1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/namei.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)
12 * linux/fs/minix/namei.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 * Directory entry file type support and forward compatibility hooks
19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20 * Hash Tree Directory indexing (c)
21 * Daniel Phillips, 2001
22 * Hash Tree Directory indexing porting
23 * Christopher Li, 2002
24 * Hash Tree Directory indexing cleanup
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include <linux/iversion.h>
38 #include <linux/unicode.h>
40 #include "ext4_jbd2.h"
45 #include <trace/events/ext4.h>
47 * define how far ahead to read directories while searching them.
49 #define NAMEI_RA_CHUNKS 2
50 #define NAMEI_RA_BLOCKS 4
51 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
53 static struct buffer_head *ext4_append(handle_t *handle,
57 struct ext4_map_blocks map;
58 struct buffer_head *bh;
61 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
62 ((inode->i_size >> 10) >=
63 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
64 return ERR_PTR(-ENOSPC);
66 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
71 * We're appending new directory block. Make sure the block is not
72 * allocated yet, otherwise we will end up corrupting the
75 err = ext4_map_blocks(NULL, inode, &map, 0);
79 EXT4_ERROR_INODE(inode, "Logical block already allocated");
80 return ERR_PTR(-EFSCORRUPTED);
83 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
86 inode->i_size += inode->i_sb->s_blocksize;
87 EXT4_I(inode)->i_disksize = inode->i_size;
88 err = ext4_mark_inode_dirty(handle, inode);
91 BUFFER_TRACE(bh, "get_write_access");
92 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
100 ext4_std_error(inode->i_sb, err);
104 static int ext4_dx_csum_verify(struct inode *inode,
105 struct ext4_dir_entry *dirent);
108 * Hints to ext4_read_dirblock regarding whether we expect a directory
109 * block being read to be an index block, or a block containing
110 * directory entries (and if the latter, whether it was found via a
111 * logical block in an htree index block). This is used to control
112 * what sort of sanity checkinig ext4_read_dirblock() will do on the
113 * directory block read from the storage device. EITHER will means
114 * the caller doesn't know what kind of directory block will be read,
115 * so no specific verification will be done.
118 EITHER, INDEX, DIRENT, DIRENT_HTREE
121 #define ext4_read_dirblock(inode, block, type) \
122 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
124 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
126 dirblock_type_t type,
130 struct buffer_head *bh;
131 struct ext4_dir_entry *dirent;
134 if (block >= inode->i_size >> inode->i_blkbits) {
135 ext4_error_inode(inode, func, line, block,
136 "Attempting to read directory block (%u) that is past i_size (%llu)",
137 block, inode->i_size);
138 return ERR_PTR(-EFSCORRUPTED);
141 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
144 bh = ext4_bread(NULL, inode, block, 0);
146 __ext4_warning(inode->i_sb, func, line,
147 "inode #%lu: lblock %lu: comm %s: "
148 "error %ld reading directory block",
149 inode->i_ino, (unsigned long)block,
150 current->comm, PTR_ERR(bh));
154 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
155 ext4_error_inode(inode, func, line, block,
156 "Directory hole found for htree %s block",
157 (type == INDEX) ? "index" : "leaf");
158 return ERR_PTR(-EFSCORRUPTED);
162 dirent = (struct ext4_dir_entry *) bh->b_data;
163 /* Determine whether or not we have an index block */
167 else if (ext4_rec_len_from_disk(dirent->rec_len,
168 inode->i_sb->s_blocksize) ==
169 inode->i_sb->s_blocksize)
172 if (!is_dx_block && type == INDEX) {
173 ext4_error_inode(inode, func, line, block,
174 "directory leaf block found instead of index block");
176 return ERR_PTR(-EFSCORRUPTED);
178 if (!ext4_has_metadata_csum(inode->i_sb) ||
183 * An empty leaf block can get mistaken for a index block; for
184 * this reason, we can only check the index checksum when the
185 * caller is sure it should be an index block.
187 if (is_dx_block && type == INDEX) {
188 if (ext4_dx_csum_verify(inode, dirent) &&
189 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
190 set_buffer_verified(bh);
192 ext4_error_inode_err(inode, func, line, block,
194 "Directory index failed checksum");
196 return ERR_PTR(-EFSBADCRC);
200 if (ext4_dirblock_csum_verify(inode, bh) &&
201 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
202 set_buffer_verified(bh);
204 ext4_error_inode_err(inode, func, line, block,
206 "Directory block failed checksum");
208 return ERR_PTR(-EFSBADCRC);
215 #define dxtrace(command) command
217 #define dxtrace(command)
241 * dx_root_info is laid out so that if it should somehow get overlaid by a
242 * dirent the two low bits of the hash version will be zero. Therefore, the
243 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
248 struct fake_dirent dot;
250 struct fake_dirent dotdot;
254 __le32 reserved_zero;
256 u8 info_length; /* 8 */
261 struct dx_entry entries[];
266 struct fake_dirent fake;
267 struct dx_entry entries[];
273 struct buffer_head *bh;
274 struct dx_entry *entries;
286 * This goes at the end of each htree block.
290 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
293 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
294 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
295 static inline unsigned dx_get_hash(struct dx_entry *entry);
296 static void dx_set_hash(struct dx_entry *entry, unsigned value);
297 static unsigned dx_get_count(struct dx_entry *entries);
298 static unsigned dx_get_limit(struct dx_entry *entries);
299 static void dx_set_count(struct dx_entry *entries, unsigned value);
300 static void dx_set_limit(struct dx_entry *entries, unsigned value);
301 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
302 static unsigned dx_node_limit(struct inode *dir);
303 static struct dx_frame *dx_probe(struct ext4_filename *fname,
305 struct dx_hash_info *hinfo,
306 struct dx_frame *frame);
307 static void dx_release(struct dx_frame *frames);
308 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
309 struct dx_hash_info *hinfo,
310 struct dx_map_entry *map_tail);
311 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
312 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
313 char *to, struct dx_map_entry *offsets,
314 int count, unsigned int blocksize);
315 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
316 unsigned int blocksize);
317 static void dx_insert_block(struct dx_frame *frame,
318 u32 hash, ext4_lblk_t block);
319 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
320 struct dx_frame *frame,
321 struct dx_frame *frames,
323 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
324 struct ext4_filename *fname,
325 struct ext4_dir_entry_2 **res_dir);
326 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
327 struct inode *dir, struct inode *inode);
329 /* checksumming functions */
330 void ext4_initialize_dirent_tail(struct buffer_head *bh,
331 unsigned int blocksize)
333 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
335 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
336 t->det_rec_len = ext4_rec_len_to_disk(
337 sizeof(struct ext4_dir_entry_tail), blocksize);
338 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
341 /* Walk through a dirent block to find a checksum "dirent" at the tail */
342 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
343 struct buffer_head *bh)
345 struct ext4_dir_entry_tail *t;
348 struct ext4_dir_entry *d, *top;
350 d = (struct ext4_dir_entry *)bh->b_data;
351 top = (struct ext4_dir_entry *)(bh->b_data +
352 (EXT4_BLOCK_SIZE(inode->i_sb) -
353 sizeof(struct ext4_dir_entry_tail)));
354 while (d < top && d->rec_len)
355 d = (struct ext4_dir_entry *)(((void *)d) +
356 le16_to_cpu(d->rec_len));
361 t = (struct ext4_dir_entry_tail *)d;
363 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
366 if (t->det_reserved_zero1 ||
367 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
368 t->det_reserved_zero2 ||
369 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
375 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
377 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
378 struct ext4_inode_info *ei = EXT4_I(inode);
381 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
382 return cpu_to_le32(csum);
385 #define warn_no_space_for_csum(inode) \
386 __warn_no_space_for_csum((inode), __func__, __LINE__)
388 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
391 __ext4_warning_inode(inode, func, line,
392 "No space for directory leaf checksum. Please run e2fsck -D.");
395 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
397 struct ext4_dir_entry_tail *t;
399 if (!ext4_has_metadata_csum(inode->i_sb))
402 t = get_dirent_tail(inode, bh);
404 warn_no_space_for_csum(inode);
408 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
409 (char *)t - bh->b_data))
415 static void ext4_dirblock_csum_set(struct inode *inode,
416 struct buffer_head *bh)
418 struct ext4_dir_entry_tail *t;
420 if (!ext4_has_metadata_csum(inode->i_sb))
423 t = get_dirent_tail(inode, bh);
425 warn_no_space_for_csum(inode);
429 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
430 (char *)t - bh->b_data);
433 int ext4_handle_dirty_dirblock(handle_t *handle,
435 struct buffer_head *bh)
437 ext4_dirblock_csum_set(inode, bh);
438 return ext4_handle_dirty_metadata(handle, inode, bh);
441 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
442 struct ext4_dir_entry *dirent,
445 struct ext4_dir_entry *dp;
446 struct dx_root_info *root;
449 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
451 else if (le16_to_cpu(dirent->rec_len) == 12) {
452 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
453 if (le16_to_cpu(dp->rec_len) !=
454 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
456 root = (struct dx_root_info *)(((void *)dp + 12));
457 if (root->reserved_zero ||
458 root->info_length != sizeof(struct dx_root_info))
465 *offset = count_offset;
466 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
469 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
470 int count_offset, int count, struct dx_tail *t)
472 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
473 struct ext4_inode_info *ei = EXT4_I(inode);
476 __u32 dummy_csum = 0;
477 int offset = offsetof(struct dx_tail, dt_checksum);
479 size = count_offset + (count * sizeof(struct dx_entry));
480 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
481 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
482 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
484 return cpu_to_le32(csum);
487 static int ext4_dx_csum_verify(struct inode *inode,
488 struct ext4_dir_entry *dirent)
490 struct dx_countlimit *c;
492 int count_offset, limit, count;
494 if (!ext4_has_metadata_csum(inode->i_sb))
497 c = get_dx_countlimit(inode, dirent, &count_offset);
499 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
502 limit = le16_to_cpu(c->limit);
503 count = le16_to_cpu(c->count);
504 if (count_offset + (limit * sizeof(struct dx_entry)) >
505 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
506 warn_no_space_for_csum(inode);
509 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
511 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
517 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
519 struct dx_countlimit *c;
521 int count_offset, limit, count;
523 if (!ext4_has_metadata_csum(inode->i_sb))
526 c = get_dx_countlimit(inode, dirent, &count_offset);
528 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
531 limit = le16_to_cpu(c->limit);
532 count = le16_to_cpu(c->count);
533 if (count_offset + (limit * sizeof(struct dx_entry)) >
534 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
535 warn_no_space_for_csum(inode);
538 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
540 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
543 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
545 struct buffer_head *bh)
547 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
548 return ext4_handle_dirty_metadata(handle, inode, bh);
552 * p is at least 6 bytes before the end of page
554 static inline struct ext4_dir_entry_2 *
555 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
557 return (struct ext4_dir_entry_2 *)((char *)p +
558 ext4_rec_len_from_disk(p->rec_len, blocksize));
562 * Future: use high four bits of block for coalesce-on-delete flags
563 * Mask them off for now.
566 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
568 return le32_to_cpu(entry->block) & 0x0fffffff;
571 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
573 entry->block = cpu_to_le32(value);
576 static inline unsigned dx_get_hash(struct dx_entry *entry)
578 return le32_to_cpu(entry->hash);
581 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
583 entry->hash = cpu_to_le32(value);
586 static inline unsigned dx_get_count(struct dx_entry *entries)
588 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
591 static inline unsigned dx_get_limit(struct dx_entry *entries)
593 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
596 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
598 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
601 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
603 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
606 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
608 unsigned int entry_space = dir->i_sb->s_blocksize -
609 ext4_dir_rec_len(1, NULL) -
610 ext4_dir_rec_len(2, NULL) - infosize;
612 if (ext4_has_metadata_csum(dir->i_sb))
613 entry_space -= sizeof(struct dx_tail);
614 return entry_space / sizeof(struct dx_entry);
617 static inline unsigned dx_node_limit(struct inode *dir)
619 unsigned int entry_space = dir->i_sb->s_blocksize -
620 ext4_dir_rec_len(0, dir);
622 if (ext4_has_metadata_csum(dir->i_sb))
623 entry_space -= sizeof(struct dx_tail);
624 return entry_space / sizeof(struct dx_entry);
631 static void dx_show_index(char * label, struct dx_entry *entries)
633 int i, n = dx_get_count (entries);
634 printk(KERN_DEBUG "%s index", label);
635 for (i = 0; i < n; i++) {
636 printk(KERN_CONT " %x->%lu",
637 i ? dx_get_hash(entries + i) : 0,
638 (unsigned long)dx_get_block(entries + i));
640 printk(KERN_CONT "\n");
650 static struct stats dx_show_leaf(struct inode *dir,
651 struct dx_hash_info *hinfo,
652 struct ext4_dir_entry_2 *de,
653 int size, int show_names)
655 unsigned names = 0, space = 0;
656 char *base = (char *) de;
657 struct dx_hash_info h = *hinfo;
660 while ((char *) de < base + size)
666 #ifdef CONFIG_FS_ENCRYPTION
669 struct fscrypt_str fname_crypto_str =
675 if (!IS_ENCRYPTED(dir)) {
676 /* Directory is not encrypted */
677 ext4fs_dirhash(dir, de->name,
679 printk("%*.s:(U)%x.%u ", len,
681 (unsigned) ((char *) de
684 struct fscrypt_str de_name =
685 FSTR_INIT(name, len);
687 /* Directory is encrypted */
688 res = fscrypt_fname_alloc_buffer(
689 len, &fname_crypto_str);
691 printk(KERN_WARNING "Error "
695 res = fscrypt_fname_disk_to_usr(dir,
699 printk(KERN_WARNING "Error "
700 "converting filename "
706 name = fname_crypto_str.name;
707 len = fname_crypto_str.len;
709 if (IS_CASEFOLDED(dir))
710 h.hash = EXT4_DIRENT_HASH(de);
712 ext4fs_dirhash(dir, de->name,
714 printk("%*.s:(E)%x.%u ", len, name,
715 h.hash, (unsigned) ((char *) de
717 fscrypt_fname_free_buffer(
721 int len = de->name_len;
722 char *name = de->name;
723 ext4fs_dirhash(dir, de->name, de->name_len, &h);
724 printk("%*.s:%x.%u ", len, name, h.hash,
725 (unsigned) ((char *) de - base));
728 space += ext4_dir_rec_len(de->name_len, dir);
731 de = ext4_next_entry(de, size);
733 printk(KERN_CONT "(%i)\n", names);
734 return (struct stats) { names, space, 1 };
737 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
738 struct dx_entry *entries, int levels)
740 unsigned blocksize = dir->i_sb->s_blocksize;
741 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
743 struct buffer_head *bh;
744 printk("%i indexed blocks...\n", count);
745 for (i = 0; i < count; i++, entries++)
747 ext4_lblk_t block = dx_get_block(entries);
748 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
749 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
751 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
752 bh = ext4_bread(NULL,dir, block, 0);
753 if (!bh || IS_ERR(bh))
756 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
757 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
758 bh->b_data, blocksize, 0);
759 names += stats.names;
760 space += stats.space;
761 bcount += stats.bcount;
765 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
766 levels ? "" : " ", names, space/bcount,
767 (space/bcount)*100/blocksize);
768 return (struct stats) { names, space, bcount};
772 * Linear search cross check
774 static inline void htree_rep_invariant_check(struct dx_entry *at,
775 struct dx_entry *target,
776 u32 hash, unsigned int n)
779 dxtrace(printk(KERN_CONT ","));
780 if (dx_get_hash(++at) > hash) {
785 ASSERT(at == target - 1);
788 static inline void htree_rep_invariant_check(struct dx_entry *at,
789 struct dx_entry *target,
790 u32 hash, unsigned int n)
793 #endif /* DX_DEBUG */
796 * Probe for a directory leaf block to search.
798 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
799 * error in the directory index, and the caller should fall back to
800 * searching the directory normally. The callers of dx_probe **MUST**
801 * check for this error code, and make sure it never gets reflected
804 static struct dx_frame *
805 dx_probe(struct ext4_filename *fname, struct inode *dir,
806 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
808 unsigned count, indirect, level, i;
809 struct dx_entry *at, *entries, *p, *q, *m;
810 struct dx_root *root;
811 struct dx_frame *frame = frame_in;
812 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
815 ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
817 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
818 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
819 if (IS_ERR(frame->bh))
820 return (struct dx_frame *) frame->bh;
822 root = (struct dx_root *) frame->bh->b_data;
823 if (root->info.hash_version != DX_HASH_TEA &&
824 root->info.hash_version != DX_HASH_HALF_MD4 &&
825 root->info.hash_version != DX_HASH_LEGACY &&
826 root->info.hash_version != DX_HASH_SIPHASH) {
827 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
828 root->info.hash_version);
831 if (ext4_hash_in_dirent(dir)) {
832 if (root->info.hash_version != DX_HASH_SIPHASH) {
833 ext4_warning_inode(dir,
834 "Hash in dirent, but hash is not SIPHASH");
838 if (root->info.hash_version == DX_HASH_SIPHASH) {
839 ext4_warning_inode(dir,
840 "Hash code is SIPHASH, but hash not in dirent");
845 hinfo = &fname->hinfo;
846 hinfo->hash_version = root->info.hash_version;
847 if (hinfo->hash_version <= DX_HASH_TEA)
848 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
849 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
850 /* hash is already computed for encrypted casefolded directory */
851 if (fname && fname_name(fname) &&
852 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)))
853 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
856 if (root->info.unused_flags & 1) {
857 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
858 root->info.unused_flags);
862 indirect = root->info.indirect_levels;
863 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
864 ext4_warning(dir->i_sb,
865 "Directory (ino: %lu) htree depth %#06x exceed"
866 "supported value", dir->i_ino,
867 ext4_dir_htree_level(dir->i_sb));
868 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
869 ext4_warning(dir->i_sb, "Enable large directory "
870 "feature to access it");
875 entries = (struct dx_entry *)(((char *)&root->info) +
876 root->info.info_length);
878 if (dx_get_limit(entries) != dx_root_limit(dir,
879 root->info.info_length)) {
880 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
881 dx_get_limit(entries),
882 dx_root_limit(dir, root->info.info_length));
886 dxtrace(printk("Look up %x", hash));
890 count = dx_get_count(entries);
891 if (!count || count > dx_get_limit(entries)) {
892 ext4_warning_inode(dir,
893 "dx entry: count %u beyond limit %u",
894 count, dx_get_limit(entries));
899 q = entries + count - 1;
902 dxtrace(printk(KERN_CONT "."));
903 if (dx_get_hash(m) > hash)
909 htree_rep_invariant_check(entries, p, hash, count - 1);
912 dxtrace(printk(KERN_CONT " %x->%u\n",
913 at == entries ? 0 : dx_get_hash(at),
915 frame->entries = entries;
918 block = dx_get_block(at);
919 for (i = 0; i <= level; i++) {
920 if (blocks[i] == block) {
921 ext4_warning_inode(dir,
922 "dx entry: tree cycle block %u points back to block %u",
923 blocks[level], block);
927 if (++level > indirect)
929 blocks[level] = block;
931 frame->bh = ext4_read_dirblock(dir, block, INDEX);
932 if (IS_ERR(frame->bh)) {
933 ret_err = (struct dx_frame *) frame->bh;
938 entries = ((struct dx_node *) frame->bh->b_data)->entries;
940 if (dx_get_limit(entries) != dx_node_limit(dir)) {
941 ext4_warning_inode(dir,
942 "dx entry: limit %u != node limit %u",
943 dx_get_limit(entries), dx_node_limit(dir));
948 while (frame >= frame_in) {
953 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
954 ext4_warning_inode(dir,
955 "Corrupt directory, running e2fsck is recommended");
959 static void dx_release(struct dx_frame *frames)
961 struct dx_root_info *info;
963 unsigned int indirect_levels;
965 if (frames[0].bh == NULL)
968 info = &((struct dx_root *)frames[0].bh->b_data)->info;
969 /* save local copy, "info" may be freed after brelse() */
970 indirect_levels = info->indirect_levels;
971 for (i = 0; i <= indirect_levels; i++) {
972 if (frames[i].bh == NULL)
974 brelse(frames[i].bh);
980 * This function increments the frame pointer to search the next leaf
981 * block, and reads in the necessary intervening nodes if the search
982 * should be necessary. Whether or not the search is necessary is
983 * controlled by the hash parameter. If the hash value is even, then
984 * the search is only continued if the next block starts with that
985 * hash value. This is used if we are searching for a specific file.
987 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
989 * This function returns 1 if the caller should continue to search,
990 * or 0 if it should not. If there is an error reading one of the
991 * index blocks, it will a negative error code.
993 * If start_hash is non-null, it will be filled in with the starting
994 * hash of the next page.
996 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
997 struct dx_frame *frame,
998 struct dx_frame *frames,
1002 struct buffer_head *bh;
1008 * Find the next leaf page by incrementing the frame pointer.
1009 * If we run out of entries in the interior node, loop around and
1010 * increment pointer in the parent node. When we break out of
1011 * this loop, num_frames indicates the number of interior
1012 * nodes need to be read.
1015 if (++(p->at) < p->entries + dx_get_count(p->entries))
1024 * If the hash is 1, then continue only if the next page has a
1025 * continuation hash of any value. This is used for readdir
1026 * handling. Otherwise, check to see if the hash matches the
1027 * desired continuation hash. If it doesn't, return since
1028 * there's no point to read in the successive index pages.
1030 bhash = dx_get_hash(p->at);
1032 *start_hash = bhash;
1033 if ((hash & 1) == 0) {
1034 if ((bhash & ~1) != hash)
1038 * If the hash is HASH_NB_ALWAYS, we always go to the next
1039 * block so no check is necessary
1041 while (num_frames--) {
1042 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1048 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1055 * This function fills a red-black tree with information from a
1056 * directory block. It returns the number directory entries loaded
1057 * into the tree. If there is an error it is returned in err.
1059 static int htree_dirblock_to_tree(struct file *dir_file,
1060 struct inode *dir, ext4_lblk_t block,
1061 struct dx_hash_info *hinfo,
1062 __u32 start_hash, __u32 start_minor_hash)
1064 struct buffer_head *bh;
1065 struct ext4_dir_entry_2 *de, *top;
1066 int err = 0, count = 0;
1067 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1068 int csum = ext4_has_metadata_csum(dir->i_sb);
1070 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1071 (unsigned long)block));
1072 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1076 de = (struct ext4_dir_entry_2 *) bh->b_data;
1077 /* csum entries are not larger in the casefolded encrypted case */
1078 top = (struct ext4_dir_entry_2 *) ((char *) de +
1079 dir->i_sb->s_blocksize -
1081 csum ? NULL : dir));
1082 /* Check if the directory is encrypted */
1083 if (IS_ENCRYPTED(dir)) {
1084 err = fscrypt_prepare_readdir(dir);
1089 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1097 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1098 if (ext4_check_dir_entry(dir, NULL, de, bh,
1099 bh->b_data, bh->b_size,
1100 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1101 + ((char *)de - bh->b_data))) {
1102 /* silently ignore the rest of the block */
1105 if (ext4_hash_in_dirent(dir)) {
1106 if (de->name_len && de->inode) {
1107 hinfo->hash = EXT4_DIRENT_HASH(de);
1108 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1111 hinfo->minor_hash = 0;
1114 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1116 if ((hinfo->hash < start_hash) ||
1117 ((hinfo->hash == start_hash) &&
1118 (hinfo->minor_hash < start_minor_hash)))
1122 if (!IS_ENCRYPTED(dir)) {
1123 tmp_str.name = de->name;
1124 tmp_str.len = de->name_len;
1125 err = ext4_htree_store_dirent(dir_file,
1126 hinfo->hash, hinfo->minor_hash, de,
1129 int save_len = fname_crypto_str.len;
1130 struct fscrypt_str de_name = FSTR_INIT(de->name,
1133 /* Directory is encrypted */
1134 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1135 hinfo->minor_hash, &de_name,
1141 err = ext4_htree_store_dirent(dir_file,
1142 hinfo->hash, hinfo->minor_hash, de,
1144 fname_crypto_str.len = save_len;
1154 fscrypt_fname_free_buffer(&fname_crypto_str);
1160 * This function fills a red-black tree with information from a
1161 * directory. We start scanning the directory in hash order, starting
1162 * at start_hash and start_minor_hash.
1164 * This function returns the number of entries inserted into the tree,
1165 * or a negative error code.
1167 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1168 __u32 start_minor_hash, __u32 *next_hash)
1170 struct dx_hash_info hinfo;
1171 struct ext4_dir_entry_2 *de;
1172 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1178 struct fscrypt_str tmp_str;
1180 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1181 start_hash, start_minor_hash));
1182 dir = file_inode(dir_file);
1183 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1184 if (ext4_hash_in_dirent(dir))
1185 hinfo.hash_version = DX_HASH_SIPHASH;
1187 hinfo.hash_version =
1188 EXT4_SB(dir->i_sb)->s_def_hash_version;
1189 if (hinfo.hash_version <= DX_HASH_TEA)
1190 hinfo.hash_version +=
1191 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1192 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1193 if (ext4_has_inline_data(dir)) {
1194 int has_inline_data = 1;
1195 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1199 if (has_inline_data) {
1204 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1205 start_hash, start_minor_hash);
1209 hinfo.hash = start_hash;
1210 hinfo.minor_hash = 0;
1211 frame = dx_probe(NULL, dir, &hinfo, frames);
1213 return PTR_ERR(frame);
1215 /* Add '.' and '..' from the htree header */
1216 if (!start_hash && !start_minor_hash) {
1217 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1218 tmp_str.name = de->name;
1219 tmp_str.len = de->name_len;
1220 err = ext4_htree_store_dirent(dir_file, 0, 0,
1226 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1227 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1228 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1229 tmp_str.name = de->name;
1230 tmp_str.len = de->name_len;
1231 err = ext4_htree_store_dirent(dir_file, 2, 0,
1239 if (fatal_signal_pending(current)) {
1244 block = dx_get_block(frame->at);
1245 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1246 start_hash, start_minor_hash);
1253 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1254 frame, frames, &hashval);
1255 *next_hash = hashval;
1261 * Stop if: (a) there are no more entries, or
1262 * (b) we have inserted at least one entry and the
1263 * next hash value is not a continuation
1266 (count && ((hashval & 1) == 0)))
1270 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1271 "next hash: %x\n", count, *next_hash));
1278 static inline int search_dirblock(struct buffer_head *bh,
1280 struct ext4_filename *fname,
1281 unsigned int offset,
1282 struct ext4_dir_entry_2 **res_dir)
1284 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1285 fname, offset, res_dir);
1289 * Directory block splitting, compacting
1293 * Create map of hash values, offsets, and sizes, stored at end of block.
1294 * Returns number of entries mapped.
1296 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1297 struct dx_hash_info *hinfo,
1298 struct dx_map_entry *map_tail)
1301 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1302 unsigned int buflen = bh->b_size;
1303 char *base = bh->b_data;
1304 struct dx_hash_info h = *hinfo;
1306 if (ext4_has_metadata_csum(dir->i_sb))
1307 buflen -= sizeof(struct ext4_dir_entry_tail);
1309 while ((char *) de < base + buflen) {
1310 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1311 ((char *)de) - base))
1312 return -EFSCORRUPTED;
1313 if (de->name_len && de->inode) {
1314 if (ext4_hash_in_dirent(dir))
1315 h.hash = EXT4_DIRENT_HASH(de);
1317 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1319 map_tail->hash = h.hash;
1320 map_tail->offs = ((char *) de - base)>>2;
1321 map_tail->size = le16_to_cpu(de->rec_len);
1325 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1330 /* Sort map by hash value */
1331 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1333 struct dx_map_entry *p, *q, *top = map + count - 1;
1335 /* Combsort until bubble sort doesn't suck */
1337 count = count*10/13;
1338 if (count - 9 < 2) /* 9, 10 -> 11 */
1340 for (p = top, q = p - count; q >= map; p--, q--)
1341 if (p->hash < q->hash)
1344 /* Garden variety bubble sort */
1349 if (q[1].hash >= q[0].hash)
1357 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1359 struct dx_entry *entries = frame->entries;
1360 struct dx_entry *old = frame->at, *new = old + 1;
1361 int count = dx_get_count(entries);
1363 ASSERT(count < dx_get_limit(entries));
1364 ASSERT(old < entries + count);
1365 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1366 dx_set_hash(new, hash);
1367 dx_set_block(new, block);
1368 dx_set_count(entries, count + 1);
1371 #if IS_ENABLED(CONFIG_UNICODE)
1373 * Test whether a case-insensitive directory entry matches the filename
1374 * being searched for. If quick is set, assume the name being looked up
1375 * is already in the casefolded form.
1377 * Returns: 0 if the directory entry matches, more than 0 if it
1378 * doesn't match or less than zero on error.
1380 static int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1381 u8 *de_name, size_t de_name_len, bool quick)
1383 const struct super_block *sb = parent->i_sb;
1384 const struct unicode_map *um = sb->s_encoding;
1385 struct fscrypt_str decrypted_name = FSTR_INIT(NULL, de_name_len);
1386 struct qstr entry = QSTR_INIT(de_name, de_name_len);
1389 if (IS_ENCRYPTED(parent)) {
1390 const struct fscrypt_str encrypted_name =
1391 FSTR_INIT(de_name, de_name_len);
1393 decrypted_name.name = kmalloc(de_name_len, GFP_KERNEL);
1394 if (!decrypted_name.name)
1396 ret = fscrypt_fname_disk_to_usr(parent, 0, 0, &encrypted_name,
1400 entry.name = decrypted_name.name;
1401 entry.len = decrypted_name.len;
1405 ret = utf8_strncasecmp_folded(um, name, &entry);
1407 ret = utf8_strncasecmp(um, name, &entry);
1409 /* Handle invalid character sequence as either an error
1410 * or as an opaque byte sequence.
1412 if (sb_has_strict_encoding(sb))
1414 else if (name->len != entry.len)
1417 ret = !!memcmp(name->name, entry.name, entry.len);
1420 kfree(decrypted_name.name);
1424 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1425 struct ext4_filename *name)
1427 struct fscrypt_str *cf_name = &name->cf_name;
1428 struct dx_hash_info *hinfo = &name->hinfo;
1431 if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding ||
1432 (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1433 cf_name->name = NULL;
1437 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1441 len = utf8_casefold(dir->i_sb->s_encoding,
1442 iname, cf_name->name,
1445 kfree(cf_name->name);
1446 cf_name->name = NULL;
1448 cf_name->len = (unsigned) len;
1449 if (!IS_ENCRYPTED(dir))
1452 hinfo->hash_version = DX_HASH_SIPHASH;
1455 ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1457 ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1463 * Test whether a directory entry matches the filename being searched for.
1465 * Return: %true if the directory entry matches, otherwise %false.
1467 static bool ext4_match(struct inode *parent,
1468 const struct ext4_filename *fname,
1469 struct ext4_dir_entry_2 *de)
1471 struct fscrypt_name f;
1476 f.usr_fname = fname->usr_fname;
1477 f.disk_name = fname->disk_name;
1478 #ifdef CONFIG_FS_ENCRYPTION
1479 f.crypto_buf = fname->crypto_buf;
1482 #if IS_ENABLED(CONFIG_UNICODE)
1483 if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent) &&
1484 (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1485 if (fname->cf_name.name) {
1486 struct qstr cf = {.name = fname->cf_name.name,
1487 .len = fname->cf_name.len};
1488 if (IS_ENCRYPTED(parent)) {
1489 if (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1490 fname->hinfo.minor_hash !=
1491 EXT4_DIRENT_MINOR_HASH(de)) {
1496 return !ext4_ci_compare(parent, &cf, de->name,
1497 de->name_len, true);
1499 return !ext4_ci_compare(parent, fname->usr_fname, de->name,
1500 de->name_len, false);
1504 return fscrypt_match_name(&f, de->name, de->name_len);
1508 * Returns 0 if not found, -1 on failure, and 1 on success
1510 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1511 struct inode *dir, struct ext4_filename *fname,
1512 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1514 struct ext4_dir_entry_2 * de;
1518 de = (struct ext4_dir_entry_2 *)search_buf;
1519 dlimit = search_buf + buf_size;
1520 while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1521 /* this code is executed quadratically often */
1522 /* do minimal checking `by hand' */
1523 if (de->name + de->name_len <= dlimit &&
1524 ext4_match(dir, fname, de)) {
1525 /* found a match - just to be sure, do
1527 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1533 /* prevent looping on a bad block */
1534 de_len = ext4_rec_len_from_disk(de->rec_len,
1535 dir->i_sb->s_blocksize);
1539 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1544 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1545 struct ext4_dir_entry *de)
1547 struct super_block *sb = dir->i_sb;
1553 if (de->inode == 0 &&
1554 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1561 * __ext4_find_entry()
1563 * finds an entry in the specified directory with the wanted name. It
1564 * returns the cache buffer in which the entry was found, and the entry
1565 * itself (as a parameter - res_dir). It does NOT read the inode of the
1566 * entry - you'll have to do that yourself if you want to.
1568 * The returned buffer_head has ->b_count elevated. The caller is expected
1569 * to brelse() it when appropriate.
1571 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1572 struct ext4_filename *fname,
1573 struct ext4_dir_entry_2 **res_dir,
1576 struct super_block *sb;
1577 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1578 struct buffer_head *bh, *ret = NULL;
1579 ext4_lblk_t start, block;
1580 const u8 *name = fname->usr_fname->name;
1581 size_t ra_max = 0; /* Number of bh's in the readahead
1583 size_t ra_ptr = 0; /* Current index into readahead
1585 ext4_lblk_t nblocks;
1586 int i, namelen, retval;
1590 namelen = fname->usr_fname->len;
1591 if (namelen > EXT4_NAME_LEN)
1594 if (ext4_has_inline_data(dir)) {
1595 int has_inline_data = 1;
1596 ret = ext4_find_inline_entry(dir, fname, res_dir,
1598 if (has_inline_data) {
1601 goto cleanup_and_exit;
1605 if ((namelen <= 2) && (name[0] == '.') &&
1606 (name[1] == '.' || name[1] == '\0')) {
1608 * "." or ".." will only be in the first block
1609 * NFS may look up ".."; "." should be handled by the VFS
1616 ret = ext4_dx_find_entry(dir, fname, res_dir);
1618 * On success, or if the error was file not found,
1619 * return. Otherwise, fall back to doing a search the
1620 * old fashioned way.
1622 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1623 goto cleanup_and_exit;
1624 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1628 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1631 goto cleanup_and_exit;
1633 start = EXT4_I(dir)->i_dir_start_lookup;
1634 if (start >= nblocks)
1640 * We deal with the read-ahead logic here.
1643 if (ra_ptr >= ra_max) {
1644 /* Refill the readahead buffer */
1647 ra_max = start - block;
1649 ra_max = nblocks - block;
1650 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1651 retval = ext4_bread_batch(dir, block, ra_max,
1652 false /* wait */, bh_use);
1654 ret = ERR_PTR(retval);
1656 goto cleanup_and_exit;
1659 if ((bh = bh_use[ra_ptr++]) == NULL)
1662 if (!buffer_uptodate(bh)) {
1663 EXT4_ERROR_INODE_ERR(dir, EIO,
1664 "reading directory lblock %lu",
1665 (unsigned long) block);
1667 ret = ERR_PTR(-EIO);
1668 goto cleanup_and_exit;
1670 if (!buffer_verified(bh) &&
1671 !is_dx_internal_node(dir, block,
1672 (struct ext4_dir_entry *)bh->b_data) &&
1673 !ext4_dirblock_csum_verify(dir, bh)) {
1674 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1675 "checksumming directory "
1676 "block %lu", (unsigned long)block);
1678 ret = ERR_PTR(-EFSBADCRC);
1679 goto cleanup_and_exit;
1681 set_buffer_verified(bh);
1682 i = search_dirblock(bh, dir, fname,
1683 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1685 EXT4_I(dir)->i_dir_start_lookup = block;
1687 goto cleanup_and_exit;
1691 goto cleanup_and_exit;
1694 if (++block >= nblocks)
1696 } while (block != start);
1699 * If the directory has grown while we were searching, then
1700 * search the last part of the directory before giving up.
1703 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1704 if (block < nblocks) {
1710 /* Clean up the read-ahead blocks */
1711 for (; ra_ptr < ra_max; ra_ptr++)
1712 brelse(bh_use[ra_ptr]);
1716 static struct buffer_head *ext4_find_entry(struct inode *dir,
1717 const struct qstr *d_name,
1718 struct ext4_dir_entry_2 **res_dir,
1722 struct ext4_filename fname;
1723 struct buffer_head *bh;
1725 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1729 return ERR_PTR(err);
1731 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1733 ext4_fname_free_filename(&fname);
1737 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1738 struct dentry *dentry,
1739 struct ext4_dir_entry_2 **res_dir)
1742 struct ext4_filename fname;
1743 struct buffer_head *bh;
1745 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1746 generic_set_encrypted_ci_d_ops(dentry);
1750 return ERR_PTR(err);
1752 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1754 ext4_fname_free_filename(&fname);
1758 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1759 struct ext4_filename *fname,
1760 struct ext4_dir_entry_2 **res_dir)
1762 struct super_block * sb = dir->i_sb;
1763 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1764 struct buffer_head *bh;
1768 #ifdef CONFIG_FS_ENCRYPTION
1771 frame = dx_probe(fname, dir, NULL, frames);
1773 return (struct buffer_head *) frame;
1775 block = dx_get_block(frame->at);
1776 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1780 retval = search_dirblock(bh, dir, fname,
1781 block << EXT4_BLOCK_SIZE_BITS(sb),
1787 bh = ERR_PTR(ERR_BAD_DX_DIR);
1791 /* Check to see if we should continue to search */
1792 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1795 ext4_warning_inode(dir,
1796 "error %d reading directory index block",
1798 bh = ERR_PTR(retval);
1801 } while (retval == 1);
1805 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1811 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1813 struct inode *inode;
1814 struct ext4_dir_entry_2 *de;
1815 struct buffer_head *bh;
1817 if (dentry->d_name.len > EXT4_NAME_LEN)
1818 return ERR_PTR(-ENAMETOOLONG);
1820 bh = ext4_lookup_entry(dir, dentry, &de);
1822 return ERR_CAST(bh);
1825 __u32 ino = le32_to_cpu(de->inode);
1827 if (!ext4_valid_inum(dir->i_sb, ino)) {
1828 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1829 return ERR_PTR(-EFSCORRUPTED);
1831 if (unlikely(ino == dir->i_ino)) {
1832 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1834 return ERR_PTR(-EFSCORRUPTED);
1836 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1837 if (inode == ERR_PTR(-ESTALE)) {
1838 EXT4_ERROR_INODE(dir,
1839 "deleted inode referenced: %u",
1841 return ERR_PTR(-EFSCORRUPTED);
1843 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1844 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1845 !fscrypt_has_permitted_context(dir, inode)) {
1846 ext4_warning(inode->i_sb,
1847 "Inconsistent encryption contexts: %lu/%lu",
1848 dir->i_ino, inode->i_ino);
1850 return ERR_PTR(-EPERM);
1854 #if IS_ENABLED(CONFIG_UNICODE)
1855 if (!inode && IS_CASEFOLDED(dir)) {
1856 /* Eventually we want to call d_add_ci(dentry, NULL)
1857 * for negative dentries in the encoding case as
1858 * well. For now, prevent the negative dentry
1859 * from being cached.
1864 return d_splice_alias(inode, dentry);
1868 struct dentry *ext4_get_parent(struct dentry *child)
1871 struct ext4_dir_entry_2 * de;
1872 struct buffer_head *bh;
1874 bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1876 return ERR_CAST(bh);
1878 return ERR_PTR(-ENOENT);
1879 ino = le32_to_cpu(de->inode);
1882 if (!ext4_valid_inum(child->d_sb, ino)) {
1883 EXT4_ERROR_INODE(d_inode(child),
1884 "bad parent inode number: %u", ino);
1885 return ERR_PTR(-EFSCORRUPTED);
1888 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1892 * Move count entries from end of map between two memory locations.
1893 * Returns pointer to last entry moved.
1895 static struct ext4_dir_entry_2 *
1896 dx_move_dirents(struct inode *dir, char *from, char *to,
1897 struct dx_map_entry *map, int count,
1900 unsigned rec_len = 0;
1903 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1904 (from + (map->offs<<2));
1905 rec_len = ext4_dir_rec_len(de->name_len, dir);
1907 memcpy (to, de, rec_len);
1908 ((struct ext4_dir_entry_2 *) to)->rec_len =
1909 ext4_rec_len_to_disk(rec_len, blocksize);
1911 /* wipe dir_entry excluding the rec_len field */
1913 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1915 offsetof(struct ext4_dir_entry_2,
1921 return (struct ext4_dir_entry_2 *) (to - rec_len);
1925 * Compact each dir entry in the range to the minimal rec_len.
1926 * Returns pointer to last entry in range.
1928 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1929 unsigned int blocksize)
1931 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1932 unsigned rec_len = 0;
1935 while ((char*)de < base + blocksize) {
1936 next = ext4_next_entry(de, blocksize);
1937 if (de->inode && de->name_len) {
1938 rec_len = ext4_dir_rec_len(de->name_len, dir);
1940 memmove(to, de, rec_len);
1941 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1943 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1951 * Split a full leaf block to make room for a new dir entry.
1952 * Allocate a new block, and move entries so that they are approx. equally full.
1953 * Returns pointer to de in block into which the new entry will be inserted.
1955 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1956 struct buffer_head **bh,struct dx_frame *frame,
1957 struct dx_hash_info *hinfo)
1959 unsigned blocksize = dir->i_sb->s_blocksize;
1962 struct buffer_head *bh2;
1963 ext4_lblk_t newblock;
1965 struct dx_map_entry *map;
1966 char *data1 = (*bh)->b_data, *data2;
1967 unsigned split, move, size;
1968 struct ext4_dir_entry_2 *de = NULL, *de2;
1972 if (ext4_has_metadata_csum(dir->i_sb))
1973 csum_size = sizeof(struct ext4_dir_entry_tail);
1975 bh2 = ext4_append(handle, dir, &newblock);
1979 return (struct ext4_dir_entry_2 *) bh2;
1982 BUFFER_TRACE(*bh, "get_write_access");
1983 err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
1988 BUFFER_TRACE(frame->bh, "get_write_access");
1989 err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
1994 data2 = bh2->b_data;
1996 /* create map in the end of data2 block */
1997 map = (struct dx_map_entry *) (data2 + blocksize);
1998 count = dx_make_map(dir, *bh, hinfo, map);
2004 dx_sort_map(map, count);
2005 /* Ensure that neither split block is over half full */
2008 for (i = count-1; i >= 0; i--) {
2009 /* is more than half of this entry in 2nd half of the block? */
2010 if (size + map[i].size/2 > blocksize/2)
2012 size += map[i].size;
2016 * map index at which we will split
2018 * If the sum of active entries didn't exceed half the block size, just
2019 * split it in half by count; each resulting block will have at least
2020 * half the space free.
2023 split = count - move;
2027 hash2 = map[split].hash;
2028 continued = hash2 == map[split - 1].hash;
2029 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2030 (unsigned long)dx_get_block(frame->at),
2031 hash2, split, count-split));
2033 /* Fancy dance to stay within two buffers */
2034 de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2036 de = dx_pack_dirents(dir, data1, blocksize);
2037 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2040 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2044 ext4_initialize_dirent_tail(*bh, blocksize);
2045 ext4_initialize_dirent_tail(bh2, blocksize);
2048 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2050 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2053 /* Which block gets the new entry? */
2054 if (hinfo->hash >= hash2) {
2058 dx_insert_block(frame, hash2 + continued, newblock);
2059 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2062 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2066 dxtrace(dx_show_index("frame", frame->entries));
2073 ext4_std_error(dir->i_sb, err);
2074 return ERR_PTR(err);
2077 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2078 struct buffer_head *bh,
2079 void *buf, int buf_size,
2080 struct ext4_filename *fname,
2081 struct ext4_dir_entry_2 **dest_de)
2083 struct ext4_dir_entry_2 *de;
2084 unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2086 unsigned int offset = 0;
2090 top = buf + buf_size - reclen;
2091 while ((char *) de <= top) {
2092 if (ext4_check_dir_entry(dir, NULL, de, bh,
2093 buf, buf_size, offset))
2094 return -EFSCORRUPTED;
2095 if (ext4_match(dir, fname, de))
2097 nlen = ext4_dir_rec_len(de->name_len, dir);
2098 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2099 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2101 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2104 if ((char *) de > top)
2111 void ext4_insert_dentry(struct inode *dir,
2112 struct inode *inode,
2113 struct ext4_dir_entry_2 *de,
2115 struct ext4_filename *fname)
2120 nlen = ext4_dir_rec_len(de->name_len, dir);
2121 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2123 struct ext4_dir_entry_2 *de1 =
2124 (struct ext4_dir_entry_2 *)((char *)de + nlen);
2125 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2126 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2129 de->file_type = EXT4_FT_UNKNOWN;
2130 de->inode = cpu_to_le32(inode->i_ino);
2131 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2132 de->name_len = fname_len(fname);
2133 memcpy(de->name, fname_name(fname), fname_len(fname));
2134 if (ext4_hash_in_dirent(dir)) {
2135 struct dx_hash_info *hinfo = &fname->hinfo;
2137 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2138 EXT4_DIRENT_HASHES(de)->minor_hash =
2139 cpu_to_le32(hinfo->minor_hash);
2144 * Add a new entry into a directory (leaf) block. If de is non-NULL,
2145 * it points to a directory entry which is guaranteed to be large
2146 * enough for new directory entry. If de is NULL, then
2147 * add_dirent_to_buf will attempt search the directory block for
2148 * space. It will return -ENOSPC if no space is available, and -EIO
2149 * and -EEXIST if directory entry already exists.
2151 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2153 struct inode *inode, struct ext4_dir_entry_2 *de,
2154 struct buffer_head *bh)
2156 unsigned int blocksize = dir->i_sb->s_blocksize;
2160 if (ext4_has_metadata_csum(inode->i_sb))
2161 csum_size = sizeof(struct ext4_dir_entry_tail);
2164 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2165 blocksize - csum_size, fname, &de);
2169 BUFFER_TRACE(bh, "get_write_access");
2170 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2173 ext4_std_error(dir->i_sb, err);
2177 /* By now the buffer is marked for journaling */
2178 ext4_insert_dentry(dir, inode, de, blocksize, fname);
2181 * XXX shouldn't update any times until successful
2182 * completion of syscall, but too many callers depend
2185 * XXX similarly, too many callers depend on
2186 * ext4_new_inode() setting the times, but error
2187 * recovery deletes the inode, so the worst that can
2188 * happen is that the times are slightly out of date
2189 * and/or different from the directory change time.
2191 dir->i_mtime = dir->i_ctime = current_time(dir);
2192 ext4_update_dx_flag(dir);
2193 inode_inc_iversion(dir);
2194 err2 = ext4_mark_inode_dirty(handle, dir);
2195 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2196 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2198 ext4_std_error(dir->i_sb, err);
2199 return err ? err : err2;
2203 * This converts a one block unindexed directory to a 3 block indexed
2204 * directory, and adds the dentry to the indexed directory.
2206 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2208 struct inode *inode, struct buffer_head *bh)
2210 struct buffer_head *bh2;
2211 struct dx_root *root;
2212 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2213 struct dx_entry *entries;
2214 struct ext4_dir_entry_2 *de, *de2;
2220 struct fake_dirent *fde;
2223 if (ext4_has_metadata_csum(inode->i_sb))
2224 csum_size = sizeof(struct ext4_dir_entry_tail);
2226 blocksize = dir->i_sb->s_blocksize;
2227 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2228 BUFFER_TRACE(bh, "get_write_access");
2229 retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2232 ext4_std_error(dir->i_sb, retval);
2236 root = (struct dx_root *) bh->b_data;
2238 /* The 0th block becomes the root, move the dirents out */
2239 fde = &root->dotdot;
2240 de = (struct ext4_dir_entry_2 *)((char *)fde +
2241 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2242 if ((char *) de >= (((char *) root) + blocksize)) {
2243 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2245 return -EFSCORRUPTED;
2247 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2249 /* Allocate new block for the 0th block's dirents */
2250 bh2 = ext4_append(handle, dir, &block);
2253 return PTR_ERR(bh2);
2255 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2256 data2 = bh2->b_data;
2258 memcpy(data2, de, len);
2259 memset(de, 0, len); /* wipe old data */
2260 de = (struct ext4_dir_entry_2 *) data2;
2262 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2263 if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2264 (data2 + (blocksize - csum_size) -
2268 return -EFSCORRUPTED;
2272 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2273 (char *) de, blocksize);
2276 ext4_initialize_dirent_tail(bh2, blocksize);
2278 /* Initialize the root; the dot dirents already exist */
2279 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2280 de->rec_len = ext4_rec_len_to_disk(
2281 blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2282 memset (&root->info, 0, sizeof(root->info));
2283 root->info.info_length = sizeof(root->info);
2284 if (ext4_hash_in_dirent(dir))
2285 root->info.hash_version = DX_HASH_SIPHASH;
2287 root->info.hash_version =
2288 EXT4_SB(dir->i_sb)->s_def_hash_version;
2290 entries = root->entries;
2291 dx_set_block(entries, 1);
2292 dx_set_count(entries, 1);
2293 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2295 /* Initialize as for dx_probe */
2296 fname->hinfo.hash_version = root->info.hash_version;
2297 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2298 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2299 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2301 /* casefolded encrypted hashes are computed on fname setup */
2302 if (!ext4_hash_in_dirent(dir))
2303 ext4fs_dirhash(dir, fname_name(fname),
2304 fname_len(fname), &fname->hinfo);
2306 memset(frames, 0, sizeof(frames));
2308 frame->entries = entries;
2309 frame->at = entries;
2312 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2315 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2319 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2321 retval = PTR_ERR(de);
2325 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2328 * Even if the block split failed, we have to properly write
2329 * out all the changes we did so far. Otherwise we can end up
2330 * with corrupted filesystem.
2333 ext4_mark_inode_dirty(handle, dir);
2342 * adds a file entry to the specified directory, using the same
2343 * semantics as ext4_find_entry(). It returns NULL if it failed.
2345 * NOTE!! The inode part of 'de' is left at 0 - which means you
2346 * may not sleep between calling this and putting something into
2347 * the entry, as someone else might have used it while you slept.
2349 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2350 struct inode *inode)
2352 struct inode *dir = d_inode(dentry->d_parent);
2353 struct buffer_head *bh = NULL;
2354 struct ext4_dir_entry_2 *de;
2355 struct super_block *sb;
2356 struct ext4_filename fname;
2360 ext4_lblk_t block, blocks;
2363 if (ext4_has_metadata_csum(inode->i_sb))
2364 csum_size = sizeof(struct ext4_dir_entry_tail);
2367 blocksize = sb->s_blocksize;
2368 if (!dentry->d_name.len)
2371 if (fscrypt_is_nokey_name(dentry))
2374 #if IS_ENABLED(CONFIG_UNICODE)
2375 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2376 sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2380 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2384 if (ext4_has_inline_data(dir)) {
2385 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2395 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2396 if (!retval || (retval != ERR_BAD_DX_DIR))
2398 /* Can we just ignore htree data? */
2399 if (ext4_has_metadata_csum(sb)) {
2400 EXT4_ERROR_INODE(dir,
2401 "Directory has corrupted htree index.");
2402 retval = -EFSCORRUPTED;
2405 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2407 retval = ext4_mark_inode_dirty(handle, dir);
2408 if (unlikely(retval))
2411 blocks = dir->i_size >> sb->s_blocksize_bits;
2412 for (block = 0; block < blocks; block++) {
2413 bh = ext4_read_dirblock(dir, block, DIRENT);
2415 bh = ext4_bread(handle, dir, block,
2416 EXT4_GET_BLOCKS_CREATE);
2417 goto add_to_new_block;
2420 retval = PTR_ERR(bh);
2424 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2426 if (retval != -ENOSPC)
2429 if (blocks == 1 && !dx_fallback &&
2430 ext4_has_feature_dir_index(sb)) {
2431 retval = make_indexed_dir(handle, &fname, dir,
2433 bh = NULL; /* make_indexed_dir releases bh */
2438 bh = ext4_append(handle, dir, &block);
2441 retval = PTR_ERR(bh);
2445 de = (struct ext4_dir_entry_2 *) bh->b_data;
2447 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2450 ext4_initialize_dirent_tail(bh, blocksize);
2452 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2454 ext4_fname_free_filename(&fname);
2457 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2462 * Returns 0 for success, or a negative error value
2464 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2465 struct inode *dir, struct inode *inode)
2467 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2468 struct dx_entry *entries, *at;
2469 struct buffer_head *bh;
2470 struct super_block *sb = dir->i_sb;
2471 struct ext4_dir_entry_2 *de;
2477 frame = dx_probe(fname, dir, NULL, frames);
2479 return PTR_ERR(frame);
2480 entries = frame->entries;
2482 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2489 BUFFER_TRACE(bh, "get_write_access");
2490 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2494 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2499 /* Block full, should compress but for now just split */
2500 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2501 dx_get_count(entries), dx_get_limit(entries)));
2502 /* Need to split index? */
2503 if (dx_get_count(entries) == dx_get_limit(entries)) {
2504 ext4_lblk_t newblock;
2505 int levels = frame - frames + 1;
2506 unsigned int icount;
2508 struct dx_entry *entries2;
2509 struct dx_node *node2;
2510 struct buffer_head *bh2;
2512 while (frame > frames) {
2513 if (dx_get_count((frame - 1)->entries) <
2514 dx_get_limit((frame - 1)->entries)) {
2518 frame--; /* split higher index block */
2520 entries = frame->entries;
2523 if (add_level && levels == ext4_dir_htree_level(sb)) {
2524 ext4_warning(sb, "Directory (ino: %lu) index full, "
2525 "reach max htree level :%d",
2526 dir->i_ino, levels);
2527 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2528 ext4_warning(sb, "Large directory feature is "
2529 "not enabled on this "
2535 icount = dx_get_count(entries);
2536 bh2 = ext4_append(handle, dir, &newblock);
2541 node2 = (struct dx_node *)(bh2->b_data);
2542 entries2 = node2->entries;
2543 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2544 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2546 BUFFER_TRACE(frame->bh, "get_write_access");
2547 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2552 unsigned icount1 = icount/2, icount2 = icount - icount1;
2553 unsigned hash2 = dx_get_hash(entries + icount1);
2554 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2557 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2558 err = ext4_journal_get_write_access(handle, sb,
2564 memcpy((char *) entries2, (char *) (entries + icount1),
2565 icount2 * sizeof(struct dx_entry));
2566 dx_set_count(entries, icount1);
2567 dx_set_count(entries2, icount2);
2568 dx_set_limit(entries2, dx_node_limit(dir));
2570 /* Which index block gets the new entry? */
2571 if (at - entries >= icount1) {
2572 frame->at = at - entries - icount1 + entries2;
2573 frame->entries = entries = entries2;
2574 swap(frame->bh, bh2);
2576 dx_insert_block((frame - 1), hash2, newblock);
2577 dxtrace(dx_show_index("node", frame->entries));
2578 dxtrace(dx_show_index("node",
2579 ((struct dx_node *) bh2->b_data)->entries));
2580 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2584 err = ext4_handle_dirty_dx_node(handle, dir,
2588 err = ext4_handle_dirty_dx_node(handle, dir,
2593 struct dx_root *dxroot;
2594 memcpy((char *) entries2, (char *) entries,
2595 icount * sizeof(struct dx_entry));
2596 dx_set_limit(entries2, dx_node_limit(dir));
2599 dx_set_count(entries, 1);
2600 dx_set_block(entries + 0, newblock);
2601 dxroot = (struct dx_root *)frames[0].bh->b_data;
2602 dxroot->info.indirect_levels += 1;
2603 dxtrace(printk(KERN_DEBUG
2604 "Creating %d level index...\n",
2605 dxroot->info.indirect_levels));
2606 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2609 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2615 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2620 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2624 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2628 /* @restart is true means htree-path has been changed, we need to
2629 * repeat dx_probe() to find out valid htree-path
2631 if (restart && err == 0)
2637 * ext4_generic_delete_entry deletes a directory entry by merging it
2638 * with the previous entry
2640 int ext4_generic_delete_entry(struct inode *dir,
2641 struct ext4_dir_entry_2 *de_del,
2642 struct buffer_head *bh,
2647 struct ext4_dir_entry_2 *de, *pde;
2648 unsigned int blocksize = dir->i_sb->s_blocksize;
2654 while (i < buf_size - csum_size) {
2655 if (ext4_check_dir_entry(dir, NULL, de, bh,
2656 entry_buf, buf_size, i))
2657 return -EFSCORRUPTED;
2660 pde->rec_len = ext4_rec_len_to_disk(
2661 ext4_rec_len_from_disk(pde->rec_len,
2663 ext4_rec_len_from_disk(de->rec_len,
2667 /* wipe entire dir_entry */
2668 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2671 /* wipe dir_entry excluding the rec_len field */
2673 memset(&de->name_len, 0,
2674 ext4_rec_len_from_disk(de->rec_len,
2676 offsetof(struct ext4_dir_entry_2,
2680 inode_inc_iversion(dir);
2683 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2685 de = ext4_next_entry(de, blocksize);
2690 static int ext4_delete_entry(handle_t *handle,
2692 struct ext4_dir_entry_2 *de_del,
2693 struct buffer_head *bh)
2695 int err, csum_size = 0;
2697 if (ext4_has_inline_data(dir)) {
2698 int has_inline_data = 1;
2699 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2701 if (has_inline_data)
2705 if (ext4_has_metadata_csum(dir->i_sb))
2706 csum_size = sizeof(struct ext4_dir_entry_tail);
2708 BUFFER_TRACE(bh, "get_write_access");
2709 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2714 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2715 dir->i_sb->s_blocksize, csum_size);
2719 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2720 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2727 ext4_std_error(dir->i_sb, err);
2732 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2733 * since this indicates that nlinks count was previously 1 to avoid overflowing
2734 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2735 * that subdirectory link counts are not being maintained accurately.
2737 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2738 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2739 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2740 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2742 static void ext4_inc_count(struct inode *inode)
2746 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2747 set_nlink(inode, 1);
2751 * If a directory had nlink == 1, then we should let it be 1. This indicates
2752 * directory has >EXT4_LINK_MAX subdirs.
2754 static void ext4_dec_count(struct inode *inode)
2756 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2762 * Add non-directory inode to a directory. On success, the inode reference is
2763 * consumed by dentry is instantiation. This is also indicated by clearing of
2764 * *inodep pointer. On failure, the caller is responsible for dropping the
2765 * inode reference in the safe context.
2767 static int ext4_add_nondir(handle_t *handle,
2768 struct dentry *dentry, struct inode **inodep)
2770 struct inode *dir = d_inode(dentry->d_parent);
2771 struct inode *inode = *inodep;
2772 int err = ext4_add_entry(handle, dentry, inode);
2774 err = ext4_mark_inode_dirty(handle, inode);
2775 if (IS_DIRSYNC(dir))
2776 ext4_handle_sync(handle);
2777 d_instantiate_new(dentry, inode);
2782 ext4_orphan_add(handle, inode);
2783 unlock_new_inode(inode);
2788 * By the time this is called, we already have created
2789 * the directory cache entry for the new file, but it
2790 * is so far negative - it has no inode.
2792 * If the create succeeds, we fill in the inode information
2793 * with d_instantiate().
2795 static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2796 struct dentry *dentry, umode_t mode, bool excl)
2798 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
2800 struct inode *inode;
2801 int err, credits, retries = 0;
2803 err = dquot_initialize(dir);
2807 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2808 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2810 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2811 0, NULL, EXT4_HT_DIR, credits);
2812 handle = ext4_journal_current_handle();
2813 err = PTR_ERR(inode);
2814 if (!IS_ERR(inode)) {
2815 inode->i_op = &ext4_file_inode_operations;
2816 inode->i_fop = &ext4_file_operations;
2817 ext4_set_aops(inode);
2818 err = ext4_add_nondir(handle, dentry, &inode);
2820 ext4_fc_track_create(handle, dentry);
2823 ext4_journal_stop(handle);
2824 if (!IS_ERR_OR_NULL(inode))
2826 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2831 static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2832 struct dentry *dentry, umode_t mode, dev_t rdev)
2834 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
2836 struct inode *inode;
2837 int err, credits, retries = 0;
2839 err = dquot_initialize(dir);
2843 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2844 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2846 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2847 0, NULL, EXT4_HT_DIR, credits);
2848 handle = ext4_journal_current_handle();
2849 err = PTR_ERR(inode);
2850 if (!IS_ERR(inode)) {
2851 init_special_inode(inode, inode->i_mode, rdev);
2852 inode->i_op = &ext4_special_inode_operations;
2853 err = ext4_add_nondir(handle, dentry, &inode);
2855 ext4_fc_track_create(handle, dentry);
2858 ext4_journal_stop(handle);
2859 if (!IS_ERR_OR_NULL(inode))
2861 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2866 static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2867 struct file *file, umode_t mode)
2869 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
2871 struct inode *inode;
2872 int err, retries = 0;
2874 err = dquot_initialize(dir);
2879 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode,
2882 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2883 4 + EXT4_XATTR_TRANS_BLOCKS);
2884 handle = ext4_journal_current_handle();
2885 err = PTR_ERR(inode);
2886 if (!IS_ERR(inode)) {
2887 inode->i_op = &ext4_file_inode_operations;
2888 inode->i_fop = &ext4_file_operations;
2889 ext4_set_aops(inode);
2890 d_tmpfile(file, inode);
2891 err = ext4_orphan_add(handle, inode);
2893 goto err_unlock_inode;
2894 mark_inode_dirty(inode);
2895 unlock_new_inode(inode);
2898 ext4_journal_stop(handle);
2899 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2901 return finish_open_simple(file, err);
2903 ext4_journal_stop(handle);
2904 unlock_new_inode(inode);
2908 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2909 struct ext4_dir_entry_2 *de,
2910 int blocksize, int csum_size,
2911 unsigned int parent_ino, int dotdot_real_len)
2913 de->inode = cpu_to_le32(inode->i_ino);
2915 de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2917 strcpy(de->name, ".");
2918 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2920 de = ext4_next_entry(de, blocksize);
2921 de->inode = cpu_to_le32(parent_ino);
2923 if (!dotdot_real_len)
2924 de->rec_len = ext4_rec_len_to_disk(blocksize -
2925 (csum_size + ext4_dir_rec_len(1, NULL)),
2928 de->rec_len = ext4_rec_len_to_disk(
2929 ext4_dir_rec_len(de->name_len, NULL),
2931 strcpy(de->name, "..");
2932 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2934 return ext4_next_entry(de, blocksize);
2937 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2938 struct inode *inode)
2940 struct buffer_head *dir_block = NULL;
2941 struct ext4_dir_entry_2 *de;
2942 ext4_lblk_t block = 0;
2943 unsigned int blocksize = dir->i_sb->s_blocksize;
2947 if (ext4_has_metadata_csum(dir->i_sb))
2948 csum_size = sizeof(struct ext4_dir_entry_tail);
2950 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2951 err = ext4_try_create_inline_dir(handle, dir, inode);
2952 if (err < 0 && err != -ENOSPC)
2959 dir_block = ext4_append(handle, inode, &block);
2960 if (IS_ERR(dir_block))
2961 return PTR_ERR(dir_block);
2962 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2963 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2964 set_nlink(inode, 2);
2966 ext4_initialize_dirent_tail(dir_block, blocksize);
2968 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2969 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2972 set_buffer_verified(dir_block);
2978 static int ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2979 struct dentry *dentry, umode_t mode)
2981 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
2983 struct inode *inode;
2984 int err, err2 = 0, credits, retries = 0;
2986 if (EXT4_DIR_LINK_MAX(dir))
2989 err = dquot_initialize(dir);
2993 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2994 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2996 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode,
2998 0, NULL, EXT4_HT_DIR, credits);
2999 handle = ext4_journal_current_handle();
3000 err = PTR_ERR(inode);
3004 inode->i_op = &ext4_dir_inode_operations;
3005 inode->i_fop = &ext4_dir_operations;
3006 err = ext4_init_new_dir(handle, dir, inode);
3008 goto out_clear_inode;
3009 err = ext4_mark_inode_dirty(handle, inode);
3011 err = ext4_add_entry(handle, dentry, inode);
3015 ext4_orphan_add(handle, inode);
3016 unlock_new_inode(inode);
3017 err2 = ext4_mark_inode_dirty(handle, inode);
3020 ext4_journal_stop(handle);
3024 ext4_inc_count(dir);
3026 ext4_update_dx_flag(dir);
3027 err = ext4_mark_inode_dirty(handle, dir);
3029 goto out_clear_inode;
3030 d_instantiate_new(dentry, inode);
3031 ext4_fc_track_create(handle, dentry);
3032 if (IS_DIRSYNC(dir))
3033 ext4_handle_sync(handle);
3037 ext4_journal_stop(handle);
3039 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3045 * routine to check that the specified directory is empty (for rmdir)
3047 bool ext4_empty_dir(struct inode *inode)
3049 unsigned int offset;
3050 struct buffer_head *bh;
3051 struct ext4_dir_entry_2 *de;
3052 struct super_block *sb;
3054 if (ext4_has_inline_data(inode)) {
3055 int has_inline_data = 1;
3058 ret = empty_inline_dir(inode, &has_inline_data);
3059 if (has_inline_data)
3064 if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3065 ext4_dir_rec_len(2, NULL)) {
3066 EXT4_ERROR_INODE(inode, "invalid size");
3069 /* The first directory block must not be a hole,
3070 * so treat it as DIRENT_HTREE
3072 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3076 de = (struct ext4_dir_entry_2 *) bh->b_data;
3077 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3079 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3080 ext4_warning_inode(inode, "directory missing '.'");
3084 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3085 de = ext4_next_entry(de, sb->s_blocksize);
3086 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3088 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3089 ext4_warning_inode(inode, "directory missing '..'");
3093 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3094 while (offset < inode->i_size) {
3095 if (!(offset & (sb->s_blocksize - 1))) {
3096 unsigned int lblock;
3098 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3099 bh = ext4_read_dirblock(inode, lblock, EITHER);
3101 offset += sb->s_blocksize;
3107 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3108 (offset & (sb->s_blocksize - 1)));
3109 if (ext4_check_dir_entry(inode, NULL, de, bh,
3110 bh->b_data, bh->b_size, offset) ||
3111 le32_to_cpu(de->inode)) {
3115 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3121 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3124 struct inode *inode;
3125 struct buffer_head *bh;
3126 struct ext4_dir_entry_2 *de;
3127 handle_t *handle = NULL;
3129 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3132 /* Initialize quotas before so that eventual writes go in
3133 * separate transaction */
3134 retval = dquot_initialize(dir);
3137 retval = dquot_initialize(d_inode(dentry));
3142 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3148 inode = d_inode(dentry);
3150 retval = -EFSCORRUPTED;
3151 if (le32_to_cpu(de->inode) != inode->i_ino)
3154 retval = -ENOTEMPTY;
3155 if (!ext4_empty_dir(inode))
3158 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3159 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3160 if (IS_ERR(handle)) {
3161 retval = PTR_ERR(handle);
3166 if (IS_DIRSYNC(dir))
3167 ext4_handle_sync(handle);
3169 retval = ext4_delete_entry(handle, dir, de, bh);
3172 if (!EXT4_DIR_LINK_EMPTY(inode))
3173 ext4_warning_inode(inode,
3174 "empty directory '%.*s' has too many links (%u)",
3175 dentry->d_name.len, dentry->d_name.name,
3177 inode_inc_iversion(inode);
3179 /* There's no need to set i_disksize: the fact that i_nlink is
3180 * zero will ensure that the right thing happens during any
3183 ext4_orphan_add(handle, inode);
3184 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3185 retval = ext4_mark_inode_dirty(handle, inode);
3188 ext4_dec_count(dir);
3189 ext4_update_dx_flag(dir);
3190 ext4_fc_track_unlink(handle, dentry);
3191 retval = ext4_mark_inode_dirty(handle, dir);
3193 #if IS_ENABLED(CONFIG_UNICODE)
3194 /* VFS negative dentries are incompatible with Encoding and
3195 * Case-insensitiveness. Eventually we'll want avoid
3196 * invalidating the dentries here, alongside with returning the
3197 * negative dentries at ext4_lookup(), when it is better
3198 * supported by the VFS for the CI case.
3200 if (IS_CASEFOLDED(dir))
3201 d_invalidate(dentry);
3207 ext4_journal_stop(handle);
3211 int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3212 struct inode *inode,
3213 struct dentry *dentry /* NULL during fast_commit recovery */)
3215 int retval = -ENOENT;
3216 struct buffer_head *bh;
3217 struct ext4_dir_entry_2 *de;
3219 int skip_remove_dentry = 0;
3222 * Keep this outside the transaction; it may have to set up the
3223 * directory's encryption key, which isn't GFP_NOFS-safe.
3225 bh = ext4_find_entry(dir, d_name, &de, NULL);
3232 if (le32_to_cpu(de->inode) != inode->i_ino) {
3234 * It's okay if we find dont find dentry which matches
3235 * the inode. That's because it might have gotten
3236 * renamed to a different inode number
3238 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3239 skip_remove_dentry = 1;
3244 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3245 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3246 if (IS_ERR(handle)) {
3247 retval = PTR_ERR(handle);
3251 if (IS_DIRSYNC(dir))
3252 ext4_handle_sync(handle);
3254 if (!skip_remove_dentry) {
3255 retval = ext4_delete_entry(handle, dir, de, bh);
3258 dir->i_ctime = dir->i_mtime = current_time(dir);
3259 ext4_update_dx_flag(dir);
3260 retval = ext4_mark_inode_dirty(handle, dir);
3266 if (inode->i_nlink == 0)
3267 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3268 d_name->len, d_name->name);
3271 if (!inode->i_nlink)
3272 ext4_orphan_add(handle, inode);
3273 inode->i_ctime = current_time(inode);
3274 retval = ext4_mark_inode_dirty(handle, inode);
3275 if (dentry && !retval)
3276 ext4_fc_track_unlink(handle, dentry);
3278 ext4_journal_stop(handle);
3284 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3288 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3291 trace_ext4_unlink_enter(dir, dentry);
3293 * Initialize quotas before so that eventual writes go
3294 * in separate transaction
3296 retval = dquot_initialize(dir);
3299 retval = dquot_initialize(d_inode(dentry));
3303 retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3304 #if IS_ENABLED(CONFIG_UNICODE)
3305 /* VFS negative dentries are incompatible with Encoding and
3306 * Case-insensitiveness. Eventually we'll want avoid
3307 * invalidating the dentries here, alongside with returning the
3308 * negative dentries at ext4_lookup(), when it is better
3309 * supported by the VFS for the CI case.
3311 if (IS_CASEFOLDED(dir))
3312 d_invalidate(dentry);
3316 trace_ext4_unlink_exit(dentry, retval);
3320 static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3321 struct fscrypt_str *disk_link)
3323 struct buffer_head *bh;
3327 bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3331 BUFFER_TRACE(bh, "get_write_access");
3332 err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3336 kaddr = (char *)bh->b_data;
3337 memcpy(kaddr, disk_link->name, disk_link->len);
3338 inode->i_size = disk_link->len - 1;
3339 EXT4_I(inode)->i_disksize = inode->i_size;
3340 err = ext4_handle_dirty_metadata(handle, inode, bh);
3346 static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3347 struct dentry *dentry, const char *symname)
3349 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
3351 struct inode *inode;
3352 int err, len = strlen(symname);
3354 struct fscrypt_str disk_link;
3357 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3360 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3365 err = dquot_initialize(dir);
3370 * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3371 * directory. +3 for inode, inode bitmap, group descriptor allocation.
3372 * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3375 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3376 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3378 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO,
3379 &dentry->d_name, 0, NULL,
3380 EXT4_HT_DIR, credits);
3381 handle = ext4_journal_current_handle();
3382 if (IS_ERR(inode)) {
3384 ext4_journal_stop(handle);
3385 err = PTR_ERR(inode);
3389 if (IS_ENCRYPTED(inode)) {
3390 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3392 goto err_drop_inode;
3393 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3395 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3396 inode->i_op = &ext4_symlink_inode_operations;
3398 inode->i_op = &ext4_fast_symlink_inode_operations;
3399 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3403 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3404 /* alloc symlink block and fill it */
3405 err = ext4_init_symlink_block(handle, inode, &disk_link);
3407 goto err_drop_inode;
3409 /* clear the extent format for fast symlink */
3410 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3411 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3413 inode->i_size = disk_link.len - 1;
3414 EXT4_I(inode)->i_disksize = inode->i_size;
3416 err = ext4_add_nondir(handle, dentry, &inode);
3418 ext4_journal_stop(handle);
3424 ext4_orphan_add(handle, inode);
3425 unlock_new_inode(inode);
3427 ext4_journal_stop(handle);
3430 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3432 if (disk_link.name != (unsigned char *)symname)
3433 kfree(disk_link.name);
3437 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3440 int err, retries = 0;
3442 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3443 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3444 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3446 return PTR_ERR(handle);
3448 if (IS_DIRSYNC(dir))
3449 ext4_handle_sync(handle);
3451 inode->i_ctime = current_time(inode);
3452 ext4_inc_count(inode);
3455 err = ext4_add_entry(handle, dentry, inode);
3457 err = ext4_mark_inode_dirty(handle, inode);
3458 /* this can happen only for tmpfile being
3459 * linked the first time
3461 if (inode->i_nlink == 1)
3462 ext4_orphan_del(handle, inode);
3463 d_instantiate(dentry, inode);
3464 ext4_fc_track_link(handle, dentry);
3469 ext4_journal_stop(handle);
3470 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3475 static int ext4_link(struct dentry *old_dentry,
3476 struct inode *dir, struct dentry *dentry)
3478 struct inode *inode = d_inode(old_dentry);
3481 if (inode->i_nlink >= EXT4_LINK_MAX)
3484 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3488 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3489 (!projid_eq(EXT4_I(dir)->i_projid,
3490 EXT4_I(old_dentry->d_inode)->i_projid)))
3493 err = dquot_initialize(dir);
3496 return __ext4_link(dir, inode, dentry);
3500 * Try to find buffer head where contains the parent block.
3501 * It should be the inode block if it is inlined or the 1st block
3502 * if it is a normal dir.
3504 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3505 struct inode *inode,
3507 struct ext4_dir_entry_2 **parent_de,
3510 struct buffer_head *bh;
3512 if (!ext4_has_inline_data(inode)) {
3513 struct ext4_dir_entry_2 *de;
3514 unsigned int offset;
3516 /* The first directory block must not be a hole, so
3517 * treat it as DIRENT_HTREE
3519 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3521 *retval = PTR_ERR(bh);
3525 de = (struct ext4_dir_entry_2 *) bh->b_data;
3526 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3528 le32_to_cpu(de->inode) != inode->i_ino ||
3529 strcmp(".", de->name)) {
3530 EXT4_ERROR_INODE(inode, "directory missing '.'");
3532 *retval = -EFSCORRUPTED;
3535 offset = ext4_rec_len_from_disk(de->rec_len,
3536 inode->i_sb->s_blocksize);
3537 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3538 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3539 bh->b_size, offset) ||
3540 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3541 EXT4_ERROR_INODE(inode, "directory missing '..'");
3543 *retval = -EFSCORRUPTED;
3552 return ext4_get_first_inline_block(inode, parent_de, retval);
3555 struct ext4_renament {
3557 struct dentry *dentry;
3558 struct inode *inode;
3560 int dir_nlink_delta;
3562 /* entry for "dentry" */
3563 struct buffer_head *bh;
3564 struct ext4_dir_entry_2 *de;
3567 /* entry for ".." in inode if it's a directory */
3568 struct buffer_head *dir_bh;
3569 struct ext4_dir_entry_2 *parent_de;
3573 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3577 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3578 &retval, &ent->parent_de,
3582 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3583 return -EFSCORRUPTED;
3584 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3585 return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3586 ent->dir_bh, EXT4_JTR_NONE);
3589 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3594 ent->parent_de->inode = cpu_to_le32(dir_ino);
3595 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3596 if (!ent->dir_inlined) {
3597 if (is_dx(ent->inode)) {
3598 retval = ext4_handle_dirty_dx_node(handle,
3602 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3606 retval = ext4_mark_inode_dirty(handle, ent->inode);
3609 ext4_std_error(ent->dir->i_sb, retval);
3615 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3616 unsigned ino, unsigned file_type)
3618 int retval, retval2;
3620 BUFFER_TRACE(ent->bh, "get write access");
3621 retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3625 ent->de->inode = cpu_to_le32(ino);
3626 if (ext4_has_feature_filetype(ent->dir->i_sb))
3627 ent->de->file_type = file_type;
3628 inode_inc_iversion(ent->dir);
3629 ent->dir->i_ctime = ent->dir->i_mtime =
3630 current_time(ent->dir);
3631 retval = ext4_mark_inode_dirty(handle, ent->dir);
3632 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3633 if (!ent->inlined) {
3634 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3635 if (unlikely(retval2)) {
3636 ext4_std_error(ent->dir->i_sb, retval2);
3643 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3644 unsigned ino, unsigned file_type)
3646 struct ext4_renament old = *ent;
3650 * old->de could have moved from under us during make indexed dir,
3651 * so the old->de may no longer valid and need to find it again
3652 * before reset old inode info.
3654 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3656 retval = PTR_ERR(old.bh);
3660 ext4_std_error(old.dir->i_sb, retval);
3664 ext4_setent(handle, &old, ino, file_type);
3668 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3669 const struct qstr *d_name)
3671 int retval = -ENOENT;
3672 struct buffer_head *bh;
3673 struct ext4_dir_entry_2 *de;
3675 bh = ext4_find_entry(dir, d_name, &de, NULL);
3679 retval = ext4_delete_entry(handle, dir, de, bh);
3685 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3690 * ent->de could have moved from under us during htree split, so make
3691 * sure that we are deleting the right entry. We might also be pointing
3692 * to a stale entry in the unused part of ent->bh so just checking inum
3693 * and the name isn't enough.
3695 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3696 ent->de->name_len != ent->dentry->d_name.len ||
3697 strncmp(ent->de->name, ent->dentry->d_name.name,
3698 ent->de->name_len) ||
3700 retval = ext4_find_delete_entry(handle, ent->dir,
3701 &ent->dentry->d_name);
3703 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3704 if (retval == -ENOENT) {
3705 retval = ext4_find_delete_entry(handle, ent->dir,
3706 &ent->dentry->d_name);
3711 ext4_warning_inode(ent->dir,
3712 "Deleting old file: nlink %d, error=%d",
3713 ent->dir->i_nlink, retval);
3717 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3719 if (ent->dir_nlink_delta) {
3720 if (ent->dir_nlink_delta == -1)
3721 ext4_dec_count(ent->dir);
3723 ext4_inc_count(ent->dir);
3724 ext4_mark_inode_dirty(handle, ent->dir);
3728 static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns,
3729 struct ext4_renament *ent,
3730 int credits, handle_t **h)
3737 * for inode block, sb block, group summaries,
3740 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3741 EXT4_XATTR_TRANS_BLOCKS + 4);
3743 wh = ext4_new_inode_start_handle(mnt_userns, ent->dir,
3744 S_IFCHR | WHITEOUT_MODE,
3745 &ent->dentry->d_name, 0, NULL,
3746 EXT4_HT_DIR, credits);
3748 handle = ext4_journal_current_handle();
3751 ext4_journal_stop(handle);
3752 if (PTR_ERR(wh) == -ENOSPC &&
3753 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3757 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3758 wh->i_op = &ext4_special_inode_operations;
3764 * Anybody can rename anything with this: the permission checks are left to the
3765 * higher-level routines.
3767 * n.b. old_{dentry,inode) refers to the source dentry/inode
3768 * while new_{dentry,inode) refers to the destination dentry/inode
3769 * This comes from rename(const char *oldpath, const char *newpath)
3771 static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
3772 struct dentry *old_dentry, struct inode *new_dir,
3773 struct dentry *new_dentry, unsigned int flags)
3775 handle_t *handle = NULL;
3776 struct ext4_renament old = {
3778 .dentry = old_dentry,
3779 .inode = d_inode(old_dentry),
3781 struct ext4_renament new = {
3783 .dentry = new_dentry,
3784 .inode = d_inode(new_dentry),
3788 struct inode *whiteout = NULL;
3792 if (new.inode && new.inode->i_nlink == 0) {
3793 EXT4_ERROR_INODE(new.inode,
3794 "target of rename is already freed");
3795 return -EFSCORRUPTED;
3798 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3799 (!projid_eq(EXT4_I(new_dir)->i_projid,
3800 EXT4_I(old_dentry->d_inode)->i_projid)))
3803 retval = dquot_initialize(old.dir);
3806 retval = dquot_initialize(old.inode);
3809 retval = dquot_initialize(new.dir);
3813 /* Initialize quotas before so that eventual writes go
3814 * in separate transaction */
3816 retval = dquot_initialize(new.inode);
3821 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3823 return PTR_ERR(old.bh);
3825 * Check for inode number is _not_ due to possible IO errors.
3826 * We might rmdir the source, keep it as pwd of some process
3827 * and merrily kill the link to whatever was created under the
3828 * same name. Goodbye sticky bit ;-<
3831 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3834 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3835 &new.de, &new.inlined);
3836 if (IS_ERR(new.bh)) {
3837 retval = PTR_ERR(new.bh);
3847 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3848 ext4_alloc_da_blocks(old.inode);
3850 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3851 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3852 if (!(flags & RENAME_WHITEOUT)) {
3853 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3854 if (IS_ERR(handle)) {
3855 retval = PTR_ERR(handle);
3859 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle);
3860 if (IS_ERR(whiteout)) {
3861 retval = PTR_ERR(whiteout);
3866 old_file_type = old.de->file_type;
3867 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3868 ext4_handle_sync(handle);
3870 if (S_ISDIR(old.inode->i_mode)) {
3872 retval = -ENOTEMPTY;
3873 if (!ext4_empty_dir(new.inode))
3877 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3880 retval = ext4_rename_dir_prepare(handle, &old);
3885 * If we're renaming a file within an inline_data dir and adding or
3886 * setting the new dirent causes a conversion from inline_data to
3887 * extents/blockmap, we need to force the dirent delete code to
3888 * re-read the directory, or else we end up trying to delete a dirent
3889 * from what is now the extent tree root (or a block map).
3891 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3892 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3896 * Do this before adding a new entry, so the old entry is sure
3897 * to be still pointing to the valid old entry.
3899 retval = ext4_setent(handle, &old, whiteout->i_ino,
3903 retval = ext4_mark_inode_dirty(handle, whiteout);
3904 if (unlikely(retval))
3909 retval = ext4_add_entry(handle, new.dentry, old.inode);
3913 retval = ext4_setent(handle, &new,
3914 old.inode->i_ino, old_file_type);
3919 force_reread = !ext4_test_inode_flag(new.dir,
3920 EXT4_INODE_INLINE_DATA);
3923 * Like most other Unix systems, set the ctime for inodes on a
3926 old.inode->i_ctime = current_time(old.inode);
3927 retval = ext4_mark_inode_dirty(handle, old.inode);
3928 if (unlikely(retval))
3935 ext4_rename_delete(handle, &old, force_reread);
3939 ext4_dec_count(new.inode);
3940 new.inode->i_ctime = current_time(new.inode);
3942 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3943 ext4_update_dx_flag(old.dir);
3945 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3949 ext4_dec_count(old.dir);
3951 /* checked ext4_empty_dir above, can't have another
3952 * parent, ext4_dec_count() won't work for many-linked
3954 clear_nlink(new.inode);
3956 ext4_inc_count(new.dir);
3957 ext4_update_dx_flag(new.dir);
3958 retval = ext4_mark_inode_dirty(handle, new.dir);
3959 if (unlikely(retval))
3963 retval = ext4_mark_inode_dirty(handle, old.dir);
3964 if (unlikely(retval))
3967 if (S_ISDIR(old.inode->i_mode)) {
3969 * We disable fast commits here that's because the
3970 * replay code is not yet capable of changing dot dot
3971 * dirents in directories.
3973 ext4_fc_mark_ineligible(old.inode->i_sb,
3974 EXT4_FC_REASON_RENAME_DIR, handle);
3976 struct super_block *sb = old.inode->i_sb;
3979 ext4_fc_track_unlink(handle, new.dentry);
3980 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
3981 !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
3982 !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
3983 __ext4_fc_track_link(handle, old.inode, new.dentry);
3984 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3986 __ext4_fc_track_create(handle, whiteout,
3992 retval = ext4_mark_inode_dirty(handle, new.inode);
3993 if (unlikely(retval))
3995 if (!new.inode->i_nlink)
3996 ext4_orphan_add(handle, new.inode);
4003 ext4_resetent(handle, &old,
4004 old.inode->i_ino, old_file_type);
4005 drop_nlink(whiteout);
4006 ext4_orphan_add(handle, whiteout);
4008 unlock_new_inode(whiteout);
4009 ext4_journal_stop(handle);
4012 ext4_journal_stop(handle);
4021 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4022 struct inode *new_dir, struct dentry *new_dentry)
4024 handle_t *handle = NULL;
4025 struct ext4_renament old = {
4027 .dentry = old_dentry,
4028 .inode = d_inode(old_dentry),
4030 struct ext4_renament new = {
4032 .dentry = new_dentry,
4033 .inode = d_inode(new_dentry),
4037 struct timespec64 ctime;
4039 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4040 !projid_eq(EXT4_I(new_dir)->i_projid,
4041 EXT4_I(old_dentry->d_inode)->i_projid)) ||
4042 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4043 !projid_eq(EXT4_I(old_dir)->i_projid,
4044 EXT4_I(new_dentry->d_inode)->i_projid)))
4047 retval = dquot_initialize(old.dir);
4050 retval = dquot_initialize(new.dir);
4054 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4055 &old.de, &old.inlined);
4057 return PTR_ERR(old.bh);
4059 * Check for inode number is _not_ due to possible IO errors.
4060 * We might rmdir the source, keep it as pwd of some process
4061 * and merrily kill the link to whatever was created under the
4062 * same name. Goodbye sticky bit ;-<
4065 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4068 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4069 &new.de, &new.inlined);
4070 if (IS_ERR(new.bh)) {
4071 retval = PTR_ERR(new.bh);
4076 /* RENAME_EXCHANGE case: old *and* new must both exist */
4077 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4080 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4081 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4082 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4083 if (IS_ERR(handle)) {
4084 retval = PTR_ERR(handle);
4089 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4090 ext4_handle_sync(handle);
4092 if (S_ISDIR(old.inode->i_mode)) {
4094 retval = ext4_rename_dir_prepare(handle, &old);
4098 if (S_ISDIR(new.inode->i_mode)) {
4100 retval = ext4_rename_dir_prepare(handle, &new);
4106 * Other than the special case of overwriting a directory, parents'
4107 * nlink only needs to be modified if this is a cross directory rename.
4109 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4110 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4111 new.dir_nlink_delta = -old.dir_nlink_delta;
4113 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4114 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4118 new_file_type = new.de->file_type;
4119 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4123 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4128 * Like most other Unix systems, set the ctime for inodes on a
4131 ctime = current_time(old.inode);
4132 old.inode->i_ctime = ctime;
4133 new.inode->i_ctime = ctime;
4134 retval = ext4_mark_inode_dirty(handle, old.inode);
4135 if (unlikely(retval))
4137 retval = ext4_mark_inode_dirty(handle, new.inode);
4138 if (unlikely(retval))
4140 ext4_fc_mark_ineligible(new.inode->i_sb,
4141 EXT4_FC_REASON_CROSS_RENAME, handle);
4143 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4148 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4152 ext4_update_dir_count(handle, &old);
4153 ext4_update_dir_count(handle, &new);
4162 ext4_journal_stop(handle);
4166 static int ext4_rename2(struct mnt_idmap *idmap,
4167 struct inode *old_dir, struct dentry *old_dentry,
4168 struct inode *new_dir, struct dentry *new_dentry,
4171 struct user_namespace *mnt_userns = mnt_idmap_owner(idmap);
4174 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4177 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4180 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4185 if (flags & RENAME_EXCHANGE) {
4186 return ext4_cross_rename(old_dir, old_dentry,
4187 new_dir, new_dentry);
4190 return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags);
4194 * directories can handle most operations...
4196 const struct inode_operations ext4_dir_inode_operations = {
4197 .create = ext4_create,
4198 .lookup = ext4_lookup,
4200 .unlink = ext4_unlink,
4201 .symlink = ext4_symlink,
4202 .mkdir = ext4_mkdir,
4203 .rmdir = ext4_rmdir,
4204 .mknod = ext4_mknod,
4205 .tmpfile = ext4_tmpfile,
4206 .rename = ext4_rename2,
4207 .setattr = ext4_setattr,
4208 .getattr = ext4_getattr,
4209 .listxattr = ext4_listxattr,
4210 .get_inode_acl = ext4_get_acl,
4211 .set_acl = ext4_set_acl,
4212 .fiemap = ext4_fiemap,
4213 .fileattr_get = ext4_fileattr_get,
4214 .fileattr_set = ext4_fileattr_set,
4217 const struct inode_operations ext4_special_inode_operations = {
4218 .setattr = ext4_setattr,
4219 .getattr = ext4_getattr,
4220 .listxattr = ext4_listxattr,
4221 .get_inode_acl = ext4_get_acl,
4222 .set_acl = ext4_set_acl,