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"
33 * ialloc.c contains the inodes allocation and deallocation routines
37 * The free inodes are managed by bitmaps. A file system contains several
38 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
39 * block for inodes, N blocks for the inode table and data blocks.
41 * The file system contains group descriptors which are located after the
42 * super block. Each descriptor contains the number of the bitmap block and
43 * the free blocks count in the block.
47 * To avoid calling the atomic setbit hundreds or thousands of times, we only
48 * need to use it within a single byte (to ensure we get endianness right).
49 * We can use memset for the rest of the bitmap as there are no other users.
51 void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
55 if (start_bit >= end_bit)
58 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
59 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
60 ext4_set_bit(i, bitmap);
62 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
65 /* Initializes an uninitialized inode bitmap */
66 unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
67 ext4_group_t block_group,
68 struct ext4_group_desc *gdp)
70 struct ext4_sb_info *sbi = EXT4_SB(sb);
72 J_ASSERT_BH(bh, buffer_locked(bh));
74 /* If checksum is bad mark all blocks and inodes use to prevent
75 * allocation, essentially implementing a per-group read-only flag. */
76 if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
77 ext4_error(sb, __func__, "Checksum bad for group %u",
79 ext4_free_blks_set(sb, gdp, 0);
80 ext4_free_inodes_set(sb, gdp, 0);
81 ext4_itable_unused_set(sb, gdp, 0);
82 memset(bh->b_data, 0xff, sb->s_blocksize);
86 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
87 mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
90 return EXT4_INODES_PER_GROUP(sb);
94 * Read the inode allocation bitmap for a given block_group, reading
95 * into the specified slot in the superblock's bitmap cache.
97 * Return buffer_head of bitmap on success or NULL.
99 static struct buffer_head *
100 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
102 struct ext4_group_desc *desc;
103 struct buffer_head *bh = NULL;
104 ext4_fsblk_t bitmap_blk;
106 desc = ext4_get_group_desc(sb, block_group, NULL);
109 bitmap_blk = ext4_inode_bitmap(sb, desc);
110 bh = sb_getblk(sb, bitmap_blk);
112 ext4_error(sb, __func__,
113 "Cannot read inode bitmap - "
114 "block_group = %u, inode_bitmap = %llu",
115 block_group, bitmap_blk);
118 if (bitmap_uptodate(bh))
122 if (bitmap_uptodate(bh)) {
126 spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
127 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
128 ext4_init_inode_bitmap(sb, bh, block_group, desc);
129 set_bitmap_uptodate(bh);
130 set_buffer_uptodate(bh);
131 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
135 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
136 if (buffer_uptodate(bh)) {
138 * if not uninit if bh is uptodate,
139 * bitmap is also uptodate
141 set_bitmap_uptodate(bh);
146 * submit the buffer_head for read. We can
147 * safely mark the bitmap as uptodate now.
148 * We do it here so the bitmap uptodate bit
149 * get set with buffer lock held.
151 set_bitmap_uptodate(bh);
152 if (bh_submit_read(bh) < 0) {
154 ext4_error(sb, __func__,
155 "Cannot read inode bitmap - "
156 "block_group = %u, inode_bitmap = %llu",
157 block_group, bitmap_blk);
164 * NOTE! When we get the inode, we're the only people
165 * that have access to it, and as such there are no
166 * race conditions we have to worry about. The inode
167 * is not on the hash-lists, and it cannot be reached
168 * through the filesystem because the directory entry
169 * has been deleted earlier.
171 * HOWEVER: we must make sure that we get no aliases,
172 * which means that we have to call "clear_inode()"
173 * _before_ we mark the inode not in use in the inode
174 * bitmaps. Otherwise a newly created file might use
175 * the same inode number (not actually the same pointer
176 * though), and then we'd have two inodes sharing the
177 * same inode number and space on the harddisk.
179 void ext4_free_inode(handle_t *handle, struct inode *inode)
181 struct super_block *sb = inode->i_sb;
184 struct buffer_head *bitmap_bh = NULL;
185 struct buffer_head *bh2;
186 ext4_group_t block_group;
188 struct ext4_group_desc *gdp;
189 struct ext4_super_block *es;
190 struct ext4_sb_info *sbi;
191 int fatal = 0, err, count, cleared;
193 if (atomic_read(&inode->i_count) > 1) {
194 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
195 atomic_read(&inode->i_count));
198 if (inode->i_nlink) {
199 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
204 printk(KERN_ERR "ext4_free_inode: inode on "
205 "nonexistent device\n");
211 ext4_debug("freeing inode %lu\n", ino);
212 trace_mark(ext4_free_inode,
213 "dev %s ino %lu mode %d uid %lu gid %lu bocks %llu",
214 sb->s_id, inode->i_ino, inode->i_mode,
215 (unsigned long) inode->i_uid, (unsigned long) inode->i_gid,
216 (unsigned long long) inode->i_blocks);
219 * Note: we must free any quota before locking the superblock,
220 * as writing the quota to disk may need the lock as well.
223 ext4_xattr_delete_inode(handle, inode);
224 vfs_dq_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 */
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, "ext4_free_inode",
235 "reserved or nonexistent inode %lu", ino);
238 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
239 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
240 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
244 BUFFER_TRACE(bitmap_bh, "get_write_access");
245 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
249 /* Ok, now we can actually update the inode bitmaps.. */
250 spin_lock(sb_bgl_lock(sbi, block_group));
251 cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
252 spin_unlock(sb_bgl_lock(sbi, block_group));
254 ext4_error(sb, "ext4_free_inode",
255 "bit already cleared for inode %lu", ino);
257 gdp = ext4_get_group_desc(sb, block_group, &bh2);
259 BUFFER_TRACE(bh2, "get_write_access");
260 fatal = ext4_journal_get_write_access(handle, bh2);
261 if (fatal) goto error_return;
264 spin_lock(sb_bgl_lock(sbi, block_group));
265 count = ext4_free_inodes_count(sb, gdp) + 1;
266 ext4_free_inodes_set(sb, gdp, count);
268 count = ext4_used_dirs_count(sb, gdp) - 1;
269 ext4_used_dirs_set(sb, gdp, count);
270 if (sbi->s_log_groups_per_flex) {
273 f = ext4_flex_group(sbi, block_group);
274 atomic_dec(&sbi->s_flex_groups[f].free_inodes);
278 gdp->bg_checksum = ext4_group_desc_csum(sbi,
280 spin_unlock(sb_bgl_lock(sbi, block_group));
281 percpu_counter_inc(&sbi->s_freeinodes_counter);
283 percpu_counter_dec(&sbi->s_dirs_counter);
285 if (sbi->s_log_groups_per_flex) {
288 f = ext4_flex_group(sbi, block_group);
289 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
292 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
293 err = ext4_handle_dirty_metadata(handle, NULL, bh2);
294 if (!fatal) fatal = err;
296 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
297 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
303 ext4_std_error(sb, fatal);
307 * There are two policies for allocating an inode. If the new inode is
308 * a directory, then a forward search is made for a block group with both
309 * free space and a low directory-to-inode ratio; if that fails, then of
310 * the groups with above-average free space, that group with the fewest
311 * directories already is chosen.
313 * For other inodes, search forward from the parent directory\'s block
314 * group to find a free inode.
316 static int find_group_dir(struct super_block *sb, struct inode *parent,
317 ext4_group_t *best_group)
319 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
320 unsigned int freei, avefreei;
321 struct ext4_group_desc *desc, *best_desc = NULL;
325 freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
326 avefreei = freei / ngroups;
328 for (group = 0; group < ngroups; group++) {
329 desc = ext4_get_group_desc(sb, group, NULL);
330 if (!desc || !ext4_free_inodes_count(sb, desc))
332 if (ext4_free_inodes_count(sb, desc) < avefreei)
335 (ext4_free_blks_count(sb, desc) >
336 ext4_free_blks_count(sb, best_desc))) {
345 #define free_block_ratio 10
347 static int find_group_flex(struct super_block *sb, struct inode *parent,
348 ext4_group_t *best_group)
350 struct ext4_sb_info *sbi = EXT4_SB(sb);
351 struct ext4_group_desc *desc;
352 struct buffer_head *bh;
353 struct flex_groups *flex_group = sbi->s_flex_groups;
354 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
355 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
356 ext4_group_t ngroups = sbi->s_groups_count;
357 int flex_size = ext4_flex_bg_size(sbi);
358 ext4_group_t best_flex = parent_fbg_group;
359 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
360 int flexbg_free_blocks;
361 int flex_freeb_ratio;
362 ext4_group_t n_fbg_groups;
365 n_fbg_groups = (sbi->s_groups_count + flex_size - 1) >>
366 sbi->s_log_groups_per_flex;
368 find_close_to_parent:
369 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
370 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
371 if (atomic_read(&flex_group[best_flex].free_inodes) &&
372 flex_freeb_ratio > free_block_ratio)
375 if (best_flex && best_flex == parent_fbg_group) {
377 goto find_close_to_parent;
380 for (i = 0; i < n_fbg_groups; i++) {
381 if (i == parent_fbg_group || i == parent_fbg_group - 1)
384 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
385 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
387 if (flex_freeb_ratio > free_block_ratio &&
388 (atomic_read(&flex_group[i].free_inodes))) {
393 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
394 ((atomic_read(&flex_group[i].free_blocks) >
395 atomic_read(&flex_group[best_flex].free_blocks)) &&
396 atomic_read(&flex_group[i].free_inodes)))
400 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
401 !atomic_read(&flex_group[best_flex].free_blocks))
405 for (i = best_flex * flex_size; i < ngroups &&
406 i < (best_flex + 1) * flex_size; i++) {
407 desc = ext4_get_group_desc(sb, i, &bh);
408 if (ext4_free_inodes_count(sb, desc)) {
426 * Helper function for Orlov's allocator; returns critical information
427 * for a particular block group or flex_bg. If flex_size is 1, then g
428 * is a block group number; otherwise it is flex_bg number.
430 void get_orlov_stats(struct super_block *sb, ext4_group_t g,
431 int flex_size, struct orlov_stats *stats)
433 struct ext4_group_desc *desc;
434 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
437 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
438 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
439 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
443 desc = ext4_get_group_desc(sb, g, NULL);
445 stats->free_inodes = ext4_free_inodes_count(sb, desc);
446 stats->free_blocks = ext4_free_blks_count(sb, desc);
447 stats->used_dirs = ext4_used_dirs_count(sb, desc);
449 stats->free_inodes = 0;
450 stats->free_blocks = 0;
451 stats->used_dirs = 0;
456 * Orlov's allocator for directories.
458 * We always try to spread first-level directories.
460 * If there are blockgroups with both free inodes and free blocks counts
461 * not worse than average we return one with smallest directory count.
462 * Otherwise we simply return a random group.
464 * For the rest rules look so:
466 * It's OK to put directory into a group unless
467 * it has too many directories already (max_dirs) or
468 * it has too few free inodes left (min_inodes) or
469 * it has too few free blocks left (min_blocks) or
470 * Parent's group is preferred, if it doesn't satisfy these
471 * conditions we search cyclically through the rest. If none
472 * of the groups look good we just look for a group with more
473 * free inodes than average (starting at parent's group).
476 static int find_group_orlov(struct super_block *sb, struct inode *parent,
477 ext4_group_t *group, int mode)
479 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
480 struct ext4_sb_info *sbi = EXT4_SB(sb);
481 ext4_group_t ngroups = sbi->s_groups_count;
482 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
483 unsigned int freei, avefreei;
484 ext4_fsblk_t freeb, avefreeb;
486 int max_dirs, min_inodes;
487 ext4_grpblk_t min_blocks;
488 ext4_group_t i, grp, g;
489 struct ext4_group_desc *desc;
490 struct orlov_stats stats;
491 int flex_size = ext4_flex_bg_size(sbi);
494 ngroups = (ngroups + flex_size - 1) >>
495 sbi->s_log_groups_per_flex;
496 parent_group >>= sbi->s_log_groups_per_flex;
499 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
500 avefreei = freei / ngroups;
501 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
503 do_div(avefreeb, ngroups);
504 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
507 ((parent == sb->s_root->d_inode) ||
508 (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) {
509 int best_ndir = inodes_per_group;
512 get_random_bytes(&grp, sizeof(grp));
513 parent_group = (unsigned)grp % ngroups;
514 for (i = 0; i < ngroups; i++) {
515 g = (parent_group + i) % ngroups;
516 get_orlov_stats(sb, g, flex_size, &stats);
517 if (!stats.free_inodes)
519 if (stats.used_dirs >= best_ndir)
521 if (stats.free_inodes < avefreei)
523 if (stats.free_blocks < avefreeb)
527 best_ndir = stats.used_dirs;
532 if (flex_size == 1) {
538 * We pack inodes at the beginning of the flexgroup's
539 * inode tables. Block allocation decisions will do
540 * something similar, although regular files will
541 * start at 2nd block group of the flexgroup. See
542 * ext4_ext_find_goal() and ext4_find_near().
545 for (i = 0; i < flex_size; i++) {
546 if (grp+i >= sbi->s_groups_count)
548 desc = ext4_get_group_desc(sb, grp+i, NULL);
549 if (desc && ext4_free_inodes_count(sb, desc)) {
557 max_dirs = ndirs / ngroups + inodes_per_group / 16;
558 min_inodes = avefreei - inodes_per_group*flex_size / 4;
561 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
564 * Start looking in the flex group where we last allocated an
565 * inode for this parent directory
567 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
568 parent_group = EXT4_I(parent)->i_last_alloc_group;
570 parent_group >>= sbi->s_log_groups_per_flex;
573 for (i = 0; i < ngroups; i++) {
574 grp = (parent_group + i) % ngroups;
575 get_orlov_stats(sb, grp, flex_size, &stats);
576 if (stats.used_dirs >= max_dirs)
578 if (stats.free_inodes < min_inodes)
580 if (stats.free_blocks < min_blocks)
586 ngroups = sbi->s_groups_count;
587 avefreei = freei / ngroups;
589 parent_group = EXT4_I(parent)->i_block_group;
590 for (i = 0; i < ngroups; i++) {
591 grp = (parent_group + i) % ngroups;
592 desc = ext4_get_group_desc(sb, grp, NULL);
593 if (desc && ext4_free_inodes_count(sb, desc) &&
594 ext4_free_inodes_count(sb, desc) >= avefreei) {
602 * The free-inodes counter is approximate, and for really small
603 * filesystems the above test can fail to find any blockgroups
612 static int find_group_other(struct super_block *sb, struct inode *parent,
613 ext4_group_t *group, int mode)
615 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
616 ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
617 struct ext4_group_desc *desc;
618 ext4_group_t i, last;
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);
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_blks_count(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_blks_count(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 sb_bgl_lock
712 * and clear the uninit flag. The inode bitmap update
713 * and group desc uninit flag clear should be done
714 * after holding sb_bgl_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_desc *gdp = ext4_get_group_desc(sb, group, NULL);
725 spin_lock(sb_bgl_lock(sbi, group));
726 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
727 /* not a free inode */
732 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
733 ino > EXT4_INODES_PER_GROUP(sb)) {
734 spin_unlock(sb_bgl_lock(sbi, group));
735 ext4_error(sb, __func__,
736 "reserved inode or inode > inodes count - "
737 "block_group = %u, inode=%lu", group,
738 ino + group * EXT4_INODES_PER_GROUP(sb));
741 /* If we didn't allocate from within the initialized part of the inode
742 * table then we need to initialize up to this inode. */
743 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
745 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
746 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
747 /* When marking the block group with
748 * ~EXT4_BG_INODE_UNINIT we don't want to depend
749 * on the value of bg_itable_unused even though
750 * mke2fs could have initialized the same for us.
751 * Instead we calculated the value below
756 free = EXT4_INODES_PER_GROUP(sb) -
757 ext4_itable_unused_count(sb, gdp);
761 * Check the relative inode number against the last used
762 * relative inode number in this group. if it is greater
763 * we need to update the bg_itable_unused count
767 ext4_itable_unused_set(sb, gdp,
768 (EXT4_INODES_PER_GROUP(sb) - ino));
770 count = ext4_free_inodes_count(sb, gdp) - 1;
771 ext4_free_inodes_set(sb, gdp, count);
773 count = ext4_used_dirs_count(sb, gdp) + 1;
774 ext4_used_dirs_set(sb, gdp, count);
775 if (sbi->s_log_groups_per_flex) {
776 ext4_group_t f = ext4_flex_group(sbi, group);
778 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
781 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
783 spin_unlock(sb_bgl_lock(sbi, group));
788 * There are two policies for allocating an inode. If the new inode is
789 * a directory, then a forward search is made for a block group with both
790 * free space and a low directory-to-inode ratio; if that fails, then of
791 * the groups with above-average free space, that group with the fewest
792 * directories already is chosen.
794 * For other inodes, search forward from the parent directory's block
795 * group to find a free inode.
797 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode)
799 struct super_block *sb;
800 struct buffer_head *inode_bitmap_bh = NULL;
801 struct buffer_head *group_desc_bh;
802 ext4_group_t group = 0;
803 unsigned long ino = 0;
805 struct ext4_group_desc *gdp = NULL;
806 struct ext4_super_block *es;
807 struct ext4_inode_info *ei;
808 struct ext4_sb_info *sbi;
814 ext4_group_t flex_group;
816 /* Cannot create files in a deleted directory */
817 if (!dir || !dir->i_nlink)
818 return ERR_PTR(-EPERM);
821 trace_mark(ext4_request_inode, "dev %s dir %lu mode %d", sb->s_id,
823 inode = new_inode(sb);
825 return ERR_PTR(-ENOMEM);
831 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
832 ret2 = find_group_flex(sb, dir, &group);
834 ret2 = find_group_other(sb, dir, &group, mode);
835 if (ret2 == 0 && once) {
837 printk(KERN_NOTICE "ext4: find_group_flex "
838 "failed, fallback succeeded dir %lu\n",
846 if (test_opt(sb, OLDALLOC))
847 ret2 = find_group_dir(sb, dir, &group);
849 ret2 = find_group_orlov(sb, dir, &group, mode);
851 ret2 = find_group_other(sb, dir, &group, mode);
854 EXT4_I(dir)->i_last_alloc_group = group;
859 for (i = 0; i < sbi->s_groups_count; i++) {
862 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
866 brelse(inode_bitmap_bh);
867 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
868 if (!inode_bitmap_bh)
873 repeat_in_this_group:
874 ino = ext4_find_next_zero_bit((unsigned long *)
875 inode_bitmap_bh->b_data,
876 EXT4_INODES_PER_GROUP(sb), ino);
878 if (ino < EXT4_INODES_PER_GROUP(sb)) {
880 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
881 err = ext4_journal_get_write_access(handle,
886 BUFFER_TRACE(group_desc_bh, "get_write_access");
887 err = ext4_journal_get_write_access(handle,
891 if (!ext4_claim_inode(sb, inode_bitmap_bh,
894 BUFFER_TRACE(inode_bitmap_bh,
895 "call ext4_handle_dirty_metadata");
896 err = ext4_handle_dirty_metadata(handle,
901 /* zero bit is inode number 1*/
906 ext4_handle_release_buffer(handle, inode_bitmap_bh);
907 ext4_handle_release_buffer(handle, group_desc_bh);
909 if (++ino < EXT4_INODES_PER_GROUP(sb))
910 goto repeat_in_this_group;
914 * This case is possible in concurrent environment. It is very
915 * rare. We cannot repeat the find_group_xxx() call because
916 * that will simply return the same blockgroup, because the
917 * group descriptor metadata has not yet been updated.
918 * So we just go onto the next blockgroup.
920 if (++group == sbi->s_groups_count)
927 /* We may have to initialize the block bitmap if it isn't already */
928 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
929 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
930 struct buffer_head *block_bitmap_bh;
932 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
933 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
934 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
936 brelse(block_bitmap_bh);
941 spin_lock(sb_bgl_lock(sbi, group));
942 /* recheck and clear flag under lock if we still need to */
943 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
944 free = ext4_free_blocks_after_init(sb, group, gdp);
945 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
946 ext4_free_blks_set(sb, gdp, free);
947 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
950 spin_unlock(sb_bgl_lock(sbi, group));
952 /* Don't need to dirty bitmap block if we didn't change it */
954 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
955 err = ext4_handle_dirty_metadata(handle,
956 NULL, block_bitmap_bh);
959 brelse(block_bitmap_bh);
963 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
964 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
968 percpu_counter_dec(&sbi->s_freeinodes_counter);
970 percpu_counter_inc(&sbi->s_dirs_counter);
973 if (sbi->s_log_groups_per_flex) {
974 flex_group = ext4_flex_group(sbi, group);
975 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
978 inode->i_uid = current_fsuid();
979 if (test_opt(sb, GRPID))
980 inode->i_gid = dir->i_gid;
981 else if (dir->i_mode & S_ISGID) {
982 inode->i_gid = dir->i_gid;
986 inode->i_gid = current_fsgid();
987 inode->i_mode = mode;
989 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
990 /* This is the optimal IO size (for stat), not the fs block size */
992 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
993 ext4_current_time(inode);
995 memset(ei->i_data, 0, sizeof(ei->i_data));
996 ei->i_dir_start_lookup = 0;
1000 * Don't inherit extent flag from directory, amongst others. We set
1001 * extent flag on newly created directory and file only if -o extent
1002 * mount option is specified
1005 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1008 ei->i_block_group = group;
1009 ei->i_last_alloc_group = ~0;
1011 ext4_set_inode_flags(inode);
1012 if (IS_DIRSYNC(inode))
1013 ext4_handle_sync(handle);
1014 if (insert_inode_locked(inode) < 0) {
1018 spin_lock(&sbi->s_next_gen_lock);
1019 inode->i_generation = sbi->s_next_generation++;
1020 spin_unlock(&sbi->s_next_gen_lock);
1022 ei->i_state = EXT4_STATE_NEW;
1024 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1027 if (vfs_dq_alloc_inode(inode)) {
1032 err = ext4_init_acl(handle, inode, dir);
1034 goto fail_free_drop;
1036 err = ext4_init_security(handle, inode, dir);
1038 goto fail_free_drop;
1040 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1041 /* set extent flag only for directory, file and normal symlink*/
1042 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1043 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
1044 ext4_ext_tree_init(handle, inode);
1048 err = ext4_mark_inode_dirty(handle, inode);
1050 ext4_std_error(sb, err);
1051 goto fail_free_drop;
1054 ext4_debug("allocating inode %lu\n", inode->i_ino);
1055 trace_mark(ext4_allocate_inode, "dev %s ino %lu dir %lu mode %d",
1056 sb->s_id, inode->i_ino, dir->i_ino, mode);
1059 ext4_std_error(sb, err);
1064 brelse(inode_bitmap_bh);
1068 vfs_dq_free_inode(inode);
1072 inode->i_flags |= S_NOQUOTA;
1074 unlock_new_inode(inode);
1076 brelse(inode_bitmap_bh);
1077 return ERR_PTR(err);
1080 /* Verify that we are loading a valid orphan from disk */
1081 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1083 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1084 ext4_group_t block_group;
1086 struct buffer_head *bitmap_bh;
1087 struct inode *inode = NULL;
1090 /* Error cases - e2fsck has already cleaned up for us */
1091 if (ino > max_ino) {
1092 ext4_warning(sb, __func__,
1093 "bad orphan ino %lu! e2fsck was run?", ino);
1097 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1098 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1099 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1101 ext4_warning(sb, __func__,
1102 "inode bitmap error for orphan %lu", ino);
1106 /* Having the inode bit set should be a 100% indicator that this
1107 * is a valid orphan (no e2fsck run on fs). Orphans also include
1108 * inodes that were being truncated, so we can't check i_nlink==0.
1110 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1113 inode = ext4_iget(sb, ino);
1118 * If the orphans has i_nlinks > 0 then it should be able to be
1119 * truncated, otherwise it won't be removed from the orphan list
1120 * during processing and an infinite loop will result.
1122 if (inode->i_nlink && !ext4_can_truncate(inode))
1125 if (NEXT_ORPHAN(inode) > max_ino)
1131 err = PTR_ERR(inode);
1134 ext4_warning(sb, __func__,
1135 "bad orphan inode %lu! e2fsck was run?", ino);
1136 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1137 bit, (unsigned long long)bitmap_bh->b_blocknr,
1138 ext4_test_bit(bit, bitmap_bh->b_data));
1139 printk(KERN_NOTICE "inode=%p\n", inode);
1141 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1142 is_bad_inode(inode));
1143 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1144 NEXT_ORPHAN(inode));
1145 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1146 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1147 /* Avoid freeing blocks if we got a bad deleted inode */
1148 if (inode->i_nlink == 0)
1149 inode->i_blocks = 0;
1154 return ERR_PTR(err);
1157 unsigned long ext4_count_free_inodes(struct super_block *sb)
1159 unsigned long desc_count;
1160 struct ext4_group_desc *gdp;
1163 struct ext4_super_block *es;
1164 unsigned long bitmap_count, x;
1165 struct buffer_head *bitmap_bh = NULL;
1167 es = EXT4_SB(sb)->s_es;
1171 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1172 gdp = ext4_get_group_desc(sb, i, NULL);
1175 desc_count += ext4_free_inodes_count(sb, gdp);
1177 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1181 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1182 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1183 i, ext4_free_inodes_count(sb, gdp), x);
1187 printk(KERN_DEBUG "ext4_count_free_inodes: "
1188 "stored = %u, computed = %lu, %lu\n",
1189 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1193 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1194 gdp = ext4_get_group_desc(sb, i, NULL);
1197 desc_count += ext4_free_inodes_count(sb, gdp);
1204 /* Called at mount-time, super-block is locked */
1205 unsigned long ext4_count_dirs(struct super_block * sb)
1207 unsigned long count = 0;
1210 for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1211 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1214 count += ext4_used_dirs_count(sb, gdp);