2 * linux/fs/ext4/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/jbd2.h>
18 #include <linux/stat.h>
19 #include <linux/string.h>
20 #include <linux/quotaops.h>
21 #include <linux/buffer_head.h>
22 #include <linux/random.h>
23 #include <linux/bitops.h>
24 #include <linux/blkdev.h>
25 #include <asm/byteorder.h>
28 #include "ext4_jbd2.h"
32 #include <trace/events/ext4.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.
49 * To avoid calling the atomic setbit hundreds or thousands of times, we only
50 * need to use it within a single byte (to ensure we get endianness right).
51 * We can use memset for the rest of the bitmap as there are no other users.
53 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
57 if (start_bit >= end_bit)
60 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
61 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
62 ext4_set_bit(i, bitmap);
64 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
67 /* Initializes an uninitialized inode bitmap */
68 static unsigned ext4_init_inode_bitmap(struct super_block *sb,
69 struct buffer_head *bh,
70 ext4_group_t block_group,
71 struct ext4_group_desc *gdp)
73 struct ext4_sb_info *sbi = EXT4_SB(sb);
75 J_ASSERT_BH(bh, buffer_locked(bh));
77 /* If checksum is bad mark all blocks and inodes use to prevent
78 * allocation, essentially implementing a per-group read-only flag. */
79 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
80 ext4_error(sb, "Checksum bad for group %u", block_group);
81 ext4_free_group_clusters_set(sb, gdp, 0);
82 ext4_free_inodes_set(sb, gdp, 0);
83 ext4_itable_unused_set(sb, gdp, 0);
84 memset(bh->b_data, 0xff, sb->s_blocksize);
88 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
89 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
92 return EXT4_INODES_PER_GROUP(sb);
95 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
98 set_buffer_uptodate(bh);
99 set_bitmap_uptodate(bh);
106 * Read the inode allocation bitmap for a given block_group, reading
107 * into the specified slot in the superblock's bitmap cache.
109 * Return buffer_head of bitmap on success or NULL.
111 static struct buffer_head *
112 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
114 struct ext4_group_desc *desc;
115 struct buffer_head *bh = NULL;
116 ext4_fsblk_t bitmap_blk;
118 desc = ext4_get_group_desc(sb, block_group, NULL);
122 bitmap_blk = ext4_inode_bitmap(sb, desc);
123 bh = sb_getblk(sb, bitmap_blk);
125 ext4_error(sb, "Cannot read inode bitmap - "
126 "block_group = %u, inode_bitmap = %llu",
127 block_group, bitmap_blk);
130 if (bitmap_uptodate(bh))
134 if (bitmap_uptodate(bh)) {
139 ext4_lock_group(sb, block_group);
140 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
141 ext4_init_inode_bitmap(sb, bh, block_group, desc);
142 set_bitmap_uptodate(bh);
143 set_buffer_uptodate(bh);
144 ext4_unlock_group(sb, block_group);
148 ext4_unlock_group(sb, block_group);
150 if (buffer_uptodate(bh)) {
152 * if not uninit if bh is uptodate,
153 * bitmap is also uptodate
155 set_bitmap_uptodate(bh);
160 * submit the buffer_head for reading
162 trace_ext4_load_inode_bitmap(sb, block_group);
163 bh->b_end_io = ext4_end_bitmap_read;
167 if (!buffer_uptodate(bh)) {
169 ext4_error(sb, "Cannot read inode bitmap - "
170 "block_group = %u, inode_bitmap = %llu",
171 block_group, bitmap_blk);
178 * NOTE! When we get the inode, we're the only people
179 * that have access to it, and as such there are no
180 * race conditions we have to worry about. The inode
181 * is not on the hash-lists, and it cannot be reached
182 * through the filesystem because the directory entry
183 * has been deleted earlier.
185 * HOWEVER: we must make sure that we get no aliases,
186 * which means that we have to call "clear_inode()"
187 * _before_ we mark the inode not in use in the inode
188 * bitmaps. Otherwise a newly created file might use
189 * the same inode number (not actually the same pointer
190 * though), and then we'd have two inodes sharing the
191 * same inode number and space on the harddisk.
193 void ext4_free_inode(handle_t *handle, struct inode *inode)
195 struct super_block *sb = inode->i_sb;
198 struct buffer_head *bitmap_bh = NULL;
199 struct buffer_head *bh2;
200 ext4_group_t block_group;
202 struct ext4_group_desc *gdp;
203 struct ext4_super_block *es;
204 struct ext4_sb_info *sbi;
205 int fatal = 0, err, count, cleared;
207 if (atomic_read(&inode->i_count) > 1) {
208 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
209 atomic_read(&inode->i_count));
212 if (inode->i_nlink) {
213 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
218 printk(KERN_ERR "ext4_free_inode: inode on "
219 "nonexistent device\n");
225 ext4_debug("freeing inode %lu\n", ino);
226 trace_ext4_free_inode(inode);
229 * Note: we must free any quota before locking the superblock,
230 * as writing the quota to disk may need the lock as well.
232 dquot_initialize(inode);
233 ext4_xattr_delete_inode(handle, inode);
234 dquot_free_inode(inode);
237 is_directory = S_ISDIR(inode->i_mode);
239 /* Do this BEFORE marking the inode not in use or returning an error */
240 ext4_clear_inode(inode);
242 es = EXT4_SB(sb)->s_es;
243 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
244 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
247 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
248 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
249 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
253 BUFFER_TRACE(bitmap_bh, "get_write_access");
254 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
259 gdp = ext4_get_group_desc(sb, block_group, &bh2);
261 BUFFER_TRACE(bh2, "get_write_access");
262 fatal = ext4_journal_get_write_access(handle, bh2);
264 ext4_lock_group(sb, block_group);
265 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
266 if (fatal || !cleared) {
267 ext4_unlock_group(sb, block_group);
271 count = ext4_free_inodes_count(sb, gdp) + 1;
272 ext4_free_inodes_set(sb, gdp, count);
274 count = ext4_used_dirs_count(sb, gdp) - 1;
275 ext4_used_dirs_set(sb, gdp, count);
276 percpu_counter_dec(&sbi->s_dirs_counter);
278 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
279 ext4_unlock_group(sb, block_group);
281 percpu_counter_inc(&sbi->s_freeinodes_counter);
282 if (sbi->s_log_groups_per_flex) {
283 ext4_group_t f = ext4_flex_group(sbi, block_group);
285 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
287 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
289 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
290 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
293 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
294 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
297 ext4_mark_super_dirty(sb);
299 ext4_error(sb, "bit already cleared for inode %lu", ino);
303 ext4_std_error(sb, fatal);
313 * Helper function for Orlov's allocator; returns critical information
314 * for a particular block group or flex_bg. If flex_size is 1, then g
315 * is a block group number; otherwise it is flex_bg number.
317 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
318 int flex_size, struct orlov_stats *stats)
320 struct ext4_group_desc *desc;
321 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
324 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
325 stats->free_clusters = atomic_read(&flex_group[g].free_clusters);
326 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
330 desc = ext4_get_group_desc(sb, g, NULL);
332 stats->free_inodes = ext4_free_inodes_count(sb, desc);
333 stats->free_clusters = ext4_free_group_clusters(sb, desc);
334 stats->used_dirs = ext4_used_dirs_count(sb, desc);
336 stats->free_inodes = 0;
337 stats->free_clusters = 0;
338 stats->used_dirs = 0;
343 * Orlov's allocator for directories.
345 * We always try to spread first-level directories.
347 * If there are blockgroups with both free inodes and free blocks counts
348 * not worse than average we return one with smallest directory count.
349 * Otherwise we simply return a random group.
351 * For the rest rules look so:
353 * It's OK to put directory into a group unless
354 * it has too many directories already (max_dirs) or
355 * it has too few free inodes left (min_inodes) or
356 * it has too few free blocks left (min_blocks) or
357 * Parent's group is preferred, if it doesn't satisfy these
358 * conditions we search cyclically through the rest. If none
359 * of the groups look good we just look for a group with more
360 * free inodes than average (starting at parent's group).
363 static int find_group_orlov(struct super_block *sb, struct inode *parent,
364 ext4_group_t *group, umode_t mode,
365 const struct qstr *qstr)
367 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
368 struct ext4_sb_info *sbi = EXT4_SB(sb);
369 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
370 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
371 unsigned int freei, avefreei, grp_free;
372 ext4_fsblk_t freeb, avefreec;
374 int max_dirs, min_inodes;
375 ext4_grpblk_t min_clusters;
376 ext4_group_t i, grp, g, ngroups;
377 struct ext4_group_desc *desc;
378 struct orlov_stats stats;
379 int flex_size = ext4_flex_bg_size(sbi);
380 struct dx_hash_info hinfo;
382 ngroups = real_ngroups;
384 ngroups = (real_ngroups + flex_size - 1) >>
385 sbi->s_log_groups_per_flex;
386 parent_group >>= sbi->s_log_groups_per_flex;
389 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
390 avefreei = freei / ngroups;
391 freeb = EXT4_C2B(sbi,
392 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
394 do_div(avefreec, ngroups);
395 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
398 ((parent == sb->s_root->d_inode) ||
399 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
400 int best_ndir = inodes_per_group;
404 hinfo.hash_version = DX_HASH_HALF_MD4;
405 hinfo.seed = sbi->s_hash_seed;
406 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
409 get_random_bytes(&grp, sizeof(grp));
410 parent_group = (unsigned)grp % ngroups;
411 for (i = 0; i < ngroups; i++) {
412 g = (parent_group + i) % ngroups;
413 get_orlov_stats(sb, g, flex_size, &stats);
414 if (!stats.free_inodes)
416 if (stats.used_dirs >= best_ndir)
418 if (stats.free_inodes < avefreei)
420 if (stats.free_clusters < avefreec)
424 best_ndir = stats.used_dirs;
429 if (flex_size == 1) {
435 * We pack inodes at the beginning of the flexgroup's
436 * inode tables. Block allocation decisions will do
437 * something similar, although regular files will
438 * start at 2nd block group of the flexgroup. See
439 * ext4_ext_find_goal() and ext4_find_near().
442 for (i = 0; i < flex_size; i++) {
443 if (grp+i >= real_ngroups)
445 desc = ext4_get_group_desc(sb, grp+i, NULL);
446 if (desc && ext4_free_inodes_count(sb, desc)) {
454 max_dirs = ndirs / ngroups + inodes_per_group / 16;
455 min_inodes = avefreei - inodes_per_group*flex_size / 4;
458 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
461 * Start looking in the flex group where we last allocated an
462 * inode for this parent directory
464 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
465 parent_group = EXT4_I(parent)->i_last_alloc_group;
467 parent_group >>= sbi->s_log_groups_per_flex;
470 for (i = 0; i < ngroups; i++) {
471 grp = (parent_group + i) % ngroups;
472 get_orlov_stats(sb, grp, flex_size, &stats);
473 if (stats.used_dirs >= max_dirs)
475 if (stats.free_inodes < min_inodes)
477 if (stats.free_clusters < min_clusters)
483 ngroups = real_ngroups;
484 avefreei = freei / ngroups;
486 parent_group = EXT4_I(parent)->i_block_group;
487 for (i = 0; i < ngroups; i++) {
488 grp = (parent_group + i) % ngroups;
489 desc = ext4_get_group_desc(sb, grp, NULL);
490 grp_free = ext4_free_inodes_count(sb, desc);
491 if (desc && grp_free && grp_free >= avefreei) {
499 * The free-inodes counter is approximate, and for really small
500 * filesystems the above test can fail to find any blockgroups
509 static int find_group_other(struct super_block *sb, struct inode *parent,
510 ext4_group_t *group, umode_t mode)
512 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
513 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
514 struct ext4_group_desc *desc;
515 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
518 * Try to place the inode is the same flex group as its
519 * parent. If we can't find space, use the Orlov algorithm to
520 * find another flex group, and store that information in the
521 * parent directory's inode information so that use that flex
522 * group for future allocations.
528 parent_group &= ~(flex_size-1);
529 last = parent_group + flex_size;
532 for (i = parent_group; i < last; i++) {
533 desc = ext4_get_group_desc(sb, i, NULL);
534 if (desc && ext4_free_inodes_count(sb, desc)) {
539 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
541 parent_group = EXT4_I(parent)->i_last_alloc_group;
545 * If this didn't work, use the Orlov search algorithm
546 * to find a new flex group; we pass in the mode to
547 * avoid the topdir algorithms.
549 *group = parent_group + flex_size;
550 if (*group > ngroups)
552 return find_group_orlov(sb, parent, group, mode, NULL);
556 * Try to place the inode in its parent directory
558 *group = parent_group;
559 desc = ext4_get_group_desc(sb, *group, NULL);
560 if (desc && ext4_free_inodes_count(sb, desc) &&
561 ext4_free_group_clusters(sb, desc))
565 * We're going to place this inode in a different blockgroup from its
566 * parent. We want to cause files in a common directory to all land in
567 * the same blockgroup. But we want files which are in a different
568 * directory which shares a blockgroup with our parent to land in a
569 * different blockgroup.
571 * So add our directory's i_ino into the starting point for the hash.
573 *group = (*group + parent->i_ino) % ngroups;
576 * Use a quadratic hash to find a group with a free inode and some free
579 for (i = 1; i < ngroups; i <<= 1) {
581 if (*group >= ngroups)
583 desc = ext4_get_group_desc(sb, *group, NULL);
584 if (desc && ext4_free_inodes_count(sb, desc) &&
585 ext4_free_group_clusters(sb, desc))
590 * That failed: try linear search for a free inode, even if that group
591 * has no free blocks.
593 *group = parent_group;
594 for (i = 0; i < ngroups; i++) {
595 if (++*group >= ngroups)
597 desc = ext4_get_group_desc(sb, *group, NULL);
598 if (desc && ext4_free_inodes_count(sb, desc))
606 * There are two policies for allocating an inode. If the new inode is
607 * a directory, then a forward search is made for a block group with both
608 * free space and a low directory-to-inode ratio; if that fails, then of
609 * the groups with above-average free space, that group with the fewest
610 * directories already is chosen.
612 * For other inodes, search forward from the parent directory's block
613 * group to find a free inode.
615 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, umode_t mode,
616 const struct qstr *qstr, __u32 goal, uid_t *owner)
618 struct super_block *sb;
619 struct buffer_head *inode_bitmap_bh = NULL;
620 struct buffer_head *group_desc_bh;
621 ext4_group_t ngroups, group = 0;
622 unsigned long ino = 0;
624 struct ext4_group_desc *gdp = NULL;
625 struct ext4_inode_info *ei;
626 struct ext4_sb_info *sbi;
630 ext4_group_t flex_group;
632 /* Cannot create files in a deleted directory */
633 if (!dir || !dir->i_nlink)
634 return ERR_PTR(-EPERM);
637 ngroups = ext4_get_groups_count(sb);
638 trace_ext4_request_inode(dir, mode);
639 inode = new_inode(sb);
641 return ERR_PTR(-ENOMEM);
646 goal = sbi->s_inode_goal;
648 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
649 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
650 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
656 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
658 ret2 = find_group_other(sb, dir, &group, mode);
661 EXT4_I(dir)->i_last_alloc_group = group;
667 * Normally we will only go through one pass of this loop,
668 * unless we get unlucky and it turns out the group we selected
669 * had its last inode grabbed by someone else.
671 for (i = 0; i < ngroups; i++, ino = 0) {
674 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
678 brelse(inode_bitmap_bh);
679 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
680 if (!inode_bitmap_bh)
683 repeat_in_this_group:
684 ino = ext4_find_next_zero_bit((unsigned long *)
685 inode_bitmap_bh->b_data,
686 EXT4_INODES_PER_GROUP(sb), ino);
687 if (ino >= EXT4_INODES_PER_GROUP(sb)) {
688 if (++group == ngroups)
692 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
693 ext4_error(sb, "reserved inode found cleared - "
694 "inode=%lu", ino + 1);
697 ext4_lock_group(sb, group);
698 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
699 ext4_unlock_group(sb, group);
700 ino++; /* the inode bitmap is zero-based */
702 goto got; /* we grabbed the inode! */
703 if (ino < EXT4_INODES_PER_GROUP(sb))
704 goto repeat_in_this_group;
710 /* We may have to initialize the block bitmap if it isn't already */
711 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
712 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
713 struct buffer_head *block_bitmap_bh;
715 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
716 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
717 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
719 brelse(block_bitmap_bh);
723 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
724 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
725 brelse(block_bitmap_bh);
727 /* recheck and clear flag under lock if we still need to */
728 ext4_lock_group(sb, group);
729 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
730 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
731 ext4_free_group_clusters_set(sb, gdp,
732 ext4_free_clusters_after_init(sb, group, gdp));
733 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
736 ext4_unlock_group(sb, group);
742 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
743 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
747 BUFFER_TRACE(group_desc_bh, "get_write_access");
748 err = ext4_journal_get_write_access(handle, group_desc_bh);
752 /* Update the relevant bg descriptor fields */
753 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
755 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
757 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
758 ext4_lock_group(sb, group); /* while we modify the bg desc */
759 free = EXT4_INODES_PER_GROUP(sb) -
760 ext4_itable_unused_count(sb, gdp);
761 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
762 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
766 * Check the relative inode number against the last used
767 * relative inode number in this group. if it is greater
768 * we need to update the bg_itable_unused count
771 ext4_itable_unused_set(sb, gdp,
772 (EXT4_INODES_PER_GROUP(sb) - ino));
773 up_read(&grp->alloc_sem);
775 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
777 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
778 if (sbi->s_log_groups_per_flex) {
779 ext4_group_t f = ext4_flex_group(sbi, group);
781 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
784 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
785 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
786 ext4_unlock_group(sb, group);
789 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
790 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
794 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
795 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
799 percpu_counter_dec(&sbi->s_freeinodes_counter);
801 percpu_counter_inc(&sbi->s_dirs_counter);
802 ext4_mark_super_dirty(sb);
804 if (sbi->s_log_groups_per_flex) {
805 flex_group = ext4_flex_group(sbi, group);
806 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
809 inode->i_mode = mode;
810 inode->i_uid = owner[0];
811 inode->i_gid = owner[1];
812 } else if (test_opt(sb, GRPID)) {
813 inode->i_mode = mode;
814 inode->i_uid = current_fsuid();
815 inode->i_gid = dir->i_gid;
817 inode_init_owner(inode, dir, mode);
819 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
820 /* This is the optimal IO size (for stat), not the fs block size */
822 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
823 ext4_current_time(inode);
825 memset(ei->i_data, 0, sizeof(ei->i_data));
826 ei->i_dir_start_lookup = 0;
829 /* Don't inherit extent flag from directory, amongst others. */
831 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
834 ei->i_block_group = group;
835 ei->i_last_alloc_group = ~0;
837 ext4_set_inode_flags(inode);
838 if (IS_DIRSYNC(inode))
839 ext4_handle_sync(handle);
840 if (insert_inode_locked(inode) < 0) {
842 * Likely a bitmap corruption causing inode to be allocated
848 spin_lock(&sbi->s_next_gen_lock);
849 inode->i_generation = sbi->s_next_generation++;
850 spin_unlock(&sbi->s_next_gen_lock);
852 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
853 ext4_set_inode_state(inode, EXT4_STATE_NEW);
855 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
858 dquot_initialize(inode);
859 err = dquot_alloc_inode(inode);
863 err = ext4_init_acl(handle, inode, dir);
867 err = ext4_init_security(handle, inode, dir, qstr);
871 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
872 /* set extent flag only for directory, file and normal symlink*/
873 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
874 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
875 ext4_ext_tree_init(handle, inode);
879 if (ext4_handle_valid(handle)) {
880 ei->i_sync_tid = handle->h_transaction->t_tid;
881 ei->i_datasync_tid = handle->h_transaction->t_tid;
884 err = ext4_mark_inode_dirty(handle, inode);
886 ext4_std_error(sb, err);
890 ext4_debug("allocating inode %lu\n", inode->i_ino);
891 trace_ext4_allocate_inode(inode, dir, mode);
894 ext4_std_error(sb, err);
899 brelse(inode_bitmap_bh);
903 dquot_free_inode(inode);
907 inode->i_flags |= S_NOQUOTA;
909 unlock_new_inode(inode);
911 brelse(inode_bitmap_bh);
915 /* Verify that we are loading a valid orphan from disk */
916 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
918 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
919 ext4_group_t block_group;
921 struct buffer_head *bitmap_bh;
922 struct inode *inode = NULL;
925 /* Error cases - e2fsck has already cleaned up for us */
927 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
931 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
932 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
933 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
935 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
939 /* Having the inode bit set should be a 100% indicator that this
940 * is a valid orphan (no e2fsck run on fs). Orphans also include
941 * inodes that were being truncated, so we can't check i_nlink==0.
943 if (!ext4_test_bit(bit, bitmap_bh->b_data))
946 inode = ext4_iget(sb, ino);
951 * If the orphans has i_nlinks > 0 then it should be able to be
952 * truncated, otherwise it won't be removed from the orphan list
953 * during processing and an infinite loop will result.
955 if (inode->i_nlink && !ext4_can_truncate(inode))
958 if (NEXT_ORPHAN(inode) > max_ino)
964 err = PTR_ERR(inode);
967 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
968 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
969 bit, (unsigned long long)bitmap_bh->b_blocknr,
970 ext4_test_bit(bit, bitmap_bh->b_data));
971 printk(KERN_NOTICE "inode=%p\n", inode);
973 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
974 is_bad_inode(inode));
975 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
977 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
978 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
979 /* Avoid freeing blocks if we got a bad deleted inode */
980 if (inode->i_nlink == 0)
989 unsigned long ext4_count_free_inodes(struct super_block *sb)
991 unsigned long desc_count;
992 struct ext4_group_desc *gdp;
993 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
995 struct ext4_super_block *es;
996 unsigned long bitmap_count, x;
997 struct buffer_head *bitmap_bh = NULL;
999 es = EXT4_SB(sb)->s_es;
1003 for (i = 0; i < ngroups; i++) {
1004 gdp = ext4_get_group_desc(sb, i, NULL);
1007 desc_count += ext4_free_inodes_count(sb, gdp);
1009 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1013 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1014 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1015 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1019 printk(KERN_DEBUG "ext4_count_free_inodes: "
1020 "stored = %u, computed = %lu, %lu\n",
1021 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1025 for (i = 0; i < ngroups; i++) {
1026 gdp = ext4_get_group_desc(sb, i, NULL);
1029 desc_count += ext4_free_inodes_count(sb, gdp);
1036 /* Called at mount-time, super-block is locked */
1037 unsigned long ext4_count_dirs(struct super_block * sb)
1039 unsigned long count = 0;
1040 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1042 for (i = 0; i < ngroups; i++) {
1043 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1046 count += ext4_used_dirs_count(sb, gdp);
1052 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1053 * inode table. Must be called without any spinlock held. The only place
1054 * where it is called from on active part of filesystem is ext4lazyinit
1055 * thread, so we do not need any special locks, however we have to prevent
1056 * inode allocation from the current group, so we take alloc_sem lock, to
1057 * block ext4_new_inode() until we are finished.
1059 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1062 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1063 struct ext4_sb_info *sbi = EXT4_SB(sb);
1064 struct ext4_group_desc *gdp = NULL;
1065 struct buffer_head *group_desc_bh;
1068 int num, ret = 0, used_blks = 0;
1070 /* This should not happen, but just to be sure check this */
1071 if (sb->s_flags & MS_RDONLY) {
1076 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1081 * We do not need to lock this, because we are the only one
1082 * handling this flag.
1084 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1087 handle = ext4_journal_start_sb(sb, 1);
1088 if (IS_ERR(handle)) {
1089 ret = PTR_ERR(handle);
1093 down_write(&grp->alloc_sem);
1095 * If inode bitmap was already initialized there may be some
1096 * used inodes so we need to skip blocks with used inodes in
1099 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1100 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1101 ext4_itable_unused_count(sb, gdp)),
1102 sbi->s_inodes_per_block);
1104 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1105 ext4_error(sb, "Something is wrong with group %u: "
1106 "used itable blocks: %d; "
1107 "itable unused count: %u",
1109 ext4_itable_unused_count(sb, gdp));
1114 blk = ext4_inode_table(sb, gdp) + used_blks;
1115 num = sbi->s_itb_per_group - used_blks;
1117 BUFFER_TRACE(group_desc_bh, "get_write_access");
1118 ret = ext4_journal_get_write_access(handle,
1124 * Skip zeroout if the inode table is full. But we set the ZEROED
1125 * flag anyway, because obviously, when it is full it does not need
1128 if (unlikely(num == 0))
1131 ext4_debug("going to zero out inode table in group %d\n",
1133 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1137 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1140 ext4_lock_group(sb, group);
1141 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1142 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1143 ext4_unlock_group(sb, group);
1145 BUFFER_TRACE(group_desc_bh,
1146 "call ext4_handle_dirty_metadata");
1147 ret = ext4_handle_dirty_metadata(handle, NULL,
1151 up_write(&grp->alloc_sem);
1152 ext4_journal_stop(handle);