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 <linux/cred.h>
26 #include <asm/byteorder.h>
29 #include "ext4_jbd2.h"
33 #include <trace/events/ext4.h>
36 * ialloc.c contains the inodes allocation and deallocation routines
40 * The free inodes are managed by bitmaps. A file system contains several
41 * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap
42 * block for inodes, N blocks for the inode table and data blocks.
44 * The file system contains group descriptors which are located after the
45 * super block. Each descriptor contains the number of the bitmap block and
46 * the free blocks count in the block.
50 * To avoid calling the atomic setbit hundreds or thousands of times, we only
51 * need to use it within a single byte (to ensure we get endianness right).
52 * We can use memset for the rest of the bitmap as there are no other users.
54 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
58 if (start_bit >= end_bit)
61 ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
62 for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
63 ext4_set_bit(i, bitmap);
65 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
68 /* Initializes an uninitialized inode bitmap */
69 static int ext4_init_inode_bitmap(struct super_block *sb,
70 struct buffer_head *bh,
71 ext4_group_t block_group,
72 struct ext4_group_desc *gdp)
74 struct ext4_group_info *grp;
75 struct ext4_sb_info *sbi = EXT4_SB(sb);
76 J_ASSERT_BH(bh, buffer_locked(bh));
78 /* If checksum is bad mark all blocks and inodes use to prevent
79 * allocation, essentially implementing a per-group read-only flag. */
80 if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
81 grp = ext4_get_group_info(sb, block_group);
82 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
83 percpu_counter_sub(&sbi->s_freeclusters_counter,
85 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
86 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
88 count = ext4_free_inodes_count(sb, gdp);
89 percpu_counter_sub(&sbi->s_freeinodes_counter,
92 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
96 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
97 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
99 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
100 EXT4_INODES_PER_GROUP(sb) / 8);
101 ext4_group_desc_csum_set(sb, block_group, gdp);
106 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
109 set_buffer_uptodate(bh);
110 set_bitmap_uptodate(bh);
116 static int ext4_validate_inode_bitmap(struct super_block *sb,
117 struct ext4_group_desc *desc,
118 ext4_group_t block_group,
119 struct buffer_head *bh)
122 struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
123 struct ext4_sb_info *sbi = EXT4_SB(sb);
125 if (buffer_verified(bh))
127 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
128 return -EFSCORRUPTED;
130 ext4_lock_group(sb, block_group);
131 blk = ext4_inode_bitmap(sb, desc);
132 if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
133 EXT4_INODES_PER_GROUP(sb) / 8)) {
134 ext4_unlock_group(sb, block_group);
135 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
136 "inode_bitmap = %llu", block_group, blk);
137 grp = ext4_get_group_info(sb, block_group);
138 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
140 count = ext4_free_inodes_count(sb, desc);
141 percpu_counter_sub(&sbi->s_freeinodes_counter,
144 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
147 set_buffer_verified(bh);
148 ext4_unlock_group(sb, block_group);
153 * Read the inode allocation bitmap for a given block_group, reading
154 * into the specified slot in the superblock's bitmap cache.
156 * Return buffer_head of bitmap on success or NULL.
158 static struct buffer_head *
159 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
161 struct ext4_group_desc *desc;
162 struct buffer_head *bh = NULL;
163 ext4_fsblk_t bitmap_blk;
166 desc = ext4_get_group_desc(sb, block_group, NULL);
168 return ERR_PTR(-EFSCORRUPTED);
170 bitmap_blk = ext4_inode_bitmap(sb, desc);
171 bh = sb_getblk(sb, bitmap_blk);
173 ext4_error(sb, "Cannot read inode bitmap - "
174 "block_group = %u, inode_bitmap = %llu",
175 block_group, bitmap_blk);
176 return ERR_PTR(-EIO);
178 if (bitmap_uptodate(bh))
182 if (bitmap_uptodate(bh)) {
187 ext4_lock_group(sb, block_group);
188 if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
189 err = ext4_init_inode_bitmap(sb, bh, block_group, desc);
190 set_bitmap_uptodate(bh);
191 set_buffer_uptodate(bh);
192 set_buffer_verified(bh);
193 ext4_unlock_group(sb, block_group);
196 ext4_error(sb, "Failed to init inode bitmap for group "
197 "%u: %d", block_group, err);
202 ext4_unlock_group(sb, block_group);
204 if (buffer_uptodate(bh)) {
206 * if not uninit if bh is uptodate,
207 * bitmap is also uptodate
209 set_bitmap_uptodate(bh);
214 * submit the buffer_head for reading
216 trace_ext4_load_inode_bitmap(sb, block_group);
217 bh->b_end_io = ext4_end_bitmap_read;
219 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
221 if (!buffer_uptodate(bh)) {
223 ext4_error(sb, "Cannot read inode bitmap - "
224 "block_group = %u, inode_bitmap = %llu",
225 block_group, bitmap_blk);
226 return ERR_PTR(-EIO);
230 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
240 * NOTE! When we get the inode, we're the only people
241 * that have access to it, and as such there are no
242 * race conditions we have to worry about. The inode
243 * is not on the hash-lists, and it cannot be reached
244 * through the filesystem because the directory entry
245 * has been deleted earlier.
247 * HOWEVER: we must make sure that we get no aliases,
248 * which means that we have to call "clear_inode()"
249 * _before_ we mark the inode not in use in the inode
250 * bitmaps. Otherwise a newly created file might use
251 * the same inode number (not actually the same pointer
252 * though), and then we'd have two inodes sharing the
253 * same inode number and space on the harddisk.
255 void ext4_free_inode(handle_t *handle, struct inode *inode)
257 struct super_block *sb = inode->i_sb;
260 struct buffer_head *bitmap_bh = NULL;
261 struct buffer_head *bh2;
262 ext4_group_t block_group;
264 struct ext4_group_desc *gdp;
265 struct ext4_super_block *es;
266 struct ext4_sb_info *sbi;
267 int fatal = 0, err, count, cleared;
268 struct ext4_group_info *grp;
271 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
272 "nonexistent device\n", __func__, __LINE__);
275 if (atomic_read(&inode->i_count) > 1) {
276 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
277 __func__, __LINE__, inode->i_ino,
278 atomic_read(&inode->i_count));
281 if (inode->i_nlink) {
282 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
283 __func__, __LINE__, inode->i_ino, inode->i_nlink);
289 ext4_debug("freeing inode %lu\n", ino);
290 trace_ext4_free_inode(inode);
293 * Note: we must free any quota before locking the superblock,
294 * as writing the quota to disk may need the lock as well.
296 dquot_initialize(inode);
297 dquot_free_inode(inode);
300 is_directory = S_ISDIR(inode->i_mode);
302 /* Do this BEFORE marking the inode not in use or returning an error */
303 ext4_clear_inode(inode);
305 es = EXT4_SB(sb)->s_es;
306 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
307 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
310 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
311 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
312 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
313 /* Don't bother if the inode bitmap is corrupt. */
314 grp = ext4_get_group_info(sb, block_group);
315 if (IS_ERR(bitmap_bh)) {
316 fatal = PTR_ERR(bitmap_bh);
320 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
321 fatal = -EFSCORRUPTED;
325 BUFFER_TRACE(bitmap_bh, "get_write_access");
326 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
331 gdp = ext4_get_group_desc(sb, block_group, &bh2);
333 BUFFER_TRACE(bh2, "get_write_access");
334 fatal = ext4_journal_get_write_access(handle, bh2);
336 ext4_lock_group(sb, block_group);
337 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
338 if (fatal || !cleared) {
339 ext4_unlock_group(sb, block_group);
343 count = ext4_free_inodes_count(sb, gdp) + 1;
344 ext4_free_inodes_set(sb, gdp, count);
346 count = ext4_used_dirs_count(sb, gdp) - 1;
347 ext4_used_dirs_set(sb, gdp, count);
348 percpu_counter_dec(&sbi->s_dirs_counter);
350 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
351 EXT4_INODES_PER_GROUP(sb) / 8);
352 ext4_group_desc_csum_set(sb, block_group, gdp);
353 ext4_unlock_group(sb, block_group);
355 percpu_counter_inc(&sbi->s_freeinodes_counter);
356 if (sbi->s_log_groups_per_flex) {
357 ext4_group_t f = ext4_flex_group(sbi, block_group);
359 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
361 atomic_dec(&sbi->s_flex_groups[f].used_dirs);
363 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
364 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
367 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
368 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
372 ext4_error(sb, "bit already cleared for inode %lu", ino);
373 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
375 count = ext4_free_inodes_count(sb, gdp);
376 percpu_counter_sub(&sbi->s_freeinodes_counter,
379 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
384 ext4_std_error(sb, fatal);
394 * Helper function for Orlov's allocator; returns critical information
395 * for a particular block group or flex_bg. If flex_size is 1, then g
396 * is a block group number; otherwise it is flex_bg number.
398 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
399 int flex_size, struct orlov_stats *stats)
401 struct ext4_group_desc *desc;
402 struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
405 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
406 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
407 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
411 desc = ext4_get_group_desc(sb, g, NULL);
413 stats->free_inodes = ext4_free_inodes_count(sb, desc);
414 stats->free_clusters = ext4_free_group_clusters(sb, desc);
415 stats->used_dirs = ext4_used_dirs_count(sb, desc);
417 stats->free_inodes = 0;
418 stats->free_clusters = 0;
419 stats->used_dirs = 0;
424 * Orlov's allocator for directories.
426 * We always try to spread first-level directories.
428 * If there are blockgroups with both free inodes and free blocks counts
429 * not worse than average we return one with smallest directory count.
430 * Otherwise we simply return a random group.
432 * For the rest rules look so:
434 * It's OK to put directory into a group unless
435 * it has too many directories already (max_dirs) or
436 * it has too few free inodes left (min_inodes) or
437 * it has too few free blocks left (min_blocks) or
438 * Parent's group is preferred, if it doesn't satisfy these
439 * conditions we search cyclically through the rest. If none
440 * of the groups look good we just look for a group with more
441 * free inodes than average (starting at parent's group).
444 static int find_group_orlov(struct super_block *sb, struct inode *parent,
445 ext4_group_t *group, umode_t mode,
446 const struct qstr *qstr)
448 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
449 struct ext4_sb_info *sbi = EXT4_SB(sb);
450 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
451 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
452 unsigned int freei, avefreei, grp_free;
453 ext4_fsblk_t freeb, avefreec;
455 int max_dirs, min_inodes;
456 ext4_grpblk_t min_clusters;
457 ext4_group_t i, grp, g, ngroups;
458 struct ext4_group_desc *desc;
459 struct orlov_stats stats;
460 int flex_size = ext4_flex_bg_size(sbi);
461 struct dx_hash_info hinfo;
463 ngroups = real_ngroups;
465 ngroups = (real_ngroups + flex_size - 1) >>
466 sbi->s_log_groups_per_flex;
467 parent_group >>= sbi->s_log_groups_per_flex;
470 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
471 avefreei = freei / ngroups;
472 freeb = EXT4_C2B(sbi,
473 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
475 do_div(avefreec, ngroups);
476 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
479 ((parent == d_inode(sb->s_root)) ||
480 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
481 int best_ndir = inodes_per_group;
485 hinfo.hash_version = DX_HASH_HALF_MD4;
486 hinfo.seed = sbi->s_hash_seed;
487 ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
491 parent_group = (unsigned)grp % ngroups;
492 for (i = 0; i < ngroups; i++) {
493 g = (parent_group + i) % ngroups;
494 get_orlov_stats(sb, g, flex_size, &stats);
495 if (!stats.free_inodes)
497 if (stats.used_dirs >= best_ndir)
499 if (stats.free_inodes < avefreei)
501 if (stats.free_clusters < avefreec)
505 best_ndir = stats.used_dirs;
510 if (flex_size == 1) {
516 * We pack inodes at the beginning of the flexgroup's
517 * inode tables. Block allocation decisions will do
518 * something similar, although regular files will
519 * start at 2nd block group of the flexgroup. See
520 * ext4_ext_find_goal() and ext4_find_near().
523 for (i = 0; i < flex_size; i++) {
524 if (grp+i >= real_ngroups)
526 desc = ext4_get_group_desc(sb, grp+i, NULL);
527 if (desc && ext4_free_inodes_count(sb, desc)) {
535 max_dirs = ndirs / ngroups + inodes_per_group / 16;
536 min_inodes = avefreei - inodes_per_group*flex_size / 4;
539 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
542 * Start looking in the flex group where we last allocated an
543 * inode for this parent directory
545 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
546 parent_group = EXT4_I(parent)->i_last_alloc_group;
548 parent_group >>= sbi->s_log_groups_per_flex;
551 for (i = 0; i < ngroups; i++) {
552 grp = (parent_group + i) % ngroups;
553 get_orlov_stats(sb, grp, flex_size, &stats);
554 if (stats.used_dirs >= max_dirs)
556 if (stats.free_inodes < min_inodes)
558 if (stats.free_clusters < min_clusters)
564 ngroups = real_ngroups;
565 avefreei = freei / ngroups;
567 parent_group = EXT4_I(parent)->i_block_group;
568 for (i = 0; i < ngroups; i++) {
569 grp = (parent_group + i) % ngroups;
570 desc = ext4_get_group_desc(sb, grp, NULL);
572 grp_free = ext4_free_inodes_count(sb, desc);
573 if (grp_free && grp_free >= avefreei) {
582 * The free-inodes counter is approximate, and for really small
583 * filesystems the above test can fail to find any blockgroups
592 static int find_group_other(struct super_block *sb, struct inode *parent,
593 ext4_group_t *group, umode_t mode)
595 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
596 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
597 struct ext4_group_desc *desc;
598 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
601 * Try to place the inode is the same flex group as its
602 * parent. If we can't find space, use the Orlov algorithm to
603 * find another flex group, and store that information in the
604 * parent directory's inode information so that use that flex
605 * group for future allocations.
611 parent_group &= ~(flex_size-1);
612 last = parent_group + flex_size;
615 for (i = parent_group; i < last; i++) {
616 desc = ext4_get_group_desc(sb, i, NULL);
617 if (desc && ext4_free_inodes_count(sb, desc)) {
622 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
624 parent_group = EXT4_I(parent)->i_last_alloc_group;
628 * If this didn't work, use the Orlov search algorithm
629 * to find a new flex group; we pass in the mode to
630 * avoid the topdir algorithms.
632 *group = parent_group + flex_size;
633 if (*group > ngroups)
635 return find_group_orlov(sb, parent, group, mode, NULL);
639 * Try to place the inode in its parent directory
641 *group = parent_group;
642 desc = ext4_get_group_desc(sb, *group, NULL);
643 if (desc && ext4_free_inodes_count(sb, desc) &&
644 ext4_free_group_clusters(sb, desc))
648 * We're going to place this inode in a different blockgroup from its
649 * parent. We want to cause files in a common directory to all land in
650 * the same blockgroup. But we want files which are in a different
651 * directory which shares a blockgroup with our parent to land in a
652 * different blockgroup.
654 * So add our directory's i_ino into the starting point for the hash.
656 *group = (*group + parent->i_ino) % ngroups;
659 * Use a quadratic hash to find a group with a free inode and some free
662 for (i = 1; i < ngroups; i <<= 1) {
664 if (*group >= ngroups)
666 desc = ext4_get_group_desc(sb, *group, NULL);
667 if (desc && ext4_free_inodes_count(sb, desc) &&
668 ext4_free_group_clusters(sb, desc))
673 * That failed: try linear search for a free inode, even if that group
674 * has no free blocks.
676 *group = parent_group;
677 for (i = 0; i < ngroups; i++) {
678 if (++*group >= ngroups)
680 desc = ext4_get_group_desc(sb, *group, NULL);
681 if (desc && ext4_free_inodes_count(sb, desc))
689 * In no journal mode, if an inode has recently been deleted, we want
690 * to avoid reusing it until we're reasonably sure the inode table
691 * block has been written back to disk. (Yes, these values are
692 * somewhat arbitrary...)
694 #define RECENTCY_MIN 5
695 #define RECENTCY_DIRTY 300
697 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
699 struct ext4_group_desc *gdp;
700 struct ext4_inode *raw_inode;
701 struct buffer_head *bh;
702 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
704 int recentcy = RECENTCY_MIN;
707 gdp = ext4_get_group_desc(sb, group, NULL);
711 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
712 (ino / inodes_per_block));
713 if (!bh || !buffer_uptodate(bh))
715 * If the block is not in the buffer cache, then it
716 * must have been written out.
720 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
721 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
723 /* i_dtime is only 32 bits on disk, but we only care about relative
724 * times in the range of a few minutes (i.e. long enough to sync a
725 * recently-deleted inode to disk), so using the low 32 bits of the
726 * clock (a 68 year range) is enough, see time_before32() */
727 dtime = le32_to_cpu(raw_inode->i_dtime);
728 now = ktime_get_real_seconds();
729 if (buffer_dirty(bh))
730 recentcy += RECENTCY_DIRTY;
732 if (dtime && time_before32(dtime, now) &&
733 time_before32(now, dtime + recentcy))
740 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
741 struct buffer_head *bitmap, unsigned long *ino)
744 *ino = ext4_find_next_zero_bit((unsigned long *)
746 EXT4_INODES_PER_GROUP(sb), *ino);
747 if (*ino >= EXT4_INODES_PER_GROUP(sb))
750 if ((EXT4_SB(sb)->s_journal == NULL) &&
751 recently_deleted(sb, group, *ino)) {
753 if (*ino < EXT4_INODES_PER_GROUP(sb))
762 * There are two policies for allocating an inode. If the new inode is
763 * a directory, then a forward search is made for a block group with both
764 * free space and a low directory-to-inode ratio; if that fails, then of
765 * the groups with above-average free space, that group with the fewest
766 * directories already is chosen.
768 * For other inodes, search forward from the parent directory's block
769 * group to find a free inode.
771 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
772 umode_t mode, const struct qstr *qstr,
773 __u32 goal, uid_t *owner, __u32 i_flags,
774 int handle_type, unsigned int line_no,
777 struct super_block *sb;
778 struct buffer_head *inode_bitmap_bh = NULL;
779 struct buffer_head *group_desc_bh;
780 ext4_group_t ngroups, group = 0;
781 unsigned long ino = 0;
783 struct ext4_group_desc *gdp = NULL;
784 struct ext4_inode_info *ei;
785 struct ext4_sb_info *sbi;
789 ext4_group_t flex_group;
790 struct ext4_group_info *grp;
793 /* Cannot create files in a deleted directory */
794 if (!dir || !dir->i_nlink)
795 return ERR_PTR(-EPERM);
800 if (unlikely(ext4_forced_shutdown(sbi)))
801 return ERR_PTR(-EIO);
803 if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
804 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
805 !(i_flags & EXT4_EA_INODE_FL)) {
806 err = fscrypt_get_encryption_info(dir);
809 if (!fscrypt_has_encryption_key(dir))
810 return ERR_PTR(-ENOKEY);
814 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
815 #ifdef CONFIG_EXT4_FS_POSIX_ACL
816 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
819 int acl_size = p->a_count * sizeof(ext4_acl_entry);
821 nblocks += (S_ISDIR(mode) ? 2 : 1) *
822 __ext4_xattr_set_credits(sb, NULL /* inode */,
823 NULL /* block_bh */, acl_size,
824 true /* is_create */);
825 posix_acl_release(p);
829 #ifdef CONFIG_SECURITY
831 int num_security_xattrs = 1;
833 #ifdef CONFIG_INTEGRITY
834 num_security_xattrs++;
837 * We assume that security xattrs are never
838 * more than 1k. In practice they are under
841 nblocks += num_security_xattrs *
842 __ext4_xattr_set_credits(sb, NULL /* inode */,
843 NULL /* block_bh */, 1024,
844 true /* is_create */);
848 nblocks += __ext4_xattr_set_credits(sb,
849 NULL /* inode */, NULL /* block_bh */,
850 FSCRYPT_SET_CONTEXT_MAX_SIZE,
851 true /* is_create */);
854 ngroups = ext4_get_groups_count(sb);
855 trace_ext4_request_inode(dir, mode);
856 inode = new_inode(sb);
858 return ERR_PTR(-ENOMEM);
862 * Initialize owners and quota early so that we don't have to account
863 * for quota initialization worst case in standard inode creating
867 inode->i_mode = mode;
868 i_uid_write(inode, owner[0]);
869 i_gid_write(inode, owner[1]);
870 } else if (test_opt(sb, GRPID)) {
871 inode->i_mode = mode;
872 inode->i_uid = current_fsuid();
873 inode->i_gid = dir->i_gid;
875 inode_init_owner(inode, dir, mode);
877 if (ext4_has_feature_project(sb) &&
878 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
879 ei->i_projid = EXT4_I(dir)->i_projid;
881 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
883 err = dquot_initialize(inode);
888 goal = sbi->s_inode_goal;
890 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
891 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
892 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
898 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
900 ret2 = find_group_other(sb, dir, &group, mode);
903 EXT4_I(dir)->i_last_alloc_group = group;
909 * Normally we will only go through one pass of this loop,
910 * unless we get unlucky and it turns out the group we selected
911 * had its last inode grabbed by someone else.
913 for (i = 0; i < ngroups; i++, ino = 0) {
916 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
921 * Check free inodes count before loading bitmap.
923 if (ext4_free_inodes_count(sb, gdp) == 0)
926 grp = ext4_get_group_info(sb, group);
927 /* Skip groups with already-known suspicious inode tables */
928 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
931 brelse(inode_bitmap_bh);
932 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
933 /* Skip groups with suspicious inode tables */
934 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
935 IS_ERR(inode_bitmap_bh)) {
936 inode_bitmap_bh = NULL;
940 repeat_in_this_group:
941 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
945 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
946 ext4_error(sb, "reserved inode found cleared - "
947 "inode=%lu", ino + 1);
952 BUG_ON(nblocks <= 0);
953 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
954 handle_type, nblocks,
956 if (IS_ERR(handle)) {
957 err = PTR_ERR(handle);
958 ext4_std_error(sb, err);
962 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
963 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
965 ext4_std_error(sb, err);
968 ext4_lock_group(sb, group);
969 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
971 /* Someone already took the bit. Repeat the search
974 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
976 ext4_set_bit(ino, inode_bitmap_bh->b_data);
979 ret2 = 1; /* we didn't grab the inode */
982 ext4_unlock_group(sb, group);
983 ino++; /* the inode bitmap is zero-based */
985 goto got; /* we grabbed the inode! */
987 if (ino < EXT4_INODES_PER_GROUP(sb))
988 goto repeat_in_this_group;
990 if (++group == ngroups)
997 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
998 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1000 ext4_std_error(sb, err);
1004 BUFFER_TRACE(group_desc_bh, "get_write_access");
1005 err = ext4_journal_get_write_access(handle, group_desc_bh);
1007 ext4_std_error(sb, err);
1011 /* We may have to initialize the block bitmap if it isn't already */
1012 if (ext4_has_group_desc_csum(sb) &&
1013 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1014 struct buffer_head *block_bitmap_bh;
1016 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1017 if (IS_ERR(block_bitmap_bh)) {
1018 err = PTR_ERR(block_bitmap_bh);
1021 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1022 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1024 brelse(block_bitmap_bh);
1025 ext4_std_error(sb, err);
1029 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1030 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1032 /* recheck and clear flag under lock if we still need to */
1033 ext4_lock_group(sb, group);
1034 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1035 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1036 ext4_free_group_clusters_set(sb, gdp,
1037 ext4_free_clusters_after_init(sb, group, gdp));
1038 ext4_block_bitmap_csum_set(sb, group, gdp,
1040 ext4_group_desc_csum_set(sb, group, gdp);
1042 ext4_unlock_group(sb, group);
1043 brelse(block_bitmap_bh);
1046 ext4_std_error(sb, err);
1051 /* Update the relevant bg descriptor fields */
1052 if (ext4_has_group_desc_csum(sb)) {
1054 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1056 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1057 ext4_lock_group(sb, group); /* while we modify the bg desc */
1058 free = EXT4_INODES_PER_GROUP(sb) -
1059 ext4_itable_unused_count(sb, gdp);
1060 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1061 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1065 * Check the relative inode number against the last used
1066 * relative inode number in this group. if it is greater
1067 * we need to update the bg_itable_unused count
1070 ext4_itable_unused_set(sb, gdp,
1071 (EXT4_INODES_PER_GROUP(sb) - ino));
1072 up_read(&grp->alloc_sem);
1074 ext4_lock_group(sb, group);
1077 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1078 if (S_ISDIR(mode)) {
1079 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1080 if (sbi->s_log_groups_per_flex) {
1081 ext4_group_t f = ext4_flex_group(sbi, group);
1083 atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1086 if (ext4_has_group_desc_csum(sb)) {
1087 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1088 EXT4_INODES_PER_GROUP(sb) / 8);
1089 ext4_group_desc_csum_set(sb, group, gdp);
1091 ext4_unlock_group(sb, group);
1093 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1094 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1096 ext4_std_error(sb, err);
1100 percpu_counter_dec(&sbi->s_freeinodes_counter);
1102 percpu_counter_inc(&sbi->s_dirs_counter);
1104 if (sbi->s_log_groups_per_flex) {
1105 flex_group = ext4_flex_group(sbi, group);
1106 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1109 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1110 /* This is the optimal IO size (for stat), not the fs block size */
1111 inode->i_blocks = 0;
1112 inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1113 current_time(inode);
1115 memset(ei->i_data, 0, sizeof(ei->i_data));
1116 ei->i_dir_start_lookup = 0;
1119 /* Don't inherit extent flag from directory, amongst others. */
1121 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1122 ei->i_flags |= i_flags;
1125 ei->i_block_group = group;
1126 ei->i_last_alloc_group = ~0;
1128 ext4_set_inode_flags(inode);
1129 if (IS_DIRSYNC(inode))
1130 ext4_handle_sync(handle);
1131 if (insert_inode_locked(inode) < 0) {
1133 * Likely a bitmap corruption causing inode to be allocated
1137 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1141 spin_lock(&sbi->s_next_gen_lock);
1142 inode->i_generation = sbi->s_next_generation++;
1143 spin_unlock(&sbi->s_next_gen_lock);
1145 /* Precompute checksum seed for inode metadata */
1146 if (ext4_has_metadata_csum(sb)) {
1148 __le32 inum = cpu_to_le32(inode->i_ino);
1149 __le32 gen = cpu_to_le32(inode->i_generation);
1150 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1152 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1156 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1157 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1159 ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1160 ei->i_inline_off = 0;
1161 if (ext4_has_feature_inline_data(sb))
1162 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1164 err = dquot_alloc_inode(inode);
1169 * Since the encryption xattr will always be unique, create it first so
1170 * that it's less likely to end up in an external xattr block and
1171 * prevent its deduplication.
1174 err = fscrypt_inherit_context(dir, inode, handle, true);
1176 goto fail_free_drop;
1179 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1180 err = ext4_init_acl(handle, inode, dir);
1182 goto fail_free_drop;
1184 err = ext4_init_security(handle, inode, dir, qstr);
1186 goto fail_free_drop;
1189 if (ext4_has_feature_extents(sb)) {
1190 /* set extent flag only for directory, file and normal symlink*/
1191 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1192 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1193 ext4_ext_tree_init(handle, inode);
1197 if (ext4_handle_valid(handle)) {
1198 ei->i_sync_tid = handle->h_transaction->t_tid;
1199 ei->i_datasync_tid = handle->h_transaction->t_tid;
1202 err = ext4_mark_inode_dirty(handle, inode);
1204 ext4_std_error(sb, err);
1205 goto fail_free_drop;
1208 ext4_debug("allocating inode %lu\n", inode->i_ino);
1209 trace_ext4_allocate_inode(inode, dir, mode);
1210 brelse(inode_bitmap_bh);
1214 dquot_free_inode(inode);
1217 unlock_new_inode(inode);
1220 inode->i_flags |= S_NOQUOTA;
1222 brelse(inode_bitmap_bh);
1223 return ERR_PTR(err);
1226 /* Verify that we are loading a valid orphan from disk */
1227 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1229 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1230 ext4_group_t block_group;
1232 struct buffer_head *bitmap_bh = NULL;
1233 struct inode *inode = NULL;
1234 int err = -EFSCORRUPTED;
1236 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1239 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1240 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1241 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1242 if (IS_ERR(bitmap_bh)) {
1243 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1244 ino, PTR_ERR(bitmap_bh));
1245 return (struct inode *) bitmap_bh;
1248 /* Having the inode bit set should be a 100% indicator that this
1249 * is a valid orphan (no e2fsck run on fs). Orphans also include
1250 * inodes that were being truncated, so we can't check i_nlink==0.
1252 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1255 inode = ext4_iget(sb, ino);
1256 if (IS_ERR(inode)) {
1257 err = PTR_ERR(inode);
1258 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1264 * If the orphans has i_nlinks > 0 then it should be able to
1265 * be truncated, otherwise it won't be removed from the orphan
1266 * list during processing and an infinite loop will result.
1267 * Similarly, it must not be a bad inode.
1269 if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1270 is_bad_inode(inode))
1273 if (NEXT_ORPHAN(inode) > max_ino)
1279 ext4_error(sb, "bad orphan inode %lu", ino);
1281 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1282 bit, (unsigned long long)bitmap_bh->b_blocknr,
1283 ext4_test_bit(bit, bitmap_bh->b_data));
1285 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1286 is_bad_inode(inode));
1287 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1288 NEXT_ORPHAN(inode));
1289 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1290 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1291 /* Avoid freeing blocks if we got a bad deleted inode */
1292 if (inode->i_nlink == 0)
1293 inode->i_blocks = 0;
1297 return ERR_PTR(err);
1300 unsigned long ext4_count_free_inodes(struct super_block *sb)
1302 unsigned long desc_count;
1303 struct ext4_group_desc *gdp;
1304 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1306 struct ext4_super_block *es;
1307 unsigned long bitmap_count, x;
1308 struct buffer_head *bitmap_bh = NULL;
1310 es = EXT4_SB(sb)->s_es;
1314 for (i = 0; i < ngroups; i++) {
1315 gdp = ext4_get_group_desc(sb, i, NULL);
1318 desc_count += ext4_free_inodes_count(sb, gdp);
1320 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1321 if (IS_ERR(bitmap_bh)) {
1326 x = ext4_count_free(bitmap_bh->b_data,
1327 EXT4_INODES_PER_GROUP(sb) / 8);
1328 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1329 (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1333 printk(KERN_DEBUG "ext4_count_free_inodes: "
1334 "stored = %u, computed = %lu, %lu\n",
1335 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1339 for (i = 0; i < ngroups; i++) {
1340 gdp = ext4_get_group_desc(sb, i, NULL);
1343 desc_count += ext4_free_inodes_count(sb, gdp);
1350 /* Called at mount-time, super-block is locked */
1351 unsigned long ext4_count_dirs(struct super_block * sb)
1353 unsigned long count = 0;
1354 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1356 for (i = 0; i < ngroups; i++) {
1357 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1360 count += ext4_used_dirs_count(sb, gdp);
1366 * Zeroes not yet zeroed inode table - just write zeroes through the whole
1367 * inode table. Must be called without any spinlock held. The only place
1368 * where it is called from on active part of filesystem is ext4lazyinit
1369 * thread, so we do not need any special locks, however we have to prevent
1370 * inode allocation from the current group, so we take alloc_sem lock, to
1371 * block ext4_new_inode() until we are finished.
1373 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1376 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1377 struct ext4_sb_info *sbi = EXT4_SB(sb);
1378 struct ext4_group_desc *gdp = NULL;
1379 struct buffer_head *group_desc_bh;
1382 int num, ret = 0, used_blks = 0;
1384 /* This should not happen, but just to be sure check this */
1385 if (sb_rdonly(sb)) {
1390 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1395 * We do not need to lock this, because we are the only one
1396 * handling this flag.
1398 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1401 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1402 if (IS_ERR(handle)) {
1403 ret = PTR_ERR(handle);
1407 down_write(&grp->alloc_sem);
1409 * If inode bitmap was already initialized there may be some
1410 * used inodes so we need to skip blocks with used inodes in
1413 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1414 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1415 ext4_itable_unused_count(sb, gdp)),
1416 sbi->s_inodes_per_block);
1418 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1419 ext4_error(sb, "Something is wrong with group %u: "
1420 "used itable blocks: %d; "
1421 "itable unused count: %u",
1423 ext4_itable_unused_count(sb, gdp));
1428 blk = ext4_inode_table(sb, gdp) + used_blks;
1429 num = sbi->s_itb_per_group - used_blks;
1431 BUFFER_TRACE(group_desc_bh, "get_write_access");
1432 ret = ext4_journal_get_write_access(handle,
1438 * Skip zeroout if the inode table is full. But we set the ZEROED
1439 * flag anyway, because obviously, when it is full it does not need
1442 if (unlikely(num == 0))
1445 ext4_debug("going to zero out inode table in group %d\n",
1447 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1451 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1454 ext4_lock_group(sb, group);
1455 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1456 ext4_group_desc_csum_set(sb, group, gdp);
1457 ext4_unlock_group(sb, group);
1459 BUFFER_TRACE(group_desc_bh,
1460 "call ext4_handle_dirty_metadata");
1461 ret = ext4_handle_dirty_metadata(handle, NULL,
1465 up_write(&grp->alloc_sem);
1466 ext4_journal_stop(handle);