data blocks.
31 * The file system contains group descriptors which are located after the
32 * super block. Each descriptor contains the number of the bitmap block and
33 * the free blocks count in the block.
38 * Read the inode allocation bitmap for a given block_group, reading
39 * into the specified slot in the superblock's bitmap cache.
41 * Return buffer_head of bitmap on success or NULL.
43 static struct buffer_head
*
44 read_inode_bitmap(struct super_block
* sb
, unsigned long block_group
)
46 struct ext3_group_desc
*desc
;
47 struct buffer_head
*bh
= NULL
;
49 desc
= ext3_get_group_desc(sb
, block_group
, NULL
);
53 bh
= sb_bread(sb
, le32_to_cpu(desc
->bg_inode_bitmap
));
55 ext3_error(sb
, "read_inode_bitmap",
56 "Cannot read inode bitmap - "
57 "block_group = %lu, inode_bitmap = %u",
58 block_group
, le32_to_cpu(desc
->bg_inode_bitmap
));
64 * NOTE! When we get the inode, we're the only people
65 * that have access to it, and as such there are no
66 * race conditions we have to worry about. The inode
67 * is not on the hash-lists, and it cannot be reached
68 * through the filesystem because the directory entry
69 * has been deleted earlier.
71 * HOWEVER: we must make sure that we get no aliases,
72 * which means that we have to call "clear_inode()"
73 * _before_ we mark the inode not in use in the inode
74 * bitmaps. Otherwise a newly created file might use
75 * the same inode number (not actually the same pointer
76 * though), and then we'd have two inodes sharing the
77 * same inode number and space on the harddisk.
79 void ext3_free_inode (handle_t
*handle
, struct inode
* inode
)
81 struct super_block
* sb
= inode
->i_sb
;
84 struct buffer_head
*bitmap_bh
= NULL
;
85 struct buffer_head
*bh2
;
86 unsigned long block_group
;
88 struct ext3_group_desc
* gdp
;
89 struct ext3_super_block
* es
;
90 struct ext3_sb_info
*sbi
;
93 if (atomic_read(&inode
->i_count
) > 1) {
94 printk ("ext3_free_inode: inode has count=%d\n",
95 atomic_read(&inode
->i_count
));
99 printk ("ext3_free_inode: inode has nlink=%d\n",
104 printk("ext3_free_inode: inode on nonexistent device\n");
110 ext3_debug ("freeing inode %lu\n", ino
);
111 trace_ext3_free_inode(inode
);
113 is_directory
= S_ISDIR(inode
->i_mode
);
115 es
= EXT3_SB(sb
)->s_es
;
116 if (ino
< EXT3_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
117 ext3_error (sb
, "ext3_free_inode",
118 "reserved or nonexistent inode %lu", ino
);
121 block_group
= (ino
- 1) / EXT3_INODES_PER_GROUP(sb
);
122 bit
= (ino
- 1) % EXT3_INODES_PER_GROUP(sb
);
123 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
127 BUFFER_TRACE(bitmap_bh
, "get_write_access");
128 fatal
= ext3_journal_get_write_access(handle
, bitmap_bh
);
132 /* Ok, now we can actually update the inode bitmaps.. */
133 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi
, block_group
),
134 bit
, bitmap_bh
->b_data
))
135 ext3_error (sb
, "ext3_free_inode",
136 "bit already cleared for inode %lu", ino
);
138 gdp
= ext3_get_group_desc (sb
, block_group
, &bh2
);
140 BUFFER_TRACE(bh2
, "get_write_access");
141 fatal
= ext3_journal_get_write_access(handle
, bh2
);
142 if (fatal
) goto error_return
;
145 spin_lock(sb_bgl_lock(sbi
, block_group
));
146 le16_add_cpu(&gdp
->bg_free_inodes_count
, 1);
148 le16_add_cpu(&gdp
->bg_used_dirs_count
, -1);
149 spin_unlock(sb_bgl_lock(sbi
, block_group
));
150 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
152 percpu_counter_dec(&sbi
->s_dirs_counter
);
155 BUFFER_TRACE(bh2
, "call ext3_journal_dirty_metadata");
156 err
= ext3_journal_dirty_metadata(handle
, bh2
);
157 if (!fatal
) fatal
= err
;
159 BUFFER_TRACE(bitmap_bh
, "call ext3_journal_dirty_metadata");
160 err
= ext3_journal_dirty_metadata(handle
, bitmap_bh
);
166 ext3_std_error(sb
, fatal
);
170 * Orlov's allocator for directories.
172 * We always try to spread first-level directories.
174 * If there are blockgroups with both free inodes and free blocks counts
175 * not worse than average we return one with smallest directory count.
176 * Otherwise we simply return a random group.
178 * For the rest rules look so:
180 * It's OK to put directory into a group unless
181 * it has too many directories already (max_dirs) or
182 * it has too few free inodes left (min_inodes) or
183 * it has too few free blocks left (min_blocks).
184 * Parent's group is preferred, if it doesn't satisfy these
185 * conditions we search cyclically through the rest. If none
186 * of the groups look good we just look for a group with more
187 * free inodes than average (starting at parent's group).
189 * Debt is incremented each time we allocate a directory and decremented
190 * when we allocate an inode, within 0--255.
193 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
)
195 int parent_group
= EXT3_I(parent
)->i_block_group
;
196 struct ext3_sb_info
*sbi
= EXT3_SB(sb
);
197 int ngroups
= sbi
->s_groups_count
;
198 int inodes_per_group
= EXT3_INODES_PER_GROUP(sb
);
199 unsigned int freei
, avefreei
;
200 ext3_fsblk_t freeb
, avefreeb
;
202 int max_dirs
, min_inodes
;
203 ext3_grpblk_t min_blocks
;
205 struct ext3_group_desc
*desc
;
207 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
208 avefreei
= freei
/ ngroups
;
209 freeb
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
210 avefreeb
= freeb
/ ngroups
;
211 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
213 if ((parent
== sb
->s_root
->d_inode
) ||
214 (EXT3_I(parent
)->i_flags
& EXT3_TOPDIR_FL
)) {
215 int best_ndir
= inodes_per_group
;
218 group
= prandom_u32();
219 parent_group
= (unsigned)group
% ngroups
;
220 for (i
= 0; i
< ngroups
; i
++) {
221 group
= (parent_group
+ i
) % ngroups
;
222 desc
= ext3_get_group_desc (sb
, group
, NULL
);
223 if (!desc
|| !desc
->bg_free_inodes_count
)
225 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= best_ndir
)
227 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
229 if (le16_to_cpu(desc
->bg_free_blocks_count
) < avefreeb
)
232 best_ndir
= le16_to_cpu(desc
->bg_used_dirs_count
);
239 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
240 min_inodes
= avefreei
- inodes_per_group
/ 4;
241 min_blocks
= avefreeb
- EXT3_BLOCKS_PER_GROUP(sb
) / 4;
243 for (i
= 0; i
< ngroups
; i
++) {
244 group
= (parent_group
+ i
) % ngroups
;
245 desc
= ext3_get_group_desc (sb
, group
, NULL
);
246 if (!desc
|| !desc
->bg_free_inodes_count
)
248 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= max_dirs
)
250 if (le16_to_cpu(desc
->bg_free_inodes_count
) < min_inodes
)
252 if (le16_to_cpu(desc
->bg_free_blocks_count
) < min_blocks
)
258 for (i
= 0; i
< ngroups
; i
++) {
259 group
= (parent_group
+ i
) % ngroups
;
260 desc
= ext3_get_group_desc (sb
, group
, NULL
);
261 if (!desc
|| !desc
->bg_free_inodes_count
)
263 if (le16_to_cpu(desc
->bg_free_inodes_count
) >= avefreei
)
269 * The free-inodes counter is approximate, and for really small
270 * filesystems the above test can fail to find any blockgroups
279 static int find_group_other(struct super_block
*sb
, struct inode
*parent
)
281 int parent_group
= EXT3_I(parent
)->i_block_group
;
282 int ngroups
= EXT3_SB(sb
)->s_groups_count
;
283 struct ext3_group_desc
*desc
;
287 * Try to place the inode in its parent directory
289 group
= parent_group
;
290 desc
= ext3_get_group_desc (sb
, group
, NULL
);
291 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
292 le16_to_cpu(desc
->bg_free_blocks_count
))
296 * We're going to place this inode in a different blockgroup from its
297 * parent. We want to cause files in a common directory to all land in
298 * the same blockgroup. But we want files which are in a different
299 * directory which shares a blockgroup with our parent to land in a
300 * different blockgroup.
302 * So add our directory's i_ino into the starting point for the hash.
304 group
= (group
+ parent
->i_ino
) % ngroups
;
307 * Use a quadratic hash to find a group with a free inode and some free
310 for (i
= 1; i
< ngroups
; i
<<= 1) {
312 if (group
>= ngroups
)
314 desc
= ext3_get_group_desc (sb
, group
, NULL
);
315 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
316 le16_to_cpu(desc
->bg_free_blocks_count
))
321 * That failed: try linear search for a free inode, even if that group
322 * has no free blocks.
324 group
= parent_group
;
325 for (i
= 0; i
< ngroups
; i
++) {
326 if (++group
>= ngroups
)
328 desc
= ext3_get_group_desc (sb
, group
, NULL
);
329 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
))
337 * There are two policies for allocating an inode. If the new inode is
338 * a directory, then a forward search is made for a block group with both
339 * free space and a low directory-to-inode ratio; if that fails, then of
340 * the groups with above-average free space, that group with the fewest
341 * directories already is chosen.
343 * For other inodes, search forward from the parent directory's block
344 * group to find a free inode.
346 struct inode
*ext3_new_inode(handle_t
*handle
, struct inode
* dir
,
347 const struct qstr
*qstr
, umode_t mode
)
349 struct super_block
*sb
;
350 struct buffer_head
*bitmap_bh
= NULL
;
351 struct buffer_head
*bh2
;
353 unsigned long ino
= 0;
354 struct inode
* inode
;
355 struct ext3_group_desc
* gdp
= NULL
;
356 struct ext3_super_block
* es
;
357 struct ext3_inode_info
*ei
;
358 struct ext3_sb_info
*sbi
;
363 /* Cannot create files in a deleted directory */
364 if (!dir
|| !dir
->i_nlink
)
365 return ERR_PTR(-EPERM
);
368 trace_ext3_request_inode(dir
, mode
);
369 inode
= new_inode(sb
);
371 return ERR_PTR(-ENOMEM
);
377 group
= find_group_orlov(sb
, dir
);
379 group
= find_group_other(sb
, dir
);
385 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
388 gdp
= ext3_get_group_desc(sb
, group
, &bh2
);
393 bitmap_bh
= read_inode_bitmap(sb
, group
);
399 repeat_in_this_group
:
400 ino
= ext3_find_next_zero_bit((unsigned long *)
401 bitmap_bh
->b_data
, EXT3_INODES_PER_GROUP(sb
), ino
);
402 if (ino
< EXT3_INODES_PER_GROUP(sb
)) {
404 BUFFER_TRACE(bitmap_bh
, "get_write_access");
405 err
= ext3_journal_get_write_access(handle
, bitmap_bh
);
409 if (!ext3_set_bit_atomic(sb_bgl_lock(sbi
, group
),
410 ino
, bitmap_bh
->b_data
)) {
412 BUFFER_TRACE(bitmap_bh
,
413 "call ext3_journal_dirty_metadata");
414 err
= ext3_journal_dirty_metadata(handle
,
421 journal_release_buffer(handle
, bitmap_bh
);
423 if (++ino
< EXT3_INODES_PER_GROUP(sb
))
424 goto repeat_in_this_group
;
428 * This case is possible in concurrent environment. It is very
429 * rare. We cannot repeat the find_group_xxx() call because
430 * that will simply return the same blockgroup, because the
431 * group descriptor metadata has not yet been updated.
432 * So we just go onto the next blockgroup.
434 if (++group
== sbi
->s_groups_count
)
441 ino
+= group
* EXT3_INODES_PER_GROUP(sb
) + 1;
442 if (ino
< EXT3_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
443 ext3_error (sb
, "ext3_new_inode",
444 "reserved inode or inode > inodes count - "
445 "block_group = %d, inode=%lu", group
, ino
);
450 BUFFER_TRACE(bh2
, "get_write_access");
451 err
= ext3_journal_get_write_access(handle
, bh2
);
453 spin_lock(sb_bgl_lock(sbi
, group
));
454 le16_add_cpu(&gdp
->bg_free_inodes_count
, -1);
456 le16_add_cpu(&gdp
->bg_used_dirs_count
, 1);
458 spin_unlock(sb_bgl_lock(sbi
, group
));
459 BUFFER_TRACE(bh2
, "call ext3_journal_dirty_metadata");
460 err
= ext3_journal_dirty_metadata(handle
, bh2
);
463 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
465 percpu_counter_inc(&sbi
->s_dirs_counter
);
468 if (test_opt(sb
, GRPID
)) {
469 inode
->i_mode
= mode
;
470 inode
->i_uid
= current_fsuid();
471 inode
->i_gid
= dir
->i_gid
;
473 inode_init_owner(inode
, dir
, mode
);
476 /* This is the optimal IO size (for stat), not the fs block size */
478 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
480 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
481 ei
->i_dir_start_lookup
= 0;
485 ext3_mask_flags(mode
, EXT3_I(dir
)->i_flags
& EXT3_FL_INHERITED
);
486 #ifdef EXT3_FRAGMENTS
494 ei
->i_block_alloc_info
= NULL
;
495 ei
->i_block_group
= group
;
497 ext3_set_inode_flags(inode
);
498 if (IS_DIRSYNC(inode
))
500 if (insert_inode_locked(inode
) < 0) {
502 * Likely a bitmap corruption causing inode to be allocated
508 spin_lock(&sbi
->s_next_gen_lock
);
509 inode
->i_generation
= sbi
->s_next_generation
++;
510 spin_unlock(&sbi
->s_next_gen_lock
);
512 ei
->i_state_flags
= 0;
513 ext3_set_inode_state(inode
, EXT3_STATE_NEW
);
515 /* See comment in ext3_iget for explanation */
516 if (ino
>= EXT3_FIRST_INO(sb
) + 1 &&
517 EXT3_INODE_SIZE(sb
) > EXT3_GOOD_OLD_INODE_SIZE
) {
519 sizeof(struct ext3_inode
) - EXT3_GOOD_OLD_INODE_SIZE
;
521 ei
->i_extra_isize
= 0;
525 dquot_initialize(inode
);
526 err
= dquot_alloc_inode(inode
);
530 err
= ext3_init_acl(handle
, inode
, dir
);
534 err
= ext3_init_security(handle
, inode
, dir
, qstr
);
538 err
= ext3_mark_inode_dirty(handle
, inode
);
540 ext3_std_error(sb
, err
);
544 ext3_debug("allocating inode %lu\n", inode
->i_ino
);
545 trace_ext3_allocate_inode(inode
, dir
, mode
);
548 ext3_std_error(sb
, err
);
557 dquot_free_inode(inode
);
561 inode
->i_flags
|= S_NOQUOTA
;
563 unlock_new_inode(inode
);
569 /* Verify that we are loading a valid orphan from disk */
570 struct inode
*ext3_orphan_get(struct super_block
*sb
, unsigned long ino
)
572 unsigned long max_ino
= le32_to_cpu(EXT3_SB(sb
)->s_es
->s_inodes_count
);
573 unsigned long block_group
;
575 struct buffer_head
*bitmap_bh
;
576 struct inode
*inode
= NULL
;
579 /* Error cases - e2fsck has already cleaned up for us */
581 ext3_warning(sb
, __func__
,
582 "bad orphan ino %lu! e2fsck was run?", ino
);
586 block_group
= (ino
- 1) / EXT3_INODES_PER_GROUP(sb
);
587 bit
= (ino
- 1) % EXT3_INODES_PER_GROUP(sb
);
588 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
590 ext3_warning(sb
, __func__
,
591 "inode bitmap error for orphan %lu", ino
);
595 /* Having the inode bit set should be a 100% indicator that this
596 * is a valid orphan (no e2fsck run on fs). Orphans also include
597 * inodes that were being truncated, so we can't check i_nlink==0.
599 if (!ext3_test_bit(bit
, bitmap_bh
->b_data
))
602 inode
= ext3_iget(sb
, ino
);
607 * If the orphans has i_nlinks > 0 then it should be able to be
608 * truncated, otherwise it won't be removed from the orphan list
609 * during processing and an infinite loop will result.
611 if (inode
->i_nlink
&& !ext3_can_truncate(inode
))
614 if (NEXT_ORPHAN(inode
) > max_ino
)
620 err
= PTR_ERR(inode
);
623 ext3_warning(sb
, __func__
,
624 "bad orphan inode %lu! e2fsck was run?", ino
);
625 printk(KERN_NOTICE
"ext3_test_bit(bit=%d, block=%llu) = %d\n",
626 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
627 ext3_test_bit(bit
, bitmap_bh
->b_data
));
628 printk(KERN_NOTICE
"inode=%p\n", inode
);
630 printk(KERN_NOTICE
"is_bad_inode(inode)=%d\n",
631 is_bad_inode(inode
));
632 printk(KERN_NOTICE
"NEXT_ORPHAN(inode)=%u\n",
634 printk(KERN_NOTICE
"max_ino=%lu\n", max_ino
);
635 printk(KERN_NOTICE
"i_nlink=%u\n", inode
->i_nlink
);
636 /* Avoid freeing blocks if we got a bad deleted inode */
637 if (inode
->i_nlink
== 0)
646 unsigned long ext3_count_free_inodes (struct super_block
* sb
)
648 unsigned long desc_count
;
649 struct ext3_group_desc
*gdp
;
652 struct ext3_super_block
*es
;
653 unsigned long bitmap_count
, x
;
654 struct buffer_head
*bitmap_bh
= NULL
;
656 es
= EXT3_SB(sb
)->s_es
;
660 for (i
= 0; i
< EXT3_SB(sb
)->s_groups_count
; i
++) {
661 gdp
= ext3_get_group_desc (sb
, i
, NULL
);
664 desc_count
+= le16_to_cpu(gdp
->bg_free_inodes_count
);
666 bitmap_bh
= read_inode_bitmap(sb
, i
);
670 x
= ext3_count_free(bitmap_bh
, EXT3_INODES_PER_GROUP(sb
) / 8);
671 printk("group %d: stored = %d, counted = %lu\n",
672 i
, le16_to_cpu(gdp
->bg_free_inodes_count
), x
);
676 printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n",
677 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
681 for (i
= 0; i
< EXT3_SB(sb
)->s_groups_count
; i
++) {
682 gdp
= ext3_get_group_desc (sb
, i
, NULL
);
685 desc_count
+= le16_to_cpu(gdp
->bg_free_inodes_count
);
692 /* Called at mount-time, super-block is locked */
693 unsigned long ext3_count_dirs (struct super_block
* sb
)
695 unsigned long count
= 0;
698 for (i
= 0; i
< EXT3_SB(sb
)->s_groups_count
; i
++) {
699 struct ext3_group_desc
*gdp
= ext3_get_group_desc (sb
, i
, NULL
);
702 count
+= le16_to_cpu(gdp
->bg_used_dirs_count
);