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 buffer_head *bh;
60 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
61 ((inode->i_size >> 10) >=
62 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
63 return ERR_PTR(-ENOSPC);
65 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
67 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
70 inode->i_size += inode->i_sb->s_blocksize;
71 EXT4_I(inode)->i_disksize = inode->i_size;
72 BUFFER_TRACE(bh, "get_write_access");
73 err = ext4_journal_get_write_access(handle, bh);
76 ext4_std_error(inode->i_sb, err);
82 static int ext4_dx_csum_verify(struct inode *inode,
83 struct ext4_dir_entry *dirent);
86 * Hints to ext4_read_dirblock regarding whether we expect a directory
87 * block being read to be an index block, or a block containing
88 * directory entries (and if the latter, whether it was found via a
89 * logical block in an htree index block). This is used to control
90 * what sort of sanity checkinig ext4_read_dirblock() will do on the
91 * directory block read from the storage device. EITHER will means
92 * the caller doesn't know what kind of directory block will be read,
93 * so no specific verification will be done.
96 EITHER, INDEX, DIRENT, DIRENT_HTREE
99 #define ext4_read_dirblock(inode, block, type) \
100 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
104 dirblock_type_t type,
108 struct buffer_head *bh;
109 struct ext4_dir_entry *dirent;
112 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
115 bh = ext4_bread(NULL, inode, block, 0);
117 __ext4_warning(inode->i_sb, func, line,
118 "inode #%lu: lblock %lu: comm %s: "
119 "error %ld reading directory block",
120 inode->i_ino, (unsigned long)block,
121 current->comm, PTR_ERR(bh));
125 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
126 ext4_error_inode(inode, func, line, block,
127 "Directory hole found for htree %s block",
128 (type == INDEX) ? "index" : "leaf");
129 return ERR_PTR(-EFSCORRUPTED);
133 dirent = (struct ext4_dir_entry *) bh->b_data;
134 /* Determine whether or not we have an index block */
138 else if (ext4_rec_len_from_disk(dirent->rec_len,
139 inode->i_sb->s_blocksize) ==
140 inode->i_sb->s_blocksize)
143 if (!is_dx_block && type == INDEX) {
144 ext4_error_inode(inode, func, line, block,
145 "directory leaf block found instead of index block");
147 return ERR_PTR(-EFSCORRUPTED);
149 if (!ext4_has_metadata_csum(inode->i_sb) ||
154 * An empty leaf block can get mistaken for a index block; for
155 * this reason, we can only check the index checksum when the
156 * caller is sure it should be an index block.
158 if (is_dx_block && type == INDEX) {
159 if (ext4_dx_csum_verify(inode, dirent) &&
160 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
161 set_buffer_verified(bh);
163 ext4_error_inode_err(inode, func, line, block,
165 "Directory index failed checksum");
167 return ERR_PTR(-EFSBADCRC);
171 if (ext4_dirblock_csum_verify(inode, bh) &&
172 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
173 set_buffer_verified(bh);
175 ext4_error_inode_err(inode, func, line, block,
177 "Directory block failed checksum");
179 return ERR_PTR(-EFSBADCRC);
186 #define dxtrace(command) command
188 #define dxtrace(command)
212 * dx_root_info is laid out so that if it should somehow get overlaid by a
213 * dirent the two low bits of the hash version will be zero. Therefore, the
214 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
219 struct fake_dirent dot;
221 struct fake_dirent dotdot;
225 __le32 reserved_zero;
227 u8 info_length; /* 8 */
232 struct dx_entry entries[];
237 struct fake_dirent fake;
238 struct dx_entry entries[];
244 struct buffer_head *bh;
245 struct dx_entry *entries;
257 * This goes at the end of each htree block.
261 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
264 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
265 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
266 static inline unsigned dx_get_hash(struct dx_entry *entry);
267 static void dx_set_hash(struct dx_entry *entry, unsigned value);
268 static unsigned dx_get_count(struct dx_entry *entries);
269 static unsigned dx_get_limit(struct dx_entry *entries);
270 static void dx_set_count(struct dx_entry *entries, unsigned value);
271 static void dx_set_limit(struct dx_entry *entries, unsigned value);
272 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
273 static unsigned dx_node_limit(struct inode *dir);
274 static struct dx_frame *dx_probe(struct ext4_filename *fname,
276 struct dx_hash_info *hinfo,
277 struct dx_frame *frame);
278 static void dx_release(struct dx_frame *frames);
279 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
280 unsigned blocksize, struct dx_hash_info *hinfo,
281 struct dx_map_entry map[]);
282 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
283 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
284 struct dx_map_entry *offsets, int count, unsigned blocksize);
285 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
286 static void dx_insert_block(struct dx_frame *frame,
287 u32 hash, ext4_lblk_t block);
288 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
289 struct dx_frame *frame,
290 struct dx_frame *frames,
292 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
293 struct ext4_filename *fname,
294 struct ext4_dir_entry_2 **res_dir);
295 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
296 struct inode *dir, struct inode *inode);
298 /* checksumming functions */
299 void ext4_initialize_dirent_tail(struct buffer_head *bh,
300 unsigned int blocksize)
302 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
304 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
305 t->det_rec_len = ext4_rec_len_to_disk(
306 sizeof(struct ext4_dir_entry_tail), blocksize);
307 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
310 /* Walk through a dirent block to find a checksum "dirent" at the tail */
311 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
312 struct buffer_head *bh)
314 struct ext4_dir_entry_tail *t;
317 struct ext4_dir_entry *d, *top;
319 d = (struct ext4_dir_entry *)bh->b_data;
320 top = (struct ext4_dir_entry *)(bh->b_data +
321 (EXT4_BLOCK_SIZE(inode->i_sb) -
322 sizeof(struct ext4_dir_entry_tail)));
323 while (d < top && d->rec_len)
324 d = (struct ext4_dir_entry *)(((void *)d) +
325 le16_to_cpu(d->rec_len));
330 t = (struct ext4_dir_entry_tail *)d;
332 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
335 if (t->det_reserved_zero1 ||
336 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
337 t->det_reserved_zero2 ||
338 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
344 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
346 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
347 struct ext4_inode_info *ei = EXT4_I(inode);
350 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
351 return cpu_to_le32(csum);
354 #define warn_no_space_for_csum(inode) \
355 __warn_no_space_for_csum((inode), __func__, __LINE__)
357 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
360 __ext4_warning_inode(inode, func, line,
361 "No space for directory leaf checksum. Please run e2fsck -D.");
364 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
366 struct ext4_dir_entry_tail *t;
368 if (!ext4_has_metadata_csum(inode->i_sb))
371 t = get_dirent_tail(inode, bh);
373 warn_no_space_for_csum(inode);
377 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
378 (char *)t - bh->b_data))
384 static void ext4_dirblock_csum_set(struct inode *inode,
385 struct buffer_head *bh)
387 struct ext4_dir_entry_tail *t;
389 if (!ext4_has_metadata_csum(inode->i_sb))
392 t = get_dirent_tail(inode, bh);
394 warn_no_space_for_csum(inode);
398 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
399 (char *)t - bh->b_data);
402 int ext4_handle_dirty_dirblock(handle_t *handle,
404 struct buffer_head *bh)
406 ext4_dirblock_csum_set(inode, bh);
407 return ext4_handle_dirty_metadata(handle, inode, bh);
410 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
411 struct ext4_dir_entry *dirent,
414 struct ext4_dir_entry *dp;
415 struct dx_root_info *root;
418 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
420 else if (le16_to_cpu(dirent->rec_len) == 12) {
421 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
422 if (le16_to_cpu(dp->rec_len) !=
423 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
425 root = (struct dx_root_info *)(((void *)dp + 12));
426 if (root->reserved_zero ||
427 root->info_length != sizeof(struct dx_root_info))
434 *offset = count_offset;
435 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
438 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
439 int count_offset, int count, struct dx_tail *t)
441 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
442 struct ext4_inode_info *ei = EXT4_I(inode);
445 __u32 dummy_csum = 0;
446 int offset = offsetof(struct dx_tail, dt_checksum);
448 size = count_offset + (count * sizeof(struct dx_entry));
449 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
450 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
451 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
453 return cpu_to_le32(csum);
456 static int ext4_dx_csum_verify(struct inode *inode,
457 struct ext4_dir_entry *dirent)
459 struct dx_countlimit *c;
461 int count_offset, limit, count;
463 if (!ext4_has_metadata_csum(inode->i_sb))
466 c = get_dx_countlimit(inode, dirent, &count_offset);
468 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
471 limit = le16_to_cpu(c->limit);
472 count = le16_to_cpu(c->count);
473 if (count_offset + (limit * sizeof(struct dx_entry)) >
474 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
475 warn_no_space_for_csum(inode);
478 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
480 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
486 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
488 struct dx_countlimit *c;
490 int count_offset, limit, count;
492 if (!ext4_has_metadata_csum(inode->i_sb))
495 c = get_dx_countlimit(inode, dirent, &count_offset);
497 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
500 limit = le16_to_cpu(c->limit);
501 count = le16_to_cpu(c->count);
502 if (count_offset + (limit * sizeof(struct dx_entry)) >
503 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
504 warn_no_space_for_csum(inode);
507 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
509 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
512 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
514 struct buffer_head *bh)
516 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
517 return ext4_handle_dirty_metadata(handle, inode, bh);
521 * p is at least 6 bytes before the end of page
523 static inline struct ext4_dir_entry_2 *
524 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
526 return (struct ext4_dir_entry_2 *)((char *)p +
527 ext4_rec_len_from_disk(p->rec_len, blocksize));
531 * Future: use high four bits of block for coalesce-on-delete flags
532 * Mask them off for now.
535 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
537 return le32_to_cpu(entry->block) & 0x0fffffff;
540 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
542 entry->block = cpu_to_le32(value);
545 static inline unsigned dx_get_hash(struct dx_entry *entry)
547 return le32_to_cpu(entry->hash);
550 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
552 entry->hash = cpu_to_le32(value);
555 static inline unsigned dx_get_count(struct dx_entry *entries)
557 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
560 static inline unsigned dx_get_limit(struct dx_entry *entries)
562 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
565 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
567 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
570 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
572 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
575 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
577 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
578 EXT4_DIR_REC_LEN(2) - infosize;
580 if (ext4_has_metadata_csum(dir->i_sb))
581 entry_space -= sizeof(struct dx_tail);
582 return entry_space / sizeof(struct dx_entry);
585 static inline unsigned dx_node_limit(struct inode *dir)
587 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
589 if (ext4_has_metadata_csum(dir->i_sb))
590 entry_space -= sizeof(struct dx_tail);
591 return entry_space / sizeof(struct dx_entry);
598 static void dx_show_index(char * label, struct dx_entry *entries)
600 int i, n = dx_get_count (entries);
601 printk(KERN_DEBUG "%s index", label);
602 for (i = 0; i < n; i++) {
603 printk(KERN_CONT " %x->%lu",
604 i ? dx_get_hash(entries + i) : 0,
605 (unsigned long)dx_get_block(entries + i));
607 printk(KERN_CONT "\n");
617 static struct stats dx_show_leaf(struct inode *dir,
618 struct dx_hash_info *hinfo,
619 struct ext4_dir_entry_2 *de,
620 int size, int show_names)
622 unsigned names = 0, space = 0;
623 char *base = (char *) de;
624 struct dx_hash_info h = *hinfo;
627 while ((char *) de < base + size)
633 #ifdef CONFIG_FS_ENCRYPTION
636 struct fscrypt_str fname_crypto_str =
642 if (!IS_ENCRYPTED(dir)) {
643 /* Directory is not encrypted */
644 ext4fs_dirhash(dir, de->name,
646 printk("%*.s:(U)%x.%u ", len,
648 (unsigned) ((char *) de
651 struct fscrypt_str de_name =
652 FSTR_INIT(name, len);
654 /* Directory is encrypted */
655 res = fscrypt_fname_alloc_buffer(
656 len, &fname_crypto_str);
658 printk(KERN_WARNING "Error "
662 res = fscrypt_fname_disk_to_usr(dir,
666 printk(KERN_WARNING "Error "
667 "converting filename "
673 name = fname_crypto_str.name;
674 len = fname_crypto_str.len;
676 ext4fs_dirhash(dir, de->name,
678 printk("%*.s:(E)%x.%u ", len, name,
679 h.hash, (unsigned) ((char *) de
681 fscrypt_fname_free_buffer(
685 int len = de->name_len;
686 char *name = de->name;
687 ext4fs_dirhash(dir, de->name, de->name_len, &h);
688 printk("%*.s:%x.%u ", len, name, h.hash,
689 (unsigned) ((char *) de - base));
692 space += EXT4_DIR_REC_LEN(de->name_len);
695 de = ext4_next_entry(de, size);
697 printk(KERN_CONT "(%i)\n", names);
698 return (struct stats) { names, space, 1 };
701 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
702 struct dx_entry *entries, int levels)
704 unsigned blocksize = dir->i_sb->s_blocksize;
705 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
707 struct buffer_head *bh;
708 printk("%i indexed blocks...\n", count);
709 for (i = 0; i < count; i++, entries++)
711 ext4_lblk_t block = dx_get_block(entries);
712 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
713 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
715 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
716 bh = ext4_bread(NULL,dir, block, 0);
717 if (!bh || IS_ERR(bh))
720 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
721 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
722 bh->b_data, blocksize, 0);
723 names += stats.names;
724 space += stats.space;
725 bcount += stats.bcount;
729 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
730 levels ? "" : " ", names, space/bcount,
731 (space/bcount)*100/blocksize);
732 return (struct stats) { names, space, bcount};
736 * Linear search cross check
738 static inline void htree_rep_invariant_check(struct dx_entry *at,
739 struct dx_entry *target,
740 u32 hash, unsigned int n)
743 dxtrace(printk(KERN_CONT ","));
744 if (dx_get_hash(++at) > hash) {
749 ASSERT(at == target - 1);
752 static inline void htree_rep_invariant_check(struct dx_entry *at,
753 struct dx_entry *target,
754 u32 hash, unsigned int n)
757 #endif /* DX_DEBUG */
760 * Probe for a directory leaf block to search.
762 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
763 * error in the directory index, and the caller should fall back to
764 * searching the directory normally. The callers of dx_probe **MUST**
765 * check for this error code, and make sure it never gets reflected
768 static struct dx_frame *
769 dx_probe(struct ext4_filename *fname, struct inode *dir,
770 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
772 unsigned count, indirect;
773 struct dx_entry *at, *entries, *p, *q, *m;
774 struct dx_root *root;
775 struct dx_frame *frame = frame_in;
776 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
779 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
780 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
781 if (IS_ERR(frame->bh))
782 return (struct dx_frame *) frame->bh;
784 root = (struct dx_root *) frame->bh->b_data;
785 if (root->info.hash_version != DX_HASH_TEA &&
786 root->info.hash_version != DX_HASH_HALF_MD4 &&
787 root->info.hash_version != DX_HASH_LEGACY) {
788 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
789 root->info.hash_version);
793 hinfo = &fname->hinfo;
794 hinfo->hash_version = root->info.hash_version;
795 if (hinfo->hash_version <= DX_HASH_TEA)
796 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
797 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
798 if (fname && fname_name(fname))
799 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
802 if (root->info.unused_flags & 1) {
803 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
804 root->info.unused_flags);
808 indirect = root->info.indirect_levels;
809 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
810 ext4_warning(dir->i_sb,
811 "Directory (ino: %lu) htree depth %#06x exceed"
812 "supported value", dir->i_ino,
813 ext4_dir_htree_level(dir->i_sb));
814 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
815 ext4_warning(dir->i_sb, "Enable large directory "
816 "feature to access it");
821 entries = (struct dx_entry *)(((char *)&root->info) +
822 root->info.info_length);
824 if (dx_get_limit(entries) != dx_root_limit(dir,
825 root->info.info_length)) {
826 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
827 dx_get_limit(entries),
828 dx_root_limit(dir, root->info.info_length));
832 dxtrace(printk("Look up %x", hash));
834 count = dx_get_count(entries);
835 if (!count || count > dx_get_limit(entries)) {
836 ext4_warning_inode(dir,
837 "dx entry: count %u beyond limit %u",
838 count, dx_get_limit(entries));
843 q = entries + count - 1;
846 dxtrace(printk(KERN_CONT "."));
847 if (dx_get_hash(m) > hash)
853 htree_rep_invariant_check(entries, p, hash, count - 1);
856 dxtrace(printk(KERN_CONT " %x->%u\n",
857 at == entries ? 0 : dx_get_hash(at),
859 frame->entries = entries;
864 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
865 if (IS_ERR(frame->bh)) {
866 ret_err = (struct dx_frame *) frame->bh;
870 entries = ((struct dx_node *) frame->bh->b_data)->entries;
872 if (dx_get_limit(entries) != dx_node_limit(dir)) {
873 ext4_warning_inode(dir,
874 "dx entry: limit %u != node limit %u",
875 dx_get_limit(entries), dx_node_limit(dir));
880 while (frame >= frame_in) {
885 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
886 ext4_warning_inode(dir,
887 "Corrupt directory, running e2fsck is recommended");
891 static void dx_release(struct dx_frame *frames)
893 struct dx_root_info *info;
895 unsigned int indirect_levels;
897 if (frames[0].bh == NULL)
900 info = &((struct dx_root *)frames[0].bh->b_data)->info;
901 /* save local copy, "info" may be freed after brelse() */
902 indirect_levels = info->indirect_levels;
903 for (i = 0; i <= indirect_levels; i++) {
904 if (frames[i].bh == NULL)
906 brelse(frames[i].bh);
912 * This function increments the frame pointer to search the next leaf
913 * block, and reads in the necessary intervening nodes if the search
914 * should be necessary. Whether or not the search is necessary is
915 * controlled by the hash parameter. If the hash value is even, then
916 * the search is only continued if the next block starts with that
917 * hash value. This is used if we are searching for a specific file.
919 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
921 * This function returns 1 if the caller should continue to search,
922 * or 0 if it should not. If there is an error reading one of the
923 * index blocks, it will a negative error code.
925 * If start_hash is non-null, it will be filled in with the starting
926 * hash of the next page.
928 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
929 struct dx_frame *frame,
930 struct dx_frame *frames,
934 struct buffer_head *bh;
940 * Find the next leaf page by incrementing the frame pointer.
941 * If we run out of entries in the interior node, loop around and
942 * increment pointer in the parent node. When we break out of
943 * this loop, num_frames indicates the number of interior
944 * nodes need to be read.
947 if (++(p->at) < p->entries + dx_get_count(p->entries))
956 * If the hash is 1, then continue only if the next page has a
957 * continuation hash of any value. This is used for readdir
958 * handling. Otherwise, check to see if the hash matches the
959 * desired contiuation hash. If it doesn't, return since
960 * there's no point to read in the successive index pages.
962 bhash = dx_get_hash(p->at);
965 if ((hash & 1) == 0) {
966 if ((bhash & ~1) != hash)
970 * If the hash is HASH_NB_ALWAYS, we always go to the next
971 * block so no check is necessary
973 while (num_frames--) {
974 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
980 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
987 * This function fills a red-black tree with information from a
988 * directory block. It returns the number directory entries loaded
989 * into the tree. If there is an error it is returned in err.
991 static int htree_dirblock_to_tree(struct file *dir_file,
992 struct inode *dir, ext4_lblk_t block,
993 struct dx_hash_info *hinfo,
994 __u32 start_hash, __u32 start_minor_hash)
996 struct buffer_head *bh;
997 struct ext4_dir_entry_2 *de, *top;
998 int err = 0, count = 0;
999 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1001 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1002 (unsigned long)block));
1003 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1007 de = (struct ext4_dir_entry_2 *) bh->b_data;
1008 top = (struct ext4_dir_entry_2 *) ((char *) de +
1009 dir->i_sb->s_blocksize -
1010 EXT4_DIR_REC_LEN(0));
1011 /* Check if the directory is encrypted */
1012 if (IS_ENCRYPTED(dir)) {
1013 err = fscrypt_prepare_readdir(dir);
1018 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1026 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1027 if (ext4_check_dir_entry(dir, NULL, de, bh,
1028 bh->b_data, bh->b_size,
1029 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1030 + ((char *)de - bh->b_data))) {
1031 /* silently ignore the rest of the block */
1034 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1035 if ((hinfo->hash < start_hash) ||
1036 ((hinfo->hash == start_hash) &&
1037 (hinfo->minor_hash < start_minor_hash)))
1041 if (!IS_ENCRYPTED(dir)) {
1042 tmp_str.name = de->name;
1043 tmp_str.len = de->name_len;
1044 err = ext4_htree_store_dirent(dir_file,
1045 hinfo->hash, hinfo->minor_hash, de,
1048 int save_len = fname_crypto_str.len;
1049 struct fscrypt_str de_name = FSTR_INIT(de->name,
1052 /* Directory is encrypted */
1053 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1054 hinfo->minor_hash, &de_name,
1060 err = ext4_htree_store_dirent(dir_file,
1061 hinfo->hash, hinfo->minor_hash, de,
1063 fname_crypto_str.len = save_len;
1073 fscrypt_fname_free_buffer(&fname_crypto_str);
1079 * This function fills a red-black tree with information from a
1080 * directory. We start scanning the directory in hash order, starting
1081 * at start_hash and start_minor_hash.
1083 * This function returns the number of entries inserted into the tree,
1084 * or a negative error code.
1086 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1087 __u32 start_minor_hash, __u32 *next_hash)
1089 struct dx_hash_info hinfo;
1090 struct ext4_dir_entry_2 *de;
1091 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1097 struct fscrypt_str tmp_str;
1099 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1100 start_hash, start_minor_hash));
1101 dir = file_inode(dir_file);
1102 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1103 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1104 if (hinfo.hash_version <= DX_HASH_TEA)
1105 hinfo.hash_version +=
1106 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1107 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1108 if (ext4_has_inline_data(dir)) {
1109 int has_inline_data = 1;
1110 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1114 if (has_inline_data) {
1119 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1120 start_hash, start_minor_hash);
1124 hinfo.hash = start_hash;
1125 hinfo.minor_hash = 0;
1126 frame = dx_probe(NULL, dir, &hinfo, frames);
1128 return PTR_ERR(frame);
1130 /* Add '.' and '..' from the htree header */
1131 if (!start_hash && !start_minor_hash) {
1132 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1133 tmp_str.name = de->name;
1134 tmp_str.len = de->name_len;
1135 err = ext4_htree_store_dirent(dir_file, 0, 0,
1141 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1142 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1143 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1144 tmp_str.name = de->name;
1145 tmp_str.len = de->name_len;
1146 err = ext4_htree_store_dirent(dir_file, 2, 0,
1154 if (fatal_signal_pending(current)) {
1159 block = dx_get_block(frame->at);
1160 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1161 start_hash, start_minor_hash);
1168 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1169 frame, frames, &hashval);
1170 *next_hash = hashval;
1176 * Stop if: (a) there are no more entries, or
1177 * (b) we have inserted at least one entry and the
1178 * next hash value is not a continuation
1181 (count && ((hashval & 1) == 0)))
1185 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1186 "next hash: %x\n", count, *next_hash));
1193 static inline int search_dirblock(struct buffer_head *bh,
1195 struct ext4_filename *fname,
1196 unsigned int offset,
1197 struct ext4_dir_entry_2 **res_dir)
1199 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1200 fname, offset, res_dir);
1204 * Directory block splitting, compacting
1208 * Create map of hash values, offsets, and sizes, stored at end of block.
1209 * Returns number of entries mapped.
1211 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1212 unsigned blocksize, struct dx_hash_info *hinfo,
1213 struct dx_map_entry *map_tail)
1216 char *base = (char *) de;
1217 struct dx_hash_info h = *hinfo;
1219 while ((char *) de < base + blocksize) {
1220 if (de->name_len && de->inode) {
1221 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1223 map_tail->hash = h.hash;
1224 map_tail->offs = ((char *) de - base)>>2;
1225 map_tail->size = le16_to_cpu(de->rec_len);
1229 /* XXX: do we need to check rec_len == 0 case? -Chris */
1230 de = ext4_next_entry(de, blocksize);
1235 /* Sort map by hash value */
1236 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1238 struct dx_map_entry *p, *q, *top = map + count - 1;
1240 /* Combsort until bubble sort doesn't suck */
1242 count = count*10/13;
1243 if (count - 9 < 2) /* 9, 10 -> 11 */
1245 for (p = top, q = p - count; q >= map; p--, q--)
1246 if (p->hash < q->hash)
1249 /* Garden variety bubble sort */
1254 if (q[1].hash >= q[0].hash)
1262 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1264 struct dx_entry *entries = frame->entries;
1265 struct dx_entry *old = frame->at, *new = old + 1;
1266 int count = dx_get_count(entries);
1268 ASSERT(count < dx_get_limit(entries));
1269 ASSERT(old < entries + count);
1270 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1271 dx_set_hash(new, hash);
1272 dx_set_block(new, block);
1273 dx_set_count(entries, count + 1);
1276 #ifdef CONFIG_UNICODE
1278 * Test whether a case-insensitive directory entry matches the filename
1279 * being searched for. If quick is set, assume the name being looked up
1280 * is already in the casefolded form.
1282 * Returns: 0 if the directory entry matches, more than 0 if it
1283 * doesn't match or less than zero on error.
1285 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1286 const struct qstr *entry, bool quick)
1288 const struct super_block *sb = parent->i_sb;
1289 const struct unicode_map *um = sb->s_encoding;
1293 ret = utf8_strncasecmp_folded(um, name, entry);
1295 ret = utf8_strncasecmp(um, name, entry);
1298 /* Handle invalid character sequence as either an error
1299 * or as an opaque byte sequence.
1301 if (sb_has_strict_encoding(sb))
1304 if (name->len != entry->len)
1307 return !!memcmp(name->name, entry->name, name->len);
1313 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1314 struct fscrypt_str *cf_name)
1318 if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) {
1319 cf_name->name = NULL;
1323 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1327 len = utf8_casefold(dir->i_sb->s_encoding,
1328 iname, cf_name->name,
1331 kfree(cf_name->name);
1332 cf_name->name = NULL;
1335 cf_name->len = (unsigned) len;
1341 * Test whether a directory entry matches the filename being searched for.
1343 * Return: %true if the directory entry matches, otherwise %false.
1345 static inline bool ext4_match(const struct inode *parent,
1346 const struct ext4_filename *fname,
1347 const struct ext4_dir_entry_2 *de)
1349 struct fscrypt_name f;
1350 #ifdef CONFIG_UNICODE
1351 const struct qstr entry = {.name = de->name, .len = de->name_len};
1357 f.usr_fname = fname->usr_fname;
1358 f.disk_name = fname->disk_name;
1359 #ifdef CONFIG_FS_ENCRYPTION
1360 f.crypto_buf = fname->crypto_buf;
1363 #ifdef CONFIG_UNICODE
1364 if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) {
1365 if (fname->cf_name.name) {
1366 struct qstr cf = {.name = fname->cf_name.name,
1367 .len = fname->cf_name.len};
1368 return !ext4_ci_compare(parent, &cf, &entry, true);
1370 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1375 return fscrypt_match_name(&f, de->name, de->name_len);
1379 * Returns 0 if not found, -1 on failure, and 1 on success
1381 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1382 struct inode *dir, struct ext4_filename *fname,
1383 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1385 struct ext4_dir_entry_2 * de;
1389 de = (struct ext4_dir_entry_2 *)search_buf;
1390 dlimit = search_buf + buf_size;
1391 while ((char *) de < dlimit) {
1392 /* this code is executed quadratically often */
1393 /* do minimal checking `by hand' */
1394 if ((char *) de + de->name_len <= dlimit &&
1395 ext4_match(dir, fname, de)) {
1396 /* found a match - just to be sure, do
1398 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1404 /* prevent looping on a bad block */
1405 de_len = ext4_rec_len_from_disk(de->rec_len,
1406 dir->i_sb->s_blocksize);
1410 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1415 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1416 struct ext4_dir_entry *de)
1418 struct super_block *sb = dir->i_sb;
1424 if (de->inode == 0 &&
1425 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1432 * __ext4_find_entry()
1434 * finds an entry in the specified directory with the wanted name. It
1435 * returns the cache buffer in which the entry was found, and the entry
1436 * itself (as a parameter - res_dir). It does NOT read the inode of the
1437 * entry - you'll have to do that yourself if you want to.
1439 * The returned buffer_head has ->b_count elevated. The caller is expected
1440 * to brelse() it when appropriate.
1442 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1443 struct ext4_filename *fname,
1444 struct ext4_dir_entry_2 **res_dir,
1447 struct super_block *sb;
1448 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1449 struct buffer_head *bh, *ret = NULL;
1450 ext4_lblk_t start, block;
1451 const u8 *name = fname->usr_fname->name;
1452 size_t ra_max = 0; /* Number of bh's in the readahead
1454 size_t ra_ptr = 0; /* Current index into readahead
1456 ext4_lblk_t nblocks;
1457 int i, namelen, retval;
1461 namelen = fname->usr_fname->len;
1462 if (namelen > EXT4_NAME_LEN)
1465 if (ext4_has_inline_data(dir)) {
1466 int has_inline_data = 1;
1467 ret = ext4_find_inline_entry(dir, fname, res_dir,
1469 if (has_inline_data) {
1472 goto cleanup_and_exit;
1476 if ((namelen <= 2) && (name[0] == '.') &&
1477 (name[1] == '.' || name[1] == '\0')) {
1479 * "." or ".." will only be in the first block
1480 * NFS may look up ".."; "." should be handled by the VFS
1487 ret = ext4_dx_find_entry(dir, fname, res_dir);
1489 * On success, or if the error was file not found,
1490 * return. Otherwise, fall back to doing a search the
1491 * old fashioned way.
1493 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1494 goto cleanup_and_exit;
1495 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1499 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1502 goto cleanup_and_exit;
1504 start = EXT4_I(dir)->i_dir_start_lookup;
1505 if (start >= nblocks)
1511 * We deal with the read-ahead logic here.
1514 if (ra_ptr >= ra_max) {
1515 /* Refill the readahead buffer */
1518 ra_max = start - block;
1520 ra_max = nblocks - block;
1521 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1522 retval = ext4_bread_batch(dir, block, ra_max,
1523 false /* wait */, bh_use);
1525 ret = ERR_PTR(retval);
1527 goto cleanup_and_exit;
1530 if ((bh = bh_use[ra_ptr++]) == NULL)
1533 if (!buffer_uptodate(bh)) {
1534 EXT4_ERROR_INODE_ERR(dir, EIO,
1535 "reading directory lblock %lu",
1536 (unsigned long) block);
1538 ret = ERR_PTR(-EIO);
1539 goto cleanup_and_exit;
1541 if (!buffer_verified(bh) &&
1542 !is_dx_internal_node(dir, block,
1543 (struct ext4_dir_entry *)bh->b_data) &&
1544 !ext4_dirblock_csum_verify(dir, bh)) {
1545 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1546 "checksumming directory "
1547 "block %lu", (unsigned long)block);
1549 ret = ERR_PTR(-EFSBADCRC);
1550 goto cleanup_and_exit;
1552 set_buffer_verified(bh);
1553 i = search_dirblock(bh, dir, fname,
1554 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1556 EXT4_I(dir)->i_dir_start_lookup = block;
1558 goto cleanup_and_exit;
1562 goto cleanup_and_exit;
1565 if (++block >= nblocks)
1567 } while (block != start);
1570 * If the directory has grown while we were searching, then
1571 * search the last part of the directory before giving up.
1574 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1575 if (block < nblocks) {
1581 /* Clean up the read-ahead blocks */
1582 for (; ra_ptr < ra_max; ra_ptr++)
1583 brelse(bh_use[ra_ptr]);
1587 static struct buffer_head *ext4_find_entry(struct inode *dir,
1588 const struct qstr *d_name,
1589 struct ext4_dir_entry_2 **res_dir,
1593 struct ext4_filename fname;
1594 struct buffer_head *bh;
1596 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1600 return ERR_PTR(err);
1602 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1604 ext4_fname_free_filename(&fname);
1608 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1609 struct dentry *dentry,
1610 struct ext4_dir_entry_2 **res_dir)
1613 struct ext4_filename fname;
1614 struct buffer_head *bh;
1616 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1617 generic_set_encrypted_ci_d_ops(dentry);
1621 return ERR_PTR(err);
1623 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1625 ext4_fname_free_filename(&fname);
1629 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1630 struct ext4_filename *fname,
1631 struct ext4_dir_entry_2 **res_dir)
1633 struct super_block * sb = dir->i_sb;
1634 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1635 struct buffer_head *bh;
1639 #ifdef CONFIG_FS_ENCRYPTION
1642 frame = dx_probe(fname, dir, NULL, frames);
1644 return (struct buffer_head *) frame;
1646 block = dx_get_block(frame->at);
1647 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1651 retval = search_dirblock(bh, dir, fname,
1652 block << EXT4_BLOCK_SIZE_BITS(sb),
1658 bh = ERR_PTR(ERR_BAD_DX_DIR);
1662 /* Check to see if we should continue to search */
1663 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1666 ext4_warning_inode(dir,
1667 "error %d reading directory index block",
1669 bh = ERR_PTR(retval);
1672 } while (retval == 1);
1676 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1682 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1684 struct inode *inode;
1685 struct ext4_dir_entry_2 *de;
1686 struct buffer_head *bh;
1688 if (dentry->d_name.len > EXT4_NAME_LEN)
1689 return ERR_PTR(-ENAMETOOLONG);
1691 bh = ext4_lookup_entry(dir, dentry, &de);
1693 return ERR_CAST(bh);
1696 __u32 ino = le32_to_cpu(de->inode);
1698 if (!ext4_valid_inum(dir->i_sb, ino)) {
1699 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1700 return ERR_PTR(-EFSCORRUPTED);
1702 if (unlikely(ino == dir->i_ino)) {
1703 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1705 return ERR_PTR(-EFSCORRUPTED);
1707 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1708 if (inode == ERR_PTR(-ESTALE)) {
1709 EXT4_ERROR_INODE(dir,
1710 "deleted inode referenced: %u",
1712 return ERR_PTR(-EFSCORRUPTED);
1714 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1715 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1716 !fscrypt_has_permitted_context(dir, inode)) {
1717 ext4_warning(inode->i_sb,
1718 "Inconsistent encryption contexts: %lu/%lu",
1719 dir->i_ino, inode->i_ino);
1721 return ERR_PTR(-EPERM);
1725 #ifdef CONFIG_UNICODE
1726 if (!inode && IS_CASEFOLDED(dir)) {
1727 /* Eventually we want to call d_add_ci(dentry, NULL)
1728 * for negative dentries in the encoding case as
1729 * well. For now, prevent the negative dentry
1730 * from being cached.
1735 return d_splice_alias(inode, dentry);
1739 struct dentry *ext4_get_parent(struct dentry *child)
1742 struct ext4_dir_entry_2 * de;
1743 struct buffer_head *bh;
1745 bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1747 return ERR_CAST(bh);
1749 return ERR_PTR(-ENOENT);
1750 ino = le32_to_cpu(de->inode);
1753 if (!ext4_valid_inum(child->d_sb, ino)) {
1754 EXT4_ERROR_INODE(d_inode(child),
1755 "bad parent inode number: %u", ino);
1756 return ERR_PTR(-EFSCORRUPTED);
1759 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1763 * Move count entries from end of map between two memory locations.
1764 * Returns pointer to last entry moved.
1766 static struct ext4_dir_entry_2 *
1767 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1770 unsigned rec_len = 0;
1773 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1774 (from + (map->offs<<2));
1775 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1776 memcpy (to, de, rec_len);
1777 ((struct ext4_dir_entry_2 *) to)->rec_len =
1778 ext4_rec_len_to_disk(rec_len, blocksize);
1783 return (struct ext4_dir_entry_2 *) (to - rec_len);
1787 * Compact each dir entry in the range to the minimal rec_len.
1788 * Returns pointer to last entry in range.
1790 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1792 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1793 unsigned rec_len = 0;
1796 while ((char*)de < base + blocksize) {
1797 next = ext4_next_entry(de, blocksize);
1798 if (de->inode && de->name_len) {
1799 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1801 memmove(to, de, rec_len);
1802 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1804 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1812 * Split a full leaf block to make room for a new dir entry.
1813 * Allocate a new block, and move entries so that they are approx. equally full.
1814 * Returns pointer to de in block into which the new entry will be inserted.
1816 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1817 struct buffer_head **bh,struct dx_frame *frame,
1818 struct dx_hash_info *hinfo)
1820 unsigned blocksize = dir->i_sb->s_blocksize;
1821 unsigned count, continued;
1822 struct buffer_head *bh2;
1823 ext4_lblk_t newblock;
1825 struct dx_map_entry *map;
1826 char *data1 = (*bh)->b_data, *data2;
1827 unsigned split, move, size;
1828 struct ext4_dir_entry_2 *de = NULL, *de2;
1832 if (ext4_has_metadata_csum(dir->i_sb))
1833 csum_size = sizeof(struct ext4_dir_entry_tail);
1835 bh2 = ext4_append(handle, dir, &newblock);
1839 return (struct ext4_dir_entry_2 *) bh2;
1842 BUFFER_TRACE(*bh, "get_write_access");
1843 err = ext4_journal_get_write_access(handle, *bh);
1847 BUFFER_TRACE(frame->bh, "get_write_access");
1848 err = ext4_journal_get_write_access(handle, frame->bh);
1852 data2 = bh2->b_data;
1854 /* create map in the end of data2 block */
1855 map = (struct dx_map_entry *) (data2 + blocksize);
1856 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1857 blocksize, hinfo, map);
1859 dx_sort_map(map, count);
1860 /* Ensure that neither split block is over half full */
1863 for (i = count-1; i >= 0; i--) {
1864 /* is more than half of this entry in 2nd half of the block? */
1865 if (size + map[i].size/2 > blocksize/2)
1867 size += map[i].size;
1871 * map index at which we will split
1873 * If the sum of active entries didn't exceed half the block size, just
1874 * split it in half by count; each resulting block will have at least
1875 * half the space free.
1878 split = count - move;
1882 hash2 = map[split].hash;
1883 continued = hash2 == map[split - 1].hash;
1884 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1885 (unsigned long)dx_get_block(frame->at),
1886 hash2, split, count-split));
1888 /* Fancy dance to stay within two buffers */
1889 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1891 de = dx_pack_dirents(data1, blocksize);
1892 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1895 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1899 ext4_initialize_dirent_tail(*bh, blocksize);
1900 ext4_initialize_dirent_tail(bh2, blocksize);
1903 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1905 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1908 /* Which block gets the new entry? */
1909 if (hinfo->hash >= hash2) {
1913 dx_insert_block(frame, hash2 + continued, newblock);
1914 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1917 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1921 dxtrace(dx_show_index("frame", frame->entries));
1928 ext4_std_error(dir->i_sb, err);
1929 return ERR_PTR(err);
1932 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1933 struct buffer_head *bh,
1934 void *buf, int buf_size,
1935 struct ext4_filename *fname,
1936 struct ext4_dir_entry_2 **dest_de)
1938 struct ext4_dir_entry_2 *de;
1939 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1941 unsigned int offset = 0;
1944 de = (struct ext4_dir_entry_2 *)buf;
1945 top = buf + buf_size - reclen;
1946 while ((char *) de <= top) {
1947 if (ext4_check_dir_entry(dir, NULL, de, bh,
1948 buf, buf_size, offset))
1949 return -EFSCORRUPTED;
1950 if (ext4_match(dir, fname, de))
1952 nlen = EXT4_DIR_REC_LEN(de->name_len);
1953 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1954 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1956 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1959 if ((char *) de > top)
1966 void ext4_insert_dentry(struct inode *inode,
1967 struct ext4_dir_entry_2 *de,
1969 struct ext4_filename *fname)
1974 nlen = EXT4_DIR_REC_LEN(de->name_len);
1975 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1977 struct ext4_dir_entry_2 *de1 =
1978 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1979 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1980 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1983 de->file_type = EXT4_FT_UNKNOWN;
1984 de->inode = cpu_to_le32(inode->i_ino);
1985 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1986 de->name_len = fname_len(fname);
1987 memcpy(de->name, fname_name(fname), fname_len(fname));
1991 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1992 * it points to a directory entry which is guaranteed to be large
1993 * enough for new directory entry. If de is NULL, then
1994 * add_dirent_to_buf will attempt search the directory block for
1995 * space. It will return -ENOSPC if no space is available, and -EIO
1996 * and -EEXIST if directory entry already exists.
1998 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2000 struct inode *inode, struct ext4_dir_entry_2 *de,
2001 struct buffer_head *bh)
2003 unsigned int blocksize = dir->i_sb->s_blocksize;
2007 if (ext4_has_metadata_csum(inode->i_sb))
2008 csum_size = sizeof(struct ext4_dir_entry_tail);
2011 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2012 blocksize - csum_size, fname, &de);
2016 BUFFER_TRACE(bh, "get_write_access");
2017 err = ext4_journal_get_write_access(handle, bh);
2019 ext4_std_error(dir->i_sb, err);
2023 /* By now the buffer is marked for journaling */
2024 ext4_insert_dentry(inode, de, blocksize, fname);
2027 * XXX shouldn't update any times until successful
2028 * completion of syscall, but too many callers depend
2031 * XXX similarly, too many callers depend on
2032 * ext4_new_inode() setting the times, but error
2033 * recovery deletes the inode, so the worst that can
2034 * happen is that the times are slightly out of date
2035 * and/or different from the directory change time.
2037 dir->i_mtime = dir->i_ctime = current_time(dir);
2038 ext4_update_dx_flag(dir);
2039 inode_inc_iversion(dir);
2040 err2 = ext4_mark_inode_dirty(handle, dir);
2041 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2042 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2044 ext4_std_error(dir->i_sb, err);
2045 return err ? err : err2;
2049 * This converts a one block unindexed directory to a 3 block indexed
2050 * directory, and adds the dentry to the indexed directory.
2052 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2054 struct inode *inode, struct buffer_head *bh)
2056 struct buffer_head *bh2;
2057 struct dx_root *root;
2058 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2059 struct dx_entry *entries;
2060 struct ext4_dir_entry_2 *de, *de2;
2066 struct fake_dirent *fde;
2069 if (ext4_has_metadata_csum(inode->i_sb))
2070 csum_size = sizeof(struct ext4_dir_entry_tail);
2072 blocksize = dir->i_sb->s_blocksize;
2073 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2074 BUFFER_TRACE(bh, "get_write_access");
2075 retval = ext4_journal_get_write_access(handle, bh);
2077 ext4_std_error(dir->i_sb, retval);
2081 root = (struct dx_root *) bh->b_data;
2083 /* The 0th block becomes the root, move the dirents out */
2084 fde = &root->dotdot;
2085 de = (struct ext4_dir_entry_2 *)((char *)fde +
2086 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2087 if ((char *) de >= (((char *) root) + blocksize)) {
2088 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2090 return -EFSCORRUPTED;
2092 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2094 /* Allocate new block for the 0th block's dirents */
2095 bh2 = ext4_append(handle, dir, &block);
2098 return PTR_ERR(bh2);
2100 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2101 data2 = bh2->b_data;
2103 memcpy(data2, de, len);
2104 de = (struct ext4_dir_entry_2 *) data2;
2106 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2108 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2109 (char *) de, blocksize);
2112 ext4_initialize_dirent_tail(bh2, blocksize);
2114 /* Initialize the root; the dot dirents already exist */
2115 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2116 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2118 memset (&root->info, 0, sizeof(root->info));
2119 root->info.info_length = sizeof(root->info);
2120 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2121 entries = root->entries;
2122 dx_set_block(entries, 1);
2123 dx_set_count(entries, 1);
2124 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2126 /* Initialize as for dx_probe */
2127 fname->hinfo.hash_version = root->info.hash_version;
2128 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2129 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2130 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2131 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2133 memset(frames, 0, sizeof(frames));
2135 frame->entries = entries;
2136 frame->at = entries;
2139 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2142 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2146 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2148 retval = PTR_ERR(de);
2152 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2155 * Even if the block split failed, we have to properly write
2156 * out all the changes we did so far. Otherwise we can end up
2157 * with corrupted filesystem.
2160 ext4_mark_inode_dirty(handle, dir);
2169 * adds a file entry to the specified directory, using the same
2170 * semantics as ext4_find_entry(). It returns NULL if it failed.
2172 * NOTE!! The inode part of 'de' is left at 0 - which means you
2173 * may not sleep between calling this and putting something into
2174 * the entry, as someone else might have used it while you slept.
2176 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2177 struct inode *inode)
2179 struct inode *dir = d_inode(dentry->d_parent);
2180 struct buffer_head *bh = NULL;
2181 struct ext4_dir_entry_2 *de;
2182 struct super_block *sb;
2183 struct ext4_filename fname;
2187 ext4_lblk_t block, blocks;
2190 if (ext4_has_metadata_csum(inode->i_sb))
2191 csum_size = sizeof(struct ext4_dir_entry_tail);
2194 blocksize = sb->s_blocksize;
2195 if (!dentry->d_name.len)
2198 if (fscrypt_is_nokey_name(dentry))
2201 #ifdef CONFIG_UNICODE
2202 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2203 sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2207 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2211 if (ext4_has_inline_data(dir)) {
2212 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2222 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2223 if (!retval || (retval != ERR_BAD_DX_DIR))
2225 /* Can we just ignore htree data? */
2226 if (ext4_has_metadata_csum(sb)) {
2227 EXT4_ERROR_INODE(dir,
2228 "Directory has corrupted htree index.");
2229 retval = -EFSCORRUPTED;
2232 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2234 retval = ext4_mark_inode_dirty(handle, dir);
2235 if (unlikely(retval))
2238 blocks = dir->i_size >> sb->s_blocksize_bits;
2239 for (block = 0; block < blocks; block++) {
2240 bh = ext4_read_dirblock(dir, block, DIRENT);
2242 bh = ext4_bread(handle, dir, block,
2243 EXT4_GET_BLOCKS_CREATE);
2244 goto add_to_new_block;
2247 retval = PTR_ERR(bh);
2251 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2253 if (retval != -ENOSPC)
2256 if (blocks == 1 && !dx_fallback &&
2257 ext4_has_feature_dir_index(sb)) {
2258 retval = make_indexed_dir(handle, &fname, dir,
2260 bh = NULL; /* make_indexed_dir releases bh */
2265 bh = ext4_append(handle, dir, &block);
2268 retval = PTR_ERR(bh);
2272 de = (struct ext4_dir_entry_2 *) bh->b_data;
2274 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2277 ext4_initialize_dirent_tail(bh, blocksize);
2279 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2281 ext4_fname_free_filename(&fname);
2284 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2289 * Returns 0 for success, or a negative error value
2291 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2292 struct inode *dir, struct inode *inode)
2294 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2295 struct dx_entry *entries, *at;
2296 struct buffer_head *bh;
2297 struct super_block *sb = dir->i_sb;
2298 struct ext4_dir_entry_2 *de;
2304 frame = dx_probe(fname, dir, NULL, frames);
2306 return PTR_ERR(frame);
2307 entries = frame->entries;
2309 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2316 BUFFER_TRACE(bh, "get_write_access");
2317 err = ext4_journal_get_write_access(handle, bh);
2321 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2326 /* Block full, should compress but for now just split */
2327 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2328 dx_get_count(entries), dx_get_limit(entries)));
2329 /* Need to split index? */
2330 if (dx_get_count(entries) == dx_get_limit(entries)) {
2331 ext4_lblk_t newblock;
2332 int levels = frame - frames + 1;
2333 unsigned int icount;
2335 struct dx_entry *entries2;
2336 struct dx_node *node2;
2337 struct buffer_head *bh2;
2339 while (frame > frames) {
2340 if (dx_get_count((frame - 1)->entries) <
2341 dx_get_limit((frame - 1)->entries)) {
2345 frame--; /* split higher index block */
2347 entries = frame->entries;
2350 if (add_level && levels == ext4_dir_htree_level(sb)) {
2351 ext4_warning(sb, "Directory (ino: %lu) index full, "
2352 "reach max htree level :%d",
2353 dir->i_ino, levels);
2354 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2355 ext4_warning(sb, "Large directory feature is "
2356 "not enabled on this "
2362 icount = dx_get_count(entries);
2363 bh2 = ext4_append(handle, dir, &newblock);
2368 node2 = (struct dx_node *)(bh2->b_data);
2369 entries2 = node2->entries;
2370 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2371 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2373 BUFFER_TRACE(frame->bh, "get_write_access");
2374 err = ext4_journal_get_write_access(handle, frame->bh);
2378 unsigned icount1 = icount/2, icount2 = icount - icount1;
2379 unsigned hash2 = dx_get_hash(entries + icount1);
2380 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2383 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2384 err = ext4_journal_get_write_access(handle,
2389 memcpy((char *) entries2, (char *) (entries + icount1),
2390 icount2 * sizeof(struct dx_entry));
2391 dx_set_count(entries, icount1);
2392 dx_set_count(entries2, icount2);
2393 dx_set_limit(entries2, dx_node_limit(dir));
2395 /* Which index block gets the new entry? */
2396 if (at - entries >= icount1) {
2397 frame->at = at = at - entries - icount1 + entries2;
2398 frame->entries = entries = entries2;
2399 swap(frame->bh, bh2);
2401 dx_insert_block((frame - 1), hash2, newblock);
2402 dxtrace(dx_show_index("node", frame->entries));
2403 dxtrace(dx_show_index("node",
2404 ((struct dx_node *) bh2->b_data)->entries));
2405 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2409 err = ext4_handle_dirty_dx_node(handle, dir,
2413 err = ext4_handle_dirty_dx_node(handle, dir,
2418 struct dx_root *dxroot;
2419 memcpy((char *) entries2, (char *) entries,
2420 icount * sizeof(struct dx_entry));
2421 dx_set_limit(entries2, dx_node_limit(dir));
2424 dx_set_count(entries, 1);
2425 dx_set_block(entries + 0, newblock);
2426 dxroot = (struct dx_root *)frames[0].bh->b_data;
2427 dxroot->info.indirect_levels += 1;
2428 dxtrace(printk(KERN_DEBUG
2429 "Creating %d level index...\n",
2430 dxroot->info.indirect_levels));
2431 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2434 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2440 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2445 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2449 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2453 /* @restart is true means htree-path has been changed, we need to
2454 * repeat dx_probe() to find out valid htree-path
2456 if (restart && err == 0)
2462 * ext4_generic_delete_entry deletes a directory entry by merging it
2463 * with the previous entry
2465 int ext4_generic_delete_entry(struct inode *dir,
2466 struct ext4_dir_entry_2 *de_del,
2467 struct buffer_head *bh,
2472 struct ext4_dir_entry_2 *de, *pde;
2473 unsigned int blocksize = dir->i_sb->s_blocksize;
2478 de = (struct ext4_dir_entry_2 *)entry_buf;
2479 while (i < buf_size - csum_size) {
2480 if (ext4_check_dir_entry(dir, NULL, de, bh,
2481 entry_buf, buf_size, i))
2482 return -EFSCORRUPTED;
2485 pde->rec_len = ext4_rec_len_to_disk(
2486 ext4_rec_len_from_disk(pde->rec_len,
2488 ext4_rec_len_from_disk(de->rec_len,
2493 inode_inc_iversion(dir);
2496 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2498 de = ext4_next_entry(de, blocksize);
2503 static int ext4_delete_entry(handle_t *handle,
2505 struct ext4_dir_entry_2 *de_del,
2506 struct buffer_head *bh)
2508 int err, csum_size = 0;
2510 if (ext4_has_inline_data(dir)) {
2511 int has_inline_data = 1;
2512 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2514 if (has_inline_data)
2518 if (ext4_has_metadata_csum(dir->i_sb))
2519 csum_size = sizeof(struct ext4_dir_entry_tail);
2521 BUFFER_TRACE(bh, "get_write_access");
2522 err = ext4_journal_get_write_access(handle, bh);
2526 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2527 dir->i_sb->s_blocksize, csum_size);
2531 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2532 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2539 ext4_std_error(dir->i_sb, err);
2544 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2545 * since this indicates that nlinks count was previously 1 to avoid overflowing
2546 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2547 * that subdirectory link counts are not being maintained accurately.
2549 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2550 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2551 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2552 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2554 static void ext4_inc_count(struct inode *inode)
2558 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2559 set_nlink(inode, 1);
2563 * If a directory had nlink == 1, then we should let it be 1. This indicates
2564 * directory has >EXT4_LINK_MAX subdirs.
2566 static void ext4_dec_count(struct inode *inode)
2568 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2574 * Add non-directory inode to a directory. On success, the inode reference is
2575 * consumed by dentry is instantiation. This is also indicated by clearing of
2576 * *inodep pointer. On failure, the caller is responsible for dropping the
2577 * inode reference in the safe context.
2579 static int ext4_add_nondir(handle_t *handle,
2580 struct dentry *dentry, struct inode **inodep)
2582 struct inode *dir = d_inode(dentry->d_parent);
2583 struct inode *inode = *inodep;
2584 int err = ext4_add_entry(handle, dentry, inode);
2586 err = ext4_mark_inode_dirty(handle, inode);
2587 if (IS_DIRSYNC(dir))
2588 ext4_handle_sync(handle);
2589 d_instantiate_new(dentry, inode);
2594 ext4_orphan_add(handle, inode);
2595 unlock_new_inode(inode);
2600 * By the time this is called, we already have created
2601 * the directory cache entry for the new file, but it
2602 * is so far negative - it has no inode.
2604 * If the create succeeds, we fill in the inode information
2605 * with d_instantiate().
2607 static int ext4_create(struct user_namespace *mnt_userns, struct inode *dir,
2608 struct dentry *dentry, umode_t mode, bool excl)
2611 struct inode *inode;
2612 int err, credits, retries = 0;
2614 err = dquot_initialize(dir);
2618 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2619 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2621 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2622 0, NULL, EXT4_HT_DIR, credits);
2623 handle = ext4_journal_current_handle();
2624 err = PTR_ERR(inode);
2625 if (!IS_ERR(inode)) {
2626 inode->i_op = &ext4_file_inode_operations;
2627 inode->i_fop = &ext4_file_operations;
2628 ext4_set_aops(inode);
2629 err = ext4_add_nondir(handle, dentry, &inode);
2631 ext4_fc_track_create(handle, dentry);
2634 ext4_journal_stop(handle);
2635 if (!IS_ERR_OR_NULL(inode))
2637 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2642 static int ext4_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2643 struct dentry *dentry, umode_t mode, dev_t rdev)
2646 struct inode *inode;
2647 int err, credits, retries = 0;
2649 err = dquot_initialize(dir);
2653 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2654 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2656 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode, &dentry->d_name,
2657 0, NULL, EXT4_HT_DIR, credits);
2658 handle = ext4_journal_current_handle();
2659 err = PTR_ERR(inode);
2660 if (!IS_ERR(inode)) {
2661 init_special_inode(inode, inode->i_mode, rdev);
2662 inode->i_op = &ext4_special_inode_operations;
2663 err = ext4_add_nondir(handle, dentry, &inode);
2665 ext4_fc_track_create(handle, dentry);
2668 ext4_journal_stop(handle);
2669 if (!IS_ERR_OR_NULL(inode))
2671 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2676 static int ext4_tmpfile(struct user_namespace *mnt_userns, struct inode *dir,
2677 struct dentry *dentry, umode_t mode)
2680 struct inode *inode;
2681 int err, retries = 0;
2683 err = dquot_initialize(dir);
2688 inode = ext4_new_inode_start_handle(mnt_userns, dir, mode,
2691 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2692 4 + EXT4_XATTR_TRANS_BLOCKS);
2693 handle = ext4_journal_current_handle();
2694 err = PTR_ERR(inode);
2695 if (!IS_ERR(inode)) {
2696 inode->i_op = &ext4_file_inode_operations;
2697 inode->i_fop = &ext4_file_operations;
2698 ext4_set_aops(inode);
2699 d_tmpfile(dentry, inode);
2700 err = ext4_orphan_add(handle, inode);
2702 goto err_unlock_inode;
2703 mark_inode_dirty(inode);
2704 unlock_new_inode(inode);
2707 ext4_journal_stop(handle);
2708 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2712 ext4_journal_stop(handle);
2713 unlock_new_inode(inode);
2717 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2718 struct ext4_dir_entry_2 *de,
2719 int blocksize, int csum_size,
2720 unsigned int parent_ino, int dotdot_real_len)
2722 de->inode = cpu_to_le32(inode->i_ino);
2724 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2726 strcpy(de->name, ".");
2727 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2729 de = ext4_next_entry(de, blocksize);
2730 de->inode = cpu_to_le32(parent_ino);
2732 if (!dotdot_real_len)
2733 de->rec_len = ext4_rec_len_to_disk(blocksize -
2734 (csum_size + EXT4_DIR_REC_LEN(1)),
2737 de->rec_len = ext4_rec_len_to_disk(
2738 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2739 strcpy(de->name, "..");
2740 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2742 return ext4_next_entry(de, blocksize);
2745 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2746 struct inode *inode)
2748 struct buffer_head *dir_block = NULL;
2749 struct ext4_dir_entry_2 *de;
2750 ext4_lblk_t block = 0;
2751 unsigned int blocksize = dir->i_sb->s_blocksize;
2755 if (ext4_has_metadata_csum(dir->i_sb))
2756 csum_size = sizeof(struct ext4_dir_entry_tail);
2758 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2759 err = ext4_try_create_inline_dir(handle, dir, inode);
2760 if (err < 0 && err != -ENOSPC)
2767 dir_block = ext4_append(handle, inode, &block);
2768 if (IS_ERR(dir_block))
2769 return PTR_ERR(dir_block);
2770 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2771 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2772 set_nlink(inode, 2);
2774 ext4_initialize_dirent_tail(dir_block, blocksize);
2776 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2777 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2780 set_buffer_verified(dir_block);
2786 static int ext4_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2787 struct dentry *dentry, umode_t mode)
2790 struct inode *inode;
2791 int err, err2 = 0, credits, retries = 0;
2793 if (EXT4_DIR_LINK_MAX(dir))
2796 err = dquot_initialize(dir);
2800 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2801 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2803 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFDIR | mode,
2805 0, NULL, EXT4_HT_DIR, credits);
2806 handle = ext4_journal_current_handle();
2807 err = PTR_ERR(inode);
2811 inode->i_op = &ext4_dir_inode_operations;
2812 inode->i_fop = &ext4_dir_operations;
2813 err = ext4_init_new_dir(handle, dir, inode);
2815 goto out_clear_inode;
2816 err = ext4_mark_inode_dirty(handle, inode);
2818 err = ext4_add_entry(handle, dentry, inode);
2822 ext4_orphan_add(handle, inode);
2823 unlock_new_inode(inode);
2824 err2 = ext4_mark_inode_dirty(handle, inode);
2827 ext4_journal_stop(handle);
2831 ext4_inc_count(dir);
2833 ext4_update_dx_flag(dir);
2834 err = ext4_mark_inode_dirty(handle, dir);
2836 goto out_clear_inode;
2837 d_instantiate_new(dentry, inode);
2838 ext4_fc_track_create(handle, dentry);
2839 if (IS_DIRSYNC(dir))
2840 ext4_handle_sync(handle);
2844 ext4_journal_stop(handle);
2846 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2852 * routine to check that the specified directory is empty (for rmdir)
2854 bool ext4_empty_dir(struct inode *inode)
2856 unsigned int offset;
2857 struct buffer_head *bh;
2858 struct ext4_dir_entry_2 *de;
2859 struct super_block *sb;
2861 if (ext4_has_inline_data(inode)) {
2862 int has_inline_data = 1;
2865 ret = empty_inline_dir(inode, &has_inline_data);
2866 if (has_inline_data)
2871 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2872 EXT4_ERROR_INODE(inode, "invalid size");
2875 /* The first directory block must not be a hole,
2876 * so treat it as DIRENT_HTREE
2878 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2882 de = (struct ext4_dir_entry_2 *) bh->b_data;
2883 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2885 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2886 ext4_warning_inode(inode, "directory missing '.'");
2890 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2891 de = ext4_next_entry(de, sb->s_blocksize);
2892 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2894 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2895 ext4_warning_inode(inode, "directory missing '..'");
2899 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2900 while (offset < inode->i_size) {
2901 if (!(offset & (sb->s_blocksize - 1))) {
2902 unsigned int lblock;
2904 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2905 bh = ext4_read_dirblock(inode, lblock, EITHER);
2907 offset += sb->s_blocksize;
2913 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2914 (offset & (sb->s_blocksize - 1)));
2915 if (ext4_check_dir_entry(inode, NULL, de, bh,
2916 bh->b_data, bh->b_size, offset)) {
2917 offset = (offset | (sb->s_blocksize - 1)) + 1;
2920 if (le32_to_cpu(de->inode)) {
2924 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2931 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2932 * such inodes, starting at the superblock, in case we crash before the
2933 * file is closed/deleted, or in case the inode truncate spans multiple
2934 * transactions and the last transaction is not recovered after a crash.
2936 * At filesystem recovery time, we walk this list deleting unlinked
2937 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2939 * Orphan list manipulation functions must be called under i_mutex unless
2940 * we are just creating the inode or deleting it.
2942 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2944 struct super_block *sb = inode->i_sb;
2945 struct ext4_sb_info *sbi = EXT4_SB(sb);
2946 struct ext4_iloc iloc;
2950 if (!sbi->s_journal || is_bad_inode(inode))
2953 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2954 !inode_is_locked(inode));
2956 * Exit early if inode already is on orphan list. This is a big speedup
2957 * since we don't have to contend on the global s_orphan_lock.
2959 if (!list_empty(&EXT4_I(inode)->i_orphan))
2963 * Orphan handling is only valid for files with data blocks
2964 * being truncated, or files being unlinked. Note that we either
2965 * hold i_mutex, or the inode can not be referenced from outside,
2966 * so i_nlink should not be bumped due to race
2968 ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2969 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2971 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2972 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2976 err = ext4_reserve_inode_write(handle, inode, &iloc);
2980 mutex_lock(&sbi->s_orphan_lock);
2982 * Due to previous errors inode may be already a part of on-disk
2983 * orphan list. If so skip on-disk list modification.
2985 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2986 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2987 /* Insert this inode at the head of the on-disk orphan list */
2988 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2989 lock_buffer(sbi->s_sbh);
2990 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2991 ext4_superblock_csum_set(sb);
2992 unlock_buffer(sbi->s_sbh);
2995 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2996 mutex_unlock(&sbi->s_orphan_lock);
2999 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3000 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
3005 * We have to remove inode from in-memory list if
3006 * addition to on disk orphan list failed. Stray orphan
3007 * list entries can cause panics at unmount time.
3009 mutex_lock(&sbi->s_orphan_lock);
3010 list_del_init(&EXT4_I(inode)->i_orphan);
3011 mutex_unlock(&sbi->s_orphan_lock);
3016 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3017 jbd_debug(4, "orphan inode %lu will point to %d\n",
3018 inode->i_ino, NEXT_ORPHAN(inode));
3020 ext4_std_error(sb, err);
3025 * ext4_orphan_del() removes an unlinked or truncated inode from the list
3026 * of such inodes stored on disk, because it is finally being cleaned up.
3028 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3030 struct list_head *prev;
3031 struct ext4_inode_info *ei = EXT4_I(inode);
3032 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3034 struct ext4_iloc iloc;
3037 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3040 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3041 !inode_is_locked(inode));
3042 /* Do this quick check before taking global s_orphan_lock. */
3043 if (list_empty(&ei->i_orphan))
3047 /* Grab inode buffer early before taking global s_orphan_lock */
3048 err = ext4_reserve_inode_write(handle, inode, &iloc);
3051 mutex_lock(&sbi->s_orphan_lock);
3052 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3054 prev = ei->i_orphan.prev;
3055 list_del_init(&ei->i_orphan);
3057 /* If we're on an error path, we may not have a valid
3058 * transaction handle with which to update the orphan list on
3059 * disk, but we still need to remove the inode from the linked
3060 * list in memory. */
3061 if (!handle || err) {
3062 mutex_unlock(&sbi->s_orphan_lock);
3066 ino_next = NEXT_ORPHAN(inode);
3067 if (prev == &sbi->s_orphan) {
3068 jbd_debug(4, "superblock will point to %u\n", ino_next);
3069 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3070 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3072 mutex_unlock(&sbi->s_orphan_lock);
3075 lock_buffer(sbi->s_sbh);
3076 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3077 ext4_superblock_csum_set(inode->i_sb);
3078 unlock_buffer(sbi->s_sbh);
3079 mutex_unlock(&sbi->s_orphan_lock);
3080 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
3082 struct ext4_iloc iloc2;
3083 struct inode *i_prev =
3084 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3086 jbd_debug(4, "orphan inode %lu will point to %u\n",
3087 i_prev->i_ino, ino_next);
3088 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3090 mutex_unlock(&sbi->s_orphan_lock);
3093 NEXT_ORPHAN(i_prev) = ino_next;
3094 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3095 mutex_unlock(&sbi->s_orphan_lock);
3099 NEXT_ORPHAN(inode) = 0;
3100 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3102 ext4_std_error(inode->i_sb, err);
3110 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3113 struct inode *inode;
3114 struct buffer_head *bh;
3115 struct ext4_dir_entry_2 *de;
3116 handle_t *handle = NULL;
3118 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3121 /* Initialize quotas before so that eventual writes go in
3122 * separate transaction */
3123 retval = dquot_initialize(dir);
3126 retval = dquot_initialize(d_inode(dentry));
3131 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3137 inode = d_inode(dentry);
3139 retval = -EFSCORRUPTED;
3140 if (le32_to_cpu(de->inode) != inode->i_ino)
3143 retval = -ENOTEMPTY;
3144 if (!ext4_empty_dir(inode))
3147 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3148 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3149 if (IS_ERR(handle)) {
3150 retval = PTR_ERR(handle);
3155 if (IS_DIRSYNC(dir))
3156 ext4_handle_sync(handle);
3158 retval = ext4_delete_entry(handle, dir, de, bh);
3161 if (!EXT4_DIR_LINK_EMPTY(inode))
3162 ext4_warning_inode(inode,
3163 "empty directory '%.*s' has too many links (%u)",
3164 dentry->d_name.len, dentry->d_name.name,
3166 inode_inc_iversion(inode);
3168 /* There's no need to set i_disksize: the fact that i_nlink is
3169 * zero will ensure that the right thing happens during any
3172 ext4_orphan_add(handle, inode);
3173 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3174 retval = ext4_mark_inode_dirty(handle, inode);
3177 ext4_dec_count(dir);
3178 ext4_update_dx_flag(dir);
3179 ext4_fc_track_unlink(handle, dentry);
3180 retval = ext4_mark_inode_dirty(handle, dir);
3182 #ifdef CONFIG_UNICODE
3183 /* VFS negative dentries are incompatible with Encoding and
3184 * Case-insensitiveness. Eventually we'll want avoid
3185 * invalidating the dentries here, alongside with returning the
3186 * negative dentries at ext4_lookup(), when it is better
3187 * supported by the VFS for the CI case.
3189 if (IS_CASEFOLDED(dir))
3190 d_invalidate(dentry);
3196 ext4_journal_stop(handle);
3200 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3201 struct inode *inode)
3203 int retval = -ENOENT;
3204 struct buffer_head *bh;
3205 struct ext4_dir_entry_2 *de;
3206 int skip_remove_dentry = 0;
3208 bh = ext4_find_entry(dir, d_name, &de, NULL);
3215 if (le32_to_cpu(de->inode) != inode->i_ino) {
3217 * It's okay if we find dont find dentry which matches
3218 * the inode. That's because it might have gotten
3219 * renamed to a different inode number
3221 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3222 skip_remove_dentry = 1;
3227 if (IS_DIRSYNC(dir))
3228 ext4_handle_sync(handle);
3230 if (!skip_remove_dentry) {
3231 retval = ext4_delete_entry(handle, dir, de, bh);
3234 dir->i_ctime = dir->i_mtime = current_time(dir);
3235 ext4_update_dx_flag(dir);
3236 retval = ext4_mark_inode_dirty(handle, dir);
3242 if (inode->i_nlink == 0)
3243 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3244 d_name->len, d_name->name);
3247 if (!inode->i_nlink)
3248 ext4_orphan_add(handle, inode);
3249 inode->i_ctime = current_time(inode);
3250 retval = ext4_mark_inode_dirty(handle, inode);
3257 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3262 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3265 trace_ext4_unlink_enter(dir, dentry);
3267 * Initialize quotas before so that eventual writes go
3268 * in separate transaction
3270 retval = dquot_initialize(dir);
3273 retval = dquot_initialize(d_inode(dentry));
3277 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3278 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3279 if (IS_ERR(handle)) {
3280 retval = PTR_ERR(handle);
3284 retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3286 ext4_fc_track_unlink(handle, dentry);
3287 #ifdef CONFIG_UNICODE
3288 /* VFS negative dentries are incompatible with Encoding and
3289 * Case-insensitiveness. Eventually we'll want avoid
3290 * invalidating the dentries here, alongside with returning the
3291 * negative dentries at ext4_lookup(), when it is better
3292 * supported by the VFS for the CI case.
3294 if (IS_CASEFOLDED(dir))
3295 d_invalidate(dentry);
3298 ext4_journal_stop(handle);
3301 trace_ext4_unlink_exit(dentry, retval);
3305 static int ext4_symlink(struct user_namespace *mnt_userns, struct inode *dir,
3306 struct dentry *dentry, const char *symname)
3309 struct inode *inode;
3310 int err, len = strlen(symname);
3312 struct fscrypt_str disk_link;
3314 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3317 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3322 err = dquot_initialize(dir);
3326 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3328 * For non-fast symlinks, we just allocate inode and put it on
3329 * orphan list in the first transaction => we need bitmap,
3330 * group descriptor, sb, inode block, quota blocks, and
3331 * possibly selinux xattr blocks.
3333 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3334 EXT4_XATTR_TRANS_BLOCKS;
3337 * Fast symlink. We have to add entry to directory
3338 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3339 * allocate new inode (bitmap, group descriptor, inode block,
3340 * quota blocks, sb is already counted in previous macros).
3342 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3343 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3346 inode = ext4_new_inode_start_handle(mnt_userns, dir, S_IFLNK|S_IRWXUGO,
3347 &dentry->d_name, 0, NULL,
3348 EXT4_HT_DIR, credits);
3349 handle = ext4_journal_current_handle();
3350 if (IS_ERR(inode)) {
3352 ext4_journal_stop(handle);
3353 return PTR_ERR(inode);
3356 if (IS_ENCRYPTED(inode)) {
3357 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3359 goto err_drop_inode;
3360 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3363 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3364 if (!IS_ENCRYPTED(inode))
3365 inode->i_op = &ext4_symlink_inode_operations;
3366 inode_nohighmem(inode);
3367 ext4_set_aops(inode);
3369 * We cannot call page_symlink() with transaction started
3370 * because it calls into ext4_write_begin() which can wait
3371 * for transaction commit if we are running out of space
3372 * and thus we deadlock. So we have to stop transaction now
3373 * and restart it when symlink contents is written.
3375 * To keep fs consistent in case of crash, we have to put inode
3376 * to orphan list in the mean time.
3379 err = ext4_orphan_add(handle, inode);
3381 ext4_journal_stop(handle);
3384 goto err_drop_inode;
3385 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3387 goto err_drop_inode;
3389 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3390 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3392 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3393 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3394 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3395 if (IS_ERR(handle)) {
3396 err = PTR_ERR(handle);
3398 goto err_drop_inode;
3400 set_nlink(inode, 1);
3401 err = ext4_orphan_del(handle, inode);
3403 goto err_drop_inode;
3405 /* clear the extent format for fast symlink */
3406 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3407 if (!IS_ENCRYPTED(inode)) {
3408 inode->i_op = &ext4_fast_symlink_inode_operations;
3409 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3411 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3413 inode->i_size = disk_link.len - 1;
3415 EXT4_I(inode)->i_disksize = inode->i_size;
3416 err = ext4_add_nondir(handle, dentry, &inode);
3418 ext4_journal_stop(handle);
3421 goto out_free_encrypted_link;
3425 ext4_journal_stop(handle);
3427 unlock_new_inode(inode);
3429 out_free_encrypted_link:
3430 if (disk_link.name != (unsigned char *)symname)
3431 kfree(disk_link.name);
3435 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3438 int err, retries = 0;
3440 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3441 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3442 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3444 return PTR_ERR(handle);
3446 if (IS_DIRSYNC(dir))
3447 ext4_handle_sync(handle);
3449 inode->i_ctime = current_time(inode);
3450 ext4_inc_count(inode);
3453 err = ext4_add_entry(handle, dentry, inode);
3455 err = ext4_mark_inode_dirty(handle, inode);
3456 /* this can happen only for tmpfile being
3457 * linked the first time
3459 if (inode->i_nlink == 1)
3460 ext4_orphan_del(handle, inode);
3461 d_instantiate(dentry, inode);
3462 ext4_fc_track_link(handle, dentry);
3467 ext4_journal_stop(handle);
3468 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3473 static int ext4_link(struct dentry *old_dentry,
3474 struct inode *dir, struct dentry *dentry)
3476 struct inode *inode = d_inode(old_dentry);
3479 if (inode->i_nlink >= EXT4_LINK_MAX)
3482 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3486 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3487 (!projid_eq(EXT4_I(dir)->i_projid,
3488 EXT4_I(old_dentry->d_inode)->i_projid)))
3491 err = dquot_initialize(dir);
3494 return __ext4_link(dir, inode, dentry);
3498 * Try to find buffer head where contains the parent block.
3499 * It should be the inode block if it is inlined or the 1st block
3500 * if it is a normal dir.
3502 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3503 struct inode *inode,
3505 struct ext4_dir_entry_2 **parent_de,
3508 struct buffer_head *bh;
3510 if (!ext4_has_inline_data(inode)) {
3511 /* The first directory block must not be a hole, so
3512 * treat it as DIRENT_HTREE
3514 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3516 *retval = PTR_ERR(bh);
3519 *parent_de = ext4_next_entry(
3520 (struct ext4_dir_entry_2 *)bh->b_data,
3521 inode->i_sb->s_blocksize);
3526 return ext4_get_first_inline_block(inode, parent_de, retval);
3529 struct ext4_renament {
3531 struct dentry *dentry;
3532 struct inode *inode;
3534 int dir_nlink_delta;
3536 /* entry for "dentry" */
3537 struct buffer_head *bh;
3538 struct ext4_dir_entry_2 *de;
3541 /* entry for ".." in inode if it's a directory */
3542 struct buffer_head *dir_bh;
3543 struct ext4_dir_entry_2 *parent_de;
3547 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3551 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3552 &retval, &ent->parent_de,
3556 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3557 return -EFSCORRUPTED;
3558 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3559 return ext4_journal_get_write_access(handle, ent->dir_bh);
3562 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3567 ent->parent_de->inode = cpu_to_le32(dir_ino);
3568 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3569 if (!ent->dir_inlined) {
3570 if (is_dx(ent->inode)) {
3571 retval = ext4_handle_dirty_dx_node(handle,
3575 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3579 retval = ext4_mark_inode_dirty(handle, ent->inode);
3582 ext4_std_error(ent->dir->i_sb, retval);
3588 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3589 unsigned ino, unsigned file_type)
3591 int retval, retval2;
3593 BUFFER_TRACE(ent->bh, "get write access");
3594 retval = ext4_journal_get_write_access(handle, ent->bh);
3597 ent->de->inode = cpu_to_le32(ino);
3598 if (ext4_has_feature_filetype(ent->dir->i_sb))
3599 ent->de->file_type = file_type;
3600 inode_inc_iversion(ent->dir);
3601 ent->dir->i_ctime = ent->dir->i_mtime =
3602 current_time(ent->dir);
3603 retval = ext4_mark_inode_dirty(handle, ent->dir);
3604 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3605 if (!ent->inlined) {
3606 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3607 if (unlikely(retval2)) {
3608 ext4_std_error(ent->dir->i_sb, retval2);
3615 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3616 const struct qstr *d_name)
3618 int retval = -ENOENT;
3619 struct buffer_head *bh;
3620 struct ext4_dir_entry_2 *de;
3622 bh = ext4_find_entry(dir, d_name, &de, NULL);
3626 retval = ext4_delete_entry(handle, dir, de, bh);
3632 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3637 * ent->de could have moved from under us during htree split, so make
3638 * sure that we are deleting the right entry. We might also be pointing
3639 * to a stale entry in the unused part of ent->bh so just checking inum
3640 * and the name isn't enough.
3642 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3643 ent->de->name_len != ent->dentry->d_name.len ||
3644 strncmp(ent->de->name, ent->dentry->d_name.name,
3645 ent->de->name_len) ||
3647 retval = ext4_find_delete_entry(handle, ent->dir,
3648 &ent->dentry->d_name);
3650 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3651 if (retval == -ENOENT) {
3652 retval = ext4_find_delete_entry(handle, ent->dir,
3653 &ent->dentry->d_name);
3658 ext4_warning_inode(ent->dir,
3659 "Deleting old file: nlink %d, error=%d",
3660 ent->dir->i_nlink, retval);
3664 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3666 if (ent->dir_nlink_delta) {
3667 if (ent->dir_nlink_delta == -1)
3668 ext4_dec_count(ent->dir);
3670 ext4_inc_count(ent->dir);
3671 ext4_mark_inode_dirty(handle, ent->dir);
3675 static struct inode *ext4_whiteout_for_rename(struct user_namespace *mnt_userns,
3676 struct ext4_renament *ent,
3677 int credits, handle_t **h)
3684 * for inode block, sb block, group summaries,
3687 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3688 EXT4_XATTR_TRANS_BLOCKS + 4);
3690 wh = ext4_new_inode_start_handle(mnt_userns, ent->dir,
3691 S_IFCHR | WHITEOUT_MODE,
3692 &ent->dentry->d_name, 0, NULL,
3693 EXT4_HT_DIR, credits);
3695 handle = ext4_journal_current_handle();
3698 ext4_journal_stop(handle);
3699 if (PTR_ERR(wh) == -ENOSPC &&
3700 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3704 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3705 wh->i_op = &ext4_special_inode_operations;
3711 * Anybody can rename anything with this: the permission checks are left to the
3712 * higher-level routines.
3714 * n.b. old_{dentry,inode) refers to the source dentry/inode
3715 * while new_{dentry,inode) refers to the destination dentry/inode
3716 * This comes from rename(const char *oldpath, const char *newpath)
3718 static int ext4_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
3719 struct dentry *old_dentry, struct inode *new_dir,
3720 struct dentry *new_dentry, unsigned int flags)
3722 handle_t *handle = NULL;
3723 struct ext4_renament old = {
3725 .dentry = old_dentry,
3726 .inode = d_inode(old_dentry),
3728 struct ext4_renament new = {
3730 .dentry = new_dentry,
3731 .inode = d_inode(new_dentry),
3735 struct inode *whiteout = NULL;
3739 if (new.inode && new.inode->i_nlink == 0) {
3740 EXT4_ERROR_INODE(new.inode,
3741 "target of rename is already freed");
3742 return -EFSCORRUPTED;
3745 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3746 (!projid_eq(EXT4_I(new_dir)->i_projid,
3747 EXT4_I(old_dentry->d_inode)->i_projid)))
3750 retval = dquot_initialize(old.dir);
3753 retval = dquot_initialize(new.dir);
3757 /* Initialize quotas before so that eventual writes go
3758 * in separate transaction */
3760 retval = dquot_initialize(new.inode);
3765 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3767 return PTR_ERR(old.bh);
3769 * Check for inode number is _not_ due to possible IO errors.
3770 * We might rmdir the source, keep it as pwd of some process
3771 * and merrily kill the link to whatever was created under the
3772 * same name. Goodbye sticky bit ;-<
3775 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3778 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3779 &new.de, &new.inlined);
3780 if (IS_ERR(new.bh)) {
3781 retval = PTR_ERR(new.bh);
3791 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3792 ext4_alloc_da_blocks(old.inode);
3794 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3795 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3796 if (!(flags & RENAME_WHITEOUT)) {
3797 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3798 if (IS_ERR(handle)) {
3799 retval = PTR_ERR(handle);
3804 whiteout = ext4_whiteout_for_rename(mnt_userns, &old, credits, &handle);
3805 if (IS_ERR(whiteout)) {
3806 retval = PTR_ERR(whiteout);
3812 old_file_type = old.de->file_type;
3813 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3814 ext4_handle_sync(handle);
3816 if (S_ISDIR(old.inode->i_mode)) {
3818 retval = -ENOTEMPTY;
3819 if (!ext4_empty_dir(new.inode))
3823 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3826 retval = ext4_rename_dir_prepare(handle, &old);
3831 * If we're renaming a file within an inline_data dir and adding or
3832 * setting the new dirent causes a conversion from inline_data to
3833 * extents/blockmap, we need to force the dirent delete code to
3834 * re-read the directory, or else we end up trying to delete a dirent
3835 * from what is now the extent tree root (or a block map).
3837 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3838 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3842 * Do this before adding a new entry, so the old entry is sure
3843 * to be still pointing to the valid old entry.
3845 retval = ext4_setent(handle, &old, whiteout->i_ino,
3849 retval = ext4_mark_inode_dirty(handle, whiteout);
3850 if (unlikely(retval))
3854 retval = ext4_add_entry(handle, new.dentry, old.inode);
3858 retval = ext4_setent(handle, &new,
3859 old.inode->i_ino, old_file_type);
3864 force_reread = !ext4_test_inode_flag(new.dir,
3865 EXT4_INODE_INLINE_DATA);
3868 * Like most other Unix systems, set the ctime for inodes on a
3871 old.inode->i_ctime = current_time(old.inode);
3872 retval = ext4_mark_inode_dirty(handle, old.inode);
3873 if (unlikely(retval))
3880 ext4_rename_delete(handle, &old, force_reread);
3884 ext4_dec_count(new.inode);
3885 new.inode->i_ctime = current_time(new.inode);
3887 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3888 ext4_update_dx_flag(old.dir);
3890 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3894 ext4_dec_count(old.dir);
3896 /* checked ext4_empty_dir above, can't have another
3897 * parent, ext4_dec_count() won't work for many-linked
3899 clear_nlink(new.inode);
3901 ext4_inc_count(new.dir);
3902 ext4_update_dx_flag(new.dir);
3903 retval = ext4_mark_inode_dirty(handle, new.dir);
3904 if (unlikely(retval))
3908 retval = ext4_mark_inode_dirty(handle, old.dir);
3909 if (unlikely(retval))
3912 if (S_ISDIR(old.inode->i_mode)) {
3914 * We disable fast commits here that's because the
3915 * replay code is not yet capable of changing dot dot
3916 * dirents in directories.
3918 ext4_fc_mark_ineligible(old.inode->i_sb,
3919 EXT4_FC_REASON_RENAME_DIR);
3922 ext4_fc_track_unlink(handle, new.dentry);
3923 __ext4_fc_track_link(handle, old.inode, new.dentry);
3924 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3928 retval = ext4_mark_inode_dirty(handle, new.inode);
3929 if (unlikely(retval))
3931 if (!new.inode->i_nlink)
3932 ext4_orphan_add(handle, new.inode);
3939 ext4_setent(handle, &old,
3940 old.inode->i_ino, old_file_type);
3941 drop_nlink(whiteout);
3943 unlock_new_inode(whiteout);
3951 ext4_journal_stop(handle);
3955 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3956 struct inode *new_dir, struct dentry *new_dentry)
3958 handle_t *handle = NULL;
3959 struct ext4_renament old = {
3961 .dentry = old_dentry,
3962 .inode = d_inode(old_dentry),
3964 struct ext4_renament new = {
3966 .dentry = new_dentry,
3967 .inode = d_inode(new_dentry),
3971 struct timespec64 ctime;
3973 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3974 !projid_eq(EXT4_I(new_dir)->i_projid,
3975 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3976 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3977 !projid_eq(EXT4_I(old_dir)->i_projid,
3978 EXT4_I(new_dentry->d_inode)->i_projid)))
3981 retval = dquot_initialize(old.dir);
3984 retval = dquot_initialize(new.dir);
3988 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3989 &old.de, &old.inlined);
3991 return PTR_ERR(old.bh);
3993 * Check for inode number is _not_ due to possible IO errors.
3994 * We might rmdir the source, keep it as pwd of some process
3995 * and merrily kill the link to whatever was created under the
3996 * same name. Goodbye sticky bit ;-<
3999 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4002 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4003 &new.de, &new.inlined);
4004 if (IS_ERR(new.bh)) {
4005 retval = PTR_ERR(new.bh);
4010 /* RENAME_EXCHANGE case: old *and* new must both exist */
4011 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4014 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4015 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4016 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4017 if (IS_ERR(handle)) {
4018 retval = PTR_ERR(handle);
4023 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4024 ext4_handle_sync(handle);
4026 if (S_ISDIR(old.inode->i_mode)) {
4028 retval = ext4_rename_dir_prepare(handle, &old);
4032 if (S_ISDIR(new.inode->i_mode)) {
4034 retval = ext4_rename_dir_prepare(handle, &new);
4040 * Other than the special case of overwriting a directory, parents'
4041 * nlink only needs to be modified if this is a cross directory rename.
4043 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4044 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4045 new.dir_nlink_delta = -old.dir_nlink_delta;
4047 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4048 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4052 new_file_type = new.de->file_type;
4053 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4057 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4062 * Like most other Unix systems, set the ctime for inodes on a
4065 ctime = current_time(old.inode);
4066 old.inode->i_ctime = ctime;
4067 new.inode->i_ctime = ctime;
4068 retval = ext4_mark_inode_dirty(handle, old.inode);
4069 if (unlikely(retval))
4071 retval = ext4_mark_inode_dirty(handle, new.inode);
4072 if (unlikely(retval))
4074 ext4_fc_mark_ineligible(new.inode->i_sb,
4075 EXT4_FC_REASON_CROSS_RENAME);
4077 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4082 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4086 ext4_update_dir_count(handle, &old);
4087 ext4_update_dir_count(handle, &new);
4096 ext4_journal_stop(handle);
4100 static int ext4_rename2(struct user_namespace *mnt_userns,
4101 struct inode *old_dir, struct dentry *old_dentry,
4102 struct inode *new_dir, struct dentry *new_dentry,
4107 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4110 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4113 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4118 if (flags & RENAME_EXCHANGE) {
4119 return ext4_cross_rename(old_dir, old_dentry,
4120 new_dir, new_dentry);
4123 return ext4_rename(mnt_userns, old_dir, old_dentry, new_dir, new_dentry, flags);
4127 * directories can handle most operations...
4129 const struct inode_operations ext4_dir_inode_operations = {
4130 .create = ext4_create,
4131 .lookup = ext4_lookup,
4133 .unlink = ext4_unlink,
4134 .symlink = ext4_symlink,
4135 .mkdir = ext4_mkdir,
4136 .rmdir = ext4_rmdir,
4137 .mknod = ext4_mknod,
4138 .tmpfile = ext4_tmpfile,
4139 .rename = ext4_rename2,
4140 .setattr = ext4_setattr,
4141 .getattr = ext4_getattr,
4142 .listxattr = ext4_listxattr,
4143 .get_acl = ext4_get_acl,
4144 .set_acl = ext4_set_acl,
4145 .fiemap = ext4_fiemap,
4148 const struct inode_operations ext4_special_inode_operations = {
4149 .setattr = ext4_setattr,
4150 .getattr = ext4_getattr,
4151 .listxattr = ext4_listxattr,
4152 .get_acl = ext4_get_acl,
4153 .set_acl = ext4_set_acl,