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>
27 #include "ext4_jbd2.h"
32 * ialloc.c contains the inodes allocation and deallocation routines
36 * The free inodes are managed by bitmaps. A file system contains several
37 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
38 * block for inodes, N blocks for the inode table and data blocks.
40 * The file system contains group descriptors which are located after the
41 * super block. Each descriptor contains the number of the bitmap block and
42 * the free blocks count in the block.
46 * To avoid calling the atomic setbit hundreds or thousands of times, we only
47 * need to use it within a single byte (to ensure we get endianness right).
48 * We can use memset for the rest of the bitmap as there are no other users.
50 void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
54 if (start_bit >= end_bit)
57 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
58 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
59 ext4_set_bit(i, bitmap);
61 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
64 /* Initializes an uninitialized inode bitmap */
65 unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
66 ext4_group_t block_group,
67 struct ext4_group_desc *gdp)
69 struct ext4_sb_info *sbi = EXT4_SB(sb);
71 J_ASSERT_BH(bh, buffer_locked(bh));
73 /* If checksum is bad mark all blocks and inodes use to prevent
74 * allocation, essentially implementing a per-group read-only flag. */
75 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
76 ext4_error(sb, __func__, "Checksum bad for group %u",
78 ext4_free_blks_set(sb, gdp, 0);
79 ext4_free_inodes_set(sb, gdp, 0);
80 ext4_itable_unused_set(sb, gdp, 0);
81 memset(bh->b_data, 0xff, sb->s_blocksize);
85 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
86 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
89 return EXT4_INODES_PER_GROUP(sb);
93 * Read the inode allocation bitmap for a given block_group, reading
94 * into the specified slot in the superblock's bitmap cache.
96 * Return buffer_head of bitmap on success or NULL.
98 static struct buffer_head *
99 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
101 struct ext4_group_desc *desc;
102 struct buffer_head *bh = NULL;
103 ext4_fsblk_t bitmap_blk;
105 desc = ext4_get_group_desc(sb, block_group, NULL);
108 bitmap_blk = ext4_inode_bitmap(sb, desc);
109 bh = sb_getblk(sb, bitmap_blk);
111 ext4_error(sb, __func__,
112 "Cannot read inode bitmap - "
113 "block_group = %u, inode_bitmap = %llu",
114 block_group, bitmap_blk);
117 if (bitmap_uptodate(bh))
121 if (bitmap_uptodate(bh)) {
125 ext4_lock_group(sb, block_group);
126 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
127 ext4_init_inode_bitmap(sb, bh, block_group, desc);
128 set_bitmap_uptodate(bh);
129 set_buffer_uptodate(bh);
130 ext4_unlock_group(sb, block_group);
134 ext4_unlock_group(sb, block_group);
135 if (buffer_uptodate(bh)) {
137 * if not uninit if bh is uptodate,
138 * bitmap is also uptodate
140 set_bitmap_uptodate(bh);
145 * submit the buffer_head for read. We can
146 * safely mark the bitmap as uptodate now.
147 * We do it here so the bitmap uptodate bit
148 * get set with buffer lock held.
150 set_bitmap_uptodate(bh);
151 if (bh_submit_read(bh) < 0) {
153 ext4_error(sb, __func__,
154 "Cannot read inode bitmap - "
155 "block_group = %u, inode_bitmap = %llu",
156 block_group, bitmap_blk);
163 * NOTE! When we get the inode, we're the only people
164 * that have access to it, and as such there are no
165 * race conditions we have to worry about. The inode
166 * is not on the hash-lists, and it cannot be reached
167 * through the filesystem because the directory entry
168 * has been deleted earlier.
170 * HOWEVER: we must make sure that we get no aliases,
171 * which means that we have to call "clear_inode()"
172 * _before_ we mark the inode not in use in the inode
173 * bitmaps. Otherwise a newly created file might use
174 * the same inode number (not actually the same pointer
175 * though), and then we'd have two inodes sharing the
176 * same inode number and space on the harddisk.
178 void ext4_free_inode(handle_t *handle, struct inode *inode)
180 struct super_block *sb = inode->i_sb;
183 struct buffer_head *bitmap_bh = NULL;
184 struct buffer_head *bh2;
185 ext4_group_t block_group;
187 struct ext4_group_desc *gdp;
188 struct ext4_super_block *es;
189 struct ext4_sb_info *sbi;
190 int fatal = 0, err, count, cleared;
192 if (atomic_read(&inode->i_count) > 1) {
193 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
194 atomic_read(&inode->i_count));
197 if (inode->i_nlink) {
198 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
203 printk(KERN_ERR "ext4_free_inode: inode on "
204 "nonexistent device\n");
210 ext4_debug("freeing inode %lu\n", ino);
211 trace_mark(ext4_free_inode,
212 "dev %s ino %lu mode %d uid %lu gid %lu bocks %llu",
213 sb->s_id, inode->i_ino, inode->i_mode,
214 (unsigned long) inode->i_uid, (unsigned long) inode->i_gid,
215 (unsigned long long) inode->i_blocks);
218 * Note: we must free any quota before locking the superblock,
219 * as writing the quota to disk may need the lock as well.
222 ext4_xattr_delete_inode(handle, inode);
223 vfs_dq_free_inode(inode);
226 is_directory = S_ISDIR(inode->i_mode);
228 /* Do this BEFORE marking the inode not in use or returning an error */
231 es = EXT4_SB(sb)->s_es;
232 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
233 ext4_error(sb, "ext4_free_inode",
234 "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);
248 /* Ok, now we can actually update the inode bitmaps.. */
249 cleared = ext4_clear_bit_atomic(ext4_group_lock_ptr(sb, block_group),
250 bit, bitmap_bh->b_data);
252 ext4_error(sb, "ext4_free_inode",
253 "bit already cleared for inode %lu", ino);
255 gdp = ext4_get_group_desc(sb, block_group, &bh2);
257 BUFFER_TRACE(bh2, "get_write_access");
258 fatal = ext4_journal_get_write_access(handle, bh2);
259 if (fatal) goto error_return;
262 ext4_lock_group(sb, block_group);
263 count = ext4_free_inodes_count(sb, gdp) + 1;
264 ext4_free_inodes_set(sb, gdp, count);
266 count = ext4_used_dirs_count(sb, gdp) - 1;
267 ext4_used_dirs_set(sb, gdp, count);
268 if (sbi->s_log_groups_per_flex) {
271 f = ext4_flex_group(sbi, block_group);
272 atomic_dec(&sbi->s_flex_groups[f].free_inodes);
276 gdp->bg_checksum = ext4_group_desc_csum(sbi,
278 ext4_unlock_group(sb, block_group);
279 percpu_counter_inc(&sbi->s_freeinodes_counter);
281 percpu_counter_dec(&sbi->s_dirs_counter);
283 if (sbi->s_log_groups_per_flex) {
286 f = ext4_flex_group(sbi, block_group);
287 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
290 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
291 err = ext4_handle_dirty_metadata(handle, NULL, bh2);
292 if (!fatal) fatal = err;
294 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
295 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
301 ext4_std_error(sb, fatal);
305 * There are two policies for allocating an inode. If the new inode is
306 * a directory, then a forward search is made for a block group with both
307 * free space and a low directory-to-inode ratio; if that fails, then of
308 * the groups with above-average free space, that group with the fewest
309 * directories already is chosen.
311 * For other inodes, search forward from the parent directory\'s block
312 * group to find a free inode.
314 static int find_group_dir(struct super_block *sb, struct inode *parent,
315 ext4_group_t *best_group)
317 ext4_group_t ngroups = ext4_get_groups_count(sb);
318 unsigned int freei, avefreei;
319 struct ext4_group_desc *desc, *best_desc = NULL;
323 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
324 avefreei = freei / ngroups;
326 for (group = 0; group < ngroups; group++) {
327 desc = ext4_get_group_desc(sb, group, NULL);
328 if (!desc || !ext4_free_inodes_count(sb, desc))
330 if (ext4_free_inodes_count(sb, desc) < avefreei)
333 (ext4_free_blks_count(sb, desc) >
334 ext4_free_blks_count(sb, best_desc))) {
343 #define free_block_ratio 10
345 static int find_group_flex(struct super_block *sb, struct inode *parent,
346 ext4_group_t *best_group)
348 struct ext4_sb_info *sbi = EXT4_SB(sb);
349 struct ext4_group_desc *desc;
350 struct buffer_head *bh;
351 struct flex_groups *flex_group = sbi->s_flex_groups;
352 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
353 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
354 ext4_group_t ngroups = ext4_get_groups_count(sb);
355 int flex_size = ext4_flex_bg_size(sbi);
356 ext4_group_t best_flex = parent_fbg_group;
357 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
358 int flexbg_free_blocks;
359 int flex_freeb_ratio;
360 ext4_group_t n_fbg_groups;
363 n_fbg_groups = (ngroups + flex_size - 1) >>
364 sbi->s_log_groups_per_flex;
366 find_close_to_parent:
367 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
368 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
369 if (atomic_read(&flex_group[best_flex].free_inodes) &&
370 flex_freeb_ratio > free_block_ratio)
373 if (best_flex && best_flex == parent_fbg_group) {
375 goto find_close_to_parent;
378 for (i = 0; i < n_fbg_groups; i++) {
379 if (i == parent_fbg_group || i == parent_fbg_group - 1)
382 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
383 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
385 if (flex_freeb_ratio > free_block_ratio &&
386 (atomic_read(&flex_group[i].free_inodes))) {
391 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
392 ((atomic_read(&flex_group[i].free_blocks) >
393 atomic_read(&flex_group[best_flex].free_blocks)) &&
394 atomic_read(&flex_group[i].free_inodes)))
398 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
399 !atomic_read(&flex_group[best_flex].free_blocks))
403 for (i = best_flex * flex_size; i < ngroups &&
404 i < (best_flex + 1) * flex_size; i++) {
405 desc = ext4_get_group_desc(sb, i, &bh);
406 if (ext4_free_inodes_count(sb, desc)) {
424 * Helper function for Orlov's allocator; returns critical information
425 * for a particular block group or flex_bg. If flex_size is 1, then g
426 * is a block group number; otherwise it is flex_bg number.
428 void get_orlov_stats(struct super_block *sb, ext4_group_t g,
429 int flex_size, struct orlov_stats *stats)
431 struct ext4_group_desc *desc;
432 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
435 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
436 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
437 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
441 desc = ext4_get_group_desc(sb, g, NULL);
443 stats->free_inodes = ext4_free_inodes_count(sb, desc);
444 stats->free_blocks = ext4_free_blks_count(sb, desc);
445 stats->used_dirs = ext4_used_dirs_count(sb, desc);
447 stats->free_inodes = 0;
448 stats->free_blocks = 0;
449 stats->used_dirs = 0;
454 * Orlov's allocator for directories.
456 * We always try to spread first-level directories.
458 * If there are blockgroups with both free inodes and free blocks counts
459 * not worse than average we return one with smallest directory count.
460 * Otherwise we simply return a random group.
462 * For the rest rules look so:
464 * It's OK to put directory into a group unless
465 * it has too many directories already (max_dirs) or
466 * it has too few free inodes left (min_inodes) or
467 * it has too few free blocks left (min_blocks) or
468 * Parent's group is preferred, if it doesn't satisfy these
469 * conditions we search cyclically through the rest. If none
470 * of the groups look good we just look for a group with more
471 * free inodes than average (starting at parent's group).
474 static int find_group_orlov(struct super_block *sb, struct inode *parent,
475 ext4_group_t *group, int mode)
477 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
478 struct ext4_sb_info *sbi = EXT4_SB(sb);
479 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
480 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
481 unsigned int freei, avefreei;
482 ext4_fsblk_t freeb, avefreeb;
484 int max_dirs, min_inodes;
485 ext4_grpblk_t min_blocks;
486 ext4_group_t i, grp, g, ngroups;
487 struct ext4_group_desc *desc;
488 struct orlov_stats stats;
489 int flex_size = ext4_flex_bg_size(sbi);
491 ngroups = real_ngroups;
493 ngroups = (real_ngroups + flex_size - 1) >>
494 sbi->s_log_groups_per_flex;
495 parent_group >>= sbi->s_log_groups_per_flex;
498 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
499 avefreei = freei / ngroups;
500 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
502 do_div(avefreeb, ngroups);
503 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
506 ((parent == sb->s_root->d_inode) ||
507 (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) {
508 int best_ndir = inodes_per_group;
511 get_random_bytes(&grp, sizeof(grp));
512 parent_group = (unsigned)grp % ngroups;
513 for (i = 0; i < ngroups; i++) {
514 g = (parent_group + i) % ngroups;
515 get_orlov_stats(sb, g, flex_size, &stats);
516 if (!stats.free_inodes)
518 if (stats.used_dirs >= best_ndir)
520 if (stats.free_inodes < avefreei)
522 if (stats.free_blocks < avefreeb)
526 best_ndir = stats.used_dirs;
531 if (flex_size == 1) {
537 * We pack inodes at the beginning of the flexgroup's
538 * inode tables. Block allocation decisions will do
539 * something similar, although regular files will
540 * start at 2nd block group of the flexgroup. See
541 * ext4_ext_find_goal() and ext4_find_near().
544 for (i = 0; i < flex_size; i++) {
545 if (grp+i >= real_ngroups)
547 desc = ext4_get_group_desc(sb, grp+i, NULL);
548 if (desc && ext4_free_inodes_count(sb, desc)) {
556 max_dirs = ndirs / ngroups + inodes_per_group / 16;
557 min_inodes = avefreei - inodes_per_group*flex_size / 4;
560 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
563 * Start looking in the flex group where we last allocated an
564 * inode for this parent directory
566 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
567 parent_group = EXT4_I(parent)->i_last_alloc_group;
569 parent_group >>= sbi->s_log_groups_per_flex;
572 for (i = 0; i < ngroups; i++) {
573 grp = (parent_group + i) % ngroups;
574 get_orlov_stats(sb, grp, flex_size, &stats);
575 if (stats.used_dirs >= max_dirs)
577 if (stats.free_inodes < min_inodes)
579 if (stats.free_blocks < min_blocks)
585 ngroups = real_ngroups;
586 avefreei = freei / ngroups;
588 parent_group = EXT4_I(parent)->i_block_group;
589 for (i = 0; i < ngroups; i++) {
590 grp = (parent_group + i) % ngroups;
591 desc = ext4_get_group_desc(sb, grp, NULL);
592 if (desc && ext4_free_inodes_count(sb, desc) &&
593 ext4_free_inodes_count(sb, desc) >= avefreei) {
601 * The free-inodes counter is approximate, and for really small
602 * filesystems the above test can fail to find any blockgroups
611 static int find_group_other(struct super_block *sb, struct inode *parent,
612 ext4_group_t *group, int mode)
614 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
615 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
616 struct ext4_group_desc *desc;
617 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
620 * Try to place the inode is the same flex group as its
621 * parent. If we can't find space, use the Orlov algorithm to
622 * find another flex group, and store that information in the
623 * parent directory's inode information so that use that flex
624 * group for future allocations.
630 parent_group &= ~(flex_size-1);
631 last = parent_group + flex_size;
634 for (i = parent_group; i < last; i++) {
635 desc = ext4_get_group_desc(sb, i, NULL);
636 if (desc && ext4_free_inodes_count(sb, desc)) {
641 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
643 parent_group = EXT4_I(parent)->i_last_alloc_group;
647 * If this didn't work, use the Orlov search algorithm
648 * to find a new flex group; we pass in the mode to
649 * avoid the topdir algorithms.
651 *group = parent_group + flex_size;
652 if (*group > ngroups)
654 return find_group_orlov(sb, parent, group, mode);
658 * Try to place the inode in its parent directory
660 *group = parent_group;
661 desc = ext4_get_group_desc(sb, *group, NULL);
662 if (desc && ext4_free_inodes_count(sb, desc) &&
663 ext4_free_blks_count(sb, desc))
667 * We're going to place this inode in a different blockgroup from its
668 * parent. We want to cause files in a common directory to all land in
669 * the same blockgroup. But we want files which are in a different
670 * directory which shares a blockgroup with our parent to land in a
671 * different blockgroup.
673 * So add our directory's i_ino into the starting point for the hash.
675 *group = (*group + parent->i_ino) % ngroups;
678 * Use a quadratic hash to find a group with a free inode and some free
681 for (i = 1; i < ngroups; i <<= 1) {
683 if (*group >= ngroups)
685 desc = ext4_get_group_desc(sb, *group, NULL);
686 if (desc && ext4_free_inodes_count(sb, desc) &&
687 ext4_free_blks_count(sb, desc))
692 * That failed: try linear search for a free inode, even if that group
693 * has no free blocks.
695 *group = parent_group;
696 for (i = 0; i < ngroups; i++) {
697 if (++*group >= ngroups)
699 desc = ext4_get_group_desc(sb, *group, NULL);
700 if (desc && ext4_free_inodes_count(sb, desc))
708 * claim the inode from the inode bitmap. If the group
709 * is uninit we need to take the groups's ext4_group_lock
710 * and clear the uninit flag. The inode bitmap update
711 * and group desc uninit flag clear should be done
712 * after holding ext4_group_lock so that ext4_read_inode_bitmap
713 * doesn't race with the ext4_claim_inode
715 static int ext4_claim_inode(struct super_block *sb,
716 struct buffer_head *inode_bitmap_bh,
717 unsigned long ino, ext4_group_t group, int mode)
719 int free = 0, retval = 0, count;
720 struct ext4_sb_info *sbi = EXT4_SB(sb);
721 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
723 ext4_lock_group(sb, group);
724 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
725 /* not a free inode */
730 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
731 ino > EXT4_INODES_PER_GROUP(sb)) {
732 ext4_unlock_group(sb, group);
733 ext4_error(sb, __func__,
734 "reserved inode or inode > inodes count - "
735 "block_group = %u, inode=%lu", group,
736 ino + group * EXT4_INODES_PER_GROUP(sb));
739 /* If we didn't allocate from within the initialized part of the inode
740 * table then we need to initialize up to this inode. */
741 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
743 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
744 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
745 /* When marking the block group with
746 * ~EXT4_BG_INODE_UNINIT we don't want to depend
747 * on the value of bg_itable_unused even though
748 * mke2fs could have initialized the same for us.
749 * Instead we calculated the value below
754 free = EXT4_INODES_PER_GROUP(sb) -
755 ext4_itable_unused_count(sb, gdp);
759 * Check the relative inode number against the last used
760 * relative inode number in this group. if it is greater
761 * we need to update the bg_itable_unused count
765 ext4_itable_unused_set(sb, gdp,
766 (EXT4_INODES_PER_GROUP(sb) - ino));
768 count = ext4_free_inodes_count(sb, gdp) - 1;
769 ext4_free_inodes_set(sb, gdp, count);
771 count = ext4_used_dirs_count(sb, gdp) + 1;
772 ext4_used_dirs_set(sb, gdp, count);
773 if (sbi->s_log_groups_per_flex) {
774 ext4_group_t f = ext4_flex_group(sbi, group);
776 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
779 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
781 ext4_unlock_group(sb, group);
786 * There are two policies for allocating an inode. If the new inode is
787 * a directory, then a forward search is made for a block group with both
788 * free space and a low directory-to-inode ratio; if that fails, then of
789 * the groups with above-average free space, that group with the fewest
790 * directories already is chosen.
792 * For other inodes, search forward from the parent directory's block
793 * group to find a free inode.
795 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode)
797 struct super_block *sb;
798 struct buffer_head *inode_bitmap_bh = NULL;
799 struct buffer_head *group_desc_bh;
800 ext4_group_t ngroups, group = 0;
801 unsigned long ino = 0;
803 struct ext4_group_desc *gdp = NULL;
804 struct ext4_inode_info *ei;
805 struct ext4_sb_info *sbi;
811 ext4_group_t flex_group;
813 /* Cannot create files in a deleted directory */
814 if (!dir || !dir->i_nlink)
815 return ERR_PTR(-EPERM);
818 ngroups = ext4_get_groups_count(sb);
819 trace_mark(ext4_request_inode, "dev %s dir %lu mode %d", sb->s_id,
821 inode = new_inode(sb);
823 return ERR_PTR(-ENOMEM);
827 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
828 ret2 = find_group_flex(sb, dir, &group);
830 ret2 = find_group_other(sb, dir, &group, mode);
831 if (ret2 == 0 && once) {
833 printk(KERN_NOTICE "ext4: find_group_flex "
834 "failed, fallback succeeded dir %lu\n",
842 if (test_opt(sb, OLDALLOC))
843 ret2 = find_group_dir(sb, dir, &group);
845 ret2 = find_group_orlov(sb, dir, &group, mode);
847 ret2 = find_group_other(sb, dir, &group, mode);
850 EXT4_I(dir)->i_last_alloc_group = group;
855 for (i = 0; i < ngroups; i++) {
858 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
862 brelse(inode_bitmap_bh);
863 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
864 if (!inode_bitmap_bh)
869 repeat_in_this_group:
870 ino = ext4_find_next_zero_bit((unsigned long *)
871 inode_bitmap_bh->b_data,
872 EXT4_INODES_PER_GROUP(sb), ino);
874 if (ino < EXT4_INODES_PER_GROUP(sb)) {
876 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
877 err = ext4_journal_get_write_access(handle,
882 BUFFER_TRACE(group_desc_bh, "get_write_access");
883 err = ext4_journal_get_write_access(handle,
887 if (!ext4_claim_inode(sb, inode_bitmap_bh,
890 BUFFER_TRACE(inode_bitmap_bh,
891 "call ext4_handle_dirty_metadata");
892 err = ext4_handle_dirty_metadata(handle,
897 /* zero bit is inode number 1*/
902 ext4_handle_release_buffer(handle, inode_bitmap_bh);
903 ext4_handle_release_buffer(handle, group_desc_bh);
905 if (++ino < EXT4_INODES_PER_GROUP(sb))
906 goto repeat_in_this_group;
910 * This case is possible in concurrent environment. It is very
911 * rare. We cannot repeat the find_group_xxx() call because
912 * that will simply return the same blockgroup, because the
913 * group descriptor metadata has not yet been updated.
914 * So we just go onto the next blockgroup.
916 if (++group == ngroups)
923 /* We may have to initialize the block bitmap if it isn't already */
924 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
925 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
926 struct buffer_head *block_bitmap_bh;
928 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
929 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
930 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
932 brelse(block_bitmap_bh);
937 ext4_lock_group(sb, group);
938 /* recheck and clear flag under lock if we still need to */
939 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
940 free = ext4_free_blocks_after_init(sb, group, gdp);
941 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
942 ext4_free_blks_set(sb, gdp, free);
943 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
946 ext4_unlock_group(sb, group);
948 /* Don't need to dirty bitmap block if we didn't change it */
950 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
951 err = ext4_handle_dirty_metadata(handle,
952 NULL, block_bitmap_bh);
955 brelse(block_bitmap_bh);
959 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
960 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
964 percpu_counter_dec(&sbi->s_freeinodes_counter);
966 percpu_counter_inc(&sbi->s_dirs_counter);
969 if (sbi->s_log_groups_per_flex) {
970 flex_group = ext4_flex_group(sbi, group);
971 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
974 inode->i_uid = current_fsuid();
975 if (test_opt(sb, GRPID))
976 inode->i_gid = dir->i_gid;
977 else if (dir->i_mode & S_ISGID) {
978 inode->i_gid = dir->i_gid;
982 inode->i_gid = current_fsgid();
983 inode->i_mode = mode;
985 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
986 /* This is the optimal IO size (for stat), not the fs block size */
988 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
989 ext4_current_time(inode);
991 memset(ei->i_data, 0, sizeof(ei->i_data));
992 ei->i_dir_start_lookup = 0;
996 * Don't inherit extent flag from directory, amongst others. We set
997 * extent flag on newly created directory and file only if -o extent
998 * mount option is specified
1001 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1004 ei->i_block_group = group;
1005 ei->i_last_alloc_group = ~0;
1007 ext4_set_inode_flags(inode);
1008 if (IS_DIRSYNC(inode))
1009 ext4_handle_sync(handle);
1010 if (insert_inode_locked(inode) < 0) {
1014 spin_lock(&sbi->s_next_gen_lock);
1015 inode->i_generation = sbi->s_next_generation++;
1016 spin_unlock(&sbi->s_next_gen_lock);
1018 ei->i_state = EXT4_STATE_NEW;
1020 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1023 if (vfs_dq_alloc_inode(inode)) {
1028 err = ext4_init_acl(handle, inode, dir);
1030 goto fail_free_drop;
1032 err = ext4_init_security(handle, inode, dir);
1034 goto fail_free_drop;
1036 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1037 /* set extent flag only for directory, file and normal symlink*/
1038 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1039 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
1040 ext4_ext_tree_init(handle, inode);
1044 err = ext4_mark_inode_dirty(handle, inode);
1046 ext4_std_error(sb, err);
1047 goto fail_free_drop;
1050 ext4_debug("allocating inode %lu\n", inode->i_ino);
1051 trace_mark(ext4_allocate_inode, "dev %s ino %lu dir %lu mode %d",
1052 sb->s_id, inode->i_ino, dir->i_ino, mode);
1055 ext4_std_error(sb, err);
1060 brelse(inode_bitmap_bh);
1064 vfs_dq_free_inode(inode);
1068 inode->i_flags |= S_NOQUOTA;
1070 unlock_new_inode(inode);
1072 brelse(inode_bitmap_bh);
1073 return ERR_PTR(err);
1076 /* Verify that we are loading a valid orphan from disk */
1077 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1079 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1080 ext4_group_t block_group;
1082 struct buffer_head *bitmap_bh;
1083 struct inode *inode = NULL;
1086 /* Error cases - e2fsck has already cleaned up for us */
1087 if (ino > max_ino) {
1088 ext4_warning(sb, __func__,
1089 "bad orphan ino %lu! e2fsck was run?", ino);
1093 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1094 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1095 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1097 ext4_warning(sb, __func__,
1098 "inode bitmap error for orphan %lu", ino);
1102 /* Having the inode bit set should be a 100% indicator that this
1103 * is a valid orphan (no e2fsck run on fs). Orphans also include
1104 * inodes that were being truncated, so we can't check i_nlink==0.
1106 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1109 inode = ext4_iget(sb, ino);
1114 * If the orphans has i_nlinks > 0 then it should be able to be
1115 * truncated, otherwise it won't be removed from the orphan list
1116 * during processing and an infinite loop will result.
1118 if (inode->i_nlink && !ext4_can_truncate(inode))
1121 if (NEXT_ORPHAN(inode) > max_ino)
1127 err = PTR_ERR(inode);
1130 ext4_warning(sb, __func__,
1131 "bad orphan inode %lu! e2fsck was run?", ino);
1132 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1133 bit, (unsigned long long)bitmap_bh->b_blocknr,
1134 ext4_test_bit(bit, bitmap_bh->b_data));
1135 printk(KERN_NOTICE "inode=%p\n", inode);
1137 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1138 is_bad_inode(inode));
1139 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1140 NEXT_ORPHAN(inode));
1141 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1142 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1143 /* Avoid freeing blocks if we got a bad deleted inode */
1144 if (inode->i_nlink == 0)
1145 inode->i_blocks = 0;
1150 return ERR_PTR(err);
1153 unsigned long ext4_count_free_inodes(struct super_block *sb)
1155 unsigned long desc_count;
1156 struct ext4_group_desc *gdp;
1157 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1159 struct ext4_super_block *es;
1160 unsigned long bitmap_count, x;
1161 struct buffer_head *bitmap_bh = NULL;
1163 es = EXT4_SB(sb)->s_es;
1167 for (i = 0; i < ngroups; i++) {
1168 gdp = ext4_get_group_desc(sb, i, NULL);
1171 desc_count += ext4_free_inodes_count(sb, gdp);
1173 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1177 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1178 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1179 i, ext4_free_inodes_count(sb, gdp), x);
1183 printk(KERN_DEBUG "ext4_count_free_inodes: "
1184 "stored = %u, computed = %lu, %lu\n",
1185 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1189 for (i = 0; i < ngroups; i++) {
1190 gdp = ext4_get_group_desc(sb, i, NULL);
1193 desc_count += ext4_free_inodes_count(sb, gdp);
1200 /* Called at mount-time, super-block is locked */
1201 unsigned long ext4_count_dirs(struct super_block * sb)
1203 unsigned long count = 0;
1204 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1206 for (i = 0; i < ngroups; i++) {
1207 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1210 count += ext4_used_dirs_count(sb, gdp);