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 flex_groups *flex_group = sbi->s_flex_groups;
351 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
352 ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
353 ext4_group_t ngroups = ext4_get_groups_count(sb);
354 int flex_size = ext4_flex_bg_size(sbi);
355 ext4_group_t best_flex = parent_fbg_group;
356 int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
357 int flexbg_free_blocks;
358 int flex_freeb_ratio;
359 ext4_group_t n_fbg_groups;
362 n_fbg_groups = (ngroups + flex_size - 1) >>
363 sbi->s_log_groups_per_flex;
365 find_close_to_parent:
366 flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
367 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
368 if (atomic_read(&flex_group[best_flex].free_inodes) &&
369 flex_freeb_ratio > free_block_ratio)
372 if (best_flex && best_flex == parent_fbg_group) {
374 goto find_close_to_parent;
377 for (i = 0; i < n_fbg_groups; i++) {
378 if (i == parent_fbg_group || i == parent_fbg_group - 1)
381 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
382 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
384 if (flex_freeb_ratio > free_block_ratio &&
385 (atomic_read(&flex_group[i].free_inodes))) {
390 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
391 ((atomic_read(&flex_group[i].free_blocks) >
392 atomic_read(&flex_group[best_flex].free_blocks)) &&
393 atomic_read(&flex_group[i].free_inodes)))
397 if (!atomic_read(&flex_group[best_flex].free_inodes) ||
398 !atomic_read(&flex_group[best_flex].free_blocks))
402 for (i = best_flex * flex_size; i < ngroups &&
403 i < (best_flex + 1) * flex_size; i++) {
404 desc = ext4_get_group_desc(sb, i, NULL);
405 if (ext4_free_inodes_count(sb, desc)) {
423 * Helper function for Orlov's allocator; returns critical information
424 * for a particular block group or flex_bg. If flex_size is 1, then g
425 * is a block group number; otherwise it is flex_bg number.
427 void get_orlov_stats(struct super_block *sb, ext4_group_t g,
428 int flex_size, struct orlov_stats *stats)
430 struct ext4_group_desc *desc;
431 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
434 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
435 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
436 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
440 desc = ext4_get_group_desc(sb, g, NULL);
442 stats->free_inodes = ext4_free_inodes_count(sb, desc);
443 stats->free_blocks = ext4_free_blks_count(sb, desc);
444 stats->used_dirs = ext4_used_dirs_count(sb, desc);
446 stats->free_inodes = 0;
447 stats->free_blocks = 0;
448 stats->used_dirs = 0;
453 * Orlov's allocator for directories.
455 * We always try to spread first-level directories.
457 * If there are blockgroups with both free inodes and free blocks counts
458 * not worse than average we return one with smallest directory count.
459 * Otherwise we simply return a random group.
461 * For the rest rules look so:
463 * It's OK to put directory into a group unless
464 * it has too many directories already (max_dirs) or
465 * it has too few free inodes left (min_inodes) or
466 * it has too few free blocks left (min_blocks) or
467 * Parent's group is preferred, if it doesn't satisfy these
468 * conditions we search cyclically through the rest. If none
469 * of the groups look good we just look for a group with more
470 * free inodes than average (starting at parent's group).
473 static int find_group_orlov(struct super_block *sb, struct inode *parent,
474 ext4_group_t *group, int mode)
476 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
477 struct ext4_sb_info *sbi = EXT4_SB(sb);
478 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
479 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
480 unsigned int freei, avefreei;
481 ext4_fsblk_t freeb, avefreeb;
483 int max_dirs, min_inodes;
484 ext4_grpblk_t min_blocks;
485 ext4_group_t i, grp, g, ngroups;
486 struct ext4_group_desc *desc;
487 struct orlov_stats stats;
488 int flex_size = ext4_flex_bg_size(sbi);
490 ngroups = real_ngroups;
492 ngroups = (real_ngroups + flex_size - 1) >>
493 sbi->s_log_groups_per_flex;
494 parent_group >>= sbi->s_log_groups_per_flex;
497 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
498 avefreei = freei / ngroups;
499 freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
501 do_div(avefreeb, ngroups);
502 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
505 ((parent == sb->s_root->d_inode) ||
506 (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL))) {
507 int best_ndir = inodes_per_group;
510 get_random_bytes(&grp, sizeof(grp));
511 parent_group = (unsigned)grp % ngroups;
512 for (i = 0; i < ngroups; i++) {
513 g = (parent_group + i) % ngroups;
514 get_orlov_stats(sb, g, flex_size, &stats);
515 if (!stats.free_inodes)
517 if (stats.used_dirs >= best_ndir)
519 if (stats.free_inodes < avefreei)
521 if (stats.free_blocks < avefreeb)
525 best_ndir = stats.used_dirs;
530 if (flex_size == 1) {
536 * We pack inodes at the beginning of the flexgroup's
537 * inode tables. Block allocation decisions will do
538 * something similar, although regular files will
539 * start at 2nd block group of the flexgroup. See
540 * ext4_ext_find_goal() and ext4_find_near().
543 for (i = 0; i < flex_size; i++) {
544 if (grp+i >= real_ngroups)
546 desc = ext4_get_group_desc(sb, grp+i, NULL);
547 if (desc && ext4_free_inodes_count(sb, desc)) {
555 max_dirs = ndirs / ngroups + inodes_per_group / 16;
556 min_inodes = avefreei - inodes_per_group*flex_size / 4;
559 min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
562 * Start looking in the flex group where we last allocated an
563 * inode for this parent directory
565 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
566 parent_group = EXT4_I(parent)->i_last_alloc_group;
568 parent_group >>= sbi->s_log_groups_per_flex;
571 for (i = 0; i < ngroups; i++) {
572 grp = (parent_group + i) % ngroups;
573 get_orlov_stats(sb, grp, flex_size, &stats);
574 if (stats.used_dirs >= max_dirs)
576 if (stats.free_inodes < min_inodes)
578 if (stats.free_blocks < min_blocks)
584 ngroups = real_ngroups;
585 avefreei = freei / ngroups;
587 parent_group = EXT4_I(parent)->i_block_group;
588 for (i = 0; i < ngroups; i++) {
589 grp = (parent_group + i) % ngroups;
590 desc = ext4_get_group_desc(sb, grp, NULL);
591 if (desc && ext4_free_inodes_count(sb, desc) &&
592 ext4_free_inodes_count(sb, desc) >= avefreei) {
600 * The free-inodes counter is approximate, and for really small
601 * filesystems the above test can fail to find any blockgroups
610 static int find_group_other(struct super_block *sb, struct inode *parent,
611 ext4_group_t *group, int mode)
613 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
614 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
615 struct ext4_group_desc *desc;
616 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
619 * Try to place the inode is the same flex group as its
620 * parent. If we can't find space, use the Orlov algorithm to
621 * find another flex group, and store that information in the
622 * parent directory's inode information so that use that flex
623 * group for future allocations.
629 parent_group &= ~(flex_size-1);
630 last = parent_group + flex_size;
633 for (i = parent_group; i < last; i++) {
634 desc = ext4_get_group_desc(sb, i, NULL);
635 if (desc && ext4_free_inodes_count(sb, desc)) {
640 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
642 parent_group = EXT4_I(parent)->i_last_alloc_group;
646 * If this didn't work, use the Orlov search algorithm
647 * to find a new flex group; we pass in the mode to
648 * avoid the topdir algorithms.
650 *group = parent_group + flex_size;
651 if (*group > ngroups)
653 return find_group_orlov(sb, parent, group, mode);
657 * Try to place the inode in its parent directory
659 *group = parent_group;
660 desc = ext4_get_group_desc(sb, *group, NULL);
661 if (desc && ext4_free_inodes_count(sb, desc) &&
662 ext4_free_blks_count(sb, desc))
666 * We're going to place this inode in a different blockgroup from its
667 * parent. We want to cause files in a common directory to all land in
668 * the same blockgroup. But we want files which are in a different
669 * directory which shares a blockgroup with our parent to land in a
670 * different blockgroup.
672 * So add our directory's i_ino into the starting point for the hash.
674 *group = (*group + parent->i_ino) % ngroups;
677 * Use a quadratic hash to find a group with a free inode and some free
680 for (i = 1; i < ngroups; i <<= 1) {
682 if (*group >= ngroups)
684 desc = ext4_get_group_desc(sb, *group, NULL);
685 if (desc && ext4_free_inodes_count(sb, desc) &&
686 ext4_free_blks_count(sb, desc))
691 * That failed: try linear search for a free inode, even if that group
692 * has no free blocks.
694 *group = parent_group;
695 for (i = 0; i < ngroups; i++) {
696 if (++*group >= ngroups)
698 desc = ext4_get_group_desc(sb, *group, NULL);
699 if (desc && ext4_free_inodes_count(sb, desc))
707 * claim the inode from the inode bitmap. If the group
708 * is uninit we need to take the groups's ext4_group_lock
709 * and clear the uninit flag. The inode bitmap update
710 * and group desc uninit flag clear should be done
711 * after holding ext4_group_lock so that ext4_read_inode_bitmap
712 * doesn't race with the ext4_claim_inode
714 static int ext4_claim_inode(struct super_block *sb,
715 struct buffer_head *inode_bitmap_bh,
716 unsigned long ino, ext4_group_t group, int mode)
718 int free = 0, retval = 0, count;
719 struct ext4_sb_info *sbi = EXT4_SB(sb);
720 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
722 ext4_lock_group(sb, group);
723 if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
724 /* not a free inode */
729 if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
730 ino > EXT4_INODES_PER_GROUP(sb)) {
731 ext4_unlock_group(sb, group);
732 ext4_error(sb, __func__,
733 "reserved inode or inode > inodes count - "
734 "block_group = %u, inode=%lu", group,
735 ino + group * EXT4_INODES_PER_GROUP(sb));
738 /* If we didn't allocate from within the initialized part of the inode
739 * table then we need to initialize up to this inode. */
740 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
742 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
743 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
744 /* When marking the block group with
745 * ~EXT4_BG_INODE_UNINIT we don't want to depend
746 * on the value of bg_itable_unused even though
747 * mke2fs could have initialized the same for us.
748 * Instead we calculated the value below
753 free = EXT4_INODES_PER_GROUP(sb) -
754 ext4_itable_unused_count(sb, gdp);
758 * Check the relative inode number against the last used
759 * relative inode number in this group. if it is greater
760 * we need to update the bg_itable_unused count
764 ext4_itable_unused_set(sb, gdp,
765 (EXT4_INODES_PER_GROUP(sb) - ino));
767 count = ext4_free_inodes_count(sb, gdp) - 1;
768 ext4_free_inodes_set(sb, gdp, count);
770 count = ext4_used_dirs_count(sb, gdp) + 1;
771 ext4_used_dirs_set(sb, gdp, count);
772 if (sbi->s_log_groups_per_flex) {
773 ext4_group_t f = ext4_flex_group(sbi, group);
775 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
778 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
780 ext4_unlock_group(sb, group);
785 * There are two policies for allocating an inode. If the new inode is
786 * a directory, then a forward search is made for a block group with both
787 * free space and a low directory-to-inode ratio; if that fails, then of
788 * the groups with above-average free space, that group with the fewest
789 * directories already is chosen.
791 * For other inodes, search forward from the parent directory's block
792 * group to find a free inode.
794 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode)
796 struct super_block *sb;
797 struct buffer_head *inode_bitmap_bh = NULL;
798 struct buffer_head *group_desc_bh;
799 ext4_group_t ngroups, group = 0;
800 unsigned long ino = 0;
802 struct ext4_group_desc *gdp = NULL;
803 struct ext4_inode_info *ei;
804 struct ext4_sb_info *sbi;
810 ext4_group_t flex_group;
812 /* Cannot create files in a deleted directory */
813 if (!dir || !dir->i_nlink)
814 return ERR_PTR(-EPERM);
817 ngroups = ext4_get_groups_count(sb);
818 trace_mark(ext4_request_inode, "dev %s dir %lu mode %d", sb->s_id,
820 inode = new_inode(sb);
822 return ERR_PTR(-ENOMEM);
826 if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
827 ret2 = find_group_flex(sb, dir, &group);
829 ret2 = find_group_other(sb, dir, &group, mode);
830 if (ret2 == 0 && once) {
832 printk(KERN_NOTICE "ext4: find_group_flex "
833 "failed, fallback succeeded dir %lu\n",
841 if (test_opt(sb, OLDALLOC))
842 ret2 = find_group_dir(sb, dir, &group);
844 ret2 = find_group_orlov(sb, dir, &group, mode);
846 ret2 = find_group_other(sb, dir, &group, mode);
849 EXT4_I(dir)->i_last_alloc_group = group;
854 for (i = 0; i < ngroups; i++) {
857 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
861 brelse(inode_bitmap_bh);
862 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
863 if (!inode_bitmap_bh)
868 repeat_in_this_group:
869 ino = ext4_find_next_zero_bit((unsigned long *)
870 inode_bitmap_bh->b_data,
871 EXT4_INODES_PER_GROUP(sb), ino);
873 if (ino < EXT4_INODES_PER_GROUP(sb)) {
875 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
876 err = ext4_journal_get_write_access(handle,
881 BUFFER_TRACE(group_desc_bh, "get_write_access");
882 err = ext4_journal_get_write_access(handle,
886 if (!ext4_claim_inode(sb, inode_bitmap_bh,
889 BUFFER_TRACE(inode_bitmap_bh,
890 "call ext4_handle_dirty_metadata");
891 err = ext4_handle_dirty_metadata(handle,
896 /* zero bit is inode number 1*/
901 ext4_handle_release_buffer(handle, inode_bitmap_bh);
902 ext4_handle_release_buffer(handle, group_desc_bh);
904 if (++ino < EXT4_INODES_PER_GROUP(sb))
905 goto repeat_in_this_group;
909 * This case is possible in concurrent environment. It is very
910 * rare. We cannot repeat the find_group_xxx() call because
911 * that will simply return the same blockgroup, because the
912 * group descriptor metadata has not yet been updated.
913 * So we just go onto the next blockgroup.
915 if (++group == ngroups)
922 /* We may have to initialize the block bitmap if it isn't already */
923 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
924 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
925 struct buffer_head *block_bitmap_bh;
927 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
928 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
929 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
931 brelse(block_bitmap_bh);
936 ext4_lock_group(sb, group);
937 /* recheck and clear flag under lock if we still need to */
938 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
939 free = ext4_free_blocks_after_init(sb, group, gdp);
940 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
941 ext4_free_blks_set(sb, gdp, free);
942 gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
945 ext4_unlock_group(sb, group);
947 /* Don't need to dirty bitmap block if we didn't change it */
949 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
950 err = ext4_handle_dirty_metadata(handle,
951 NULL, block_bitmap_bh);
954 brelse(block_bitmap_bh);
958 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
959 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
963 percpu_counter_dec(&sbi->s_freeinodes_counter);
965 percpu_counter_inc(&sbi->s_dirs_counter);
968 if (sbi->s_log_groups_per_flex) {
969 flex_group = ext4_flex_group(sbi, group);
970 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
973 inode->i_uid = current_fsuid();
974 if (test_opt(sb, GRPID))
975 inode->i_gid = dir->i_gid;
976 else if (dir->i_mode & S_ISGID) {
977 inode->i_gid = dir->i_gid;
981 inode->i_gid = current_fsgid();
982 inode->i_mode = mode;
984 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
985 /* This is the optimal IO size (for stat), not the fs block size */
987 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
988 ext4_current_time(inode);
990 memset(ei->i_data, 0, sizeof(ei->i_data));
991 ei->i_dir_start_lookup = 0;
995 * Don't inherit extent flag from directory, amongst others. We set
996 * extent flag on newly created directory and file only if -o extent
997 * mount option is specified
1000 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1003 ei->i_block_group = group;
1004 ei->i_last_alloc_group = ~0;
1006 ext4_set_inode_flags(inode);
1007 if (IS_DIRSYNC(inode))
1008 ext4_handle_sync(handle);
1009 if (insert_inode_locked(inode) < 0) {
1013 spin_lock(&sbi->s_next_gen_lock);
1014 inode->i_generation = sbi->s_next_generation++;
1015 spin_unlock(&sbi->s_next_gen_lock);
1017 ei->i_state = EXT4_STATE_NEW;
1019 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1022 if (vfs_dq_alloc_inode(inode)) {
1027 err = ext4_init_acl(handle, inode, dir);
1029 goto fail_free_drop;
1031 err = ext4_init_security(handle, inode, dir);
1033 goto fail_free_drop;
1035 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1036 /* set extent flag only for directory, file and normal symlink*/
1037 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1038 EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
1039 ext4_ext_tree_init(handle, inode);
1043 err = ext4_mark_inode_dirty(handle, inode);
1045 ext4_std_error(sb, err);
1046 goto fail_free_drop;
1049 ext4_debug("allocating inode %lu\n", inode->i_ino);
1050 trace_mark(ext4_allocate_inode, "dev %s ino %lu dir %lu mode %d",
1051 sb->s_id, inode->i_ino, dir->i_ino, mode);
1054 ext4_std_error(sb, err);
1059 brelse(inode_bitmap_bh);
1063 vfs_dq_free_inode(inode);
1067 inode->i_flags |= S_NOQUOTA;
1069 unlock_new_inode(inode);
1071 brelse(inode_bitmap_bh);
1072 return ERR_PTR(err);
1075 /* Verify that we are loading a valid orphan from disk */
1076 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1078 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1079 ext4_group_t block_group;
1081 struct buffer_head *bitmap_bh;
1082 struct inode *inode = NULL;
1085 /* Error cases - e2fsck has already cleaned up for us */
1086 if (ino > max_ino) {
1087 ext4_warning(sb, __func__,
1088 "bad orphan ino %lu! e2fsck was run?", ino);
1092 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1093 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1094 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1096 ext4_warning(sb, __func__,
1097 "inode bitmap error for orphan %lu", ino);
1101 /* Having the inode bit set should be a 100% indicator that this
1102 * is a valid orphan (no e2fsck run on fs). Orphans also include
1103 * inodes that were being truncated, so we can't check i_nlink==0.
1105 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1108 inode = ext4_iget(sb, ino);
1113 * If the orphans has i_nlinks > 0 then it should be able to be
1114 * truncated, otherwise it won't be removed from the orphan list
1115 * during processing and an infinite loop will result.
1117 if (inode->i_nlink && !ext4_can_truncate(inode))
1120 if (NEXT_ORPHAN(inode) > max_ino)
1126 err = PTR_ERR(inode);
1129 ext4_warning(sb, __func__,
1130 "bad orphan inode %lu! e2fsck was run?", ino);
1131 printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1132 bit, (unsigned long long)bitmap_bh->b_blocknr,
1133 ext4_test_bit(bit, bitmap_bh->b_data));
1134 printk(KERN_NOTICE "inode=%p\n", inode);
1136 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1137 is_bad_inode(inode));
1138 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1139 NEXT_ORPHAN(inode));
1140 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1141 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1142 /* Avoid freeing blocks if we got a bad deleted inode */
1143 if (inode->i_nlink == 0)
1144 inode->i_blocks = 0;
1149 return ERR_PTR(err);
1152 unsigned long ext4_count_free_inodes(struct super_block *sb)
1154 unsigned long desc_count;
1155 struct ext4_group_desc *gdp;
1156 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1158 struct ext4_super_block *es;
1159 unsigned long bitmap_count, x;
1160 struct buffer_head *bitmap_bh = NULL;
1162 es = EXT4_SB(sb)->s_es;
1166 for (i = 0; i < ngroups; i++) {
1167 gdp = ext4_get_group_desc(sb, i, NULL);
1170 desc_count += ext4_free_inodes_count(sb, gdp);
1172 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1176 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1177 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1178 i, ext4_free_inodes_count(sb, gdp), x);
1182 printk(KERN_DEBUG "ext4_count_free_inodes: "
1183 "stored = %u, computed = %lu, %lu\n",
1184 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1188 for (i = 0; i < ngroups; i++) {
1189 gdp = ext4_get_group_desc(sb, i, NULL);
1192 desc_count += ext4_free_inodes_count(sb, gdp);
1199 /* Called at mount-time, super-block is locked */
1200 unsigned long ext4_count_dirs(struct super_block * sb)
1202 unsigned long count = 0;
1203 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1205 for (i = 0; i < ngroups; i++) {
1206 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1209 count += ext4_used_dirs_count(sb, gdp);