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);
96 * Read the inode allocation bitmap for a given block_group, reading
97 * into the specified slot in the superblock's bitmap cache.
99 * Return buffer_head of bitmap on success or NULL.
101 static struct buffer_head *
102 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
104 struct ext4_group_desc *desc;
105 struct buffer_head *bh = NULL;
106 ext4_fsblk_t bitmap_blk;
108 desc = ext4_get_group_desc(sb, block_group, NULL);
112 bitmap_blk = ext4_inode_bitmap(sb, desc);
113 bh = sb_getblk(sb, bitmap_blk);
115 ext4_error(sb, "Cannot read inode bitmap - "
116 "block_group = %u, inode_bitmap = %llu",
117 block_group, bitmap_blk);
120 if (bitmap_uptodate(bh))
124 if (bitmap_uptodate(bh)) {
129 ext4_lock_group(sb, block_group);
130 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
131 ext4_init_inode_bitmap(sb, bh, block_group, desc);
132 set_bitmap_uptodate(bh);
133 set_buffer_uptodate(bh);
134 ext4_unlock_group(sb, block_group);
138 ext4_unlock_group(sb, block_group);
140 if (buffer_uptodate(bh)) {
142 * if not uninit if bh is uptodate,
143 * bitmap is also uptodate
145 set_bitmap_uptodate(bh);
150 * submit the buffer_head for read. We can
151 * safely mark the bitmap as uptodate now.
152 * We do it here so the bitmap uptodate bit
153 * get set with buffer lock held.
155 trace_ext4_load_inode_bitmap(sb, block_group);
156 set_bitmap_uptodate(bh);
157 if (bh_submit_read(bh) < 0) {
159 ext4_error(sb, "Cannot read inode bitmap - "
160 "block_group = %u, inode_bitmap = %llu",
161 block_group, bitmap_blk);
168 * NOTE! When we get the inode, we're the only people
169 * that have access to it, and as such there are no
170 * race conditions we have to worry about. The inode
171 * is not on the hash-lists, and it cannot be reached
172 * through the filesystem because the directory entry
173 * has been deleted earlier.
175 * HOWEVER: we must make sure that we get no aliases,
176 * which means that we have to call "clear_inode()"
177 * _before_ we mark the inode not in use in the inode
178 * bitmaps. Otherwise a newly created file might use
179 * the same inode number (not actually the same pointer
180 * though), and then we'd have two inodes sharing the
181 * same inode number and space on the harddisk.
183 void ext4_free_inode(handle_t *handle, struct inode *inode)
185 struct super_block *sb = inode->i_sb;
188 struct buffer_head *bitmap_bh = NULL;
189 struct buffer_head *bh2;
190 ext4_group_t block_group;
192 struct ext4_group_desc *gdp;
193 struct ext4_super_block *es;
194 struct ext4_sb_info *sbi;
195 int fatal = 0, err, count, cleared;
197 if (atomic_read(&inode->i_count) > 1) {
198 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
199 atomic_read(&inode->i_count));
202 if (inode->i_nlink) {
203 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
208 printk(KERN_ERR "ext4_free_inode: inode on "
209 "nonexistent device\n");
215 ext4_debug("freeing inode %lu\n", ino);
216 trace_ext4_free_inode(inode);
219 * Note: we must free any quota before locking the superblock,
220 * as writing the quota to disk may need the lock as well.
222 dquot_initialize(inode);
223 ext4_xattr_delete_inode(handle, inode);
224 dquot_free_inode(inode);
227 is_directory = S_ISDIR(inode->i_mode);
229 /* Do this BEFORE marking the inode not in use or returning an error */
230 ext4_clear_inode(inode);
232 es = EXT4_SB(sb)->s_es;
233 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
234 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
237 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
238 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
239 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
243 BUFFER_TRACE(bitmap_bh, "get_write_access");
244 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
249 gdp = ext4_get_group_desc(sb, block_group, &bh2);
251 BUFFER_TRACE(bh2, "get_write_access");
252 fatal = ext4_journal_get_write_access(handle, bh2);
254 ext4_lock_group(sb, block_group);
255 cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
256 if (fatal || !cleared) {
257 ext4_unlock_group(sb, block_group);
261 count = ext4_free_inodes_count(sb, gdp) + 1;
262 ext4_free_inodes_set(sb, gdp, count);
264 count = ext4_used_dirs_count(sb, gdp) - 1;
265 ext4_used_dirs_set(sb, gdp, count);
266 percpu_counter_dec(&sbi->s_dirs_counter);
268 gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
269 ext4_unlock_group(sb, block_group);
271 percpu_counter_inc(&sbi->s_freeinodes_counter);
272 if (sbi->s_log_groups_per_flex) {
273 ext4_group_t f = ext4_flex_group(sbi, block_group);
275 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
277 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
279 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
280 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
283 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
284 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
287 ext4_mark_super_dirty(sb);
289 ext4_error(sb, "bit already cleared for inode %lu", ino);
293 ext4_std_error(sb, fatal);
297 * There are two policies for allocating an inode. If the new inode is
298 * a directory, then a forward search is made for a block group with both
299 * free space and a low directory-to-inode ratio; if that fails, then of
300 * the groups with above-average free space, that group with the fewest
301 * directories already is chosen.
303 * For other inodes, search forward from the parent directory\'s block
304 * group to find a free inode.
306 static int find_group_dir(struct super_block *sb, struct inode *parent,
307 ext4_group_t *best_group)
309 ext4_group_t ngroups = ext4_get_groups_count(sb);
310 unsigned int freei, avefreei;
311 struct ext4_group_desc *desc, *best_desc = NULL;
315 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
316 avefreei = freei / ngroups;
318 for (group = 0; group < ngroups; group++) {
319 desc = ext4_get_group_desc(sb, group, NULL);
320 if (!desc || !ext4_free_inodes_count(sb, desc))
322 if (ext4_free_inodes_count(sb, desc) < avefreei)
325 (ext4_free_group_clusters(sb, desc) >
326 ext4_free_group_clusters(sb, best_desc))) {
335 #define free_block_ratio 10
337 static int find_group_flex(struct super_block *sb, struct inode *parent,
338 ext4_group_t *best_group)
340 struct ext4_sb_info *sbi = EXT4_SB(sb);
341 struct ext4_group_desc *desc;
342 struct flex_groups *flex_group = sbi->s_flex_groups;
343 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
344 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
345 ext4_group_t ngroups = ext4_get_groups_count(sb);
346 int flex_size = ext4_flex_bg_size(sbi);
347 ext4_group_t best_flex = parent_fbg_group;
348 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
349 int flexbg_free_clusters;
350 int flex_freeb_ratio;
351 ext4_group_t n_fbg_groups;
354 n_fbg_groups = (ngroups + flex_size - 1) >>
355 sbi->s_log_groups_per_flex;
357 find_close_to_parent:
358 flexbg_free_clusters = atomic_read(&flex_group[best_flex].free_clusters);
359 flex_freeb_ratio = EXT4_C2B(sbi, flexbg_free_clusters) * 100 /
361 if (atomic_read(&flex_group[best_flex].free_inodes) &&
362 flex_freeb_ratio > free_block_ratio)
365 if (best_flex && best_flex == parent_fbg_group) {
367 goto find_close_to_parent;
370 for (i = 0; i < n_fbg_groups; i++) {
371 if (i == parent_fbg_group || i == parent_fbg_group - 1)
374 flexbg_free_clusters = atomic_read(&flex_group[i].free_clusters);
375 flex_freeb_ratio = EXT4_C2B(sbi, flexbg_free_clusters) * 100 /
378 if (flex_freeb_ratio > free_block_ratio &&
379 (atomic_read(&flex_group[i].free_inodes))) {
384 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
385 ((atomic_read(&flex_group[i].free_clusters) >
386 atomic_read(&flex_group[best_flex].free_clusters)) &&
387 atomic_read(&flex_group[i].free_inodes)))
391 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
392 !atomic_read(&flex_group[best_flex].free_clusters))
396 for (i = best_flex * flex_size; i < ngroups &&
397 i < (best_flex + 1) * flex_size; i++) {
398 desc = ext4_get_group_desc(sb, i, NULL);
399 if (ext4_free_inodes_count(sb, desc)) {
417 * Helper function for Orlov's allocator; returns critical information
418 * for a particular block group or flex_bg. If flex_size is 1, then g
419 * is a block group number; otherwise it is flex_bg number.
421 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
422 int flex_size, struct orlov_stats *stats)
424 struct ext4_group_desc *desc;
425 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
428 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
429 stats->free_clusters = atomic_read(&flex_group[g].free_clusters);
430 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
434 desc = ext4_get_group_desc(sb, g, NULL);
436 stats->free_inodes = ext4_free_inodes_count(sb, desc);
437 stats->free_clusters = ext4_free_group_clusters(sb, desc);
438 stats->used_dirs = ext4_used_dirs_count(sb, desc);
440 stats->free_inodes = 0;
441 stats->free_clusters = 0;
442 stats->used_dirs = 0;
447 * Orlov's allocator for directories.
449 * We always try to spread first-level directories.
451 * If there are blockgroups with both free inodes and free blocks counts
452 * not worse than average we return one with smallest directory count.
453 * Otherwise we simply return a random group.
455 * For the rest rules look so:
457 * It's OK to put directory into a group unless
458 * it has too many directories already (max_dirs) or
459 * it has too few free inodes left (min_inodes) or
460 * it has too few free blocks left (min_blocks) or
461 * Parent's group is preferred, if it doesn't satisfy these
462 * conditions we search cyclically through the rest. If none
463 * of the groups look good we just look for a group with more
464 * free inodes than average (starting at parent's group).
467 static int find_group_orlov(struct super_block *sb, struct inode *parent,
468 ext4_group_t *group, int mode,
469 const struct qstr *qstr)
471 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
472 struct ext4_sb_info *sbi = EXT4_SB(sb);
473 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
474 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
475 unsigned int freei, avefreei;
476 ext4_fsblk_t freeb, avefreec;
478 int max_dirs, min_inodes;
479 ext4_grpblk_t min_clusters;
480 ext4_group_t i, grp, g, ngroups;
481 struct ext4_group_desc *desc;
482 struct orlov_stats stats;
483 int flex_size = ext4_flex_bg_size(sbi);
484 struct dx_hash_info hinfo;
486 ngroups = real_ngroups;
488 ngroups = (real_ngroups + flex_size - 1) >>
489 sbi->s_log_groups_per_flex;
490 parent_group >>= sbi->s_log_groups_per_flex;
493 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
494 avefreei = freei / ngroups;
495 freeb = EXT4_C2B(sbi,
496 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
498 do_div(avefreec, ngroups);
499 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
502 ((parent == sb->s_root->d_inode) ||
503 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
504 int best_ndir = inodes_per_group;
508 hinfo.hash_version = DX_HASH_HALF_MD4;
509 hinfo.seed = sbi->s_hash_seed;
510 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
513 get_random_bytes(&grp, sizeof(grp));
514 parent_group = (unsigned)grp % ngroups;
515 for (i = 0; i < ngroups; i++) {
516 g = (parent_group + i) % ngroups;
517 get_orlov_stats(sb, g, flex_size, &stats);
518 if (!stats.free_inodes)
520 if (stats.used_dirs >= best_ndir)
522 if (stats.free_inodes < avefreei)
524 if (stats.free_clusters < avefreec)
528 best_ndir = stats.used_dirs;
533 if (flex_size == 1) {
539 * We pack inodes at the beginning of the flexgroup's
540 * inode tables. Block allocation decisions will do
541 * something similar, although regular files will
542 * start at 2nd block group of the flexgroup. See
543 * ext4_ext_find_goal() and ext4_find_near().
546 for (i = 0; i < flex_size; i++) {
547 if (grp+i >= real_ngroups)
549 desc = ext4_get_group_desc(sb, grp+i, NULL);
550 if (desc && ext4_free_inodes_count(sb, desc)) {
558 max_dirs = ndirs / ngroups + inodes_per_group / 16;
559 min_inodes = avefreei - inodes_per_group*flex_size / 4;
562 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
565 * Start looking in the flex group where we last allocated an
566 * inode for this parent directory
568 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
569 parent_group = EXT4_I(parent)->i_last_alloc_group;
571 parent_group >>= sbi->s_log_groups_per_flex;
574 for (i = 0; i < ngroups; i++) {
575 grp = (parent_group + i) % ngroups;
576 get_orlov_stats(sb, grp, flex_size, &stats);
577 if (stats.used_dirs >= max_dirs)
579 if (stats.free_inodes < min_inodes)
581 if (stats.free_clusters < min_clusters)
587 ngroups = real_ngroups;
588 avefreei = freei / ngroups;
590 parent_group = EXT4_I(parent)->i_block_group;
591 for (i = 0; i < ngroups; i++) {
592 grp = (parent_group + i) % ngroups;
593 desc = ext4_get_group_desc(sb, grp, NULL);
594 if (desc && ext4_free_inodes_count(sb, desc) &&
595 ext4_free_inodes_count(sb, desc) >= avefreei) {
603 * The free-inodes counter is approximate, and for really small
604 * filesystems the above test can fail to find any blockgroups
613 static int find_group_other(struct super_block *sb, struct inode *parent,
614 ext4_group_t *group, int mode)
616 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
617 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
618 struct ext4_group_desc *desc;
619 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
622 * Try to place the inode is the same flex group as its
623 * parent. If we can't find space, use the Orlov algorithm to
624 * find another flex group, and store that information in the
625 * parent directory's inode information so that use that flex
626 * group for future allocations.
632 parent_group &= ~(flex_size-1);
633 last = parent_group + flex_size;
636 for (i = parent_group; i < last; i++) {
637 desc = ext4_get_group_desc(sb, i, NULL);
638 if (desc && ext4_free_inodes_count(sb, desc)) {
643 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
645 parent_group = EXT4_I(parent)->i_last_alloc_group;
649 * If this didn't work, use the Orlov search algorithm
650 * to find a new flex group; we pass in the mode to
651 * avoid the topdir algorithms.
653 *group = parent_group + flex_size;
654 if (*group > ngroups)
656 return find_group_orlov(sb, parent, group, mode, NULL);
660 * Try to place the inode in its parent directory
662 *group = parent_group;
663 desc = ext4_get_group_desc(sb, *group, NULL);
664 if (desc && ext4_free_inodes_count(sb, desc) &&
665 ext4_free_group_clusters(sb, desc))
669 * We're going to place this inode in a different blockgroup from its
670 * parent. We want to cause files in a common directory to all land in
671 * the same blockgroup. But we want files which are in a different
672 * directory which shares a blockgroup with our parent to land in a
673 * different blockgroup.
675 * So add our directory's i_ino into the starting point for the hash.
677 *group = (*group + parent->i_ino) % ngroups;
680 * Use a quadratic hash to find a group with a free inode and some free
683 for (i = 1; i < ngroups; i <<= 1) {
685 if (*group >= ngroups)
687 desc = ext4_get_group_desc(sb, *group, NULL);
688 if (desc && ext4_free_inodes_count(sb, desc) &&
689 ext4_free_group_clusters(sb, desc))
694 * That failed: try linear search for a free inode, even if that group
695 * has no free blocks.
697 *group = parent_group;
698 for (i = 0; i < ngroups; i++) {
699 if (++*group >= ngroups)
701 desc = ext4_get_group_desc(sb, *group, NULL);
702 if (desc && ext4_free_inodes_count(sb, desc))
710 * claim the inode from the inode bitmap. If the group
711 * is uninit we need to take the groups's ext4_group_lock
712 * and clear the uninit flag. The inode bitmap update
713 * and group desc uninit flag clear should be done
714 * after holding ext4_group_lock so that ext4_read_inode_bitmap
715 * doesn't race with the ext4_claim_inode
717 static int ext4_claim_inode(struct super_block *sb,
718 struct buffer_head *inode_bitmap_bh,
719 unsigned long ino, ext4_group_t group, int mode)
721 int free = 0, retval = 0, count;
722 struct ext4_sb_info *sbi = EXT4_SB(sb);
723 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
724 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
727 * We have to be sure that new inode allocation does not race with
728 * inode table initialization, because otherwise we may end up
729 * allocating and writing new inode right before sb_issue_zeroout
730 * takes place and overwriting our new inode with zeroes. So we
731 * take alloc_sem to prevent it.
733 down_read(&grp->alloc_sem);
734 ext4_lock_group(sb, group);
735 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
736 /* not a free inode */
741 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
742 ino > EXT4_INODES_PER_GROUP(sb)) {
743 ext4_unlock_group(sb, group);
744 up_read(&grp->alloc_sem);
745 ext4_error(sb, "reserved inode or inode > inodes count - "
746 "block_group = %u, inode=%lu", group,
747 ino + group * EXT4_INODES_PER_GROUP(sb));
750 /* If we didn't allocate from within the initialized part of the inode
751 * table then we need to initialize up to this inode. */
752 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
754 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
755 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
756 /* When marking the block group with
757 * ~EXT4_BG_INODE_UNINIT we don't want to depend
758 * on the value of bg_itable_unused even though
759 * mke2fs could have initialized the same for us.
760 * Instead we calculated the value below
765 free = EXT4_INODES_PER_GROUP(sb) -
766 ext4_itable_unused_count(sb, gdp);
770 * Check the relative inode number against the last used
771 * relative inode number in this group. if it is greater
772 * we need to update the bg_itable_unused count
776 ext4_itable_unused_set(sb, gdp,
777 (EXT4_INODES_PER_GROUP(sb) - ino));
779 count = ext4_free_inodes_count(sb, gdp) - 1;
780 ext4_free_inodes_set(sb, gdp, count);
782 count = ext4_used_dirs_count(sb, gdp) + 1;
783 ext4_used_dirs_set(sb, gdp, count);
784 if (sbi->s_log_groups_per_flex) {
785 ext4_group_t f = ext4_flex_group(sbi, group);
787 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
790 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
792 ext4_unlock_group(sb, group);
793 up_read(&grp->alloc_sem);
798 * There are two policies for allocating an inode. If the new inode is
799 * a directory, then a forward search is made for a block group with both
800 * free space and a low directory-to-inode ratio; if that fails, then of
801 * the groups with above-average free space, that group with the fewest
802 * directories already is chosen.
804 * For other inodes, search forward from the parent directory's block
805 * group to find a free inode.
807 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
808 const struct qstr *qstr, __u32 goal)
810 struct super_block *sb;
811 struct buffer_head *inode_bitmap_bh = NULL;
812 struct buffer_head *group_desc_bh;
813 ext4_group_t ngroups, group = 0;
814 unsigned long ino = 0;
816 struct ext4_group_desc *gdp = NULL;
817 struct ext4_inode_info *ei;
818 struct ext4_sb_info *sbi;
823 ext4_group_t flex_group;
825 /* Cannot create files in a deleted directory */
826 if (!dir || !dir->i_nlink)
827 return ERR_PTR(-EPERM);
830 ngroups = ext4_get_groups_count(sb);
831 trace_ext4_request_inode(dir, mode);
832 inode = new_inode(sb);
834 return ERR_PTR(-ENOMEM);
839 goal = sbi->s_inode_goal;
841 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
842 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
843 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
848 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
849 ret2 = find_group_flex(sb, dir, &group);
851 ret2 = find_group_other(sb, dir, &group, mode);
852 if (ret2 == 0 && once) {
854 printk(KERN_NOTICE "ext4: find_group_flex "
855 "failed, fallback succeeded dir %lu\n",
863 if (test_opt(sb, OLDALLOC))
864 ret2 = find_group_dir(sb, dir, &group);
866 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
868 ret2 = find_group_other(sb, dir, &group, mode);
871 EXT4_I(dir)->i_last_alloc_group = group;
876 for (i = 0; i < ngroups; i++, ino = 0) {
879 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
883 brelse(inode_bitmap_bh);
884 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
885 if (!inode_bitmap_bh)
888 repeat_in_this_group:
889 ino = ext4_find_next_zero_bit((unsigned long *)
890 inode_bitmap_bh->b_data,
891 EXT4_INODES_PER_GROUP(sb), ino);
893 if (ino < EXT4_INODES_PER_GROUP(sb)) {
895 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
896 err = ext4_journal_get_write_access(handle,
901 BUFFER_TRACE(group_desc_bh, "get_write_access");
902 err = ext4_journal_get_write_access(handle,
906 if (!ext4_claim_inode(sb, inode_bitmap_bh,
909 BUFFER_TRACE(inode_bitmap_bh,
910 "call ext4_handle_dirty_metadata");
911 err = ext4_handle_dirty_metadata(handle,
916 /* zero bit is inode number 1*/
921 ext4_handle_release_buffer(handle, inode_bitmap_bh);
922 ext4_handle_release_buffer(handle, group_desc_bh);
924 if (++ino < EXT4_INODES_PER_GROUP(sb))
925 goto repeat_in_this_group;
929 * This case is possible in concurrent environment. It is very
930 * rare. We cannot repeat the find_group_xxx() call because
931 * that will simply return the same blockgroup, because the
932 * group descriptor metadata has not yet been updated.
933 * So we just go onto the next blockgroup.
935 if (++group == ngroups)
942 /* We may have to initialize the block bitmap if it isn't already */
943 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
944 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
945 struct buffer_head *block_bitmap_bh;
947 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
948 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
949 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
951 brelse(block_bitmap_bh);
955 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
956 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
957 brelse(block_bitmap_bh);
959 /* recheck and clear flag under lock if we still need to */
960 ext4_lock_group(sb, group);
961 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
962 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
963 ext4_free_group_clusters_set(sb, gdp,
964 ext4_free_clusters_after_init(sb, group, gdp));
965 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
968 ext4_unlock_group(sb, group);
973 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
974 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
978 percpu_counter_dec(&sbi->s_freeinodes_counter);
980 percpu_counter_inc(&sbi->s_dirs_counter);
981 ext4_mark_super_dirty(sb);
983 if (sbi->s_log_groups_per_flex) {
984 flex_group = ext4_flex_group(sbi, group);
985 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
988 if (test_opt(sb, GRPID)) {
989 inode->i_mode = mode;
990 inode->i_uid = current_fsuid();
991 inode->i_gid = dir->i_gid;
993 inode_init_owner(inode, dir, mode);
995 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
996 /* This is the optimal IO size (for stat), not the fs block size */
998 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
999 ext4_current_time(inode);
1001 memset(ei->i_data, 0, sizeof(ei->i_data));
1002 ei->i_dir_start_lookup = 0;
1006 * Don't inherit extent flag from directory, amongst others. We set
1007 * extent flag on newly created directory and file only if -o extent
1008 * mount option is specified
1011 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1014 ei->i_block_group = group;
1015 ei->i_last_alloc_group = ~0;
1017 ext4_set_inode_flags(inode);
1018 if (IS_DIRSYNC(inode))
1019 ext4_handle_sync(handle);
1020 if (insert_inode_locked(inode) < 0) {
1024 spin_lock(&sbi->s_next_gen_lock);
1025 inode->i_generation = sbi->s_next_generation++;
1026 spin_unlock(&sbi->s_next_gen_lock);
1028 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1029 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1031 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1034 dquot_initialize(inode);
1035 err = dquot_alloc_inode(inode);
1039 err = ext4_init_acl(handle, inode, dir);
1041 goto fail_free_drop;
1043 err = ext4_init_security(handle, inode, dir, qstr);
1045 goto fail_free_drop;
1047 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1048 /* set extent flag only for directory, file and normal symlink*/
1049 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1050 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1051 ext4_ext_tree_init(handle, inode);
1055 if (ext4_handle_valid(handle)) {
1056 ei->i_sync_tid = handle->h_transaction->t_tid;
1057 ei->i_datasync_tid = handle->h_transaction->t_tid;
1060 err = ext4_mark_inode_dirty(handle, inode);
1062 ext4_std_error(sb, err);
1063 goto fail_free_drop;
1066 ext4_debug("allocating inode %lu\n", inode->i_ino);
1067 trace_ext4_allocate_inode(inode, dir, mode);
1070 ext4_std_error(sb, err);
1075 brelse(inode_bitmap_bh);
1079 dquot_free_inode(inode);
1083 inode->i_flags |= S_NOQUOTA;
1085 unlock_new_inode(inode);
1087 brelse(inode_bitmap_bh);
1088 return ERR_PTR(err);
1091 /* Verify that we are loading a valid orphan from disk */
1092 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1094 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1095 ext4_group_t block_group;
1097 struct buffer_head *bitmap_bh;
1098 struct inode *inode = NULL;
1101 /* Error cases - e2fsck has already cleaned up for us */
1102 if (ino > max_ino) {
1103 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
1107 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1108 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1109 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1111 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
1115 /* Having the inode bit set should be a 100% indicator that this
1116 * is a valid orphan (no e2fsck run on fs). Orphans also include
1117 * inodes that were being truncated, so we can't check i_nlink==0.
1119 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1122 inode = ext4_iget(sb, ino);
1127 * If the orphans has i_nlinks > 0 then it should be able to be
1128 * truncated, otherwise it won't be removed from the orphan list
1129 * during processing and an infinite loop will result.
1131 if (inode->i_nlink && !ext4_can_truncate(inode))
1134 if (NEXT_ORPHAN(inode) > max_ino)
1140 err = PTR_ERR(inode);
1143 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
1144 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1145 bit, (unsigned long long)bitmap_bh->b_blocknr,
1146 ext4_test_bit(bit, bitmap_bh->b_data));
1147 printk(KERN_NOTICE "inode=%p\n", inode);
1149 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1150 is_bad_inode(inode));
1151 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1152 NEXT_ORPHAN(inode));
1153 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1154 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1155 /* Avoid freeing blocks if we got a bad deleted inode */
1156 if (inode->i_nlink == 0)
1157 inode->i_blocks = 0;
1162 return ERR_PTR(err);
1165 unsigned long ext4_count_free_inodes(struct super_block *sb)
1167 unsigned long desc_count;
1168 struct ext4_group_desc *gdp;
1169 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1171 struct ext4_super_block *es;
1172 unsigned long bitmap_count, x;
1173 struct buffer_head *bitmap_bh = NULL;
1175 es = EXT4_SB(sb)->s_es;
1179 for (i = 0; i < ngroups; i++) {
1180 gdp = ext4_get_group_desc(sb, i, NULL);
1183 desc_count += ext4_free_inodes_count(sb, gdp);
1185 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1189 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1190 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1191 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1195 printk(KERN_DEBUG "ext4_count_free_inodes: "
1196 "stored = %u, computed = %lu, %lu\n",
1197 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1201 for (i = 0; i < ngroups; i++) {
1202 gdp = ext4_get_group_desc(sb, i, NULL);
1205 desc_count += ext4_free_inodes_count(sb, gdp);
1212 /* Called at mount-time, super-block is locked */
1213 unsigned long ext4_count_dirs(struct super_block * sb)
1215 unsigned long count = 0;
1216 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1218 for (i = 0; i < ngroups; i++) {
1219 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1222 count += ext4_used_dirs_count(sb, gdp);
1228 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1229 * inode table. Must be called without any spinlock held. The only place
1230 * where it is called from on active part of filesystem is ext4lazyinit
1231 * thread, so we do not need any special locks, however we have to prevent
1232 * inode allocation from the current group, so we take alloc_sem lock, to
1233 * block ext4_claim_inode until we are finished.
1235 extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1238 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1239 struct ext4_sb_info *sbi = EXT4_SB(sb);
1240 struct ext4_group_desc *gdp = NULL;
1241 struct buffer_head *group_desc_bh;
1244 int num, ret = 0, used_blks = 0;
1246 /* This should not happen, but just to be sure check this */
1247 if (sb->s_flags & MS_RDONLY) {
1252 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1257 * We do not need to lock this, because we are the only one
1258 * handling this flag.
1260 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1263 handle = ext4_journal_start_sb(sb, 1);
1264 if (IS_ERR(handle)) {
1265 ret = PTR_ERR(handle);
1269 down_write(&grp->alloc_sem);
1271 * If inode bitmap was already initialized there may be some
1272 * used inodes so we need to skip blocks with used inodes in
1275 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1276 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1277 ext4_itable_unused_count(sb, gdp)),
1278 sbi->s_inodes_per_block);
1280 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1281 ext4_error(sb, "Something is wrong with group %u\n"
1282 "Used itable blocks: %d"
1283 "itable unused count: %u\n",
1285 ext4_itable_unused_count(sb, gdp));
1290 blk = ext4_inode_table(sb, gdp) + used_blks;
1291 num = sbi->s_itb_per_group - used_blks;
1293 BUFFER_TRACE(group_desc_bh, "get_write_access");
1294 ret = ext4_journal_get_write_access(handle,
1300 * Skip zeroout if the inode table is full. But we set the ZEROED
1301 * flag anyway, because obviously, when it is full it does not need
1304 if (unlikely(num == 0))
1307 ext4_debug("going to zero out inode table in group %d\n",
1309 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1313 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1316 ext4_lock_group(sb, group);
1317 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1318 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1319 ext4_unlock_group(sb, group);
1321 BUFFER_TRACE(group_desc_bh,
1322 "call ext4_handle_dirty_metadata");
1323 ret = ext4_handle_dirty_metadata(handle, NULL,
1327 up_write(&grp->alloc_sem);
1328 ext4_journal_stop(handle);