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/stat.h>
18 #include <linux/string.h>
19 #include <linux/quotaops.h>
20 #include <linux/buffer_head.h>
21 #include <linux/random.h>
22 #include <linux/bitops.h>
23 #include <linux/blkdev.h>
24 #include <asm/byteorder.h>
27 #include "ext4_jbd2.h"
31 #include <trace/events/ext4.h>
34 * ialloc.c contains the inodes allocation and deallocation routines
38 * The free inodes are managed by bitmaps. A file system contains several
39 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
40 * block for inodes, N blocks for the inode table and data blocks.
42 * The file system contains group descriptors which are located after the
43 * super block. Each descriptor contains the number of the bitmap block and
44 * the free blocks count in the block.
48 * To avoid calling the atomic setbit hundreds or thousands of times, we only
49 * need to use it within a single byte (to ensure we get endianness right).
50 * We can use memset for the rest of the bitmap as there are no other users.
52 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
56 if (start_bit >= end_bit)
59 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
60 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
61 ext4_set_bit(i, bitmap);
63 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
66 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
69 set_buffer_uptodate(bh);
70 set_bitmap_uptodate(bh);
76 static int ext4_validate_inode_bitmap(struct super_block *sb,
77 struct ext4_group_desc *desc,
78 ext4_group_t block_group,
79 struct buffer_head *bh)
82 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
83 struct ext4_sb_info *sbi = EXT4_SB(sb);
85 if (buffer_verified(bh))
87 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
90 ext4_lock_group(sb, block_group);
91 if (buffer_verified(bh))
93 blk = ext4_inode_bitmap(sb, desc);
94 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
95 EXT4_INODES_PER_GROUP(sb) / 8)) {
96 ext4_unlock_group(sb, block_group);
97 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
98 "inode_bitmap = %llu", block_group, blk);
99 grp = ext4_get_group_info(sb, block_group);
100 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
102 count = ext4_free_inodes_count(sb, desc);
103 percpu_counter_sub(&sbi->s_freeinodes_counter,
106 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
109 set_buffer_verified(bh);
111 ext4_unlock_group(sb, block_group);
116 * Read the inode allocation bitmap for a given block_group, reading
117 * into the specified slot in the superblock's bitmap cache.
119 * Return buffer_head of bitmap on success or NULL.
121 static struct buffer_head *
122 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
124 struct ext4_group_desc *desc;
125 struct ext4_sb_info *sbi = EXT4_SB(sb);
126 struct buffer_head *bh = NULL;
127 ext4_fsblk_t bitmap_blk;
130 desc = ext4_get_group_desc(sb, block_group, NULL);
132 return ERR_PTR(-EFSCORRUPTED);
134 bitmap_blk = ext4_inode_bitmap(sb, desc);
135 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
136 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
137 ext4_error(sb, "Invalid inode bitmap blk %llu in "
138 "block_group %u", bitmap_blk, block_group);
139 return ERR_PTR(-EFSCORRUPTED);
141 bh = sb_getblk(sb, bitmap_blk);
143 ext4_error(sb, "Cannot read inode bitmap - "
144 "block_group = %u, inode_bitmap = %llu",
145 block_group, bitmap_blk);
146 return ERR_PTR(-EIO);
148 if (bitmap_uptodate(bh))
152 if (bitmap_uptodate(bh)) {
157 ext4_lock_group(sb, block_group);
158 if (ext4_has_group_desc_csum(sb) &&
159 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
160 if (block_group == 0) {
161 ext4_unlock_group(sb, block_group);
163 ext4_error(sb, "Inode bitmap for bg 0 marked "
168 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
169 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
170 sb->s_blocksize * 8, bh->b_data);
171 set_bitmap_uptodate(bh);
172 set_buffer_uptodate(bh);
173 set_buffer_verified(bh);
174 ext4_unlock_group(sb, block_group);
178 ext4_unlock_group(sb, block_group);
180 if (buffer_uptodate(bh)) {
182 * if not uninit if bh is uptodate,
183 * bitmap is also uptodate
185 set_bitmap_uptodate(bh);
190 * submit the buffer_head for reading
192 trace_ext4_load_inode_bitmap(sb, block_group);
193 bh->b_end_io = ext4_end_bitmap_read;
195 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
197 if (!buffer_uptodate(bh)) {
199 ext4_error(sb, "Cannot read inode bitmap - "
200 "block_group = %u, inode_bitmap = %llu",
201 block_group, bitmap_blk);
202 return ERR_PTR(-EIO);
206 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
216 * NOTE! When we get the inode, we're the only people
217 * that have access to it, and as such there are no
218 * race conditions we have to worry about. The inode
219 * is not on the hash-lists, and it cannot be reached
220 * through the filesystem because the directory entry
221 * has been deleted earlier.
223 * HOWEVER: we must make sure that we get no aliases,
224 * which means that we have to call "clear_inode()"
225 * _before_ we mark the inode not in use in the inode
226 * bitmaps. Otherwise a newly created file might use
227 * the same inode number (not actually the same pointer
228 * though), and then we'd have two inodes sharing the
229 * same inode number and space on the harddisk.
231 void ext4_free_inode(handle_t *handle, struct inode *inode)
233 struct super_block *sb = inode->i_sb;
236 struct buffer_head *bitmap_bh = NULL;
237 struct buffer_head *bh2;
238 ext4_group_t block_group;
240 struct ext4_group_desc *gdp;
241 struct ext4_super_block *es;
242 struct ext4_sb_info *sbi;
243 int fatal = 0, err, count, cleared;
244 struct ext4_group_info *grp;
247 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
248 "nonexistent device\n", __func__, __LINE__);
251 if (atomic_read(&inode->i_count) > 1) {
252 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
253 __func__, __LINE__, inode->i_ino,
254 atomic_read(&inode->i_count));
257 if (inode->i_nlink) {
258 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
259 __func__, __LINE__, inode->i_ino, inode->i_nlink);
265 ext4_debug("freeing inode %lu\n", ino);
266 trace_ext4_free_inode(inode);
269 * Note: we must free any quota before locking the superblock,
270 * as writing the quota to disk may need the lock as well.
272 dquot_initialize(inode);
273 ext4_xattr_delete_inode(handle, inode);
274 dquot_free_inode(inode);
277 is_directory = S_ISDIR(inode->i_mode);
279 /* Do this BEFORE marking the inode not in use or returning an error */
280 ext4_clear_inode(inode);
282 es = EXT4_SB(sb)->s_es;
283 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
284 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
287 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
288 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
289 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
290 /* Don't bother if the inode bitmap is corrupt. */
291 grp = ext4_get_group_info(sb, block_group);
292 if (IS_ERR(bitmap_bh)) {
293 fatal = PTR_ERR(bitmap_bh);
297 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
298 fatal = -EFSCORRUPTED;
302 BUFFER_TRACE(bitmap_bh, "get_write_access");
303 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
308 gdp = ext4_get_group_desc(sb, block_group, &bh2);
310 BUFFER_TRACE(bh2, "get_write_access");
311 fatal = ext4_journal_get_write_access(handle, bh2);
313 ext4_lock_group(sb, block_group);
314 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
315 if (fatal || !cleared) {
316 ext4_unlock_group(sb, block_group);
320 count = ext4_free_inodes_count(sb, gdp) + 1;
321 ext4_free_inodes_set(sb, gdp, count);
323 count = ext4_used_dirs_count(sb, gdp) - 1;
324 ext4_used_dirs_set(sb, gdp, count);
325 percpu_counter_dec(&sbi->s_dirs_counter);
327 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
328 EXT4_INODES_PER_GROUP(sb) / 8);
329 ext4_group_desc_csum_set(sb, block_group, gdp);
330 ext4_unlock_group(sb, block_group);
332 percpu_counter_inc(&sbi->s_freeinodes_counter);
333 if (sbi->s_log_groups_per_flex) {
334 struct flex_groups *fg;
336 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
337 ext4_flex_group(sbi, block_group));
338 atomic_inc(&fg->free_inodes);
340 atomic_dec(&fg->used_dirs);
342 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
343 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
346 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
347 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
351 ext4_error(sb, "bit already cleared for inode %lu", ino);
352 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
354 count = ext4_free_inodes_count(sb, gdp);
355 percpu_counter_sub(&sbi->s_freeinodes_counter,
358 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
363 ext4_std_error(sb, fatal);
373 * Helper function for Orlov's allocator; returns critical information
374 * for a particular block group or flex_bg. If flex_size is 1, then g
375 * is a block group number; otherwise it is flex_bg number.
377 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
378 int flex_size, struct orlov_stats *stats)
380 struct ext4_group_desc *desc;
383 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
385 stats->free_inodes = atomic_read(&fg->free_inodes);
386 stats->free_clusters = atomic64_read(&fg->free_clusters);
387 stats->used_dirs = atomic_read(&fg->used_dirs);
391 desc = ext4_get_group_desc(sb, g, NULL);
393 stats->free_inodes = ext4_free_inodes_count(sb, desc);
394 stats->free_clusters = ext4_free_group_clusters(sb, desc);
395 stats->used_dirs = ext4_used_dirs_count(sb, desc);
397 stats->free_inodes = 0;
398 stats->free_clusters = 0;
399 stats->used_dirs = 0;
404 * Orlov's allocator for directories.
406 * We always try to spread first-level directories.
408 * If there are blockgroups with both free inodes and free clusters counts
409 * not worse than average we return one with smallest directory count.
410 * Otherwise we simply return a random group.
412 * For the rest rules look so:
414 * It's OK to put directory into a group unless
415 * it has too many directories already (max_dirs) or
416 * it has too few free inodes left (min_inodes) or
417 * it has too few free clusters left (min_clusters) or
418 * Parent's group is preferred, if it doesn't satisfy these
419 * conditions we search cyclically through the rest. If none
420 * of the groups look good we just look for a group with more
421 * free inodes than average (starting at parent's group).
424 static int find_group_orlov(struct super_block *sb, struct inode *parent,
425 ext4_group_t *group, umode_t mode,
426 const struct qstr *qstr)
428 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
429 struct ext4_sb_info *sbi = EXT4_SB(sb);
430 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
431 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
432 unsigned int freei, avefreei, grp_free;
433 ext4_fsblk_t freec, avefreec;
435 int max_dirs, min_inodes;
436 ext4_grpblk_t min_clusters;
437 ext4_group_t i, grp, g, ngroups;
438 struct ext4_group_desc *desc;
439 struct orlov_stats stats;
440 int flex_size = ext4_flex_bg_size(sbi);
441 struct dx_hash_info hinfo;
443 ngroups = real_ngroups;
445 ngroups = (real_ngroups + flex_size - 1) >>
446 sbi->s_log_groups_per_flex;
447 parent_group >>= sbi->s_log_groups_per_flex;
450 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
451 avefreei = freei / ngroups;
452 freec = percpu_counter_read_positive(&sbi->s_freeclusters_counter);
454 do_div(avefreec, ngroups);
455 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
458 ((parent == d_inode(sb->s_root)) ||
459 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
460 int best_ndir = inodes_per_group;
464 hinfo.hash_version = DX_HASH_HALF_MD4;
465 hinfo.seed = sbi->s_hash_seed;
466 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
470 parent_group = (unsigned)grp % ngroups;
471 for (i = 0; i < ngroups; i++) {
472 g = (parent_group + i) % ngroups;
473 get_orlov_stats(sb, g, flex_size, &stats);
474 if (!stats.free_inodes)
476 if (stats.used_dirs >= best_ndir)
478 if (stats.free_inodes < avefreei)
480 if (stats.free_clusters < avefreec)
484 best_ndir = stats.used_dirs;
489 if (flex_size == 1) {
495 * We pack inodes at the beginning of the flexgroup's
496 * inode tables. Block allocation decisions will do
497 * something similar, although regular files will
498 * start at 2nd block group of the flexgroup. See
499 * ext4_ext_find_goal() and ext4_find_near().
502 for (i = 0; i < flex_size; i++) {
503 if (grp+i >= real_ngroups)
505 desc = ext4_get_group_desc(sb, grp+i, NULL);
506 if (desc && ext4_free_inodes_count(sb, desc)) {
514 max_dirs = ndirs / ngroups + inodes_per_group / 16;
515 min_inodes = avefreei - inodes_per_group*flex_size / 4;
518 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
521 * Start looking in the flex group where we last allocated an
522 * inode for this parent directory
524 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
525 parent_group = EXT4_I(parent)->i_last_alloc_group;
527 parent_group >>= sbi->s_log_groups_per_flex;
530 for (i = 0; i < ngroups; i++) {
531 grp = (parent_group + i) % ngroups;
532 get_orlov_stats(sb, grp, flex_size, &stats);
533 if (stats.used_dirs >= max_dirs)
535 if (stats.free_inodes < min_inodes)
537 if (stats.free_clusters < min_clusters)
543 ngroups = real_ngroups;
544 avefreei = freei / ngroups;
546 parent_group = EXT4_I(parent)->i_block_group;
547 for (i = 0; i < ngroups; i++) {
548 grp = (parent_group + i) % ngroups;
549 desc = ext4_get_group_desc(sb, grp, NULL);
551 grp_free = ext4_free_inodes_count(sb, desc);
552 if (grp_free && grp_free >= avefreei) {
561 * The free-inodes counter is approximate, and for really small
562 * filesystems the above test can fail to find any blockgroups
571 static int find_group_other(struct super_block *sb, struct inode *parent,
572 ext4_group_t *group, umode_t mode)
574 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
575 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
576 struct ext4_group_desc *desc;
577 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
580 * Try to place the inode is the same flex group as its
581 * parent. If we can't find space, use the Orlov algorithm to
582 * find another flex group, and store that information in the
583 * parent directory's inode information so that use that flex
584 * group for future allocations.
590 parent_group &= ~(flex_size-1);
591 last = parent_group + flex_size;
594 for (i = parent_group; i < last; i++) {
595 desc = ext4_get_group_desc(sb, i, NULL);
596 if (desc && ext4_free_inodes_count(sb, desc)) {
601 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
603 parent_group = EXT4_I(parent)->i_last_alloc_group;
607 * If this didn't work, use the Orlov search algorithm
608 * to find a new flex group; we pass in the mode to
609 * avoid the topdir algorithms.
611 *group = parent_group + flex_size;
612 if (*group > ngroups)
614 return find_group_orlov(sb, parent, group, mode, NULL);
618 * Try to place the inode in its parent directory
620 *group = parent_group;
621 desc = ext4_get_group_desc(sb, *group, NULL);
622 if (desc && ext4_free_inodes_count(sb, desc) &&
623 ext4_free_group_clusters(sb, desc))
627 * We're going to place this inode in a different blockgroup from its
628 * parent. We want to cause files in a common directory to all land in
629 * the same blockgroup. But we want files which are in a different
630 * directory which shares a blockgroup with our parent to land in a
631 * different blockgroup.
633 * So add our directory's i_ino into the starting point for the hash.
635 *group = (*group + parent->i_ino) % ngroups;
638 * Use a quadratic hash to find a group with a free inode and some free
641 for (i = 1; i < ngroups; i <<= 1) {
643 if (*group >= ngroups)
645 desc = ext4_get_group_desc(sb, *group, NULL);
646 if (desc && ext4_free_inodes_count(sb, desc) &&
647 ext4_free_group_clusters(sb, desc))
652 * That failed: try linear search for a free inode, even if that group
653 * has no free blocks.
655 *group = parent_group;
656 for (i = 0; i < ngroups; i++) {
657 if (++*group >= ngroups)
659 desc = ext4_get_group_desc(sb, *group, NULL);
660 if (desc && ext4_free_inodes_count(sb, desc))
668 * In no journal mode, if an inode has recently been deleted, we want
669 * to avoid reusing it until we're reasonably sure the inode table
670 * block has been written back to disk. (Yes, these values are
671 * somewhat arbitrary...)
673 #define RECENTCY_MIN 5
674 #define RECENTCY_DIRTY 30
676 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
678 struct ext4_group_desc *gdp;
679 struct ext4_inode *raw_inode;
680 struct buffer_head *bh;
681 unsigned long dtime, now;
682 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
683 int offset, ret = 0, recentcy = RECENTCY_MIN;
685 gdp = ext4_get_group_desc(sb, group, NULL);
689 bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
690 (ino / inodes_per_block));
691 if (unlikely(!bh) || !buffer_uptodate(bh))
693 * If the block is not in the buffer cache, then it
694 * must have been written out.
698 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
699 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
700 dtime = le32_to_cpu(raw_inode->i_dtime);
702 if (buffer_dirty(bh))
703 recentcy += RECENTCY_DIRTY;
705 if (dtime && (dtime < now) && (now < dtime + recentcy))
713 * There are two policies for allocating an inode. If the new inode is
714 * a directory, then a forward search is made for a block group with both
715 * free space and a low directory-to-inode ratio; if that fails, then of
716 * the groups with above-average free space, that group with the fewest
717 * directories already is chosen.
719 * For other inodes, search forward from the parent directory's block
720 * group to find a free inode.
722 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
723 umode_t mode, const struct qstr *qstr,
724 __u32 goal, uid_t *owner, int handle_type,
725 unsigned int line_no, int nblocks)
727 struct super_block *sb;
728 struct buffer_head *inode_bitmap_bh = NULL;
729 struct buffer_head *group_desc_bh;
730 ext4_group_t ngroups, group = 0;
731 unsigned long ino = 0;
733 struct ext4_group_desc *gdp = NULL;
734 struct ext4_inode_info *ei;
735 struct ext4_sb_info *sbi;
739 ext4_group_t flex_group;
740 struct ext4_group_info *grp;
743 /* Cannot create files in a deleted directory */
744 if (!dir || !dir->i_nlink)
745 return ERR_PTR(-EPERM);
747 if ((ext4_encrypted_inode(dir) ||
748 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
749 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
750 err = fscrypt_get_encryption_info(dir);
753 if (!fscrypt_has_encryption_key(dir))
754 return ERR_PTR(-ENOKEY);
756 nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
761 ngroups = ext4_get_groups_count(sb);
762 trace_ext4_request_inode(dir, mode);
763 inode = new_inode(sb);
765 return ERR_PTR(-ENOMEM);
770 * Initialize owners and quota early so that we don't have to account
771 * for quota initialization worst case in standard inode creating
775 inode->i_mode = mode;
776 i_uid_write(inode, owner[0]);
777 i_gid_write(inode, owner[1]);
778 } else if (test_opt(sb, GRPID)) {
779 inode->i_mode = mode;
780 inode->i_uid = current_fsuid();
781 inode->i_gid = dir->i_gid;
783 inode_init_owner(inode, dir, mode);
785 if (ext4_has_feature_project(sb) &&
786 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
787 ei->i_projid = EXT4_I(dir)->i_projid;
789 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
791 err = dquot_initialize(inode);
796 goal = sbi->s_inode_goal;
798 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
799 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
800 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
806 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
808 ret2 = find_group_other(sb, dir, &group, mode);
811 EXT4_I(dir)->i_last_alloc_group = group;
817 * Normally we will only go through one pass of this loop,
818 * unless we get unlucky and it turns out the group we selected
819 * had its last inode grabbed by someone else.
821 for (i = 0; i < ngroups; i++, ino = 0) {
824 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
829 * Check free inodes count before loading bitmap.
831 if (ext4_free_inodes_count(sb, gdp) == 0) {
832 if (++group == ngroups)
837 grp = ext4_get_group_info(sb, group);
838 /* Skip groups with already-known suspicious inode tables */
839 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
840 if (++group == ngroups)
845 brelse(inode_bitmap_bh);
846 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
847 /* Skip groups with suspicious inode tables */
848 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
849 IS_ERR(inode_bitmap_bh)) {
850 inode_bitmap_bh = NULL;
851 if (++group == ngroups)
856 repeat_in_this_group:
857 ino = ext4_find_next_zero_bit((unsigned long *)
858 inode_bitmap_bh->b_data,
859 EXT4_INODES_PER_GROUP(sb), ino);
860 if (ino >= EXT4_INODES_PER_GROUP(sb))
862 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
863 ext4_error(sb, "reserved inode found cleared - "
864 "inode=%lu", ino + 1);
867 if ((EXT4_SB(sb)->s_journal == NULL) &&
868 recently_deleted(sb, group, ino)) {
873 BUG_ON(nblocks <= 0);
874 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
875 handle_type, nblocks,
877 if (IS_ERR(handle)) {
878 err = PTR_ERR(handle);
879 ext4_std_error(sb, err);
883 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
884 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
886 ext4_std_error(sb, err);
889 ext4_lock_group(sb, group);
890 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
891 ext4_unlock_group(sb, group);
892 ino++; /* the inode bitmap is zero-based */
894 goto got; /* we grabbed the inode! */
896 if (ino < EXT4_INODES_PER_GROUP(sb))
897 goto repeat_in_this_group;
899 if (++group == ngroups)
906 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
907 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
909 ext4_std_error(sb, err);
913 BUFFER_TRACE(group_desc_bh, "get_write_access");
914 err = ext4_journal_get_write_access(handle, group_desc_bh);
916 ext4_std_error(sb, err);
920 /* We may have to initialize the block bitmap if it isn't already */
921 if (ext4_has_group_desc_csum(sb) &&
922 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
923 struct buffer_head *block_bitmap_bh;
925 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
926 if (IS_ERR(block_bitmap_bh)) {
927 err = PTR_ERR(block_bitmap_bh);
930 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
931 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
933 brelse(block_bitmap_bh);
934 ext4_std_error(sb, err);
938 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
939 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
941 /* recheck and clear flag under lock if we still need to */
942 ext4_lock_group(sb, group);
943 if (ext4_has_group_desc_csum(sb) &&
944 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
945 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
946 ext4_free_group_clusters_set(sb, gdp,
947 ext4_free_clusters_after_init(sb, group, gdp));
948 ext4_block_bitmap_csum_set(sb, group, gdp,
950 ext4_group_desc_csum_set(sb, group, gdp);
952 ext4_unlock_group(sb, group);
953 brelse(block_bitmap_bh);
956 ext4_std_error(sb, err);
961 /* Update the relevant bg descriptor fields */
962 if (ext4_has_group_desc_csum(sb)) {
964 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
966 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
967 ext4_lock_group(sb, group); /* while we modify the bg desc */
968 free = EXT4_INODES_PER_GROUP(sb) -
969 ext4_itable_unused_count(sb, gdp);
970 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
971 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
975 * Check the relative inode number against the last used
976 * relative inode number in this group. if it is greater
977 * we need to update the bg_itable_unused count
980 ext4_itable_unused_set(sb, gdp,
981 (EXT4_INODES_PER_GROUP(sb) - ino));
982 up_read(&grp->alloc_sem);
984 ext4_lock_group(sb, group);
987 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
989 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
990 if (sbi->s_log_groups_per_flex) {
991 ext4_group_t f = ext4_flex_group(sbi, group);
993 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
997 if (ext4_has_group_desc_csum(sb)) {
998 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
999 EXT4_INODES_PER_GROUP(sb) / 8);
1000 ext4_group_desc_csum_set(sb, group, gdp);
1002 ext4_unlock_group(sb, group);
1004 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1005 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1007 ext4_std_error(sb, err);
1011 percpu_counter_dec(&sbi->s_freeinodes_counter);
1013 percpu_counter_inc(&sbi->s_dirs_counter);
1015 if (sbi->s_log_groups_per_flex) {
1016 flex_group = ext4_flex_group(sbi, group);
1017 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1018 flex_group)->free_inodes);
1021 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1022 /* This is the optimal IO size (for stat), not the fs block size */
1023 inode->i_blocks = 0;
1024 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1025 ext4_current_time(inode);
1027 memset(ei->i_data, 0, sizeof(ei->i_data));
1028 ei->i_dir_start_lookup = 0;
1031 /* Don't inherit extent flag from directory, amongst others. */
1033 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1036 ei->i_block_group = group;
1037 ei->i_last_alloc_group = ~0;
1039 ext4_set_inode_flags(inode);
1040 if (IS_DIRSYNC(inode))
1041 ext4_handle_sync(handle);
1042 if (insert_inode_locked(inode) < 0) {
1044 * Likely a bitmap corruption causing inode to be allocated
1048 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1052 spin_lock(&sbi->s_next_gen_lock);
1053 inode->i_generation = sbi->s_next_generation++;
1054 spin_unlock(&sbi->s_next_gen_lock);
1056 /* Precompute checksum seed for inode metadata */
1057 if (ext4_has_metadata_csum(sb)) {
1059 __le32 inum = cpu_to_le32(inode->i_ino);
1060 __le32 gen = cpu_to_le32(inode->i_generation);
1061 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1063 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1067 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1068 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1070 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1071 ei->i_inline_off = 0;
1072 if (ext4_has_feature_inline_data(sb))
1073 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1075 err = dquot_alloc_inode(inode);
1079 err = ext4_init_acl(handle, inode, dir);
1081 goto fail_free_drop;
1083 err = ext4_init_security(handle, inode, dir, qstr);
1085 goto fail_free_drop;
1087 if (ext4_has_feature_extents(sb)) {
1088 /* set extent flag only for directory, file and normal symlink*/
1089 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1090 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1091 ext4_ext_tree_init(handle, inode);
1095 if (ext4_handle_valid(handle)) {
1096 ei->i_sync_tid = handle->h_transaction->t_tid;
1097 ei->i_datasync_tid = handle->h_transaction->t_tid;
1101 /* give pointer to avoid set_context with journal ops. */
1102 err = fscrypt_inherit_context(dir, inode, &encrypt, true);
1104 goto fail_free_drop;
1107 err = ext4_mark_inode_dirty(handle, inode);
1109 ext4_std_error(sb, err);
1110 goto fail_free_drop;
1113 ext4_debug("allocating inode %lu\n", inode->i_ino);
1114 trace_ext4_allocate_inode(inode, dir, mode);
1115 brelse(inode_bitmap_bh);
1119 dquot_free_inode(inode);
1122 unlock_new_inode(inode);
1125 inode->i_flags |= S_NOQUOTA;
1127 brelse(inode_bitmap_bh);
1128 return ERR_PTR(err);
1131 /* Verify that we are loading a valid orphan from disk */
1132 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1134 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1135 ext4_group_t block_group;
1137 struct buffer_head *bitmap_bh = NULL;
1138 struct inode *inode = NULL;
1139 int err = -EFSCORRUPTED;
1141 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1144 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1145 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1146 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1147 if (IS_ERR(bitmap_bh)) {
1148 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1149 ino, PTR_ERR(bitmap_bh));
1150 return (struct inode *) bitmap_bh;
1153 /* Having the inode bit set should be a 100% indicator that this
1154 * is a valid orphan (no e2fsck run on fs). Orphans also include
1155 * inodes that were being truncated, so we can't check i_nlink==0.
1157 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1160 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1161 if (IS_ERR(inode)) {
1162 err = PTR_ERR(inode);
1163 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1169 * If the orphans has i_nlinks > 0 then it should be able to
1170 * be truncated, otherwise it won't be removed from the orphan
1171 * list during processing and an infinite loop will result.
1172 * Similarly, it must not be a bad inode.
1174 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1175 is_bad_inode(inode))
1178 if (NEXT_ORPHAN(inode) > max_ino)
1184 ext4_error(sb, "bad orphan inode %lu", ino);
1186 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1187 bit, (unsigned long long)bitmap_bh->b_blocknr,
1188 ext4_test_bit(bit, bitmap_bh->b_data));
1190 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1191 is_bad_inode(inode));
1192 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1193 NEXT_ORPHAN(inode));
1194 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1195 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1196 /* Avoid freeing blocks if we got a bad deleted inode */
1197 if (inode->i_nlink == 0)
1198 inode->i_blocks = 0;
1202 return ERR_PTR(err);
1205 unsigned long ext4_count_free_inodes(struct super_block *sb)
1207 unsigned long desc_count;
1208 struct ext4_group_desc *gdp;
1209 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1211 struct ext4_super_block *es;
1212 unsigned long bitmap_count, x;
1213 struct buffer_head *bitmap_bh = NULL;
1215 es = EXT4_SB(sb)->s_es;
1219 for (i = 0; i < ngroups; i++) {
1220 gdp = ext4_get_group_desc(sb, i, NULL);
1223 desc_count += ext4_free_inodes_count(sb, gdp);
1225 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1226 if (IS_ERR(bitmap_bh)) {
1231 x = ext4_count_free(bitmap_bh->b_data,
1232 EXT4_INODES_PER_GROUP(sb) / 8);
1233 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1234 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1238 printk(KERN_DEBUG "ext4_count_free_inodes: "
1239 "stored = %u, computed = %lu, %lu\n",
1240 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1244 for (i = 0; i < ngroups; i++) {
1245 gdp = ext4_get_group_desc(sb, i, NULL);
1248 desc_count += ext4_free_inodes_count(sb, gdp);
1255 /* Called at mount-time, super-block is locked */
1256 unsigned long ext4_count_dirs(struct super_block * sb)
1258 unsigned long count = 0;
1259 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1261 for (i = 0; i < ngroups; i++) {
1262 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1265 count += ext4_used_dirs_count(sb, gdp);
1271 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1272 * inode table. Must be called without any spinlock held. The only place
1273 * where it is called from on active part of filesystem is ext4lazyinit
1274 * thread, so we do not need any special locks, however we have to prevent
1275 * inode allocation from the current group, so we take alloc_sem lock, to
1276 * block ext4_new_inode() until we are finished.
1278 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1281 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1282 struct ext4_sb_info *sbi = EXT4_SB(sb);
1283 struct ext4_group_desc *gdp = NULL;
1284 struct buffer_head *group_desc_bh;
1287 int num, ret = 0, used_blks = 0;
1288 unsigned long used_inos = 0;
1290 /* This should not happen, but just to be sure check this */
1291 if (sb->s_flags & MS_RDONLY) {
1296 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1301 * We do not need to lock this, because we are the only one
1302 * handling this flag.
1304 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1307 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1308 if (IS_ERR(handle)) {
1309 ret = PTR_ERR(handle);
1313 down_write(&grp->alloc_sem);
1315 * If inode bitmap was already initialized there may be some
1316 * used inodes so we need to skip blocks with used inodes in
1319 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1320 used_inos = EXT4_INODES_PER_GROUP(sb) -
1321 ext4_itable_unused_count(sb, gdp);
1322 used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1324 /* Bogus inode unused count? */
1325 if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1326 ext4_error(sb, "Something is wrong with group %u: "
1327 "used itable blocks: %d; "
1328 "itable unused count: %u",
1330 ext4_itable_unused_count(sb, gdp));
1335 used_inos += group * EXT4_INODES_PER_GROUP(sb);
1337 * Are there some uninitialized inodes in the inode table
1338 * before the first normal inode?
1340 if ((used_blks != sbi->s_itb_per_group) &&
1341 (used_inos < EXT4_FIRST_INO(sb))) {
1342 ext4_error(sb, "Something is wrong with group %u: "
1343 "itable unused count: %u; "
1344 "itables initialized count: %ld",
1345 group, ext4_itable_unused_count(sb, gdp),
1352 blk = ext4_inode_table(sb, gdp) + used_blks;
1353 num = sbi->s_itb_per_group - used_blks;
1355 BUFFER_TRACE(group_desc_bh, "get_write_access");
1356 ret = ext4_journal_get_write_access(handle,
1362 * Skip zeroout if the inode table is full. But we set the ZEROED
1363 * flag anyway, because obviously, when it is full it does not need
1366 if (unlikely(num == 0))
1369 ext4_debug("going to zero out inode table in group %d\n",
1371 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1375 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1378 ext4_lock_group(sb, group);
1379 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1380 ext4_group_desc_csum_set(sb, group, gdp);
1381 ext4_unlock_group(sb, group);
1383 BUFFER_TRACE(group_desc_bh,
1384 "call ext4_handle_dirty_metadata");
1385 ret = ext4_handle_dirty_metadata(handle, NULL,
1389 up_write(&grp->alloc_sem);
1390 ext4_journal_stop(handle);