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 J_ASSERT_BH(bh, buffer_locked(bh));
75 /* If checksum is bad mark all blocks and inodes use to prevent
76 * allocation, essentially implementing a per-group read-only flag. */
77 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
78 ext4_error(sb, "Checksum bad for group %u", block_group);
79 ext4_free_group_clusters_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);
83 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
84 EXT4_INODES_PER_GROUP(sb) / 8);
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,
91 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
92 EXT4_INODES_PER_GROUP(sb) / 8);
93 ext4_group_desc_csum_set(sb, block_group, gdp);
95 return EXT4_INODES_PER_GROUP(sb);
98 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
101 set_buffer_uptodate(bh);
102 set_bitmap_uptodate(bh);
109 * Read the inode allocation bitmap for a given block_group, reading
110 * into the specified slot in the superblock's bitmap cache.
112 * Return buffer_head of bitmap on success or NULL.
114 static struct buffer_head *
115 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
117 struct ext4_group_desc *desc;
118 struct buffer_head *bh = NULL;
119 ext4_fsblk_t bitmap_blk;
120 struct ext4_group_info *grp;
122 desc = ext4_get_group_desc(sb, block_group, NULL);
126 bitmap_blk = ext4_inode_bitmap(sb, desc);
127 bh = sb_getblk(sb, bitmap_blk);
129 ext4_error(sb, "Cannot read inode bitmap - "
130 "block_group = %u, inode_bitmap = %llu",
131 block_group, bitmap_blk);
134 if (bitmap_uptodate(bh))
138 if (bitmap_uptodate(bh)) {
143 ext4_lock_group(sb, block_group);
144 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
145 ext4_init_inode_bitmap(sb, bh, block_group, desc);
146 set_bitmap_uptodate(bh);
147 set_buffer_uptodate(bh);
148 set_buffer_verified(bh);
149 ext4_unlock_group(sb, block_group);
153 ext4_unlock_group(sb, block_group);
155 if (buffer_uptodate(bh)) {
157 * if not uninit if bh is uptodate,
158 * bitmap is also uptodate
160 set_bitmap_uptodate(bh);
165 * submit the buffer_head for reading
167 trace_ext4_load_inode_bitmap(sb, block_group);
168 bh->b_end_io = ext4_end_bitmap_read;
170 submit_bh(READ | REQ_META | REQ_PRIO, bh);
172 if (!buffer_uptodate(bh)) {
174 ext4_error(sb, "Cannot read inode bitmap - "
175 "block_group = %u, inode_bitmap = %llu",
176 block_group, bitmap_blk);
181 ext4_lock_group(sb, block_group);
182 if (!buffer_verified(bh) &&
183 !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
184 EXT4_INODES_PER_GROUP(sb) / 8)) {
185 ext4_unlock_group(sb, block_group);
187 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
188 "inode_bitmap = %llu", block_group, bitmap_blk);
189 grp = ext4_get_group_info(sb, block_group);
190 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
193 ext4_unlock_group(sb, block_group);
194 set_buffer_verified(bh);
199 * NOTE! When we get the inode, we're the only people
200 * that have access to it, and as such there are no
201 * race conditions we have to worry about. The inode
202 * is not on the hash-lists, and it cannot be reached
203 * through the filesystem because the directory entry
204 * has been deleted earlier.
206 * HOWEVER: we must make sure that we get no aliases,
207 * which means that we have to call "clear_inode()"
208 * _before_ we mark the inode not in use in the inode
209 * bitmaps. Otherwise a newly created file might use
210 * the same inode number (not actually the same pointer
211 * though), and then we'd have two inodes sharing the
212 * same inode number and space on the harddisk.
214 void ext4_free_inode(handle_t *handle, struct inode *inode)
216 struct super_block *sb = inode->i_sb;
219 struct buffer_head *bitmap_bh = NULL;
220 struct buffer_head *bh2;
221 ext4_group_t block_group;
223 struct ext4_group_desc *gdp;
224 struct ext4_super_block *es;
225 struct ext4_sb_info *sbi;
226 int fatal = 0, err, count, cleared;
227 struct ext4_group_info *grp;
230 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
231 "nonexistent device\n", __func__, __LINE__);
234 if (atomic_read(&inode->i_count) > 1) {
235 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
236 __func__, __LINE__, inode->i_ino,
237 atomic_read(&inode->i_count));
240 if (inode->i_nlink) {
241 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
242 __func__, __LINE__, inode->i_ino, inode->i_nlink);
248 ext4_debug("freeing inode %lu\n", ino);
249 trace_ext4_free_inode(inode);
252 * Note: we must free any quota before locking the superblock,
253 * as writing the quota to disk may need the lock as well.
255 dquot_initialize(inode);
256 ext4_xattr_delete_inode(handle, inode);
257 dquot_free_inode(inode);
260 is_directory = S_ISDIR(inode->i_mode);
262 /* Do this BEFORE marking the inode not in use or returning an error */
263 ext4_clear_inode(inode);
265 es = EXT4_SB(sb)->s_es;
266 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
267 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
270 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
271 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
272 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
273 /* Don't bother if the inode bitmap is corrupt. */
274 grp = ext4_get_group_info(sb, block_group);
275 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) || !bitmap_bh)
278 BUFFER_TRACE(bitmap_bh, "get_write_access");
279 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
284 gdp = ext4_get_group_desc(sb, block_group, &bh2);
286 BUFFER_TRACE(bh2, "get_write_access");
287 fatal = ext4_journal_get_write_access(handle, bh2);
289 ext4_lock_group(sb, block_group);
290 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
291 if (fatal || !cleared) {
292 ext4_unlock_group(sb, block_group);
296 count = ext4_free_inodes_count(sb, gdp) + 1;
297 ext4_free_inodes_set(sb, gdp, count);
299 count = ext4_used_dirs_count(sb, gdp) - 1;
300 ext4_used_dirs_set(sb, gdp, count);
301 percpu_counter_dec(&sbi->s_dirs_counter);
303 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
304 EXT4_INODES_PER_GROUP(sb) / 8);
305 ext4_group_desc_csum_set(sb, block_group, gdp);
306 ext4_unlock_group(sb, block_group);
308 percpu_counter_inc(&sbi->s_freeinodes_counter);
309 if (sbi->s_log_groups_per_flex) {
310 ext4_group_t f = ext4_flex_group(sbi, block_group);
312 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
314 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
316 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
317 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
320 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
321 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
325 ext4_error(sb, "bit already cleared for inode %lu", ino);
326 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
331 ext4_std_error(sb, fatal);
341 * Helper function for Orlov's allocator; returns critical information
342 * for a particular block group or flex_bg. If flex_size is 1, then g
343 * is a block group number; otherwise it is flex_bg number.
345 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
346 int flex_size, struct orlov_stats *stats)
348 struct ext4_group_desc *desc;
349 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
352 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
353 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
354 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
358 desc = ext4_get_group_desc(sb, g, NULL);
360 stats->free_inodes = ext4_free_inodes_count(sb, desc);
361 stats->free_clusters = ext4_free_group_clusters(sb, desc);
362 stats->used_dirs = ext4_used_dirs_count(sb, desc);
364 stats->free_inodes = 0;
365 stats->free_clusters = 0;
366 stats->used_dirs = 0;
371 * Orlov's allocator for directories.
373 * We always try to spread first-level directories.
375 * If there are blockgroups with both free inodes and free blocks counts
376 * not worse than average we return one with smallest directory count.
377 * Otherwise we simply return a random group.
379 * For the rest rules look so:
381 * It's OK to put directory into a group unless
382 * it has too many directories already (max_dirs) or
383 * it has too few free inodes left (min_inodes) or
384 * it has too few free blocks left (min_blocks) or
385 * Parent's group is preferred, if it doesn't satisfy these
386 * conditions we search cyclically through the rest. If none
387 * of the groups look good we just look for a group with more
388 * free inodes than average (starting at parent's group).
391 static int find_group_orlov(struct super_block *sb, struct inode *parent,
392 ext4_group_t *group, umode_t mode,
393 const struct qstr *qstr)
395 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
396 struct ext4_sb_info *sbi = EXT4_SB(sb);
397 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
398 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
399 unsigned int freei, avefreei, grp_free;
400 ext4_fsblk_t freeb, avefreec;
402 int max_dirs, min_inodes;
403 ext4_grpblk_t min_clusters;
404 ext4_group_t i, grp, g, ngroups;
405 struct ext4_group_desc *desc;
406 struct orlov_stats stats;
407 int flex_size = ext4_flex_bg_size(sbi);
408 struct dx_hash_info hinfo;
410 ngroups = real_ngroups;
412 ngroups = (real_ngroups + flex_size - 1) >>
413 sbi->s_log_groups_per_flex;
414 parent_group >>= sbi->s_log_groups_per_flex;
417 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
418 avefreei = freei / ngroups;
419 freeb = EXT4_C2B(sbi,
420 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
422 do_div(avefreec, ngroups);
423 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
426 ((parent == sb->s_root->d_inode) ||
427 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
428 int best_ndir = inodes_per_group;
432 hinfo.hash_version = DX_HASH_HALF_MD4;
433 hinfo.seed = sbi->s_hash_seed;
434 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
437 get_random_bytes(&grp, sizeof(grp));
438 parent_group = (unsigned)grp % ngroups;
439 for (i = 0; i < ngroups; i++) {
440 g = (parent_group + i) % ngroups;
441 get_orlov_stats(sb, g, flex_size, &stats);
442 if (!stats.free_inodes)
444 if (stats.used_dirs >= best_ndir)
446 if (stats.free_inodes < avefreei)
448 if (stats.free_clusters < avefreec)
452 best_ndir = stats.used_dirs;
457 if (flex_size == 1) {
463 * We pack inodes at the beginning of the flexgroup's
464 * inode tables. Block allocation decisions will do
465 * something similar, although regular files will
466 * start at 2nd block group of the flexgroup. See
467 * ext4_ext_find_goal() and ext4_find_near().
470 for (i = 0; i < flex_size; i++) {
471 if (grp+i >= real_ngroups)
473 desc = ext4_get_group_desc(sb, grp+i, NULL);
474 if (desc && ext4_free_inodes_count(sb, desc)) {
482 max_dirs = ndirs / ngroups + inodes_per_group / 16;
483 min_inodes = avefreei - inodes_per_group*flex_size / 4;
486 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
489 * Start looking in the flex group where we last allocated an
490 * inode for this parent directory
492 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
493 parent_group = EXT4_I(parent)->i_last_alloc_group;
495 parent_group >>= sbi->s_log_groups_per_flex;
498 for (i = 0; i < ngroups; i++) {
499 grp = (parent_group + i) % ngroups;
500 get_orlov_stats(sb, grp, flex_size, &stats);
501 if (stats.used_dirs >= max_dirs)
503 if (stats.free_inodes < min_inodes)
505 if (stats.free_clusters < min_clusters)
511 ngroups = real_ngroups;
512 avefreei = freei / ngroups;
514 parent_group = EXT4_I(parent)->i_block_group;
515 for (i = 0; i < ngroups; i++) {
516 grp = (parent_group + i) % ngroups;
517 desc = ext4_get_group_desc(sb, grp, NULL);
519 grp_free = ext4_free_inodes_count(sb, desc);
520 if (grp_free && grp_free >= avefreei) {
529 * The free-inodes counter is approximate, and for really small
530 * filesystems the above test can fail to find any blockgroups
539 static int find_group_other(struct super_block *sb, struct inode *parent,
540 ext4_group_t *group, umode_t mode)
542 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
543 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
544 struct ext4_group_desc *desc;
545 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
548 * Try to place the inode is the same flex group as its
549 * parent. If we can't find space, use the Orlov algorithm to
550 * find another flex group, and store that information in the
551 * parent directory's inode information so that use that flex
552 * group for future allocations.
558 parent_group &= ~(flex_size-1);
559 last = parent_group + flex_size;
562 for (i = parent_group; i < last; i++) {
563 desc = ext4_get_group_desc(sb, i, NULL);
564 if (desc && ext4_free_inodes_count(sb, desc)) {
569 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
571 parent_group = EXT4_I(parent)->i_last_alloc_group;
575 * If this didn't work, use the Orlov search algorithm
576 * to find a new flex group; we pass in the mode to
577 * avoid the topdir algorithms.
579 *group = parent_group + flex_size;
580 if (*group > ngroups)
582 return find_group_orlov(sb, parent, group, mode, NULL);
586 * Try to place the inode in its parent directory
588 *group = parent_group;
589 desc = ext4_get_group_desc(sb, *group, NULL);
590 if (desc && ext4_free_inodes_count(sb, desc) &&
591 ext4_free_group_clusters(sb, desc))
595 * We're going to place this inode in a different blockgroup from its
596 * parent. We want to cause files in a common directory to all land in
597 * the same blockgroup. But we want files which are in a different
598 * directory which shares a blockgroup with our parent to land in a
599 * different blockgroup.
601 * So add our directory's i_ino into the starting point for the hash.
603 *group = (*group + parent->i_ino) % ngroups;
606 * Use a quadratic hash to find a group with a free inode and some free
609 for (i = 1; i < ngroups; i <<= 1) {
611 if (*group >= ngroups)
613 desc = ext4_get_group_desc(sb, *group, NULL);
614 if (desc && ext4_free_inodes_count(sb, desc) &&
615 ext4_free_group_clusters(sb, desc))
620 * That failed: try linear search for a free inode, even if that group
621 * has no free blocks.
623 *group = parent_group;
624 for (i = 0; i < ngroups; i++) {
625 if (++*group >= ngroups)
627 desc = ext4_get_group_desc(sb, *group, NULL);
628 if (desc && ext4_free_inodes_count(sb, desc))
636 * In no journal mode, if an inode has recently been deleted, we want
637 * to avoid reusing it until we're reasonably sure the inode table
638 * block has been written back to disk. (Yes, these values are
639 * somewhat arbitrary...)
641 #define RECENTCY_MIN 5
642 #define RECENTCY_DIRTY 30
644 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
646 struct ext4_group_desc *gdp;
647 struct ext4_inode *raw_inode;
648 struct buffer_head *bh;
649 unsigned long dtime, now;
650 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
651 int offset, ret = 0, recentcy = RECENTCY_MIN;
653 gdp = ext4_get_group_desc(sb, group, NULL);
657 bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
658 (ino / inodes_per_block));
659 if (unlikely(!bh) || !buffer_uptodate(bh))
661 * If the block is not in the buffer cache, then it
662 * must have been written out.
666 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
667 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
668 dtime = le32_to_cpu(raw_inode->i_dtime);
670 if (buffer_dirty(bh))
671 recentcy += RECENTCY_DIRTY;
673 if (dtime && (dtime < now) && (now < dtime + recentcy))
681 * There are two policies for allocating an inode. If the new inode is
682 * a directory, then a forward search is made for a block group with both
683 * free space and a low directory-to-inode ratio; if that fails, then of
684 * the groups with above-average free space, that group with the fewest
685 * directories already is chosen.
687 * For other inodes, search forward from the parent directory's block
688 * group to find a free inode.
690 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
691 umode_t mode, const struct qstr *qstr,
692 __u32 goal, uid_t *owner, int handle_type,
693 unsigned int line_no, int nblocks)
695 struct super_block *sb;
696 struct buffer_head *inode_bitmap_bh = NULL;
697 struct buffer_head *group_desc_bh;
698 ext4_group_t ngroups, group = 0;
699 unsigned long ino = 0;
701 struct ext4_group_desc *gdp = NULL;
702 struct ext4_inode_info *ei;
703 struct ext4_sb_info *sbi;
707 ext4_group_t flex_group;
708 struct ext4_group_info *grp;
710 /* Cannot create files in a deleted directory */
711 if (!dir || !dir->i_nlink)
712 return ERR_PTR(-EPERM);
715 ngroups = ext4_get_groups_count(sb);
716 trace_ext4_request_inode(dir, mode);
717 inode = new_inode(sb);
719 return ERR_PTR(-ENOMEM);
724 * Initalize owners and quota early so that we don't have to account
725 * for quota initialization worst case in standard inode creating
729 inode->i_mode = mode;
730 i_uid_write(inode, owner[0]);
731 i_gid_write(inode, owner[1]);
732 } else if (test_opt(sb, GRPID)) {
733 inode->i_mode = mode;
734 inode->i_uid = current_fsuid();
735 inode->i_gid = dir->i_gid;
737 inode_init_owner(inode, dir, mode);
738 dquot_initialize(inode);
741 goal = sbi->s_inode_goal;
743 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
744 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
745 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
751 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
753 ret2 = find_group_other(sb, dir, &group, mode);
756 EXT4_I(dir)->i_last_alloc_group = group;
762 * Normally we will only go through one pass of this loop,
763 * unless we get unlucky and it turns out the group we selected
764 * had its last inode grabbed by someone else.
766 for (i = 0; i < ngroups; i++, ino = 0) {
769 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
774 * Check free inodes count before loading bitmap.
776 if (ext4_free_inodes_count(sb, gdp) == 0) {
777 if (++group == ngroups)
782 grp = ext4_get_group_info(sb, group);
783 /* Skip groups with already-known suspicious inode tables */
784 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
785 if (++group == ngroups)
790 brelse(inode_bitmap_bh);
791 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
792 /* Skip groups with suspicious inode tables */
793 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || !inode_bitmap_bh) {
794 if (++group == ngroups)
799 repeat_in_this_group:
800 ino = ext4_find_next_zero_bit((unsigned long *)
801 inode_bitmap_bh->b_data,
802 EXT4_INODES_PER_GROUP(sb), ino);
803 if (ino >= EXT4_INODES_PER_GROUP(sb))
805 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
806 ext4_error(sb, "reserved inode found cleared - "
807 "inode=%lu", ino + 1);
810 if ((EXT4_SB(sb)->s_journal == NULL) &&
811 recently_deleted(sb, group, ino)) {
816 BUG_ON(nblocks <= 0);
817 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
818 handle_type, nblocks,
820 if (IS_ERR(handle)) {
821 err = PTR_ERR(handle);
822 ext4_std_error(sb, err);
826 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
827 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
829 ext4_std_error(sb, err);
832 ext4_lock_group(sb, group);
833 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
834 ext4_unlock_group(sb, group);
835 ino++; /* the inode bitmap is zero-based */
837 goto got; /* we grabbed the inode! */
839 if (ino < EXT4_INODES_PER_GROUP(sb))
840 goto repeat_in_this_group;
842 if (++group == ngroups)
849 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
850 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
852 ext4_std_error(sb, err);
856 /* We may have to initialize the block bitmap if it isn't already */
857 if (ext4_has_group_desc_csum(sb) &&
858 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
859 struct buffer_head *block_bitmap_bh;
861 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
862 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
863 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
865 brelse(block_bitmap_bh);
866 ext4_std_error(sb, err);
870 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
871 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
873 /* recheck and clear flag under lock if we still need to */
874 ext4_lock_group(sb, group);
875 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
876 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
877 ext4_free_group_clusters_set(sb, gdp,
878 ext4_free_clusters_after_init(sb, group, gdp));
879 ext4_block_bitmap_csum_set(sb, group, gdp,
881 ext4_group_desc_csum_set(sb, group, gdp);
883 ext4_unlock_group(sb, group);
884 brelse(block_bitmap_bh);
887 ext4_std_error(sb, err);
892 BUFFER_TRACE(group_desc_bh, "get_write_access");
893 err = ext4_journal_get_write_access(handle, group_desc_bh);
895 ext4_std_error(sb, err);
899 /* Update the relevant bg descriptor fields */
900 if (ext4_has_group_desc_csum(sb)) {
902 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
904 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
905 ext4_lock_group(sb, group); /* while we modify the bg desc */
906 free = EXT4_INODES_PER_GROUP(sb) -
907 ext4_itable_unused_count(sb, gdp);
908 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
909 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
913 * Check the relative inode number against the last used
914 * relative inode number in this group. if it is greater
915 * we need to update the bg_itable_unused count
918 ext4_itable_unused_set(sb, gdp,
919 (EXT4_INODES_PER_GROUP(sb) - ino));
920 up_read(&grp->alloc_sem);
922 ext4_lock_group(sb, group);
925 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
927 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
928 if (sbi->s_log_groups_per_flex) {
929 ext4_group_t f = ext4_flex_group(sbi, group);
931 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
934 if (ext4_has_group_desc_csum(sb)) {
935 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
936 EXT4_INODES_PER_GROUP(sb) / 8);
937 ext4_group_desc_csum_set(sb, group, gdp);
939 ext4_unlock_group(sb, group);
941 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
942 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
944 ext4_std_error(sb, err);
948 percpu_counter_dec(&sbi->s_freeinodes_counter);
950 percpu_counter_inc(&sbi->s_dirs_counter);
952 if (sbi->s_log_groups_per_flex) {
953 flex_group = ext4_flex_group(sbi, group);
954 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
957 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
958 /* This is the optimal IO size (for stat), not the fs block size */
960 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
961 ext4_current_time(inode);
963 memset(ei->i_data, 0, sizeof(ei->i_data));
964 ei->i_dir_start_lookup = 0;
967 /* Don't inherit extent flag from directory, amongst others. */
969 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
972 ei->i_block_group = group;
973 ei->i_last_alloc_group = ~0;
975 ext4_set_inode_flags(inode);
976 if (IS_DIRSYNC(inode))
977 ext4_handle_sync(handle);
978 if (insert_inode_locked(inode) < 0) {
980 * Likely a bitmap corruption causing inode to be allocated
984 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
988 spin_lock(&sbi->s_next_gen_lock);
989 inode->i_generation = sbi->s_next_generation++;
990 spin_unlock(&sbi->s_next_gen_lock);
992 /* Precompute checksum seed for inode metadata */
993 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
994 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
996 __le32 inum = cpu_to_le32(inode->i_ino);
997 __le32 gen = cpu_to_le32(inode->i_generation);
998 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1000 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1004 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1005 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1007 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1009 ei->i_inline_off = 0;
1010 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA))
1011 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1014 err = dquot_alloc_inode(inode);
1018 err = ext4_init_acl(handle, inode, dir);
1020 goto fail_free_drop;
1022 err = ext4_init_security(handle, inode, dir, qstr);
1024 goto fail_free_drop;
1026 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1027 /* set extent flag only for directory, file and normal symlink*/
1028 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1029 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1030 ext4_ext_tree_init(handle, inode);
1034 if (ext4_handle_valid(handle)) {
1035 ei->i_sync_tid = handle->h_transaction->t_tid;
1036 ei->i_datasync_tid = handle->h_transaction->t_tid;
1039 err = ext4_mark_inode_dirty(handle, inode);
1041 ext4_std_error(sb, err);
1042 goto fail_free_drop;
1045 ext4_debug("allocating inode %lu\n", inode->i_ino);
1046 trace_ext4_allocate_inode(inode, dir, mode);
1047 brelse(inode_bitmap_bh);
1051 dquot_free_inode(inode);
1054 unlock_new_inode(inode);
1057 inode->i_flags |= S_NOQUOTA;
1059 brelse(inode_bitmap_bh);
1060 return ERR_PTR(err);
1063 /* Verify that we are loading a valid orphan from disk */
1064 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1066 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1067 ext4_group_t block_group;
1069 struct buffer_head *bitmap_bh;
1070 struct inode *inode = NULL;
1073 /* Error cases - e2fsck has already cleaned up for us */
1074 if (ino > max_ino) {
1075 ext4_warning(sb, "bad orphan ino %lu! e2fsck was run?", ino);
1079 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1080 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1081 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1083 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
1087 /* Having the inode bit set should be a 100% indicator that this
1088 * is a valid orphan (no e2fsck run on fs). Orphans also include
1089 * inodes that were being truncated, so we can't check i_nlink==0.
1091 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1094 inode = ext4_iget(sb, ino);
1099 * If the orphans has i_nlinks > 0 then it should be able to be
1100 * truncated, otherwise it won't be removed from the orphan list
1101 * during processing and an infinite loop will result.
1103 if (inode->i_nlink && !ext4_can_truncate(inode))
1106 if (NEXT_ORPHAN(inode) > max_ino)
1112 err = PTR_ERR(inode);
1115 ext4_warning(sb, "bad orphan inode %lu! e2fsck was run?", ino);
1116 printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1117 bit, (unsigned long long)bitmap_bh->b_blocknr,
1118 ext4_test_bit(bit, bitmap_bh->b_data));
1119 printk(KERN_WARNING "inode=%p\n", inode);
1121 printk(KERN_WARNING "is_bad_inode(inode)=%d\n",
1122 is_bad_inode(inode));
1123 printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n",
1124 NEXT_ORPHAN(inode));
1125 printk(KERN_WARNING "max_ino=%lu\n", max_ino);
1126 printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink);
1127 /* Avoid freeing blocks if we got a bad deleted inode */
1128 if (inode->i_nlink == 0)
1129 inode->i_blocks = 0;
1134 return ERR_PTR(err);
1137 unsigned long ext4_count_free_inodes(struct super_block *sb)
1139 unsigned long desc_count;
1140 struct ext4_group_desc *gdp;
1141 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1143 struct ext4_super_block *es;
1144 unsigned long bitmap_count, x;
1145 struct buffer_head *bitmap_bh = NULL;
1147 es = EXT4_SB(sb)->s_es;
1151 for (i = 0; i < ngroups; i++) {
1152 gdp = ext4_get_group_desc(sb, i, NULL);
1155 desc_count += ext4_free_inodes_count(sb, gdp);
1157 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1161 x = ext4_count_free(bitmap_bh->b_data,
1162 EXT4_INODES_PER_GROUP(sb) / 8);
1163 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1164 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1168 printk(KERN_DEBUG "ext4_count_free_inodes: "
1169 "stored = %u, computed = %lu, %lu\n",
1170 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1174 for (i = 0; i < ngroups; i++) {
1175 gdp = ext4_get_group_desc(sb, i, NULL);
1178 desc_count += ext4_free_inodes_count(sb, gdp);
1185 /* Called at mount-time, super-block is locked */
1186 unsigned long ext4_count_dirs(struct super_block * sb)
1188 unsigned long count = 0;
1189 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1191 for (i = 0; i < ngroups; i++) {
1192 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1195 count += ext4_used_dirs_count(sb, gdp);
1201 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1202 * inode table. Must be called without any spinlock held. The only place
1203 * where it is called from on active part of filesystem is ext4lazyinit
1204 * thread, so we do not need any special locks, however we have to prevent
1205 * inode allocation from the current group, so we take alloc_sem lock, to
1206 * block ext4_new_inode() until we are finished.
1208 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1211 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1212 struct ext4_sb_info *sbi = EXT4_SB(sb);
1213 struct ext4_group_desc *gdp = NULL;
1214 struct buffer_head *group_desc_bh;
1217 int num, ret = 0, used_blks = 0;
1219 /* This should not happen, but just to be sure check this */
1220 if (sb->s_flags & MS_RDONLY) {
1225 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1230 * We do not need to lock this, because we are the only one
1231 * handling this flag.
1233 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1236 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1237 if (IS_ERR(handle)) {
1238 ret = PTR_ERR(handle);
1242 down_write(&grp->alloc_sem);
1244 * If inode bitmap was already initialized there may be some
1245 * used inodes so we need to skip blocks with used inodes in
1248 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1249 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1250 ext4_itable_unused_count(sb, gdp)),
1251 sbi->s_inodes_per_block);
1253 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1254 ext4_error(sb, "Something is wrong with group %u: "
1255 "used itable blocks: %d; "
1256 "itable unused count: %u",
1258 ext4_itable_unused_count(sb, gdp));
1263 blk = ext4_inode_table(sb, gdp) + used_blks;
1264 num = sbi->s_itb_per_group - used_blks;
1266 BUFFER_TRACE(group_desc_bh, "get_write_access");
1267 ret = ext4_journal_get_write_access(handle,
1273 * Skip zeroout if the inode table is full. But we set the ZEROED
1274 * flag anyway, because obviously, when it is full it does not need
1277 if (unlikely(num == 0))
1280 ext4_debug("going to zero out inode table in group %d\n",
1282 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1286 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1289 ext4_lock_group(sb, group);
1290 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1291 ext4_group_desc_csum_set(sb, group, gdp);
1292 ext4_unlock_group(sb, group);
1294 BUFFER_TRACE(group_desc_bh,
1295 "call ext4_handle_dirty_metadata");
1296 ret = ext4_handle_dirty_metadata(handle, NULL,
1300 up_write(&grp->alloc_sem);
1301 ext4_journal_stop(handle);