1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/ialloc.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
10 * BSD ufs-inspired inode and directory allocation by
11 * Stephen Tweedie (sct@redhat.com), 1993
12 * Big-endian to little-endian byte-swapping/bitmaps by
13 * David S. Miller (davem@caip.rutgers.edu), 1995
16 #include <linux/time.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 <linux/cred.h>
27 #include <asm/byteorder.h>
30 #include "ext4_jbd2.h"
34 #include <trace/events/ext4.h>
37 * ialloc.c contains the inodes allocation and deallocation routines
41 * The free inodes are managed by bitmaps. A file system contains several
42 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
43 * block for inodes, N blocks for the inode table and data blocks.
45 * The file system contains group descriptors which are located after the
46 * super block. Each descriptor contains the number of the bitmap block and
47 * the free blocks count in the block.
51 * To avoid calling the atomic setbit hundreds or thousands of times, we only
52 * need to use it within a single byte (to ensure we get endianness right).
53 * We can use memset for the rest of the bitmap as there are no other users.
55 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
59 if (start_bit >= end_bit)
62 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
63 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
64 ext4_set_bit(i, bitmap);
66 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
69 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
72 set_buffer_uptodate(bh);
73 set_bitmap_uptodate(bh);
79 static int ext4_validate_inode_bitmap(struct super_block *sb,
80 struct ext4_group_desc *desc,
81 ext4_group_t block_group,
82 struct buffer_head *bh)
85 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
87 if (buffer_verified(bh))
89 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
92 ext4_lock_group(sb, block_group);
93 if (buffer_verified(bh))
95 blk = ext4_inode_bitmap(sb, desc);
96 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
97 EXT4_INODES_PER_GROUP(sb) / 8)) {
98 ext4_unlock_group(sb, block_group);
99 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
100 "inode_bitmap = %llu", block_group, blk);
101 ext4_mark_group_bitmap_corrupted(sb, block_group,
102 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
105 set_buffer_verified(bh);
107 ext4_unlock_group(sb, block_group);
112 * Read the inode allocation bitmap for a given block_group, reading
113 * into the specified slot in the superblock's bitmap cache.
115 * Return buffer_head of bitmap on success or NULL.
117 static struct buffer_head *
118 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
120 struct ext4_group_desc *desc;
121 struct ext4_sb_info *sbi = EXT4_SB(sb);
122 struct buffer_head *bh = NULL;
123 ext4_fsblk_t bitmap_blk;
126 desc = ext4_get_group_desc(sb, block_group, NULL);
128 return ERR_PTR(-EFSCORRUPTED);
130 bitmap_blk = ext4_inode_bitmap(sb, desc);
131 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
132 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
133 ext4_error(sb, "Invalid inode bitmap blk %llu in "
134 "block_group %u", bitmap_blk, block_group);
135 ext4_mark_group_bitmap_corrupted(sb, block_group,
136 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
137 return ERR_PTR(-EFSCORRUPTED);
139 bh = sb_getblk(sb, bitmap_blk);
141 ext4_warning(sb, "Cannot read inode bitmap - "
142 "block_group = %u, inode_bitmap = %llu",
143 block_group, bitmap_blk);
144 return ERR_PTR(-ENOMEM);
146 if (bitmap_uptodate(bh))
150 if (bitmap_uptodate(bh)) {
155 ext4_lock_group(sb, block_group);
156 if (ext4_has_group_desc_csum(sb) &&
157 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
158 if (block_group == 0) {
159 ext4_unlock_group(sb, block_group);
161 ext4_error(sb, "Inode bitmap for bg 0 marked "
166 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
167 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
168 sb->s_blocksize * 8, bh->b_data);
169 set_bitmap_uptodate(bh);
170 set_buffer_uptodate(bh);
171 set_buffer_verified(bh);
172 ext4_unlock_group(sb, block_group);
176 ext4_unlock_group(sb, block_group);
178 if (buffer_uptodate(bh)) {
180 * if not uninit if bh is uptodate,
181 * bitmap is also uptodate
183 set_bitmap_uptodate(bh);
188 * submit the buffer_head for reading
190 trace_ext4_load_inode_bitmap(sb, block_group);
191 bh->b_end_io = ext4_end_bitmap_read;
193 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
195 if (!buffer_uptodate(bh)) {
197 ext4_error(sb, "Cannot read inode bitmap - "
198 "block_group = %u, inode_bitmap = %llu",
199 block_group, bitmap_blk);
200 ext4_mark_group_bitmap_corrupted(sb, block_group,
201 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
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 dquot_free_inode(inode);
276 is_directory = S_ISDIR(inode->i_mode);
278 /* Do this BEFORE marking the inode not in use or returning an error */
279 ext4_clear_inode(inode);
282 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
283 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
286 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
287 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
288 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
289 /* Don't bother if the inode bitmap is corrupt. */
290 grp = ext4_get_group_info(sb, block_group);
291 if (IS_ERR(bitmap_bh)) {
292 fatal = PTR_ERR(bitmap_bh);
296 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
297 fatal = -EFSCORRUPTED;
301 BUFFER_TRACE(bitmap_bh, "get_write_access");
302 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
307 gdp = ext4_get_group_desc(sb, block_group, &bh2);
309 BUFFER_TRACE(bh2, "get_write_access");
310 fatal = ext4_journal_get_write_access(handle, bh2);
312 ext4_lock_group(sb, block_group);
313 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
314 if (fatal || !cleared) {
315 ext4_unlock_group(sb, block_group);
319 count = ext4_free_inodes_count(sb, gdp) + 1;
320 ext4_free_inodes_set(sb, gdp, count);
322 count = ext4_used_dirs_count(sb, gdp) - 1;
323 ext4_used_dirs_set(sb, gdp, count);
324 percpu_counter_dec(&sbi->s_dirs_counter);
326 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
327 EXT4_INODES_PER_GROUP(sb) / 8);
328 ext4_group_desc_csum_set(sb, block_group, gdp);
329 ext4_unlock_group(sb, block_group);
331 percpu_counter_inc(&sbi->s_freeinodes_counter);
332 if (sbi->s_log_groups_per_flex) {
333 struct flex_groups *fg;
335 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
336 ext4_flex_group(sbi, block_group));
337 atomic_inc(&fg->free_inodes);
339 atomic_dec(&fg->used_dirs);
341 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
342 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
345 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
346 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
350 ext4_error(sb, "bit already cleared for inode %lu", ino);
351 ext4_mark_group_bitmap_corrupted(sb, block_group,
352 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
357 ext4_std_error(sb, fatal);
367 * Helper function for Orlov's allocator; returns critical information
368 * for a particular block group or flex_bg. If flex_size is 1, then g
369 * is a block group number; otherwise it is flex_bg number.
371 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
372 int flex_size, struct orlov_stats *stats)
374 struct ext4_group_desc *desc;
377 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
379 stats->free_inodes = atomic_read(&fg->free_inodes);
380 stats->free_clusters = atomic64_read(&fg->free_clusters);
381 stats->used_dirs = atomic_read(&fg->used_dirs);
385 desc = ext4_get_group_desc(sb, g, NULL);
387 stats->free_inodes = ext4_free_inodes_count(sb, desc);
388 stats->free_clusters = ext4_free_group_clusters(sb, desc);
389 stats->used_dirs = ext4_used_dirs_count(sb, desc);
391 stats->free_inodes = 0;
392 stats->free_clusters = 0;
393 stats->used_dirs = 0;
398 * Orlov's allocator for directories.
400 * We always try to spread first-level directories.
402 * If there are blockgroups with both free inodes and free clusters counts
403 * not worse than average we return one with smallest directory count.
404 * Otherwise we simply return a random group.
406 * For the rest rules look so:
408 * It's OK to put directory into a group unless
409 * it has too many directories already (max_dirs) or
410 * it has too few free inodes left (min_inodes) or
411 * it has too few free clusters left (min_clusters) or
412 * Parent's group is preferred, if it doesn't satisfy these
413 * conditions we search cyclically through the rest. If none
414 * of the groups look good we just look for a group with more
415 * free inodes than average (starting at parent's group).
418 static int find_group_orlov(struct super_block *sb, struct inode *parent,
419 ext4_group_t *group, umode_t mode,
420 const struct qstr *qstr)
422 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
423 struct ext4_sb_info *sbi = EXT4_SB(sb);
424 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
425 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
426 unsigned int freei, avefreei, grp_free;
427 ext4_fsblk_t freec, avefreec;
429 int max_dirs, min_inodes;
430 ext4_grpblk_t min_clusters;
431 ext4_group_t i, grp, g, ngroups;
432 struct ext4_group_desc *desc;
433 struct orlov_stats stats;
434 int flex_size = ext4_flex_bg_size(sbi);
435 struct dx_hash_info hinfo;
437 ngroups = real_ngroups;
439 ngroups = (real_ngroups + flex_size - 1) >>
440 sbi->s_log_groups_per_flex;
441 parent_group >>= sbi->s_log_groups_per_flex;
444 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
445 avefreei = freei / ngroups;
446 freec = percpu_counter_read_positive(&sbi->s_freeclusters_counter);
448 do_div(avefreec, ngroups);
449 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
452 ((parent == d_inode(sb->s_root)) ||
453 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
454 int best_ndir = inodes_per_group;
458 hinfo.hash_version = DX_HASH_HALF_MD4;
459 hinfo.seed = sbi->s_hash_seed;
460 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
464 parent_group = (unsigned)grp % ngroups;
465 for (i = 0; i < ngroups; i++) {
466 g = (parent_group + i) % ngroups;
467 get_orlov_stats(sb, g, flex_size, &stats);
468 if (!stats.free_inodes)
470 if (stats.used_dirs >= best_ndir)
472 if (stats.free_inodes < avefreei)
474 if (stats.free_clusters < avefreec)
478 best_ndir = stats.used_dirs;
483 if (flex_size == 1) {
489 * We pack inodes at the beginning of the flexgroup's
490 * inode tables. Block allocation decisions will do
491 * something similar, although regular files will
492 * start at 2nd block group of the flexgroup. See
493 * ext4_ext_find_goal() and ext4_find_near().
496 for (i = 0; i < flex_size; i++) {
497 if (grp+i >= real_ngroups)
499 desc = ext4_get_group_desc(sb, grp+i, NULL);
500 if (desc && ext4_free_inodes_count(sb, desc)) {
508 max_dirs = ndirs / ngroups + inodes_per_group / 16;
509 min_inodes = avefreei - inodes_per_group*flex_size / 4;
512 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
515 * Start looking in the flex group where we last allocated an
516 * inode for this parent directory
518 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
519 parent_group = EXT4_I(parent)->i_last_alloc_group;
521 parent_group >>= sbi->s_log_groups_per_flex;
524 for (i = 0; i < ngroups; i++) {
525 grp = (parent_group + i) % ngroups;
526 get_orlov_stats(sb, grp, flex_size, &stats);
527 if (stats.used_dirs >= max_dirs)
529 if (stats.free_inodes < min_inodes)
531 if (stats.free_clusters < min_clusters)
537 ngroups = real_ngroups;
538 avefreei = freei / ngroups;
540 parent_group = EXT4_I(parent)->i_block_group;
541 for (i = 0; i < ngroups; i++) {
542 grp = (parent_group + i) % ngroups;
543 desc = ext4_get_group_desc(sb, grp, NULL);
545 grp_free = ext4_free_inodes_count(sb, desc);
546 if (grp_free && grp_free >= avefreei) {
555 * The free-inodes counter is approximate, and for really small
556 * filesystems the above test can fail to find any blockgroups
565 static int find_group_other(struct super_block *sb, struct inode *parent,
566 ext4_group_t *group, umode_t mode)
568 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
569 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
570 struct ext4_group_desc *desc;
571 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
574 * Try to place the inode is the same flex group as its
575 * parent. If we can't find space, use the Orlov algorithm to
576 * find another flex group, and store that information in the
577 * parent directory's inode information so that use that flex
578 * group for future allocations.
584 parent_group &= ~(flex_size-1);
585 last = parent_group + flex_size;
588 for (i = parent_group; i < last; i++) {
589 desc = ext4_get_group_desc(sb, i, NULL);
590 if (desc && ext4_free_inodes_count(sb, desc)) {
595 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
597 parent_group = EXT4_I(parent)->i_last_alloc_group;
601 * If this didn't work, use the Orlov search algorithm
602 * to find a new flex group; we pass in the mode to
603 * avoid the topdir algorithms.
605 *group = parent_group + flex_size;
606 if (*group > ngroups)
608 return find_group_orlov(sb, parent, group, mode, NULL);
612 * Try to place the inode in its parent directory
614 *group = parent_group;
615 desc = ext4_get_group_desc(sb, *group, NULL);
616 if (desc && ext4_free_inodes_count(sb, desc) &&
617 ext4_free_group_clusters(sb, desc))
621 * We're going to place this inode in a different blockgroup from its
622 * parent. We want to cause files in a common directory to all land in
623 * the same blockgroup. But we want files which are in a different
624 * directory which shares a blockgroup with our parent to land in a
625 * different blockgroup.
627 * So add our directory's i_ino into the starting point for the hash.
629 *group = (*group + parent->i_ino) % ngroups;
632 * Use a quadratic hash to find a group with a free inode and some free
635 for (i = 1; i < ngroups; i <<= 1) {
637 if (*group >= ngroups)
639 desc = ext4_get_group_desc(sb, *group, NULL);
640 if (desc && ext4_free_inodes_count(sb, desc) &&
641 ext4_free_group_clusters(sb, desc))
646 * That failed: try linear search for a free inode, even if that group
647 * has no free blocks.
649 *group = parent_group;
650 for (i = 0; i < ngroups; i++) {
651 if (++*group >= ngroups)
653 desc = ext4_get_group_desc(sb, *group, NULL);
654 if (desc && ext4_free_inodes_count(sb, desc))
662 * In no journal mode, if an inode has recently been deleted, we want
663 * to avoid reusing it until we're reasonably sure the inode table
664 * block has been written back to disk. (Yes, these values are
665 * somewhat arbitrary...)
667 #define RECENTCY_MIN 60
668 #define RECENTCY_DIRTY 300
670 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
672 struct ext4_group_desc *gdp;
673 struct ext4_inode *raw_inode;
674 struct buffer_head *bh;
675 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
677 int recentcy = RECENTCY_MIN;
680 gdp = ext4_get_group_desc(sb, group, NULL);
684 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
685 (ino / inodes_per_block));
686 if (!bh || !buffer_uptodate(bh))
688 * If the block is not in the buffer cache, then it
689 * must have been written out.
693 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
694 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
696 /* i_dtime is only 32 bits on disk, but we only care about relative
697 * times in the range of a few minutes (i.e. long enough to sync a
698 * recently-deleted inode to disk), so using the low 32 bits of the
699 * clock (a 68 year range) is enough, see time_before32() */
700 dtime = le32_to_cpu(raw_inode->i_dtime);
701 now = ktime_get_real_seconds();
702 if (buffer_dirty(bh))
703 recentcy += RECENTCY_DIRTY;
705 if (dtime && time_before32(dtime, now) &&
706 time_before32(now, dtime + recentcy))
713 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
714 struct buffer_head *bitmap, unsigned long *ino)
717 *ino = ext4_find_next_zero_bit((unsigned long *)
719 EXT4_INODES_PER_GROUP(sb), *ino);
720 if (*ino >= EXT4_INODES_PER_GROUP(sb))
723 if ((EXT4_SB(sb)->s_journal == NULL) &&
724 recently_deleted(sb, group, *ino)) {
726 if (*ino < EXT4_INODES_PER_GROUP(sb))
735 * There are two policies for allocating an inode. If the new inode is
736 * a directory, then a forward search is made for a block group with both
737 * free space and a low directory-to-inode ratio; if that fails, then of
738 * the groups with above-average free space, that group with the fewest
739 * directories already is chosen.
741 * For other inodes, search forward from the parent directory's block
742 * group to find a free inode.
744 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
745 umode_t mode, const struct qstr *qstr,
746 __u32 goal, uid_t *owner, __u32 i_flags,
747 int handle_type, unsigned int line_no,
750 struct super_block *sb;
751 struct buffer_head *inode_bitmap_bh = NULL;
752 struct buffer_head *group_desc_bh;
753 ext4_group_t ngroups, group = 0;
754 unsigned long ino = 0;
756 struct ext4_group_desc *gdp = NULL;
757 struct ext4_inode_info *ei;
758 struct ext4_sb_info *sbi;
762 ext4_group_t flex_group;
763 struct ext4_group_info *grp;
766 /* Cannot create files in a deleted directory */
767 if (!dir || !dir->i_nlink)
768 return ERR_PTR(-EPERM);
773 if (unlikely(ext4_forced_shutdown(sbi)))
774 return ERR_PTR(-EIO);
776 if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
777 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
778 !(i_flags & EXT4_EA_INODE_FL)) {
779 err = fscrypt_get_encryption_info(dir);
782 if (!fscrypt_has_encryption_key(dir))
783 return ERR_PTR(-ENOKEY);
787 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
788 #ifdef CONFIG_EXT4_FS_POSIX_ACL
789 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
794 int acl_size = p->a_count * sizeof(ext4_acl_entry);
796 nblocks += (S_ISDIR(mode) ? 2 : 1) *
797 __ext4_xattr_set_credits(sb, NULL /* inode */,
798 NULL /* block_bh */, acl_size,
799 true /* is_create */);
800 posix_acl_release(p);
804 #ifdef CONFIG_SECURITY
806 int num_security_xattrs = 1;
808 #ifdef CONFIG_INTEGRITY
809 num_security_xattrs++;
812 * We assume that security xattrs are never
813 * more than 1k. In practice they are under
816 nblocks += num_security_xattrs *
817 __ext4_xattr_set_credits(sb, NULL /* inode */,
818 NULL /* block_bh */, 1024,
819 true /* is_create */);
823 nblocks += __ext4_xattr_set_credits(sb,
824 NULL /* inode */, NULL /* block_bh */,
825 FSCRYPT_SET_CONTEXT_MAX_SIZE,
826 true /* is_create */);
829 ngroups = ext4_get_groups_count(sb);
830 trace_ext4_request_inode(dir, mode);
831 inode = new_inode(sb);
833 return ERR_PTR(-ENOMEM);
837 * Initialize owners and quota early so that we don't have to account
838 * for quota initialization worst case in standard inode creating
842 inode->i_mode = mode;
843 i_uid_write(inode, owner[0]);
844 i_gid_write(inode, owner[1]);
845 } else if (test_opt(sb, GRPID)) {
846 inode->i_mode = mode;
847 inode->i_uid = current_fsuid();
848 inode->i_gid = dir->i_gid;
850 inode_init_owner(inode, dir, mode);
852 if (ext4_has_feature_project(sb) &&
853 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
854 ei->i_projid = EXT4_I(dir)->i_projid;
856 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
858 err = dquot_initialize(inode);
863 goal = sbi->s_inode_goal;
865 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
866 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
867 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
873 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
875 ret2 = find_group_other(sb, dir, &group, mode);
878 EXT4_I(dir)->i_last_alloc_group = group;
884 * Normally we will only go through one pass of this loop,
885 * unless we get unlucky and it turns out the group we selected
886 * had its last inode grabbed by someone else.
888 for (i = 0; i < ngroups; i++, ino = 0) {
891 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
896 * Check free inodes count before loading bitmap.
898 if (ext4_free_inodes_count(sb, gdp) == 0)
901 grp = ext4_get_group_info(sb, group);
902 /* Skip groups with already-known suspicious inode tables */
903 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
906 brelse(inode_bitmap_bh);
907 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
908 /* Skip groups with suspicious inode tables */
909 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
910 IS_ERR(inode_bitmap_bh)) {
911 inode_bitmap_bh = NULL;
915 repeat_in_this_group:
916 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
920 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
921 ext4_error(sb, "reserved inode found cleared - "
922 "inode=%lu", ino + 1);
923 ext4_mark_group_bitmap_corrupted(sb, group,
924 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
929 BUG_ON(nblocks <= 0);
930 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
931 handle_type, nblocks,
933 if (IS_ERR(handle)) {
934 err = PTR_ERR(handle);
935 ext4_std_error(sb, err);
939 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
940 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
942 ext4_std_error(sb, err);
945 ext4_lock_group(sb, group);
946 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
948 /* Someone already took the bit. Repeat the search
951 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
953 ext4_set_bit(ino, inode_bitmap_bh->b_data);
956 ret2 = 1; /* we didn't grab the inode */
959 ext4_unlock_group(sb, group);
960 ino++; /* the inode bitmap is zero-based */
962 goto got; /* we grabbed the inode! */
964 if (ino < EXT4_INODES_PER_GROUP(sb))
965 goto repeat_in_this_group;
967 if (++group == ngroups)
974 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
975 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
977 ext4_std_error(sb, err);
981 BUFFER_TRACE(group_desc_bh, "get_write_access");
982 err = ext4_journal_get_write_access(handle, group_desc_bh);
984 ext4_std_error(sb, err);
988 /* We may have to initialize the block bitmap if it isn't already */
989 if (ext4_has_group_desc_csum(sb) &&
990 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
991 struct buffer_head *block_bitmap_bh;
993 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
994 if (IS_ERR(block_bitmap_bh)) {
995 err = PTR_ERR(block_bitmap_bh);
998 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
999 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1001 brelse(block_bitmap_bh);
1002 ext4_std_error(sb, err);
1006 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1007 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1009 /* recheck and clear flag under lock if we still need to */
1010 ext4_lock_group(sb, group);
1011 if (ext4_has_group_desc_csum(sb) &&
1012 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1013 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1014 ext4_free_group_clusters_set(sb, gdp,
1015 ext4_free_clusters_after_init(sb, group, gdp));
1016 ext4_block_bitmap_csum_set(sb, group, gdp,
1018 ext4_group_desc_csum_set(sb, group, gdp);
1020 ext4_unlock_group(sb, group);
1021 brelse(block_bitmap_bh);
1024 ext4_std_error(sb, err);
1029 /* Update the relevant bg descriptor fields */
1030 if (ext4_has_group_desc_csum(sb)) {
1032 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1034 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1035 ext4_lock_group(sb, group); /* while we modify the bg desc */
1036 free = EXT4_INODES_PER_GROUP(sb) -
1037 ext4_itable_unused_count(sb, gdp);
1038 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1039 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1043 * Check the relative inode number against the last used
1044 * relative inode number in this group. if it is greater
1045 * we need to update the bg_itable_unused count
1048 ext4_itable_unused_set(sb, gdp,
1049 (EXT4_INODES_PER_GROUP(sb) - ino));
1050 up_read(&grp->alloc_sem);
1052 ext4_lock_group(sb, group);
1055 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1056 if (S_ISDIR(mode)) {
1057 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1058 if (sbi->s_log_groups_per_flex) {
1059 ext4_group_t f = ext4_flex_group(sbi, group);
1061 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1065 if (ext4_has_group_desc_csum(sb)) {
1066 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1067 EXT4_INODES_PER_GROUP(sb) / 8);
1068 ext4_group_desc_csum_set(sb, group, gdp);
1070 ext4_unlock_group(sb, group);
1072 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1073 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1075 ext4_std_error(sb, err);
1079 percpu_counter_dec(&sbi->s_freeinodes_counter);
1081 percpu_counter_inc(&sbi->s_dirs_counter);
1083 if (sbi->s_log_groups_per_flex) {
1084 flex_group = ext4_flex_group(sbi, group);
1085 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1086 flex_group)->free_inodes);
1089 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1090 /* This is the optimal IO size (for stat), not the fs block size */
1091 inode->i_blocks = 0;
1092 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1093 ei->i_crtime = inode->i_mtime;
1095 memset(ei->i_data, 0, sizeof(ei->i_data));
1096 ei->i_dir_start_lookup = 0;
1099 /* Don't inherit extent flag from directory, amongst others. */
1101 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1102 ei->i_flags |= i_flags;
1105 ei->i_block_group = group;
1106 ei->i_last_alloc_group = ~0;
1108 ext4_set_inode_flags(inode);
1109 if (IS_DIRSYNC(inode))
1110 ext4_handle_sync(handle);
1111 if (insert_inode_locked(inode) < 0) {
1113 * Likely a bitmap corruption causing inode to be allocated
1117 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1119 ext4_mark_group_bitmap_corrupted(sb, group,
1120 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1123 inode->i_generation = prandom_u32();
1125 /* Precompute checksum seed for inode metadata */
1126 if (ext4_has_metadata_csum(sb)) {
1128 __le32 inum = cpu_to_le32(inode->i_ino);
1129 __le32 gen = cpu_to_le32(inode->i_generation);
1130 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1132 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1136 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1137 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1139 ei->i_extra_isize = sbi->s_want_extra_isize;
1140 ei->i_inline_off = 0;
1141 if (ext4_has_feature_inline_data(sb))
1142 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1144 err = dquot_alloc_inode(inode);
1149 * Since the encryption xattr will always be unique, create it first so
1150 * that it's less likely to end up in an external xattr block and
1151 * prevent its deduplication.
1154 err = fscrypt_inherit_context(dir, inode, handle, true);
1156 goto fail_free_drop;
1159 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1160 err = ext4_init_acl(handle, inode, dir);
1162 goto fail_free_drop;
1164 err = ext4_init_security(handle, inode, dir, qstr);
1166 goto fail_free_drop;
1169 if (ext4_has_feature_extents(sb)) {
1170 /* set extent flag only for directory, file and normal symlink*/
1171 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1172 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1173 ext4_ext_tree_init(handle, inode);
1177 if (ext4_handle_valid(handle)) {
1178 ei->i_sync_tid = handle->h_transaction->t_tid;
1179 ei->i_datasync_tid = handle->h_transaction->t_tid;
1182 err = ext4_mark_inode_dirty(handle, inode);
1184 ext4_std_error(sb, err);
1185 goto fail_free_drop;
1188 ext4_debug("allocating inode %lu\n", inode->i_ino);
1189 trace_ext4_allocate_inode(inode, dir, mode);
1190 brelse(inode_bitmap_bh);
1194 dquot_free_inode(inode);
1197 unlock_new_inode(inode);
1200 inode->i_flags |= S_NOQUOTA;
1202 brelse(inode_bitmap_bh);
1203 return ERR_PTR(err);
1206 /* Verify that we are loading a valid orphan from disk */
1207 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1209 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1210 ext4_group_t block_group;
1212 struct buffer_head *bitmap_bh = NULL;
1213 struct inode *inode = NULL;
1214 int err = -EFSCORRUPTED;
1216 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1219 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1220 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1221 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1222 if (IS_ERR(bitmap_bh))
1223 return (struct inode *) bitmap_bh;
1225 /* Having the inode bit set should be a 100% indicator that this
1226 * is a valid orphan (no e2fsck run on fs). Orphans also include
1227 * inodes that were being truncated, so we can't check i_nlink==0.
1229 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1232 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1233 if (IS_ERR(inode)) {
1234 err = PTR_ERR(inode);
1235 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1241 * If the orphans has i_nlinks > 0 then it should be able to
1242 * be truncated, otherwise it won't be removed from the orphan
1243 * list during processing and an infinite loop will result.
1244 * Similarly, it must not be a bad inode.
1246 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1247 is_bad_inode(inode))
1250 if (NEXT_ORPHAN(inode) > max_ino)
1256 ext4_error(sb, "bad orphan inode %lu", ino);
1258 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1259 bit, (unsigned long long)bitmap_bh->b_blocknr,
1260 ext4_test_bit(bit, bitmap_bh->b_data));
1262 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1263 is_bad_inode(inode));
1264 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1265 NEXT_ORPHAN(inode));
1266 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1267 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1268 /* Avoid freeing blocks if we got a bad deleted inode */
1269 if (inode->i_nlink == 0)
1270 inode->i_blocks = 0;
1274 return ERR_PTR(err);
1277 unsigned long ext4_count_free_inodes(struct super_block *sb)
1279 unsigned long desc_count;
1280 struct ext4_group_desc *gdp;
1281 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1283 struct ext4_super_block *es;
1284 unsigned long bitmap_count, x;
1285 struct buffer_head *bitmap_bh = NULL;
1287 es = EXT4_SB(sb)->s_es;
1291 for (i = 0; i < ngroups; i++) {
1292 gdp = ext4_get_group_desc(sb, i, NULL);
1295 desc_count += ext4_free_inodes_count(sb, gdp);
1297 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1298 if (IS_ERR(bitmap_bh)) {
1303 x = ext4_count_free(bitmap_bh->b_data,
1304 EXT4_INODES_PER_GROUP(sb) / 8);
1305 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1306 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1310 printk(KERN_DEBUG "ext4_count_free_inodes: "
1311 "stored = %u, computed = %lu, %lu\n",
1312 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1316 for (i = 0; i < ngroups; i++) {
1317 gdp = ext4_get_group_desc(sb, i, NULL);
1320 desc_count += ext4_free_inodes_count(sb, gdp);
1327 /* Called at mount-time, super-block is locked */
1328 unsigned long ext4_count_dirs(struct super_block * sb)
1330 unsigned long count = 0;
1331 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1333 for (i = 0; i < ngroups; i++) {
1334 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1337 count += ext4_used_dirs_count(sb, gdp);
1343 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1344 * inode table. Must be called without any spinlock held. The only place
1345 * where it is called from on active part of filesystem is ext4lazyinit
1346 * thread, so we do not need any special locks, however we have to prevent
1347 * inode allocation from the current group, so we take alloc_sem lock, to
1348 * block ext4_new_inode() until we are finished.
1350 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1353 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1354 struct ext4_sb_info *sbi = EXT4_SB(sb);
1355 struct ext4_group_desc *gdp = NULL;
1356 struct buffer_head *group_desc_bh;
1359 int num, ret = 0, used_blks = 0;
1360 unsigned long used_inos = 0;
1362 /* This should not happen, but just to be sure check this */
1363 if (sb_rdonly(sb)) {
1368 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1373 * We do not need to lock this, because we are the only one
1374 * handling this flag.
1376 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1379 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1380 if (IS_ERR(handle)) {
1381 ret = PTR_ERR(handle);
1385 down_write(&grp->alloc_sem);
1387 * If inode bitmap was already initialized there may be some
1388 * used inodes so we need to skip blocks with used inodes in
1391 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1392 used_inos = EXT4_INODES_PER_GROUP(sb) -
1393 ext4_itable_unused_count(sb, gdp);
1394 used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1396 /* Bogus inode unused count? */
1397 if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1398 ext4_error(sb, "Something is wrong with group %u: "
1399 "used itable blocks: %d; "
1400 "itable unused count: %u",
1402 ext4_itable_unused_count(sb, gdp));
1407 used_inos += group * EXT4_INODES_PER_GROUP(sb);
1409 * Are there some uninitialized inodes in the inode table
1410 * before the first normal inode?
1412 if ((used_blks != sbi->s_itb_per_group) &&
1413 (used_inos < EXT4_FIRST_INO(sb))) {
1414 ext4_error(sb, "Something is wrong with group %u: "
1415 "itable unused count: %u; "
1416 "itables initialized count: %ld",
1417 group, ext4_itable_unused_count(sb, gdp),
1424 blk = ext4_inode_table(sb, gdp) + used_blks;
1425 num = sbi->s_itb_per_group - used_blks;
1427 BUFFER_TRACE(group_desc_bh, "get_write_access");
1428 ret = ext4_journal_get_write_access(handle,
1434 * Skip zeroout if the inode table is full. But we set the ZEROED
1435 * flag anyway, because obviously, when it is full it does not need
1438 if (unlikely(num == 0))
1441 ext4_debug("going to zero out inode table in group %d\n",
1443 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1447 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1450 ext4_lock_group(sb, group);
1451 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1452 ext4_group_desc_csum_set(sb, group, gdp);
1453 ext4_unlock_group(sb, group);
1455 BUFFER_TRACE(group_desc_bh,
1456 "call ext4_handle_dirty_metadata");
1457 ret = ext4_handle_dirty_metadata(handle, NULL,
1461 up_write(&grp->alloc_sem);
1462 ext4_journal_stop(handle);