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_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
99 ext4_unlock_group(sb, block_group);
100 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
101 "inode_bitmap = %llu", block_group, blk);
102 ext4_mark_group_bitmap_corrupted(sb, block_group,
103 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
106 set_buffer_verified(bh);
108 ext4_unlock_group(sb, block_group);
113 * Read the inode allocation bitmap for a given block_group, reading
114 * into the specified slot in the superblock's bitmap cache.
116 * Return buffer_head of bitmap on success or NULL.
118 static struct buffer_head *
119 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
121 struct ext4_group_desc *desc;
122 struct ext4_sb_info *sbi = EXT4_SB(sb);
123 struct buffer_head *bh = NULL;
124 ext4_fsblk_t bitmap_blk;
127 desc = ext4_get_group_desc(sb, block_group, NULL);
129 return ERR_PTR(-EFSCORRUPTED);
131 bitmap_blk = ext4_inode_bitmap(sb, desc);
132 if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
133 (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
134 ext4_error(sb, "Invalid inode bitmap blk %llu in "
135 "block_group %u", bitmap_blk, block_group);
136 ext4_mark_group_bitmap_corrupted(sb, block_group,
137 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
138 return ERR_PTR(-EFSCORRUPTED);
140 bh = sb_getblk(sb, bitmap_blk);
142 ext4_warning(sb, "Cannot read inode bitmap - "
143 "block_group = %u, inode_bitmap = %llu",
144 block_group, bitmap_blk);
145 return ERR_PTR(-ENOMEM);
147 if (bitmap_uptodate(bh))
151 if (bitmap_uptodate(bh)) {
156 ext4_lock_group(sb, block_group);
157 if (ext4_has_group_desc_csum(sb) &&
158 (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
159 if (block_group == 0) {
160 ext4_unlock_group(sb, block_group);
162 ext4_error(sb, "Inode bitmap for bg 0 marked "
167 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
168 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
169 sb->s_blocksize * 8, bh->b_data);
170 set_bitmap_uptodate(bh);
171 set_buffer_uptodate(bh);
172 set_buffer_verified(bh);
173 ext4_unlock_group(sb, block_group);
177 ext4_unlock_group(sb, block_group);
179 if (buffer_uptodate(bh)) {
181 * if not uninit if bh is uptodate,
182 * bitmap is also uptodate
184 set_bitmap_uptodate(bh);
189 * submit the buffer_head for reading
191 trace_ext4_load_inode_bitmap(sb, block_group);
192 bh->b_end_io = ext4_end_bitmap_read;
194 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
196 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
197 if (!buffer_uptodate(bh)) {
199 ext4_set_errno(sb, EIO);
200 ext4_error(sb, "Cannot read inode bitmap - "
201 "block_group = %u, inode_bitmap = %llu",
202 block_group, bitmap_blk);
203 ext4_mark_group_bitmap_corrupted(sb, block_group,
204 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
205 return ERR_PTR(-EIO);
209 err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
219 * NOTE! When we get the inode, we're the only people
220 * that have access to it, and as such there are no
221 * race conditions we have to worry about. The inode
222 * is not on the hash-lists, and it cannot be reached
223 * through the filesystem because the directory entry
224 * has been deleted earlier.
226 * HOWEVER: we must make sure that we get no aliases,
227 * which means that we have to call "clear_inode()"
228 * _before_ we mark the inode not in use in the inode
229 * bitmaps. Otherwise a newly created file might use
230 * the same inode number (not actually the same pointer
231 * though), and then we'd have two inodes sharing the
232 * same inode number and space on the harddisk.
234 void ext4_free_inode(handle_t *handle, struct inode *inode)
236 struct super_block *sb = inode->i_sb;
239 struct buffer_head *bitmap_bh = NULL;
240 struct buffer_head *bh2;
241 ext4_group_t block_group;
243 struct ext4_group_desc *gdp;
244 struct ext4_super_block *es;
245 struct ext4_sb_info *sbi;
246 int fatal = 0, err, count, cleared;
247 struct ext4_group_info *grp;
250 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
251 "nonexistent device\n", __func__, __LINE__);
254 if (atomic_read(&inode->i_count) > 1) {
255 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
256 __func__, __LINE__, inode->i_ino,
257 atomic_read(&inode->i_count));
260 if (inode->i_nlink) {
261 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
262 __func__, __LINE__, inode->i_ino, inode->i_nlink);
268 ext4_debug("freeing inode %lu\n", ino);
269 trace_ext4_free_inode(inode);
271 dquot_initialize(inode);
272 dquot_free_inode(inode);
274 is_directory = S_ISDIR(inode->i_mode);
276 /* Do this BEFORE marking the inode not in use or returning an error */
277 ext4_clear_inode(inode);
280 if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
281 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
284 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
285 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
286 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
287 /* Don't bother if the inode bitmap is corrupt. */
288 grp = ext4_get_group_info(sb, block_group);
289 if (IS_ERR(bitmap_bh)) {
290 fatal = PTR_ERR(bitmap_bh);
294 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
295 fatal = -EFSCORRUPTED;
299 BUFFER_TRACE(bitmap_bh, "get_write_access");
300 fatal = ext4_journal_get_write_access(handle, bitmap_bh);
305 gdp = ext4_get_group_desc(sb, block_group, &bh2);
307 BUFFER_TRACE(bh2, "get_write_access");
308 fatal = ext4_journal_get_write_access(handle, bh2);
310 ext4_lock_group(sb, block_group);
311 cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
312 if (fatal || !cleared) {
313 ext4_unlock_group(sb, block_group);
317 count = ext4_free_inodes_count(sb, gdp) + 1;
318 ext4_free_inodes_set(sb, gdp, count);
320 count = ext4_used_dirs_count(sb, gdp) - 1;
321 ext4_used_dirs_set(sb, gdp, count);
322 percpu_counter_dec(&sbi->s_dirs_counter);
324 ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
325 EXT4_INODES_PER_GROUP(sb) / 8);
326 ext4_group_desc_csum_set(sb, block_group, gdp);
327 ext4_unlock_group(sb, block_group);
329 percpu_counter_inc(&sbi->s_freeinodes_counter);
330 if (sbi->s_log_groups_per_flex) {
331 struct flex_groups *fg;
333 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
334 ext4_flex_group(sbi, block_group));
335 atomic_inc(&fg->free_inodes);
337 atomic_dec(&fg->used_dirs);
339 BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
340 fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
343 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
344 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
348 ext4_error(sb, "bit already cleared for inode %lu", ino);
349 ext4_mark_group_bitmap_corrupted(sb, block_group,
350 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
355 ext4_std_error(sb, fatal);
365 * Helper function for Orlov's allocator; returns critical information
366 * for a particular block group or flex_bg. If flex_size is 1, then g
367 * is a block group number; otherwise it is flex_bg number.
369 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
370 int flex_size, struct orlov_stats *stats)
372 struct ext4_group_desc *desc;
375 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
377 stats->free_inodes = atomic_read(&fg->free_inodes);
378 stats->free_clusters = atomic64_read(&fg->free_clusters);
379 stats->used_dirs = atomic_read(&fg->used_dirs);
383 desc = ext4_get_group_desc(sb, g, NULL);
385 stats->free_inodes = ext4_free_inodes_count(sb, desc);
386 stats->free_clusters = ext4_free_group_clusters(sb, desc);
387 stats->used_dirs = ext4_used_dirs_count(sb, desc);
389 stats->free_inodes = 0;
390 stats->free_clusters = 0;
391 stats->used_dirs = 0;
396 * Orlov's allocator for directories.
398 * We always try to spread first-level directories.
400 * If there are blockgroups with both free inodes and free blocks counts
401 * not worse than average we return one with smallest directory count.
402 * Otherwise we simply return a random group.
404 * For the rest rules look so:
406 * It's OK to put directory into a group unless
407 * it has too many directories already (max_dirs) or
408 * it has too few free inodes left (min_inodes) or
409 * it has too few free blocks left (min_blocks) or
410 * Parent's group is preferred, if it doesn't satisfy these
411 * conditions we search cyclically through the rest. If none
412 * of the groups look good we just look for a group with more
413 * free inodes than average (starting at parent's group).
416 static int find_group_orlov(struct super_block *sb, struct inode *parent,
417 ext4_group_t *group, umode_t mode,
418 const struct qstr *qstr)
420 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
421 struct ext4_sb_info *sbi = EXT4_SB(sb);
422 ext4_group_t real_ngroups = ext4_get_groups_count(sb);
423 int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
424 unsigned int freei, avefreei, grp_free;
425 ext4_fsblk_t freeb, avefreec;
427 int max_dirs, min_inodes;
428 ext4_grpblk_t min_clusters;
429 ext4_group_t i, grp, g, ngroups;
430 struct ext4_group_desc *desc;
431 struct orlov_stats stats;
432 int flex_size = ext4_flex_bg_size(sbi);
433 struct dx_hash_info hinfo;
435 ngroups = real_ngroups;
437 ngroups = (real_ngroups + flex_size - 1) >>
438 sbi->s_log_groups_per_flex;
439 parent_group >>= sbi->s_log_groups_per_flex;
442 freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
443 avefreei = freei / ngroups;
444 freeb = EXT4_C2B(sbi,
445 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
447 do_div(avefreec, ngroups);
448 ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
451 ((parent == d_inode(sb->s_root)) ||
452 (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
453 int best_ndir = inodes_per_group;
457 hinfo.hash_version = DX_HASH_HALF_MD4;
458 hinfo.seed = sbi->s_hash_seed;
459 ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
463 parent_group = (unsigned)grp % ngroups;
464 for (i = 0; i < ngroups; i++) {
465 g = (parent_group + i) % ngroups;
466 get_orlov_stats(sb, g, flex_size, &stats);
467 if (!stats.free_inodes)
469 if (stats.used_dirs >= best_ndir)
471 if (stats.free_inodes < avefreei)
473 if (stats.free_clusters < avefreec)
477 best_ndir = stats.used_dirs;
482 if (flex_size == 1) {
488 * We pack inodes at the beginning of the flexgroup's
489 * inode tables. Block allocation decisions will do
490 * something similar, although regular files will
491 * start at 2nd block group of the flexgroup. See
492 * ext4_ext_find_goal() and ext4_find_near().
495 for (i = 0; i < flex_size; i++) {
496 if (grp+i >= real_ngroups)
498 desc = ext4_get_group_desc(sb, grp+i, NULL);
499 if (desc && ext4_free_inodes_count(sb, desc)) {
507 max_dirs = ndirs / ngroups + inodes_per_group / 16;
508 min_inodes = avefreei - inodes_per_group*flex_size / 4;
511 min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
514 * Start looking in the flex group where we last allocated an
515 * inode for this parent directory
517 if (EXT4_I(parent)->i_last_alloc_group != ~0) {
518 parent_group = EXT4_I(parent)->i_last_alloc_group;
520 parent_group >>= sbi->s_log_groups_per_flex;
523 for (i = 0; i < ngroups; i++) {
524 grp = (parent_group + i) % ngroups;
525 get_orlov_stats(sb, grp, flex_size, &stats);
526 if (stats.used_dirs >= max_dirs)
528 if (stats.free_inodes < min_inodes)
530 if (stats.free_clusters < min_clusters)
536 ngroups = real_ngroups;
537 avefreei = freei / ngroups;
539 parent_group = EXT4_I(parent)->i_block_group;
540 for (i = 0; i < ngroups; i++) {
541 grp = (parent_group + i) % ngroups;
542 desc = ext4_get_group_desc(sb, grp, NULL);
544 grp_free = ext4_free_inodes_count(sb, desc);
545 if (grp_free && grp_free >= avefreei) {
554 * The free-inodes counter is approximate, and for really small
555 * filesystems the above test can fail to find any blockgroups
564 static int find_group_other(struct super_block *sb, struct inode *parent,
565 ext4_group_t *group, umode_t mode)
567 ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
568 ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
569 struct ext4_group_desc *desc;
570 int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
573 * Try to place the inode is the same flex group as its
574 * parent. If we can't find space, use the Orlov algorithm to
575 * find another flex group, and store that information in the
576 * parent directory's inode information so that use that flex
577 * group for future allocations.
583 parent_group &= ~(flex_size-1);
584 last = parent_group + flex_size;
587 for (i = parent_group; i < last; i++) {
588 desc = ext4_get_group_desc(sb, i, NULL);
589 if (desc && ext4_free_inodes_count(sb, desc)) {
594 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
596 parent_group = EXT4_I(parent)->i_last_alloc_group;
600 * If this didn't work, use the Orlov search algorithm
601 * to find a new flex group; we pass in the mode to
602 * avoid the topdir algorithms.
604 *group = parent_group + flex_size;
605 if (*group > ngroups)
607 return find_group_orlov(sb, parent, group, mode, NULL);
611 * Try to place the inode in its parent directory
613 *group = parent_group;
614 desc = ext4_get_group_desc(sb, *group, NULL);
615 if (desc && ext4_free_inodes_count(sb, desc) &&
616 ext4_free_group_clusters(sb, desc))
620 * We're going to place this inode in a different blockgroup from its
621 * parent. We want to cause files in a common directory to all land in
622 * the same blockgroup. But we want files which are in a different
623 * directory which shares a blockgroup with our parent to land in a
624 * different blockgroup.
626 * So add our directory's i_ino into the starting point for the hash.
628 *group = (*group + parent->i_ino) % ngroups;
631 * Use a quadratic hash to find a group with a free inode and some free
634 for (i = 1; i < ngroups; i <<= 1) {
636 if (*group >= ngroups)
638 desc = ext4_get_group_desc(sb, *group, NULL);
639 if (desc && ext4_free_inodes_count(sb, desc) &&
640 ext4_free_group_clusters(sb, desc))
645 * That failed: try linear search for a free inode, even if that group
646 * has no free blocks.
648 *group = parent_group;
649 for (i = 0; i < ngroups; i++) {
650 if (++*group >= ngroups)
652 desc = ext4_get_group_desc(sb, *group, NULL);
653 if (desc && ext4_free_inodes_count(sb, desc))
661 * In no journal mode, if an inode has recently been deleted, we want
662 * to avoid reusing it until we're reasonably sure the inode table
663 * block has been written back to disk. (Yes, these values are
664 * somewhat arbitrary...)
666 #define RECENTCY_MIN 5
667 #define RECENTCY_DIRTY 300
669 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
671 struct ext4_group_desc *gdp;
672 struct ext4_inode *raw_inode;
673 struct buffer_head *bh;
674 int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
676 int recentcy = RECENTCY_MIN;
679 gdp = ext4_get_group_desc(sb, group, NULL);
683 bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
684 (ino / inodes_per_block));
685 if (!bh || !buffer_uptodate(bh))
687 * If the block is not in the buffer cache, then it
688 * must have been written out.
692 offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
693 raw_inode = (struct ext4_inode *) (bh->b_data + offset);
695 /* i_dtime is only 32 bits on disk, but we only care about relative
696 * times in the range of a few minutes (i.e. long enough to sync a
697 * recently-deleted inode to disk), so using the low 32 bits of the
698 * clock (a 68 year range) is enough, see time_before32() */
699 dtime = le32_to_cpu(raw_inode->i_dtime);
700 now = ktime_get_real_seconds();
701 if (buffer_dirty(bh))
702 recentcy += RECENTCY_DIRTY;
704 if (dtime && time_before32(dtime, now) &&
705 time_before32(now, dtime + recentcy))
712 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
713 struct buffer_head *bitmap, unsigned long *ino)
716 *ino = ext4_find_next_zero_bit((unsigned long *)
718 EXT4_INODES_PER_GROUP(sb), *ino);
719 if (*ino >= EXT4_INODES_PER_GROUP(sb))
722 if ((EXT4_SB(sb)->s_journal == NULL) &&
723 recently_deleted(sb, group, *ino)) {
725 if (*ino < EXT4_INODES_PER_GROUP(sb))
734 * There are two policies for allocating an inode. If the new inode is
735 * a directory, then a forward search is made for a block group with both
736 * free space and a low directory-to-inode ratio; if that fails, then of
737 * the groups with above-average free space, that group with the fewest
738 * directories already is chosen.
740 * For other inodes, search forward from the parent directory's block
741 * group to find a free inode.
743 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
744 umode_t mode, const struct qstr *qstr,
745 __u32 goal, uid_t *owner, __u32 i_flags,
746 int handle_type, unsigned int line_no,
749 struct super_block *sb;
750 struct buffer_head *inode_bitmap_bh = NULL;
751 struct buffer_head *group_desc_bh;
752 ext4_group_t ngroups, group = 0;
753 unsigned long ino = 0;
755 struct ext4_group_desc *gdp = NULL;
756 struct ext4_inode_info *ei;
757 struct ext4_sb_info *sbi;
761 ext4_group_t flex_group;
762 struct ext4_group_info *grp;
765 /* Cannot create files in a deleted directory */
766 if (!dir || !dir->i_nlink)
767 return ERR_PTR(-EPERM);
772 if (unlikely(ext4_forced_shutdown(sbi)))
773 return ERR_PTR(-EIO);
775 if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
776 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
777 !(i_flags & EXT4_EA_INODE_FL)) {
778 err = fscrypt_get_encryption_info(dir);
781 if (!fscrypt_has_encryption_key(dir))
782 return ERR_PTR(-ENOKEY);
786 if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
787 #ifdef CONFIG_EXT4_FS_POSIX_ACL
788 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
793 int acl_size = p->a_count * sizeof(ext4_acl_entry);
795 nblocks += (S_ISDIR(mode) ? 2 : 1) *
796 __ext4_xattr_set_credits(sb, NULL /* inode */,
797 NULL /* block_bh */, acl_size,
798 true /* is_create */);
799 posix_acl_release(p);
803 #ifdef CONFIG_SECURITY
805 int num_security_xattrs = 1;
807 #ifdef CONFIG_INTEGRITY
808 num_security_xattrs++;
811 * We assume that security xattrs are never
812 * more than 1k. In practice they are under
815 nblocks += num_security_xattrs *
816 __ext4_xattr_set_credits(sb, NULL /* inode */,
817 NULL /* block_bh */, 1024,
818 true /* is_create */);
822 nblocks += __ext4_xattr_set_credits(sb,
823 NULL /* inode */, NULL /* block_bh */,
824 FSCRYPT_SET_CONTEXT_MAX_SIZE,
825 true /* is_create */);
828 ngroups = ext4_get_groups_count(sb);
829 trace_ext4_request_inode(dir, mode);
830 inode = new_inode(sb);
832 return ERR_PTR(-ENOMEM);
836 * Initialize owners and quota early so that we don't have to account
837 * for quota initialization worst case in standard inode creating
841 inode->i_mode = mode;
842 i_uid_write(inode, owner[0]);
843 i_gid_write(inode, owner[1]);
844 } else if (test_opt(sb, GRPID)) {
845 inode->i_mode = mode;
846 inode->i_uid = current_fsuid();
847 inode->i_gid = dir->i_gid;
849 inode_init_owner(inode, dir, mode);
851 if (ext4_has_feature_project(sb) &&
852 ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
853 ei->i_projid = EXT4_I(dir)->i_projid;
855 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
857 err = dquot_initialize(inode);
862 goal = sbi->s_inode_goal;
864 if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
865 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
866 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
872 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
874 ret2 = find_group_other(sb, dir, &group, mode);
877 EXT4_I(dir)->i_last_alloc_group = group;
883 * Normally we will only go through one pass of this loop,
884 * unless we get unlucky and it turns out the group we selected
885 * had its last inode grabbed by someone else.
887 for (i = 0; i < ngroups; i++, ino = 0) {
890 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
895 * Check free inodes count before loading bitmap.
897 if (ext4_free_inodes_count(sb, gdp) == 0)
900 grp = ext4_get_group_info(sb, group);
901 /* Skip groups with already-known suspicious inode tables */
902 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
905 brelse(inode_bitmap_bh);
906 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
907 /* Skip groups with suspicious inode tables */
908 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
909 IS_ERR(inode_bitmap_bh)) {
910 inode_bitmap_bh = NULL;
914 repeat_in_this_group:
915 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
919 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
920 ext4_error(sb, "reserved inode found cleared - "
921 "inode=%lu", ino + 1);
922 ext4_mark_group_bitmap_corrupted(sb, group,
923 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
928 BUG_ON(nblocks <= 0);
929 handle = __ext4_journal_start_sb(dir->i_sb, line_no,
930 handle_type, nblocks, 0,
931 ext4_trans_default_revoke_credits(sb));
932 if (IS_ERR(handle)) {
933 err = PTR_ERR(handle);
934 ext4_std_error(sb, err);
938 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
939 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
941 ext4_std_error(sb, err);
944 ext4_lock_group(sb, group);
945 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
947 /* Someone already took the bit. Repeat the search
950 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
952 ext4_set_bit(ino, inode_bitmap_bh->b_data);
955 ret2 = 1; /* we didn't grab the inode */
958 ext4_unlock_group(sb, group);
959 ino++; /* the inode bitmap is zero-based */
961 goto got; /* we grabbed the inode! */
963 if (ino < EXT4_INODES_PER_GROUP(sb))
964 goto repeat_in_this_group;
966 if (++group == ngroups)
973 BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
974 err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
976 ext4_std_error(sb, err);
980 BUFFER_TRACE(group_desc_bh, "get_write_access");
981 err = ext4_journal_get_write_access(handle, group_desc_bh);
983 ext4_std_error(sb, err);
987 /* We may have to initialize the block bitmap if it isn't already */
988 if (ext4_has_group_desc_csum(sb) &&
989 gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
990 struct buffer_head *block_bitmap_bh;
992 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
993 if (IS_ERR(block_bitmap_bh)) {
994 err = PTR_ERR(block_bitmap_bh);
997 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
998 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1000 brelse(block_bitmap_bh);
1001 ext4_std_error(sb, err);
1005 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1006 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1008 /* recheck and clear flag under lock if we still need to */
1009 ext4_lock_group(sb, group);
1010 if (ext4_has_group_desc_csum(sb) &&
1011 (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1012 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1013 ext4_free_group_clusters_set(sb, gdp,
1014 ext4_free_clusters_after_init(sb, group, gdp));
1015 ext4_block_bitmap_csum_set(sb, group, gdp,
1017 ext4_group_desc_csum_set(sb, group, gdp);
1019 ext4_unlock_group(sb, group);
1020 brelse(block_bitmap_bh);
1023 ext4_std_error(sb, err);
1028 /* Update the relevant bg descriptor fields */
1029 if (ext4_has_group_desc_csum(sb)) {
1031 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1033 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1034 ext4_lock_group(sb, group); /* while we modify the bg desc */
1035 free = EXT4_INODES_PER_GROUP(sb) -
1036 ext4_itable_unused_count(sb, gdp);
1037 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1038 gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1042 * Check the relative inode number against the last used
1043 * relative inode number in this group. if it is greater
1044 * we need to update the bg_itable_unused count
1047 ext4_itable_unused_set(sb, gdp,
1048 (EXT4_INODES_PER_GROUP(sb) - ino));
1049 up_read(&grp->alloc_sem);
1051 ext4_lock_group(sb, group);
1054 ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1055 if (S_ISDIR(mode)) {
1056 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1057 if (sbi->s_log_groups_per_flex) {
1058 ext4_group_t f = ext4_flex_group(sbi, group);
1060 atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1064 if (ext4_has_group_desc_csum(sb)) {
1065 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1066 EXT4_INODES_PER_GROUP(sb) / 8);
1067 ext4_group_desc_csum_set(sb, group, gdp);
1069 ext4_unlock_group(sb, group);
1071 BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1072 err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1074 ext4_std_error(sb, err);
1078 percpu_counter_dec(&sbi->s_freeinodes_counter);
1080 percpu_counter_inc(&sbi->s_dirs_counter);
1082 if (sbi->s_log_groups_per_flex) {
1083 flex_group = ext4_flex_group(sbi, group);
1084 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1085 flex_group)->free_inodes);
1088 inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1089 /* This is the optimal IO size (for stat), not the fs block size */
1090 inode->i_blocks = 0;
1091 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1092 ei->i_crtime = inode->i_mtime;
1094 memset(ei->i_data, 0, sizeof(ei->i_data));
1095 ei->i_dir_start_lookup = 0;
1098 /* Don't inherit extent flag from directory, amongst others. */
1100 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1101 ei->i_flags |= i_flags;
1104 ei->i_block_group = group;
1105 ei->i_last_alloc_group = ~0;
1107 ext4_set_inode_flags(inode);
1108 if (IS_DIRSYNC(inode))
1109 ext4_handle_sync(handle);
1110 if (insert_inode_locked(inode) < 0) {
1112 * Likely a bitmap corruption causing inode to be allocated
1116 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1118 ext4_mark_group_bitmap_corrupted(sb, group,
1119 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1122 inode->i_generation = prandom_u32();
1124 /* Precompute checksum seed for inode metadata */
1125 if (ext4_has_metadata_csum(sb)) {
1127 __le32 inum = cpu_to_le32(inode->i_ino);
1128 __le32 gen = cpu_to_le32(inode->i_generation);
1129 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1131 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1135 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1136 ext4_set_inode_state(inode, EXT4_STATE_NEW);
1138 ei->i_extra_isize = sbi->s_want_extra_isize;
1139 ei->i_inline_off = 0;
1140 if (ext4_has_feature_inline_data(sb))
1141 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1143 err = dquot_alloc_inode(inode);
1148 * Since the encryption xattr will always be unique, create it first so
1149 * that it's less likely to end up in an external xattr block and
1150 * prevent its deduplication.
1153 err = fscrypt_inherit_context(dir, inode, handle, true);
1155 goto fail_free_drop;
1158 if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1159 err = ext4_init_acl(handle, inode, dir);
1161 goto fail_free_drop;
1163 err = ext4_init_security(handle, inode, dir, qstr);
1165 goto fail_free_drop;
1168 if (ext4_has_feature_extents(sb)) {
1169 /* set extent flag only for directory, file and normal symlink*/
1170 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1171 ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1172 ext4_ext_tree_init(handle, inode);
1176 if (ext4_handle_valid(handle)) {
1177 ei->i_sync_tid = handle->h_transaction->t_tid;
1178 ei->i_datasync_tid = handle->h_transaction->t_tid;
1181 err = ext4_mark_inode_dirty(handle, inode);
1183 ext4_std_error(sb, err);
1184 goto fail_free_drop;
1187 ext4_debug("allocating inode %lu\n", inode->i_ino);
1188 trace_ext4_allocate_inode(inode, dir, mode);
1189 brelse(inode_bitmap_bh);
1193 dquot_free_inode(inode);
1196 unlock_new_inode(inode);
1199 inode->i_flags |= S_NOQUOTA;
1201 brelse(inode_bitmap_bh);
1202 return ERR_PTR(err);
1205 /* Verify that we are loading a valid orphan from disk */
1206 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1208 unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1209 ext4_group_t block_group;
1211 struct buffer_head *bitmap_bh = NULL;
1212 struct inode *inode = NULL;
1213 int err = -EFSCORRUPTED;
1215 if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1218 block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1219 bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1220 bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1221 if (IS_ERR(bitmap_bh))
1222 return ERR_CAST(bitmap_bh);
1224 /* Having the inode bit set should be a 100% indicator that this
1225 * is a valid orphan (no e2fsck run on fs). Orphans also include
1226 * inodes that were being truncated, so we can't check i_nlink==0.
1228 if (!ext4_test_bit(bit, bitmap_bh->b_data))
1231 inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1232 if (IS_ERR(inode)) {
1233 err = PTR_ERR(inode);
1234 ext4_set_errno(sb, -err);
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;
1361 /* This should not happen, but just to be sure check this */
1362 if (sb_rdonly(sb)) {
1367 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1372 * We do not need to lock this, because we are the only one
1373 * handling this flag.
1375 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1378 handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1379 if (IS_ERR(handle)) {
1380 ret = PTR_ERR(handle);
1384 down_write(&grp->alloc_sem);
1386 * If inode bitmap was already initialized there may be some
1387 * used inodes so we need to skip blocks with used inodes in
1390 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1391 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1392 ext4_itable_unused_count(sb, gdp)),
1393 sbi->s_inodes_per_block);
1395 if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1396 ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1397 ext4_itable_unused_count(sb, gdp)) <
1398 EXT4_FIRST_INO(sb)))) {
1399 ext4_error(sb, "Something is wrong with group %u: "
1400 "used itable blocks: %d; "
1401 "itable unused count: %u",
1403 ext4_itable_unused_count(sb, gdp));
1408 blk = ext4_inode_table(sb, gdp) + used_blks;
1409 num = sbi->s_itb_per_group - used_blks;
1411 BUFFER_TRACE(group_desc_bh, "get_write_access");
1412 ret = ext4_journal_get_write_access(handle,
1418 * Skip zeroout if the inode table is full. But we set the ZEROED
1419 * flag anyway, because obviously, when it is full it does not need
1422 if (unlikely(num == 0))
1425 ext4_debug("going to zero out inode table in group %d\n",
1427 ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1431 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1434 ext4_lock_group(sb, group);
1435 gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1436 ext4_group_desc_csum_set(sb, group, gdp);
1437 ext4_unlock_group(sb, group);
1439 BUFFER_TRACE(group_desc_bh,
1440 "call ext4_handle_dirty_metadata");
1441 ret = ext4_handle_dirty_metadata(handle, NULL,
1445 up_write(&grp->alloc_sem);
1446 ext4_journal_stop(handle);