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/bitops.h>
26 #include <trace/events/ext3.h>
28 #include <asm/byteorder.h>
34 * ialloc.c contains the inodes allocation and deallocation routines
38 * The free inodes are managed by bitmaps. A file system contains several
39 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
40 * block for inodes, N blocks for the inode table and data blocks.
42 * The file system contains group descriptors which are located after the
43 * super block. Each descriptor contains the number of the bitmap block and
44 * the free blocks count in the block.
49 * Read the inode allocation bitmap for a given block_group, reading
50 * into the specified slot in the superblock's bitmap cache.
52 * Return buffer_head of bitmap on success or NULL.
54 static struct buffer_head
*
55 read_inode_bitmap(struct super_block
* sb
, unsigned long block_group
)
57 struct ext3_group_desc
*desc
;
58 struct buffer_head
*bh
= NULL
;
60 desc
= ext3_get_group_desc(sb
, block_group
, NULL
);
64 bh
= sb_bread(sb
, le32_to_cpu(desc
->bg_inode_bitmap
));
66 ext3_error(sb
, "read_inode_bitmap",
67 "Cannot read inode bitmap - "
68 "block_group = %lu, inode_bitmap = %u",
69 block_group
, le32_to_cpu(desc
->bg_inode_bitmap
));
75 * NOTE! When we get the inode, we're the only people
76 * that have access to it, and as such there are no
77 * race conditions we have to worry about. The inode
78 * is not on the hash-lists, and it cannot be reached
79 * through the filesystem because the directory entry
80 * has been deleted earlier.
82 * HOWEVER: we must make sure that we get no aliases,
83 * which means that we have to call "clear_inode()"
84 * _before_ we mark the inode not in use in the inode
85 * bitmaps. Otherwise a newly created file might use
86 * the same inode number (not actually the same pointer
87 * though), and then we'd have two inodes sharing the
88 * same inode number and space on the harddisk.
90 void ext3_free_inode (handle_t
*handle
, struct inode
* inode
)
92 struct super_block
* sb
= inode
->i_sb
;
95 struct buffer_head
*bitmap_bh
= NULL
;
96 struct buffer_head
*bh2
;
97 unsigned long block_group
;
99 struct ext3_group_desc
* gdp
;
100 struct ext3_super_block
* es
;
101 struct ext3_sb_info
*sbi
;
104 if (atomic_read(&inode
->i_count
) > 1) {
105 printk ("ext3_free_inode: inode has count=%d\n",
106 atomic_read(&inode
->i_count
));
109 if (inode
->i_nlink
) {
110 printk ("ext3_free_inode: inode has nlink=%d\n",
115 printk("ext3_free_inode: inode on nonexistent device\n");
121 ext3_debug ("freeing inode %lu\n", ino
);
122 trace_ext3_free_inode(inode
);
124 is_directory
= S_ISDIR(inode
->i_mode
);
126 es
= EXT3_SB(sb
)->s_es
;
127 if (ino
< EXT3_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
128 ext3_error (sb
, "ext3_free_inode",
129 "reserved or nonexistent inode %lu", ino
);
132 block_group
= (ino
- 1) / EXT3_INODES_PER_GROUP(sb
);
133 bit
= (ino
- 1) % EXT3_INODES_PER_GROUP(sb
);
134 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
138 BUFFER_TRACE(bitmap_bh
, "get_write_access");
139 fatal
= ext3_journal_get_write_access(handle
, bitmap_bh
);
143 /* Ok, now we can actually update the inode bitmaps.. */
144 if (!ext3_clear_bit_atomic(sb_bgl_lock(sbi
, block_group
),
145 bit
, bitmap_bh
->b_data
))
146 ext3_error (sb
, "ext3_free_inode",
147 "bit already cleared for inode %lu", ino
);
149 gdp
= ext3_get_group_desc (sb
, block_group
, &bh2
);
151 BUFFER_TRACE(bh2
, "get_write_access");
152 fatal
= ext3_journal_get_write_access(handle
, bh2
);
153 if (fatal
) goto error_return
;
156 spin_lock(sb_bgl_lock(sbi
, block_group
));
157 le16_add_cpu(&gdp
->bg_free_inodes_count
, 1);
159 le16_add_cpu(&gdp
->bg_used_dirs_count
, -1);
160 spin_unlock(sb_bgl_lock(sbi
, block_group
));
161 percpu_counter_inc(&sbi
->s_freeinodes_counter
);
163 percpu_counter_dec(&sbi
->s_dirs_counter
);
166 BUFFER_TRACE(bh2
, "call ext3_journal_dirty_metadata");
167 err
= ext3_journal_dirty_metadata(handle
, bh2
);
168 if (!fatal
) fatal
= err
;
170 BUFFER_TRACE(bitmap_bh
, "call ext3_journal_dirty_metadata");
171 err
= ext3_journal_dirty_metadata(handle
, bitmap_bh
);
177 ext3_std_error(sb
, fatal
);
181 * Orlov's allocator for directories.
183 * We always try to spread first-level directories.
185 * If there are blockgroups with both free inodes and free blocks counts
186 * not worse than average we return one with smallest directory count.
187 * Otherwise we simply return a random group.
189 * For the rest rules look so:
191 * It's OK to put directory into a group unless
192 * it has too many directories already (max_dirs) or
193 * it has too few free inodes left (min_inodes) or
194 * it has too few free blocks left (min_blocks) or
195 * it's already running too large debt (max_debt).
196 * Parent's group is preferred, if it doesn't satisfy these
197 * conditions we search cyclically through the rest. If none
198 * of the groups look good we just look for a group with more
199 * free inodes than average (starting at parent's group).
201 * Debt is incremented each time we allocate a directory and decremented
202 * when we allocate an inode, within 0--255.
205 #define INODE_COST 64
206 #define BLOCK_COST 256
208 static int find_group_orlov(struct super_block
*sb
, struct inode
*parent
)
210 int parent_group
= EXT3_I(parent
)->i_block_group
;
211 struct ext3_sb_info
*sbi
= EXT3_SB(sb
);
212 struct ext3_super_block
*es
= sbi
->s_es
;
213 int ngroups
= sbi
->s_groups_count
;
214 int inodes_per_group
= EXT3_INODES_PER_GROUP(sb
);
215 unsigned int freei
, avefreei
;
216 ext3_fsblk_t freeb
, avefreeb
;
217 ext3_fsblk_t blocks_per_dir
;
219 int max_debt
, max_dirs
, min_inodes
;
220 ext3_grpblk_t min_blocks
;
222 struct ext3_group_desc
*desc
;
224 freei
= percpu_counter_read_positive(&sbi
->s_freeinodes_counter
);
225 avefreei
= freei
/ ngroups
;
226 freeb
= percpu_counter_read_positive(&sbi
->s_freeblocks_counter
);
227 avefreeb
= freeb
/ ngroups
;
228 ndirs
= percpu_counter_read_positive(&sbi
->s_dirs_counter
);
230 if ((parent
== sb
->s_root
->d_inode
) ||
231 (EXT3_I(parent
)->i_flags
& EXT3_TOPDIR_FL
)) {
232 int best_ndir
= inodes_per_group
;
235 get_random_bytes(&group
, sizeof(group
));
236 parent_group
= (unsigned)group
% ngroups
;
237 for (i
= 0; i
< ngroups
; i
++) {
238 group
= (parent_group
+ i
) % ngroups
;
239 desc
= ext3_get_group_desc (sb
, group
, NULL
);
240 if (!desc
|| !desc
->bg_free_inodes_count
)
242 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= best_ndir
)
244 if (le16_to_cpu(desc
->bg_free_inodes_count
) < avefreei
)
246 if (le16_to_cpu(desc
->bg_free_blocks_count
) < avefreeb
)
249 best_ndir
= le16_to_cpu(desc
->bg_used_dirs_count
);
256 blocks_per_dir
= (le32_to_cpu(es
->s_blocks_count
) - freeb
) / ndirs
;
258 max_dirs
= ndirs
/ ngroups
+ inodes_per_group
/ 16;
259 min_inodes
= avefreei
- inodes_per_group
/ 4;
260 min_blocks
= avefreeb
- EXT3_BLOCKS_PER_GROUP(sb
) / 4;
262 max_debt
= EXT3_BLOCKS_PER_GROUP(sb
) / max(blocks_per_dir
, (ext3_fsblk_t
)BLOCK_COST
);
263 if (max_debt
* INODE_COST
> inodes_per_group
)
264 max_debt
= inodes_per_group
/ INODE_COST
;
270 for (i
= 0; i
< ngroups
; i
++) {
271 group
= (parent_group
+ i
) % ngroups
;
272 desc
= ext3_get_group_desc (sb
, group
, NULL
);
273 if (!desc
|| !desc
->bg_free_inodes_count
)
275 if (le16_to_cpu(desc
->bg_used_dirs_count
) >= max_dirs
)
277 if (le16_to_cpu(desc
->bg_free_inodes_count
) < min_inodes
)
279 if (le16_to_cpu(desc
->bg_free_blocks_count
) < min_blocks
)
285 for (i
= 0; i
< ngroups
; i
++) {
286 group
= (parent_group
+ i
) % ngroups
;
287 desc
= ext3_get_group_desc (sb
, group
, NULL
);
288 if (!desc
|| !desc
->bg_free_inodes_count
)
290 if (le16_to_cpu(desc
->bg_free_inodes_count
) >= avefreei
)
296 * The free-inodes counter is approximate, and for really small
297 * filesystems the above test can fail to find any blockgroups
306 static int find_group_other(struct super_block
*sb
, struct inode
*parent
)
308 int parent_group
= EXT3_I(parent
)->i_block_group
;
309 int ngroups
= EXT3_SB(sb
)->s_groups_count
;
310 struct ext3_group_desc
*desc
;
314 * Try to place the inode in its parent directory
316 group
= parent_group
;
317 desc
= ext3_get_group_desc (sb
, group
, NULL
);
318 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
319 le16_to_cpu(desc
->bg_free_blocks_count
))
323 * We're going to place this inode in a different blockgroup from its
324 * parent. We want to cause files in a common directory to all land in
325 * the same blockgroup. But we want files which are in a different
326 * directory which shares a blockgroup with our parent to land in a
327 * different blockgroup.
329 * So add our directory's i_ino into the starting point for the hash.
331 group
= (group
+ parent
->i_ino
) % ngroups
;
334 * Use a quadratic hash to find a group with a free inode and some free
337 for (i
= 1; i
< ngroups
; i
<<= 1) {
339 if (group
>= ngroups
)
341 desc
= ext3_get_group_desc (sb
, group
, NULL
);
342 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
) &&
343 le16_to_cpu(desc
->bg_free_blocks_count
))
348 * That failed: try linear search for a free inode, even if that group
349 * has no free blocks.
351 group
= parent_group
;
352 for (i
= 0; i
< ngroups
; i
++) {
353 if (++group
>= ngroups
)
355 desc
= ext3_get_group_desc (sb
, group
, NULL
);
356 if (desc
&& le16_to_cpu(desc
->bg_free_inodes_count
))
364 * There are two policies for allocating an inode. If the new inode is
365 * a directory, then a forward search is made for a block group with both
366 * free space and a low directory-to-inode ratio; if that fails, then of
367 * the groups with above-average free space, that group with the fewest
368 * directories already is chosen.
370 * For other inodes, search forward from the parent directory's block
371 * group to find a free inode.
373 struct inode
*ext3_new_inode(handle_t
*handle
, struct inode
* dir
,
374 const struct qstr
*qstr
, umode_t mode
)
376 struct super_block
*sb
;
377 struct buffer_head
*bitmap_bh
= NULL
;
378 struct buffer_head
*bh2
;
380 unsigned long ino
= 0;
381 struct inode
* inode
;
382 struct ext3_group_desc
* gdp
= NULL
;
383 struct ext3_super_block
* es
;
384 struct ext3_inode_info
*ei
;
385 struct ext3_sb_info
*sbi
;
390 /* Cannot create files in a deleted directory */
391 if (!dir
|| !dir
->i_nlink
)
392 return ERR_PTR(-EPERM
);
395 trace_ext3_request_inode(dir
, mode
);
396 inode
= new_inode(sb
);
398 return ERR_PTR(-ENOMEM
);
404 group
= find_group_orlov(sb
, dir
);
406 group
= find_group_other(sb
, dir
);
412 for (i
= 0; i
< sbi
->s_groups_count
; i
++) {
415 gdp
= ext3_get_group_desc(sb
, group
, &bh2
);
420 bitmap_bh
= read_inode_bitmap(sb
, group
);
426 repeat_in_this_group
:
427 ino
= ext3_find_next_zero_bit((unsigned long *)
428 bitmap_bh
->b_data
, EXT3_INODES_PER_GROUP(sb
), ino
);
429 if (ino
< EXT3_INODES_PER_GROUP(sb
)) {
431 BUFFER_TRACE(bitmap_bh
, "get_write_access");
432 err
= ext3_journal_get_write_access(handle
, bitmap_bh
);
436 if (!ext3_set_bit_atomic(sb_bgl_lock(sbi
, group
),
437 ino
, bitmap_bh
->b_data
)) {
439 BUFFER_TRACE(bitmap_bh
,
440 "call ext3_journal_dirty_metadata");
441 err
= ext3_journal_dirty_metadata(handle
,
448 journal_release_buffer(handle
, bitmap_bh
);
450 if (++ino
< EXT3_INODES_PER_GROUP(sb
))
451 goto repeat_in_this_group
;
455 * This case is possible in concurrent environment. It is very
456 * rare. We cannot repeat the find_group_xxx() call because
457 * that will simply return the same blockgroup, because the
458 * group descriptor metadata has not yet been updated.
459 * So we just go onto the next blockgroup.
461 if (++group
== sbi
->s_groups_count
)
468 ino
+= group
* EXT3_INODES_PER_GROUP(sb
) + 1;
469 if (ino
< EXT3_FIRST_INO(sb
) || ino
> le32_to_cpu(es
->s_inodes_count
)) {
470 ext3_error (sb
, "ext3_new_inode",
471 "reserved inode or inode > inodes count - "
472 "block_group = %d, inode=%lu", group
, ino
);
477 BUFFER_TRACE(bh2
, "get_write_access");
478 err
= ext3_journal_get_write_access(handle
, bh2
);
480 spin_lock(sb_bgl_lock(sbi
, group
));
481 le16_add_cpu(&gdp
->bg_free_inodes_count
, -1);
483 le16_add_cpu(&gdp
->bg_used_dirs_count
, 1);
485 spin_unlock(sb_bgl_lock(sbi
, group
));
486 BUFFER_TRACE(bh2
, "call ext3_journal_dirty_metadata");
487 err
= ext3_journal_dirty_metadata(handle
, bh2
);
490 percpu_counter_dec(&sbi
->s_freeinodes_counter
);
492 percpu_counter_inc(&sbi
->s_dirs_counter
);
495 if (test_opt(sb
, GRPID
)) {
496 inode
->i_mode
= mode
;
497 inode
->i_uid
= current_fsuid();
498 inode
->i_gid
= dir
->i_gid
;
500 inode_init_owner(inode
, dir
, mode
);
503 /* This is the optimal IO size (for stat), not the fs block size */
505 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME_SEC
;
507 memset(ei
->i_data
, 0, sizeof(ei
->i_data
));
508 ei
->i_dir_start_lookup
= 0;
512 ext3_mask_flags(mode
, EXT3_I(dir
)->i_flags
& EXT3_FL_INHERITED
);
513 #ifdef EXT3_FRAGMENTS
521 ei
->i_block_alloc_info
= NULL
;
522 ei
->i_block_group
= group
;
524 ext3_set_inode_flags(inode
);
525 if (IS_DIRSYNC(inode
))
527 if (insert_inode_locked(inode
) < 0) {
529 * Likely a bitmap corruption causing inode to be allocated
535 spin_lock(&sbi
->s_next_gen_lock
);
536 inode
->i_generation
= sbi
->s_next_generation
++;
537 spin_unlock(&sbi
->s_next_gen_lock
);
539 ei
->i_state_flags
= 0;
540 ext3_set_inode_state(inode
, EXT3_STATE_NEW
);
542 /* See comment in ext3_iget for explanation */
543 if (ino
>= EXT3_FIRST_INO(sb
) + 1 &&
544 EXT3_INODE_SIZE(sb
) > EXT3_GOOD_OLD_INODE_SIZE
) {
546 sizeof(struct ext3_inode
) - EXT3_GOOD_OLD_INODE_SIZE
;
548 ei
->i_extra_isize
= 0;
552 dquot_initialize(inode
);
553 err
= dquot_alloc_inode(inode
);
557 err
= ext3_init_acl(handle
, inode
, dir
);
561 err
= ext3_init_security(handle
, inode
, dir
, qstr
);
565 err
= ext3_mark_inode_dirty(handle
, inode
);
567 ext3_std_error(sb
, err
);
571 ext3_debug("allocating inode %lu\n", inode
->i_ino
);
572 trace_ext3_allocate_inode(inode
, dir
, mode
);
575 ext3_std_error(sb
, err
);
584 dquot_free_inode(inode
);
588 inode
->i_flags
|= S_NOQUOTA
;
590 unlock_new_inode(inode
);
596 /* Verify that we are loading a valid orphan from disk */
597 struct inode
*ext3_orphan_get(struct super_block
*sb
, unsigned long ino
)
599 unsigned long max_ino
= le32_to_cpu(EXT3_SB(sb
)->s_es
->s_inodes_count
);
600 unsigned long block_group
;
602 struct buffer_head
*bitmap_bh
;
603 struct inode
*inode
= NULL
;
606 /* Error cases - e2fsck has already cleaned up for us */
608 ext3_warning(sb
, __func__
,
609 "bad orphan ino %lu! e2fsck was run?", ino
);
613 block_group
= (ino
- 1) / EXT3_INODES_PER_GROUP(sb
);
614 bit
= (ino
- 1) % EXT3_INODES_PER_GROUP(sb
);
615 bitmap_bh
= read_inode_bitmap(sb
, block_group
);
617 ext3_warning(sb
, __func__
,
618 "inode bitmap error for orphan %lu", ino
);
622 /* Having the inode bit set should be a 100% indicator that this
623 * is a valid orphan (no e2fsck run on fs). Orphans also include
624 * inodes that were being truncated, so we can't check i_nlink==0.
626 if (!ext3_test_bit(bit
, bitmap_bh
->b_data
))
629 inode
= ext3_iget(sb
, ino
);
634 * If the orphans has i_nlinks > 0 then it should be able to be
635 * truncated, otherwise it won't be removed from the orphan list
636 * during processing and an infinite loop will result.
638 if (inode
->i_nlink
&& !ext3_can_truncate(inode
))
641 if (NEXT_ORPHAN(inode
) > max_ino
)
647 err
= PTR_ERR(inode
);
650 ext3_warning(sb
, __func__
,
651 "bad orphan inode %lu! e2fsck was run?", ino
);
652 printk(KERN_NOTICE
"ext3_test_bit(bit=%d, block=%llu) = %d\n",
653 bit
, (unsigned long long)bitmap_bh
->b_blocknr
,
654 ext3_test_bit(bit
, bitmap_bh
->b_data
));
655 printk(KERN_NOTICE
"inode=%p\n", inode
);
657 printk(KERN_NOTICE
"is_bad_inode(inode)=%d\n",
658 is_bad_inode(inode
));
659 printk(KERN_NOTICE
"NEXT_ORPHAN(inode)=%u\n",
661 printk(KERN_NOTICE
"max_ino=%lu\n", max_ino
);
662 printk(KERN_NOTICE
"i_nlink=%u\n", inode
->i_nlink
);
663 /* Avoid freeing blocks if we got a bad deleted inode */
664 if (inode
->i_nlink
== 0)
673 unsigned long ext3_count_free_inodes (struct super_block
* sb
)
675 unsigned long desc_count
;
676 struct ext3_group_desc
*gdp
;
679 struct ext3_super_block
*es
;
680 unsigned long bitmap_count
, x
;
681 struct buffer_head
*bitmap_bh
= NULL
;
683 es
= EXT3_SB(sb
)->s_es
;
687 for (i
= 0; i
< EXT3_SB(sb
)->s_groups_count
; i
++) {
688 gdp
= ext3_get_group_desc (sb
, i
, NULL
);
691 desc_count
+= le16_to_cpu(gdp
->bg_free_inodes_count
);
693 bitmap_bh
= read_inode_bitmap(sb
, i
);
697 x
= ext3_count_free(bitmap_bh
, EXT3_INODES_PER_GROUP(sb
) / 8);
698 printk("group %d: stored = %d, counted = %lu\n",
699 i
, le16_to_cpu(gdp
->bg_free_inodes_count
), x
);
703 printk("ext3_count_free_inodes: stored = %u, computed = %lu, %lu\n",
704 le32_to_cpu(es
->s_free_inodes_count
), desc_count
, bitmap_count
);
708 for (i
= 0; i
< EXT3_SB(sb
)->s_groups_count
; i
++) {
709 gdp
= ext3_get_group_desc (sb
, i
, NULL
);
712 desc_count
+= le16_to_cpu(gdp
->bg_free_inodes_count
);
719 /* Called at mount-time, super-block is locked */
720 unsigned long ext3_count_dirs (struct super_block
* sb
)
722 unsigned long count
= 0;
725 for (i
= 0; i
< EXT3_SB(sb
)->s_groups_count
; i
++) {
726 struct ext3_group_desc
*gdp
= ext3_get_group_desc (sb
, i
, NULL
);
729 count
+= le16_to_cpu(gdp
->bg_used_dirs_count
);