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 /* Initializes an uninitialized inode bitmap */
67 static int ext4_init_inode_bitmap(struct super_block *sb,
68 struct buffer_head *bh,
69 ext4_group_t block_group,
70 struct ext4_group_desc *gdp)
72 struct ext4_group_info *grp;
73 struct ext4_sb_info *sbi = EXT4_SB(sb);
74 J_ASSERT_BH(bh, buffer_locked(bh));
76 /* If checksum is bad mark all blocks and inodes use to prevent
77 * allocation, essentially implementing a per-group read-only flag. */
78 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
79 grp = ext4_get_group_info(sb, block_group);
80 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
81 percpu_counter_sub(&sbi->s_freeclusters_counter,
83 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
84 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
86 count = ext4_free_inodes_count(sb, gdp);
87 percpu_counter_sub(&sbi->s_freeinodes_counter,
90 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
94 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
95 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
97 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
98 EXT4_INODES_PER_GROUP(sb) / 8);
99 ext4_group_desc_csum_set(sb, block_group, gdp);
104 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
107 set_buffer_uptodate(bh);
108 set_bitmap_uptodate(bh);
114 static int ext4_validate_inode_bitmap(struct super_block *sb,
115 struct ext4_group_desc *desc,
116 ext4_group_t block_group,
117 struct buffer_head *bh)
120 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
121 struct ext4_sb_info *sbi = EXT4_SB(sb);
123 if (buffer_verified(bh))
125 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
126 return -EFSCORRUPTED;
128 ext4_lock_group(sb, block_group);
129 blk = ext4_inode_bitmap(sb, desc);
130 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
131 EXT4_INODES_PER_GROUP(sb) / 8)) {
132 ext4_unlock_group(sb, block_group);
133 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
134 "inode_bitmap = %llu", block_group, blk);
135 grp = ext4_get_group_info(sb, block_group);
136 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
138 count = ext4_free_inodes_count(sb, desc);
139 percpu_counter_sub(&sbi->s_freeinodes_counter,
142 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
145 set_buffer_verified(bh);
146 ext4_unlock_group(sb, block_group);
151 * Read the inode allocation bitmap for a given block_group, reading
152 * into the specified slot in the superblock's bitmap cache.
154 * Return buffer_head of bitmap on success or NULL.
156 static struct buffer_head *
157 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
159 struct ext4_group_desc *desc;
160 struct buffer_head *bh = NULL;
161 ext4_fsblk_t bitmap_blk;
164 desc = ext4_get_group_desc(sb, block_group, NULL);
166 return ERR_PTR(-EFSCORRUPTED);
168 bitmap_blk = ext4_inode_bitmap(sb, desc);
169 bh = sb_getblk(sb, bitmap_blk);
171 ext4_error(sb, "Cannot read inode bitmap - "
172 "block_group = %u, inode_bitmap = %llu",
173 block_group, bitmap_blk);
174 return ERR_PTR(-EIO);
176 if (bitmap_uptodate(bh))
180 if (bitmap_uptodate(bh)) {
185 ext4_lock_group(sb, block_group);
186 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
187 err = ext4_init_inode_bitmap(sb, bh, block_group, desc);
188 set_bitmap_uptodate(bh);
189 set_buffer_uptodate(bh);
190 set_buffer_verified(bh);
191 ext4_unlock_group(sb, block_group);
194 ext4_error(sb, "Failed to init inode bitmap for group "
195 "%u: %d", block_group, err);
200 ext4_unlock_group(sb, block_group);
202 if (buffer_uptodate(bh)) {
204 * if not uninit if bh is uptodate,
205 * bitmap is also uptodate
207 set_bitmap_uptodate(bh);
212 * submit the buffer_head for reading
214 trace_ext4_load_inode_bitmap(sb, block_group);
215 bh->b_end_io = ext4_end_bitmap_read;
217 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
219 if (!buffer_uptodate(bh)) {
221 ext4_error(sb, "Cannot read inode bitmap - "
222 "block_group = %u, inode_bitmap = %llu",
223 block_group, bitmap_blk);
224 return ERR_PTR(-EIO);
228 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
238 * NOTE! When we get the inode, we're the only people
239 * that have access to it, and as such there are no
240 * race conditions we have to worry about. The inode
241 * is not on the hash-lists, and it cannot be reached
242 * through the filesystem because the directory entry
243 * has been deleted earlier.
245 * HOWEVER: we must make sure that we get no aliases,
246 * which means that we have to call "clear_inode()"
247 * _before_ we mark the inode not in use in the inode
248 * bitmaps. Otherwise a newly created file might use
249 * the same inode number (not actually the same pointer
250 * though), and then we'd have two inodes sharing the
251 * same inode number and space on the harddisk.
253 void ext4_free_inode(handle_t *handle, struct inode *inode)
255 struct super_block *sb = inode->i_sb;
258 struct buffer_head *bitmap_bh = NULL;
259 struct buffer_head *bh2;
260 ext4_group_t block_group;
262 struct ext4_group_desc *gdp;
263 struct ext4_super_block *es;
264 struct ext4_sb_info *sbi;
265 int fatal = 0, err, count, cleared;
266 struct ext4_group_info *grp;
269 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
270 "nonexistent device\n", __func__, __LINE__);
273 if (atomic_read(&inode->i_count) > 1) {
274 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
275 __func__, __LINE__, inode->i_ino,
276 atomic_read(&inode->i_count));
279 if (inode->i_nlink) {
280 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
281 __func__, __LINE__, inode->i_ino, inode->i_nlink);
287 ext4_debug("freeing inode %lu\n", ino);
288 trace_ext4_free_inode(inode);
291 * Note: we must free any quota before locking the superblock,
292 * as writing the quota to disk may need the lock as well.
294 dquot_initialize(inode);
295 ext4_xattr_delete_inode(handle, inode);
296 dquot_free_inode(inode);
299 is_directory = S_ISDIR(inode->i_mode);
301 /* Do this BEFORE marking the inode not in use or returning an error */
302 ext4_clear_inode(inode);
304 es = EXT4_SB(sb)->s_es;
305 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
306 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
309 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
310 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
311 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
312 /* Don't bother if the inode bitmap is corrupt. */
313 grp = ext4_get_group_info(sb, block_group);
314 if (IS_ERR(bitmap_bh)) {
315 fatal = PTR_ERR(bitmap_bh);
319 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
320 fatal = -EFSCORRUPTED;
324 BUFFER_TRACE(bitmap_bh, "get_write_access");
325 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
330 gdp = ext4_get_group_desc(sb, block_group, &bh2);
332 BUFFER_TRACE(bh2, "get_write_access");
333 fatal = ext4_journal_get_write_access(handle, bh2);
335 ext4_lock_group(sb, block_group);
336 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
337 if (fatal || !cleared) {
338 ext4_unlock_group(sb, block_group);
342 count = ext4_free_inodes_count(sb, gdp) + 1;
343 ext4_free_inodes_set(sb, gdp, count);
345 count = ext4_used_dirs_count(sb, gdp) - 1;
346 ext4_used_dirs_set(sb, gdp, count);
347 percpu_counter_dec(&sbi->s_dirs_counter);
349 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
350 EXT4_INODES_PER_GROUP(sb) / 8);
351 ext4_group_desc_csum_set(sb, block_group, gdp);
352 ext4_unlock_group(sb, block_group);
354 percpu_counter_inc(&sbi->s_freeinodes_counter);
355 if (sbi->s_log_groups_per_flex) {
356 ext4_group_t f = ext4_flex_group(sbi, block_group);
358 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
360 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
362 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
363 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
366 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
367 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
371 ext4_error(sb, "bit already cleared for inode %lu", ino);
372 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
374 count = ext4_free_inodes_count(sb, gdp);
375 percpu_counter_sub(&sbi->s_freeinodes_counter,
378 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
383 ext4_std_error(sb, fatal);
393 * Helper function for Orlov's allocator; returns critical information
394 * for a particular block group or flex_bg. If flex_size is 1, then g
395 * is a block group number; otherwise it is flex_bg number.
397 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
398 int flex_size, struct orlov_stats *stats)
400 struct ext4_group_desc *desc;
401 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
404 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
405 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
406 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
410 desc = ext4_get_group_desc(sb, g, NULL);
412 stats->free_inodes = ext4_free_inodes_count(sb, desc);
413 stats->free_clusters = ext4_free_group_clusters(sb, desc);
414 stats->used_dirs = ext4_used_dirs_count(sb, desc);
416 stats->free_inodes = 0;
417 stats->free_clusters = 0;
418 stats->used_dirs = 0;
423 * Orlov's allocator for directories.
425 * We always try to spread first-level directories.
427 * If there are blockgroups with both free inodes and free blocks counts
428 * not worse than average we return one with smallest directory count.
429 * Otherwise we simply return a random group.
431 * For the rest rules look so:
433 * It's OK to put directory into a group unless
434 * it has too many directories already (max_dirs) or
435 * it has too few free inodes left (min_inodes) or
436 * it has too few free blocks left (min_blocks) or
437 * Parent's group is preferred, if it doesn't satisfy these
438 * conditions we search cyclically through the rest. If none
439 * of the groups look good we just look for a group with more
440 * free inodes than average (starting at parent's group).
443 static int find_group_orlov(struct super_block *sb, struct inode *parent,
444 ext4_group_t *group, umode_t mode,
445 const struct qstr *qstr)
447 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
448 struct ext4_sb_info *sbi = EXT4_SB(sb);
449 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
450 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
451 unsigned int freei, avefreei, grp_free;
452 ext4_fsblk_t freeb, avefreec;
454 int max_dirs, min_inodes;
455 ext4_grpblk_t min_clusters;
456 ext4_group_t i, grp, g, ngroups;
457 struct ext4_group_desc *desc;
458 struct orlov_stats stats;
459 int flex_size = ext4_flex_bg_size(sbi);
460 struct dx_hash_info hinfo;
462 ngroups = real_ngroups;
464 ngroups = (real_ngroups + flex_size - 1) >>
465 sbi->s_log_groups_per_flex;
466 parent_group >>= sbi->s_log_groups_per_flex;
469 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
470 avefreei = freei / ngroups;
471 freeb = EXT4_C2B(sbi,
472 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
474 do_div(avefreec, ngroups);
475 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
478 ((parent == d_inode(sb->s_root)) ||
479 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
480 int best_ndir = inodes_per_group;
484 hinfo.hash_version = DX_HASH_HALF_MD4;
485 hinfo.seed = sbi->s_hash_seed;
486 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
490 parent_group = (unsigned)grp % ngroups;
491 for (i = 0; i < ngroups; i++) {
492 g = (parent_group + i) % ngroups;
493 get_orlov_stats(sb, g, flex_size, &stats);
494 if (!stats.free_inodes)
496 if (stats.used_dirs >= best_ndir)
498 if (stats.free_inodes < avefreei)
500 if (stats.free_clusters < avefreec)
504 best_ndir = stats.used_dirs;
509 if (flex_size == 1) {
515 * We pack inodes at the beginning of the flexgroup's
516 * inode tables. Block allocation decisions will do
517 * something similar, although regular files will
518 * start at 2nd block group of the flexgroup. See
519 * ext4_ext_find_goal() and ext4_find_near().
522 for (i = 0; i < flex_size; i++) {
523 if (grp+i >= real_ngroups)
525 desc = ext4_get_group_desc(sb, grp+i, NULL);
526 if (desc && ext4_free_inodes_count(sb, desc)) {
534 max_dirs = ndirs / ngroups + inodes_per_group / 16;
535 min_inodes = avefreei - inodes_per_group*flex_size / 4;
538 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
541 * Start looking in the flex group where we last allocated an
542 * inode for this parent directory
544 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
545 parent_group = EXT4_I(parent)->i_last_alloc_group;
547 parent_group >>= sbi->s_log_groups_per_flex;
550 for (i = 0; i < ngroups; i++) {
551 grp = (parent_group + i) % ngroups;
552 get_orlov_stats(sb, grp, flex_size, &stats);
553 if (stats.used_dirs >= max_dirs)
555 if (stats.free_inodes < min_inodes)
557 if (stats.free_clusters < min_clusters)
563 ngroups = real_ngroups;
564 avefreei = freei / ngroups;
566 parent_group = EXT4_I(parent)->i_block_group;
567 for (i = 0; i < ngroups; i++) {
568 grp = (parent_group + i) % ngroups;
569 desc = ext4_get_group_desc(sb, grp, NULL);
571 grp_free = ext4_free_inodes_count(sb, desc);
572 if (grp_free && grp_free >= avefreei) {
581 * The free-inodes counter is approximate, and for really small
582 * filesystems the above test can fail to find any blockgroups
591 static int find_group_other(struct super_block *sb, struct inode *parent,
592 ext4_group_t *group, umode_t mode)
594 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
595 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
596 struct ext4_group_desc *desc;
597 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
600 * Try to place the inode is the same flex group as its
601 * parent. If we can't find space, use the Orlov algorithm to
602 * find another flex group, and store that information in the
603 * parent directory's inode information so that use that flex
604 * group for future allocations.
610 parent_group &= ~(flex_size-1);
611 last = parent_group + flex_size;
614 for (i = parent_group; i < last; i++) {
615 desc = ext4_get_group_desc(sb, i, NULL);
616 if (desc && ext4_free_inodes_count(sb, desc)) {
621 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
623 parent_group = EXT4_I(parent)->i_last_alloc_group;
627 * If this didn't work, use the Orlov search algorithm
628 * to find a new flex group; we pass in the mode to
629 * avoid the topdir algorithms.
631 *group = parent_group + flex_size;
632 if (*group > ngroups)
634 return find_group_orlov(sb, parent, group, mode, NULL);
638 * Try to place the inode in its parent directory
640 *group = parent_group;
641 desc = ext4_get_group_desc(sb, *group, NULL);
642 if (desc && ext4_free_inodes_count(sb, desc) &&
643 ext4_free_group_clusters(sb, desc))
647 * We're going to place this inode in a different blockgroup from its
648 * parent. We want to cause files in a common directory to all land in
649 * the same blockgroup. But we want files which are in a different
650 * directory which shares a blockgroup with our parent to land in a
651 * different blockgroup.
653 * So add our directory's i_ino into the starting point for the hash.
655 *group = (*group + parent->i_ino) % ngroups;
658 * Use a quadratic hash to find a group with a free inode and some free
661 for (i = 1; i < ngroups; i <<= 1) {
663 if (*group >= ngroups)
665 desc = ext4_get_group_desc(sb, *group, NULL);
666 if (desc && ext4_free_inodes_count(sb, desc) &&
667 ext4_free_group_clusters(sb, desc))
672 * That failed: try linear search for a free inode, even if that group
673 * has no free blocks.
675 *group = parent_group;
676 for (i = 0; i < ngroups; i++) {
677 if (++*group >= ngroups)
679 desc = ext4_get_group_desc(sb, *group, NULL);
680 if (desc && ext4_free_inodes_count(sb, desc))
688 * In no journal mode, if an inode has recently been deleted, we want
689 * to avoid reusing it until we're reasonably sure the inode table
690 * block has been written back to disk. (Yes, these values are
691 * somewhat arbitrary...)
693 #define RECENTCY_MIN 5
694 #define RECENTCY_DIRTY 30
696 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
698 struct ext4_group_desc *gdp;
699 struct ext4_inode *raw_inode;
700 struct buffer_head *bh;
701 unsigned long dtime, now;
702 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
703 int offset, ret = 0, recentcy = RECENTCY_MIN;
705 gdp = ext4_get_group_desc(sb, group, NULL);
709 bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
710 (ino / inodes_per_block));
711 if (unlikely(!bh) || !buffer_uptodate(bh))
713 * If the block is not in the buffer cache, then it
714 * must have been written out.
718 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
719 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
720 dtime = le32_to_cpu(raw_inode->i_dtime);
722 if (buffer_dirty(bh))
723 recentcy += RECENTCY_DIRTY;
725 if (dtime && (dtime < now) && (now < dtime + recentcy))
733 * There are two policies for allocating an inode. If the new inode is
734 * a directory, then a forward search is made for a block group with both
735 * free space and a low directory-to-inode ratio; if that fails, then of
736 * the groups with above-average free space, that group with the fewest
737 * directories already is chosen.
739 * For other inodes, search forward from the parent directory's block
740 * group to find a free inode.
742 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
743 umode_t mode, const struct qstr *qstr,
744 __u32 goal, uid_t *owner, int handle_type,
745 unsigned int line_no, int nblocks)
747 struct super_block *sb;
748 struct buffer_head *inode_bitmap_bh = NULL;
749 struct buffer_head *group_desc_bh;
750 ext4_group_t ngroups, group = 0;
751 unsigned long ino = 0;
753 struct ext4_group_desc *gdp = NULL;
754 struct ext4_inode_info *ei;
755 struct ext4_sb_info *sbi;
759 ext4_group_t flex_group;
760 struct ext4_group_info *grp;
763 /* Cannot create files in a deleted directory */
764 if (!dir || !dir->i_nlink)
765 return ERR_PTR(-EPERM);
767 if ((ext4_encrypted_inode(dir) ||
768 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
769 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
770 err = fscrypt_get_encryption_info(dir);
773 if (!fscrypt_has_encryption_key(dir))
774 return ERR_PTR(-EPERM);
776 nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
781 ngroups = ext4_get_groups_count(sb);
782 trace_ext4_request_inode(dir, mode);
783 inode = new_inode(sb);
785 return ERR_PTR(-ENOMEM);
790 * Initialize owners and quota early so that we don't have to account
791 * for quota initialization worst case in standard inode creating
795 inode->i_mode = mode;
796 i_uid_write(inode, owner[0]);
797 i_gid_write(inode, owner[1]);
798 } else if (test_opt(sb, GRPID)) {
799 inode->i_mode = mode;
800 inode->i_uid = current_fsuid();
801 inode->i_gid = dir->i_gid;
803 inode_init_owner(inode, dir, mode);
805 if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_PROJECT) &&
806 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
807 ei->i_projid = EXT4_I(dir)->i_projid;
809 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
811 err = dquot_initialize(inode);
816 goal = sbi->s_inode_goal;
818 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
819 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
820 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
826 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
828 ret2 = find_group_other(sb, dir, &group, mode);
831 EXT4_I(dir)->i_last_alloc_group = group;
837 * Normally we will only go through one pass of this loop,
838 * unless we get unlucky and it turns out the group we selected
839 * had its last inode grabbed by someone else.
841 for (i = 0; i < ngroups; i++, ino = 0) {
844 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
849 * Check free inodes count before loading bitmap.
851 if (ext4_free_inodes_count(sb, gdp) == 0) {
852 if (++group == ngroups)
857 grp = ext4_get_group_info(sb, group);
858 /* Skip groups with already-known suspicious inode tables */
859 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
860 if (++group == ngroups)
865 brelse(inode_bitmap_bh);
866 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
867 /* Skip groups with suspicious inode tables */
868 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
869 IS_ERR(inode_bitmap_bh)) {
870 inode_bitmap_bh = NULL;
871 if (++group == ngroups)
876 repeat_in_this_group:
877 ino = ext4_find_next_zero_bit((unsigned long *)
878 inode_bitmap_bh->b_data,
879 EXT4_INODES_PER_GROUP(sb), ino);
880 if (ino >= EXT4_INODES_PER_GROUP(sb))
882 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
883 ext4_error(sb, "reserved inode found cleared - "
884 "inode=%lu", ino + 1);
887 if ((EXT4_SB(sb)->s_journal == NULL) &&
888 recently_deleted(sb, group, ino)) {
893 BUG_ON(nblocks <= 0);
894 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
895 handle_type, nblocks,
897 if (IS_ERR(handle)) {
898 err = PTR_ERR(handle);
899 ext4_std_error(sb, err);
903 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
904 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
906 ext4_std_error(sb, err);
909 ext4_lock_group(sb, group);
910 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
911 ext4_unlock_group(sb, group);
912 ino++; /* the inode bitmap is zero-based */
914 goto got; /* we grabbed the inode! */
916 if (ino < EXT4_INODES_PER_GROUP(sb))
917 goto repeat_in_this_group;
919 if (++group == ngroups)
926 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
927 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
929 ext4_std_error(sb, err);
933 BUFFER_TRACE(group_desc_bh, "get_write_access");
934 err = ext4_journal_get_write_access(handle, group_desc_bh);
936 ext4_std_error(sb, err);
940 /* We may have to initialize the block bitmap if it isn't already */
941 if (ext4_has_group_desc_csum(sb) &&
942 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
943 struct buffer_head *block_bitmap_bh;
945 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
946 if (IS_ERR(block_bitmap_bh)) {
947 err = PTR_ERR(block_bitmap_bh);
950 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
951 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
953 brelse(block_bitmap_bh);
954 ext4_std_error(sb, err);
958 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
959 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
961 /* recheck and clear flag under lock if we still need to */
962 ext4_lock_group(sb, group);
963 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
964 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
965 ext4_free_group_clusters_set(sb, gdp,
966 ext4_free_clusters_after_init(sb, group, gdp));
967 ext4_block_bitmap_csum_set(sb, group, gdp,
969 ext4_group_desc_csum_set(sb, group, gdp);
971 ext4_unlock_group(sb, group);
972 brelse(block_bitmap_bh);
975 ext4_std_error(sb, err);
980 /* Update the relevant bg descriptor fields */
981 if (ext4_has_group_desc_csum(sb)) {
983 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
985 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
986 ext4_lock_group(sb, group); /* while we modify the bg desc */
987 free = EXT4_INODES_PER_GROUP(sb) -
988 ext4_itable_unused_count(sb, gdp);
989 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
990 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
994 * Check the relative inode number against the last used
995 * relative inode number in this group. if it is greater
996 * we need to update the bg_itable_unused count
999 ext4_itable_unused_set(sb, gdp,
1000 (EXT4_INODES_PER_GROUP(sb) - ino));
1001 up_read(&grp->alloc_sem);
1003 ext4_lock_group(sb, group);
1006 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1007 if (S_ISDIR(mode)) {
1008 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1009 if (sbi->s_log_groups_per_flex) {
1010 ext4_group_t f = ext4_flex_group(sbi, group);
1012 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1015 if (ext4_has_group_desc_csum(sb)) {
1016 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1017 EXT4_INODES_PER_GROUP(sb) / 8);
1018 ext4_group_desc_csum_set(sb, group, gdp);
1020 ext4_unlock_group(sb, group);
1022 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1023 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1025 ext4_std_error(sb, err);
1029 percpu_counter_dec(&sbi->s_freeinodes_counter);
1031 percpu_counter_inc(&sbi->s_dirs_counter);
1033 if (sbi->s_log_groups_per_flex) {
1034 flex_group = ext4_flex_group(sbi, group);
1035 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1038 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1039 /* This is the optimal IO size (for stat), not the fs block size */
1040 inode->i_blocks = 0;
1041 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1042 ext4_current_time(inode);
1044 memset(ei->i_data, 0, sizeof(ei->i_data));
1045 ei->i_dir_start_lookup = 0;
1048 /* Don't inherit extent flag from directory, amongst others. */
1050 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1053 ei->i_block_group = group;
1054 ei->i_last_alloc_group = ~0;
1056 ext4_set_inode_flags(inode);
1057 if (IS_DIRSYNC(inode))
1058 ext4_handle_sync(handle);
1059 if (insert_inode_locked(inode) < 0) {
1061 * Likely a bitmap corruption causing inode to be allocated
1065 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1069 spin_lock(&sbi->s_next_gen_lock);
1070 inode->i_generation = sbi->s_next_generation++;
1071 spin_unlock(&sbi->s_next_gen_lock);
1073 /* Precompute checksum seed for inode metadata */
1074 if (ext4_has_metadata_csum(sb)) {
1076 __le32 inum = cpu_to_le32(inode->i_ino);
1077 __le32 gen = cpu_to_le32(inode->i_generation);
1078 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1080 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1084 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1085 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1087 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1088 ei->i_inline_off = 0;
1089 if (ext4_has_feature_inline_data(sb))
1090 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1092 err = dquot_alloc_inode(inode);
1096 err = ext4_init_acl(handle, inode, dir);
1098 goto fail_free_drop;
1100 err = ext4_init_security(handle, inode, dir, qstr);
1102 goto fail_free_drop;
1104 if (ext4_has_feature_extents(sb)) {
1105 /* set extent flag only for directory, file and normal symlink*/
1106 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1107 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1108 ext4_ext_tree_init(handle, inode);
1112 if (ext4_handle_valid(handle)) {
1113 ei->i_sync_tid = handle->h_transaction->t_tid;
1114 ei->i_datasync_tid = handle->h_transaction->t_tid;
1118 /* give pointer to avoid set_context with journal ops. */
1119 err = fscrypt_inherit_context(dir, inode, &encrypt, true);
1121 goto fail_free_drop;
1124 err = ext4_mark_inode_dirty(handle, inode);
1126 ext4_std_error(sb, err);
1127 goto fail_free_drop;
1130 ext4_debug("allocating inode %lu\n", inode->i_ino);
1131 trace_ext4_allocate_inode(inode, dir, mode);
1132 brelse(inode_bitmap_bh);
1136 dquot_free_inode(inode);
1139 unlock_new_inode(inode);
1142 inode->i_flags |= S_NOQUOTA;
1144 brelse(inode_bitmap_bh);
1145 return ERR_PTR(err);
1148 /* Verify that we are loading a valid orphan from disk */
1149 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1151 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1152 ext4_group_t block_group;
1154 struct buffer_head *bitmap_bh = NULL;
1155 struct inode *inode = NULL;
1156 int err = -EFSCORRUPTED;
1158 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1161 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1162 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1163 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1164 if (IS_ERR(bitmap_bh)) {
1165 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1166 ino, PTR_ERR(bitmap_bh));
1167 return (struct inode *) bitmap_bh;
1170 /* Having the inode bit set should be a 100% indicator that this
1171 * is a valid orphan (no e2fsck run on fs). Orphans also include
1172 * inodes that were being truncated, so we can't check i_nlink==0.
1174 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1177 inode = ext4_iget(sb, ino);
1178 if (IS_ERR(inode)) {
1179 err = PTR_ERR(inode);
1180 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1186 * If the orphans has i_nlinks > 0 then it should be able to
1187 * be truncated, otherwise it won't be removed from the orphan
1188 * list during processing and an infinite loop will result.
1189 * Similarly, it must not be a bad inode.
1191 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1192 is_bad_inode(inode))
1195 if (NEXT_ORPHAN(inode) > max_ino)
1201 ext4_error(sb, "bad orphan inode %lu", ino);
1203 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1204 bit, (unsigned long long)bitmap_bh->b_blocknr,
1205 ext4_test_bit(bit, bitmap_bh->b_data));
1207 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1208 is_bad_inode(inode));
1209 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1210 NEXT_ORPHAN(inode));
1211 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1212 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1213 /* Avoid freeing blocks if we got a bad deleted inode */
1214 if (inode->i_nlink == 0)
1215 inode->i_blocks = 0;
1219 return ERR_PTR(err);
1222 unsigned long ext4_count_free_inodes(struct super_block *sb)
1224 unsigned long desc_count;
1225 struct ext4_group_desc *gdp;
1226 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1228 struct ext4_super_block *es;
1229 unsigned long bitmap_count, x;
1230 struct buffer_head *bitmap_bh = NULL;
1232 es = EXT4_SB(sb)->s_es;
1236 for (i = 0; i < ngroups; i++) {
1237 gdp = ext4_get_group_desc(sb, i, NULL);
1240 desc_count += ext4_free_inodes_count(sb, gdp);
1242 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1243 if (IS_ERR(bitmap_bh)) {
1248 x = ext4_count_free(bitmap_bh->b_data,
1249 EXT4_INODES_PER_GROUP(sb) / 8);
1250 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1251 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1255 printk(KERN_DEBUG "ext4_count_free_inodes: "
1256 "stored = %u, computed = %lu, %lu\n",
1257 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1261 for (i = 0; i < ngroups; i++) {
1262 gdp = ext4_get_group_desc(sb, i, NULL);
1265 desc_count += ext4_free_inodes_count(sb, gdp);
1272 /* Called at mount-time, super-block is locked */
1273 unsigned long ext4_count_dirs(struct super_block * sb)
1275 unsigned long count = 0;
1276 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1278 for (i = 0; i < ngroups; i++) {
1279 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1282 count += ext4_used_dirs_count(sb, gdp);
1288 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1289 * inode table. Must be called without any spinlock held. The only place
1290 * where it is called from on active part of filesystem is ext4lazyinit
1291 * thread, so we do not need any special locks, however we have to prevent
1292 * inode allocation from the current group, so we take alloc_sem lock, to
1293 * block ext4_new_inode() until we are finished.
1295 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1298 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1299 struct ext4_sb_info *sbi = EXT4_SB(sb);
1300 struct ext4_group_desc *gdp = NULL;
1301 struct buffer_head *group_desc_bh;
1304 int num, ret = 0, used_blks = 0;
1306 /* This should not happen, but just to be sure check this */
1307 if (sb->s_flags & MS_RDONLY) {
1312 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1317 * We do not need to lock this, because we are the only one
1318 * handling this flag.
1320 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1323 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1324 if (IS_ERR(handle)) {
1325 ret = PTR_ERR(handle);
1329 down_write(&grp->alloc_sem);
1331 * If inode bitmap was already initialized there may be some
1332 * used inodes so we need to skip blocks with used inodes in
1335 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1336 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1337 ext4_itable_unused_count(sb, gdp)),
1338 sbi->s_inodes_per_block);
1340 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1341 ext4_error(sb, "Something is wrong with group %u: "
1342 "used itable blocks: %d; "
1343 "itable unused count: %u",
1345 ext4_itable_unused_count(sb, gdp));
1350 blk = ext4_inode_table(sb, gdp) + used_blks;
1351 num = sbi->s_itb_per_group - used_blks;
1353 BUFFER_TRACE(group_desc_bh, "get_write_access");
1354 ret = ext4_journal_get_write_access(handle,
1360 * Skip zeroout if the inode table is full. But we set the ZEROED
1361 * flag anyway, because obviously, when it is full it does not need
1364 if (unlikely(num == 0))
1367 ext4_debug("going to zero out inode table in group %d\n",
1369 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1373 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1376 ext4_lock_group(sb, group);
1377 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1378 ext4_group_desc_csum_set(sb, group, gdp);
1379 ext4_unlock_group(sb, group);
1381 BUFFER_TRACE(group_desc_bh,
1382 "call ext4_handle_dirty_metadata");
1383 ret = ext4_handle_dirty_metadata(handle, NULL,
1387 up_write(&grp->alloc_sem);
1388 ext4_journal_stop(handle);