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 res = fscrypt_get_encryption_info(dir);
645 printk(KERN_WARNING "Error setting up"
646 " fname crypto: %d\n", res);
648 if (!fscrypt_has_encryption_key(dir)) {
649 /* Directory is not encrypted */
650 ext4fs_dirhash(dir, de->name,
652 printk("%*.s:(U)%x.%u ", len,
654 (unsigned) ((char *) de
657 struct fscrypt_str de_name =
658 FSTR_INIT(name, len);
660 /* Directory is encrypted */
661 res = fscrypt_fname_alloc_buffer(
662 len, &fname_crypto_str);
664 printk(KERN_WARNING "Error "
668 res = fscrypt_fname_disk_to_usr(dir,
672 printk(KERN_WARNING "Error "
673 "converting filename "
679 name = fname_crypto_str.name;
680 len = fname_crypto_str.len;
682 ext4fs_dirhash(dir, de->name,
684 printk("%*.s:(E)%x.%u ", len, name,
685 h.hash, (unsigned) ((char *) de
687 fscrypt_fname_free_buffer(
691 int len = de->name_len;
692 char *name = de->name;
693 ext4fs_dirhash(dir, de->name, de->name_len, &h);
694 printk("%*.s:%x.%u ", len, name, h.hash,
695 (unsigned) ((char *) de - base));
698 space += EXT4_DIR_REC_LEN(de->name_len);
701 de = ext4_next_entry(de, size);
703 printk(KERN_CONT "(%i)\n", names);
704 return (struct stats) { names, space, 1 };
707 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
708 struct dx_entry *entries, int levels)
710 unsigned blocksize = dir->i_sb->s_blocksize;
711 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
713 struct buffer_head *bh;
714 printk("%i indexed blocks...\n", count);
715 for (i = 0; i < count; i++, entries++)
717 ext4_lblk_t block = dx_get_block(entries);
718 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
719 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
721 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
722 bh = ext4_bread(NULL,dir, block, 0);
723 if (!bh || IS_ERR(bh))
726 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
727 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
728 bh->b_data, blocksize, 0);
729 names += stats.names;
730 space += stats.space;
731 bcount += stats.bcount;
735 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
736 levels ? "" : " ", names, space/bcount,
737 (space/bcount)*100/blocksize);
738 return (struct stats) { names, space, bcount};
740 #endif /* DX_DEBUG */
743 * Probe for a directory leaf block to search.
745 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
746 * error in the directory index, and the caller should fall back to
747 * searching the directory normally. The callers of dx_probe **MUST**
748 * check for this error code, and make sure it never gets reflected
751 static struct dx_frame *
752 dx_probe(struct ext4_filename *fname, struct inode *dir,
753 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
755 unsigned count, indirect;
756 struct dx_entry *at, *entries, *p, *q, *m;
757 struct dx_root *root;
758 struct dx_frame *frame = frame_in;
759 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
762 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
763 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
764 if (IS_ERR(frame->bh))
765 return (struct dx_frame *) frame->bh;
767 root = (struct dx_root *) frame->bh->b_data;
768 if (root->info.hash_version != DX_HASH_TEA &&
769 root->info.hash_version != DX_HASH_HALF_MD4 &&
770 root->info.hash_version != DX_HASH_LEGACY) {
771 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
772 root->info.hash_version);
776 hinfo = &fname->hinfo;
777 hinfo->hash_version = root->info.hash_version;
778 if (hinfo->hash_version <= DX_HASH_TEA)
779 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
780 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
781 if (fname && fname_name(fname))
782 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
785 if (root->info.unused_flags & 1) {
786 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
787 root->info.unused_flags);
791 indirect = root->info.indirect_levels;
792 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
793 ext4_warning(dir->i_sb,
794 "Directory (ino: %lu) htree depth %#06x exceed"
795 "supported value", dir->i_ino,
796 ext4_dir_htree_level(dir->i_sb));
797 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
798 ext4_warning(dir->i_sb, "Enable large directory "
799 "feature to access it");
804 entries = (struct dx_entry *)(((char *)&root->info) +
805 root->info.info_length);
807 if (dx_get_limit(entries) != dx_root_limit(dir,
808 root->info.info_length)) {
809 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
810 dx_get_limit(entries),
811 dx_root_limit(dir, root->info.info_length));
815 dxtrace(printk("Look up %x", hash));
817 count = dx_get_count(entries);
818 if (!count || count > dx_get_limit(entries)) {
819 ext4_warning_inode(dir,
820 "dx entry: count %u beyond limit %u",
821 count, dx_get_limit(entries));
826 q = entries + count - 1;
829 dxtrace(printk(KERN_CONT "."));
830 if (dx_get_hash(m) > hash)
836 if (0) { // linear search cross check
837 unsigned n = count - 1;
841 dxtrace(printk(KERN_CONT ","));
842 if (dx_get_hash(++at) > hash)
852 dxtrace(printk(KERN_CONT " %x->%u\n",
853 at == entries ? 0 : dx_get_hash(at),
855 frame->entries = entries;
860 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
861 if (IS_ERR(frame->bh)) {
862 ret_err = (struct dx_frame *) frame->bh;
866 entries = ((struct dx_node *) frame->bh->b_data)->entries;
868 if (dx_get_limit(entries) != dx_node_limit(dir)) {
869 ext4_warning_inode(dir,
870 "dx entry: limit %u != node limit %u",
871 dx_get_limit(entries), dx_node_limit(dir));
876 while (frame >= frame_in) {
881 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
882 ext4_warning_inode(dir,
883 "Corrupt directory, running e2fsck is recommended");
887 static void dx_release(struct dx_frame *frames)
889 struct dx_root_info *info;
891 unsigned int indirect_levels;
893 if (frames[0].bh == NULL)
896 info = &((struct dx_root *)frames[0].bh->b_data)->info;
897 /* save local copy, "info" may be freed after brelse() */
898 indirect_levels = info->indirect_levels;
899 for (i = 0; i <= indirect_levels; i++) {
900 if (frames[i].bh == NULL)
902 brelse(frames[i].bh);
908 * This function increments the frame pointer to search the next leaf
909 * block, and reads in the necessary intervening nodes if the search
910 * should be necessary. Whether or not the search is necessary is
911 * controlled by the hash parameter. If the hash value is even, then
912 * the search is only continued if the next block starts with that
913 * hash value. This is used if we are searching for a specific file.
915 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
917 * This function returns 1 if the caller should continue to search,
918 * or 0 if it should not. If there is an error reading one of the
919 * index blocks, it will a negative error code.
921 * If start_hash is non-null, it will be filled in with the starting
922 * hash of the next page.
924 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
925 struct dx_frame *frame,
926 struct dx_frame *frames,
930 struct buffer_head *bh;
936 * Find the next leaf page by incrementing the frame pointer.
937 * If we run out of entries in the interior node, loop around and
938 * increment pointer in the parent node. When we break out of
939 * this loop, num_frames indicates the number of interior
940 * nodes need to be read.
943 if (++(p->at) < p->entries + dx_get_count(p->entries))
952 * If the hash is 1, then continue only if the next page has a
953 * continuation hash of any value. This is used for readdir
954 * handling. Otherwise, check to see if the hash matches the
955 * desired contiuation hash. If it doesn't, return since
956 * there's no point to read in the successive index pages.
958 bhash = dx_get_hash(p->at);
961 if ((hash & 1) == 0) {
962 if ((bhash & ~1) != hash)
966 * If the hash is HASH_NB_ALWAYS, we always go to the next
967 * block so no check is necessary
969 while (num_frames--) {
970 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
976 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
983 * This function fills a red-black tree with information from a
984 * directory block. It returns the number directory entries loaded
985 * into the tree. If there is an error it is returned in err.
987 static int htree_dirblock_to_tree(struct file *dir_file,
988 struct inode *dir, ext4_lblk_t block,
989 struct dx_hash_info *hinfo,
990 __u32 start_hash, __u32 start_minor_hash)
992 struct buffer_head *bh;
993 struct ext4_dir_entry_2 *de, *top;
994 int err = 0, count = 0;
995 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
997 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
998 (unsigned long)block));
999 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1003 de = (struct ext4_dir_entry_2 *) bh->b_data;
1004 top = (struct ext4_dir_entry_2 *) ((char *) de +
1005 dir->i_sb->s_blocksize -
1006 EXT4_DIR_REC_LEN(0));
1007 /* Check if the directory is encrypted */
1008 if (IS_ENCRYPTED(dir)) {
1009 err = fscrypt_get_encryption_info(dir);
1014 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1022 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1023 if (ext4_check_dir_entry(dir, NULL, de, bh,
1024 bh->b_data, bh->b_size,
1025 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1026 + ((char *)de - bh->b_data))) {
1027 /* silently ignore the rest of the block */
1030 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1031 if ((hinfo->hash < start_hash) ||
1032 ((hinfo->hash == start_hash) &&
1033 (hinfo->minor_hash < start_minor_hash)))
1037 if (!IS_ENCRYPTED(dir)) {
1038 tmp_str.name = de->name;
1039 tmp_str.len = de->name_len;
1040 err = ext4_htree_store_dirent(dir_file,
1041 hinfo->hash, hinfo->minor_hash, de,
1044 int save_len = fname_crypto_str.len;
1045 struct fscrypt_str de_name = FSTR_INIT(de->name,
1048 /* Directory is encrypted */
1049 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1050 hinfo->minor_hash, &de_name,
1056 err = ext4_htree_store_dirent(dir_file,
1057 hinfo->hash, hinfo->minor_hash, de,
1059 fname_crypto_str.len = save_len;
1069 fscrypt_fname_free_buffer(&fname_crypto_str);
1075 * This function fills a red-black tree with information from a
1076 * directory. We start scanning the directory in hash order, starting
1077 * at start_hash and start_minor_hash.
1079 * This function returns the number of entries inserted into the tree,
1080 * or a negative error code.
1082 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1083 __u32 start_minor_hash, __u32 *next_hash)
1085 struct dx_hash_info hinfo;
1086 struct ext4_dir_entry_2 *de;
1087 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1093 struct fscrypt_str tmp_str;
1095 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1096 start_hash, start_minor_hash));
1097 dir = file_inode(dir_file);
1098 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1099 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1100 if (hinfo.hash_version <= DX_HASH_TEA)
1101 hinfo.hash_version +=
1102 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1103 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1104 if (ext4_has_inline_data(dir)) {
1105 int has_inline_data = 1;
1106 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1110 if (has_inline_data) {
1115 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1116 start_hash, start_minor_hash);
1120 hinfo.hash = start_hash;
1121 hinfo.minor_hash = 0;
1122 frame = dx_probe(NULL, dir, &hinfo, frames);
1124 return PTR_ERR(frame);
1126 /* Add '.' and '..' from the htree header */
1127 if (!start_hash && !start_minor_hash) {
1128 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1129 tmp_str.name = de->name;
1130 tmp_str.len = de->name_len;
1131 err = ext4_htree_store_dirent(dir_file, 0, 0,
1137 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1138 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1139 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1140 tmp_str.name = de->name;
1141 tmp_str.len = de->name_len;
1142 err = ext4_htree_store_dirent(dir_file, 2, 0,
1150 if (fatal_signal_pending(current)) {
1155 block = dx_get_block(frame->at);
1156 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1157 start_hash, start_minor_hash);
1164 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1165 frame, frames, &hashval);
1166 *next_hash = hashval;
1172 * Stop if: (a) there are no more entries, or
1173 * (b) we have inserted at least one entry and the
1174 * next hash value is not a continuation
1177 (count && ((hashval & 1) == 0)))
1181 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1182 "next hash: %x\n", count, *next_hash));
1189 static inline int search_dirblock(struct buffer_head *bh,
1191 struct ext4_filename *fname,
1192 unsigned int offset,
1193 struct ext4_dir_entry_2 **res_dir)
1195 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1196 fname, offset, res_dir);
1200 * Directory block splitting, compacting
1204 * Create map of hash values, offsets, and sizes, stored at end of block.
1205 * Returns number of entries mapped.
1207 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1208 unsigned blocksize, struct dx_hash_info *hinfo,
1209 struct dx_map_entry *map_tail)
1212 char *base = (char *) de;
1213 struct dx_hash_info h = *hinfo;
1215 while ((char *) de < base + blocksize) {
1216 if (de->name_len && de->inode) {
1217 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1219 map_tail->hash = h.hash;
1220 map_tail->offs = ((char *) de - base)>>2;
1221 map_tail->size = le16_to_cpu(de->rec_len);
1225 /* XXX: do we need to check rec_len == 0 case? -Chris */
1226 de = ext4_next_entry(de, blocksize);
1231 /* Sort map by hash value */
1232 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1234 struct dx_map_entry *p, *q, *top = map + count - 1;
1236 /* Combsort until bubble sort doesn't suck */
1238 count = count*10/13;
1239 if (count - 9 < 2) /* 9, 10 -> 11 */
1241 for (p = top, q = p - count; q >= map; p--, q--)
1242 if (p->hash < q->hash)
1245 /* Garden variety bubble sort */
1250 if (q[1].hash >= q[0].hash)
1258 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1260 struct dx_entry *entries = frame->entries;
1261 struct dx_entry *old = frame->at, *new = old + 1;
1262 int count = dx_get_count(entries);
1264 ASSERT(count < dx_get_limit(entries));
1265 ASSERT(old < entries + count);
1266 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1267 dx_set_hash(new, hash);
1268 dx_set_block(new, block);
1269 dx_set_count(entries, count + 1);
1272 #ifdef CONFIG_UNICODE
1274 * Test whether a case-insensitive directory entry matches the filename
1275 * being searched for. If quick is set, assume the name being looked up
1276 * is already in the casefolded form.
1278 * Returns: 0 if the directory entry matches, more than 0 if it
1279 * doesn't match or less than zero on error.
1281 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1282 const struct qstr *entry, bool quick)
1284 const struct super_block *sb = parent->i_sb;
1285 const struct unicode_map *um = sb->s_encoding;
1289 ret = utf8_strncasecmp_folded(um, name, entry);
1291 ret = utf8_strncasecmp(um, name, entry);
1294 /* Handle invalid character sequence as either an error
1295 * or as an opaque byte sequence.
1297 if (sb_has_strict_encoding(sb))
1300 if (name->len != entry->len)
1303 return !!memcmp(name->name, entry->name, name->len);
1309 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1310 struct fscrypt_str *cf_name)
1314 if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) {
1315 cf_name->name = NULL;
1319 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1323 len = utf8_casefold(dir->i_sb->s_encoding,
1324 iname, cf_name->name,
1327 kfree(cf_name->name);
1328 cf_name->name = NULL;
1331 cf_name->len = (unsigned) len;
1337 * Test whether a directory entry matches the filename being searched for.
1339 * Return: %true if the directory entry matches, otherwise %false.
1341 static inline bool ext4_match(const struct inode *parent,
1342 const struct ext4_filename *fname,
1343 const struct ext4_dir_entry_2 *de)
1345 struct fscrypt_name f;
1346 #ifdef CONFIG_UNICODE
1347 const struct qstr entry = {.name = de->name, .len = de->name_len};
1353 f.usr_fname = fname->usr_fname;
1354 f.disk_name = fname->disk_name;
1355 #ifdef CONFIG_FS_ENCRYPTION
1356 f.crypto_buf = fname->crypto_buf;
1359 #ifdef CONFIG_UNICODE
1360 if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) {
1361 if (fname->cf_name.name) {
1362 struct qstr cf = {.name = fname->cf_name.name,
1363 .len = fname->cf_name.len};
1364 return !ext4_ci_compare(parent, &cf, &entry, true);
1366 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1371 return fscrypt_match_name(&f, de->name, de->name_len);
1375 * Returns 0 if not found, -1 on failure, and 1 on success
1377 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1378 struct inode *dir, struct ext4_filename *fname,
1379 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1381 struct ext4_dir_entry_2 * de;
1385 de = (struct ext4_dir_entry_2 *)search_buf;
1386 dlimit = search_buf + buf_size;
1387 while ((char *) de < dlimit) {
1388 /* this code is executed quadratically often */
1389 /* do minimal checking `by hand' */
1390 if ((char *) de + de->name_len <= dlimit &&
1391 ext4_match(dir, fname, de)) {
1392 /* found a match - just to be sure, do
1394 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1400 /* prevent looping on a bad block */
1401 de_len = ext4_rec_len_from_disk(de->rec_len,
1402 dir->i_sb->s_blocksize);
1406 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1411 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1412 struct ext4_dir_entry *de)
1414 struct super_block *sb = dir->i_sb;
1420 if (de->inode == 0 &&
1421 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1428 * __ext4_find_entry()
1430 * finds an entry in the specified directory with the wanted name. It
1431 * returns the cache buffer in which the entry was found, and the entry
1432 * itself (as a parameter - res_dir). It does NOT read the inode of the
1433 * entry - you'll have to do that yourself if you want to.
1435 * The returned buffer_head has ->b_count elevated. The caller is expected
1436 * to brelse() it when appropriate.
1438 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1439 struct ext4_filename *fname,
1440 struct ext4_dir_entry_2 **res_dir,
1443 struct super_block *sb;
1444 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1445 struct buffer_head *bh, *ret = NULL;
1446 ext4_lblk_t start, block;
1447 const u8 *name = fname->usr_fname->name;
1448 size_t ra_max = 0; /* Number of bh's in the readahead
1450 size_t ra_ptr = 0; /* Current index into readahead
1452 ext4_lblk_t nblocks;
1453 int i, namelen, retval;
1457 namelen = fname->usr_fname->len;
1458 if (namelen > EXT4_NAME_LEN)
1461 if (ext4_has_inline_data(dir)) {
1462 int has_inline_data = 1;
1463 ret = ext4_find_inline_entry(dir, fname, res_dir,
1465 if (has_inline_data) {
1468 goto cleanup_and_exit;
1472 if ((namelen <= 2) && (name[0] == '.') &&
1473 (name[1] == '.' || name[1] == '\0')) {
1475 * "." or ".." will only be in the first block
1476 * NFS may look up ".."; "." should be handled by the VFS
1483 ret = ext4_dx_find_entry(dir, fname, res_dir);
1485 * On success, or if the error was file not found,
1486 * return. Otherwise, fall back to doing a search the
1487 * old fashioned way.
1489 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1490 goto cleanup_and_exit;
1491 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1495 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1498 goto cleanup_and_exit;
1500 start = EXT4_I(dir)->i_dir_start_lookup;
1501 if (start >= nblocks)
1507 * We deal with the read-ahead logic here.
1510 if (ra_ptr >= ra_max) {
1511 /* Refill the readahead buffer */
1514 ra_max = start - block;
1516 ra_max = nblocks - block;
1517 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1518 retval = ext4_bread_batch(dir, block, ra_max,
1519 false /* wait */, bh_use);
1521 ret = ERR_PTR(retval);
1523 goto cleanup_and_exit;
1526 if ((bh = bh_use[ra_ptr++]) == NULL)
1529 if (!buffer_uptodate(bh)) {
1530 EXT4_ERROR_INODE_ERR(dir, EIO,
1531 "reading directory lblock %lu",
1532 (unsigned long) block);
1534 ret = ERR_PTR(-EIO);
1535 goto cleanup_and_exit;
1537 if (!buffer_verified(bh) &&
1538 !is_dx_internal_node(dir, block,
1539 (struct ext4_dir_entry *)bh->b_data) &&
1540 !ext4_dirblock_csum_verify(dir, bh)) {
1541 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1542 "checksumming directory "
1543 "block %lu", (unsigned long)block);
1545 ret = ERR_PTR(-EFSBADCRC);
1546 goto cleanup_and_exit;
1548 set_buffer_verified(bh);
1549 i = search_dirblock(bh, dir, fname,
1550 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1552 EXT4_I(dir)->i_dir_start_lookup = block;
1554 goto cleanup_and_exit;
1558 goto cleanup_and_exit;
1561 if (++block >= nblocks)
1563 } while (block != start);
1566 * If the directory has grown while we were searching, then
1567 * search the last part of the directory before giving up.
1570 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1571 if (block < nblocks) {
1577 /* Clean up the read-ahead blocks */
1578 for (; ra_ptr < ra_max; ra_ptr++)
1579 brelse(bh_use[ra_ptr]);
1583 static struct buffer_head *ext4_find_entry(struct inode *dir,
1584 const struct qstr *d_name,
1585 struct ext4_dir_entry_2 **res_dir,
1589 struct ext4_filename fname;
1590 struct buffer_head *bh;
1592 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1596 return ERR_PTR(err);
1598 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1600 ext4_fname_free_filename(&fname);
1604 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1605 struct dentry *dentry,
1606 struct ext4_dir_entry_2 **res_dir)
1609 struct ext4_filename fname;
1610 struct buffer_head *bh;
1612 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1616 return ERR_PTR(err);
1618 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1620 ext4_fname_free_filename(&fname);
1624 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1625 struct ext4_filename *fname,
1626 struct ext4_dir_entry_2 **res_dir)
1628 struct super_block * sb = dir->i_sb;
1629 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1630 struct buffer_head *bh;
1634 #ifdef CONFIG_FS_ENCRYPTION
1637 frame = dx_probe(fname, dir, NULL, frames);
1639 return (struct buffer_head *) frame;
1641 block = dx_get_block(frame->at);
1642 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1646 retval = search_dirblock(bh, dir, fname,
1647 block << EXT4_BLOCK_SIZE_BITS(sb),
1653 bh = ERR_PTR(ERR_BAD_DX_DIR);
1657 /* Check to see if we should continue to search */
1658 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1661 ext4_warning_inode(dir,
1662 "error %d reading directory index block",
1664 bh = ERR_PTR(retval);
1667 } while (retval == 1);
1671 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1677 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1679 struct inode *inode;
1680 struct ext4_dir_entry_2 *de;
1681 struct buffer_head *bh;
1683 if (dentry->d_name.len > EXT4_NAME_LEN)
1684 return ERR_PTR(-ENAMETOOLONG);
1686 bh = ext4_lookup_entry(dir, dentry, &de);
1688 return ERR_CAST(bh);
1691 __u32 ino = le32_to_cpu(de->inode);
1693 if (!ext4_valid_inum(dir->i_sb, ino)) {
1694 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1695 return ERR_PTR(-EFSCORRUPTED);
1697 if (unlikely(ino == dir->i_ino)) {
1698 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1700 return ERR_PTR(-EFSCORRUPTED);
1702 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1703 if (inode == ERR_PTR(-ESTALE)) {
1704 EXT4_ERROR_INODE(dir,
1705 "deleted inode referenced: %u",
1707 return ERR_PTR(-EFSCORRUPTED);
1709 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1710 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1711 !fscrypt_has_permitted_context(dir, inode)) {
1712 ext4_warning(inode->i_sb,
1713 "Inconsistent encryption contexts: %lu/%lu",
1714 dir->i_ino, inode->i_ino);
1716 return ERR_PTR(-EPERM);
1720 #ifdef CONFIG_UNICODE
1721 if (!inode && IS_CASEFOLDED(dir)) {
1722 /* Eventually we want to call d_add_ci(dentry, NULL)
1723 * for negative dentries in the encoding case as
1724 * well. For now, prevent the negative dentry
1725 * from being cached.
1730 return d_splice_alias(inode, dentry);
1734 struct dentry *ext4_get_parent(struct dentry *child)
1737 static const struct qstr dotdot = QSTR_INIT("..", 2);
1738 struct ext4_dir_entry_2 * de;
1739 struct buffer_head *bh;
1741 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1743 return ERR_CAST(bh);
1745 return ERR_PTR(-ENOENT);
1746 ino = le32_to_cpu(de->inode);
1749 if (!ext4_valid_inum(child->d_sb, ino)) {
1750 EXT4_ERROR_INODE(d_inode(child),
1751 "bad parent inode number: %u", ino);
1752 return ERR_PTR(-EFSCORRUPTED);
1755 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1759 * Move count entries from end of map between two memory locations.
1760 * Returns pointer to last entry moved.
1762 static struct ext4_dir_entry_2 *
1763 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1766 unsigned rec_len = 0;
1769 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1770 (from + (map->offs<<2));
1771 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1772 memcpy (to, de, rec_len);
1773 ((struct ext4_dir_entry_2 *) to)->rec_len =
1774 ext4_rec_len_to_disk(rec_len, blocksize);
1779 return (struct ext4_dir_entry_2 *) (to - rec_len);
1783 * Compact each dir entry in the range to the minimal rec_len.
1784 * Returns pointer to last entry in range.
1786 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1788 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1789 unsigned rec_len = 0;
1792 while ((char*)de < base + blocksize) {
1793 next = ext4_next_entry(de, blocksize);
1794 if (de->inode && de->name_len) {
1795 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1797 memmove(to, de, rec_len);
1798 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1800 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1808 * Split a full leaf block to make room for a new dir entry.
1809 * Allocate a new block, and move entries so that they are approx. equally full.
1810 * Returns pointer to de in block into which the new entry will be inserted.
1812 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1813 struct buffer_head **bh,struct dx_frame *frame,
1814 struct dx_hash_info *hinfo)
1816 unsigned blocksize = dir->i_sb->s_blocksize;
1817 unsigned count, continued;
1818 struct buffer_head *bh2;
1819 ext4_lblk_t newblock;
1821 struct dx_map_entry *map;
1822 char *data1 = (*bh)->b_data, *data2;
1823 unsigned split, move, size;
1824 struct ext4_dir_entry_2 *de = NULL, *de2;
1828 if (ext4_has_metadata_csum(dir->i_sb))
1829 csum_size = sizeof(struct ext4_dir_entry_tail);
1831 bh2 = ext4_append(handle, dir, &newblock);
1835 return (struct ext4_dir_entry_2 *) bh2;
1838 BUFFER_TRACE(*bh, "get_write_access");
1839 err = ext4_journal_get_write_access(handle, *bh);
1843 BUFFER_TRACE(frame->bh, "get_write_access");
1844 err = ext4_journal_get_write_access(handle, frame->bh);
1848 data2 = bh2->b_data;
1850 /* create map in the end of data2 block */
1851 map = (struct dx_map_entry *) (data2 + blocksize);
1852 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1853 blocksize, hinfo, map);
1855 dx_sort_map(map, count);
1856 /* Ensure that neither split block is over half full */
1859 for (i = count-1; i >= 0; i--) {
1860 /* is more than half of this entry in 2nd half of the block? */
1861 if (size + map[i].size/2 > blocksize/2)
1863 size += map[i].size;
1867 * map index at which we will split
1869 * If the sum of active entries didn't exceed half the block size, just
1870 * split it in half by count; each resulting block will have at least
1871 * half the space free.
1874 split = count - move;
1878 hash2 = map[split].hash;
1879 continued = hash2 == map[split - 1].hash;
1880 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1881 (unsigned long)dx_get_block(frame->at),
1882 hash2, split, count-split));
1884 /* Fancy dance to stay within two buffers */
1885 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1887 de = dx_pack_dirents(data1, blocksize);
1888 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1891 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1895 ext4_initialize_dirent_tail(*bh, blocksize);
1896 ext4_initialize_dirent_tail(bh2, blocksize);
1899 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1901 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1904 /* Which block gets the new entry? */
1905 if (hinfo->hash >= hash2) {
1909 dx_insert_block(frame, hash2 + continued, newblock);
1910 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1913 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1917 dxtrace(dx_show_index("frame", frame->entries));
1924 ext4_std_error(dir->i_sb, err);
1925 return ERR_PTR(err);
1928 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1929 struct buffer_head *bh,
1930 void *buf, int buf_size,
1931 struct ext4_filename *fname,
1932 struct ext4_dir_entry_2 **dest_de)
1934 struct ext4_dir_entry_2 *de;
1935 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1937 unsigned int offset = 0;
1940 de = (struct ext4_dir_entry_2 *)buf;
1941 top = buf + buf_size - reclen;
1942 while ((char *) de <= top) {
1943 if (ext4_check_dir_entry(dir, NULL, de, bh,
1944 buf, buf_size, offset))
1945 return -EFSCORRUPTED;
1946 if (ext4_match(dir, fname, de))
1948 nlen = EXT4_DIR_REC_LEN(de->name_len);
1949 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1950 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1952 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1955 if ((char *) de > top)
1962 void ext4_insert_dentry(struct inode *inode,
1963 struct ext4_dir_entry_2 *de,
1965 struct ext4_filename *fname)
1970 nlen = EXT4_DIR_REC_LEN(de->name_len);
1971 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1973 struct ext4_dir_entry_2 *de1 =
1974 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1975 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1976 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1979 de->file_type = EXT4_FT_UNKNOWN;
1980 de->inode = cpu_to_le32(inode->i_ino);
1981 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1982 de->name_len = fname_len(fname);
1983 memcpy(de->name, fname_name(fname), fname_len(fname));
1987 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1988 * it points to a directory entry which is guaranteed to be large
1989 * enough for new directory entry. If de is NULL, then
1990 * add_dirent_to_buf will attempt search the directory block for
1991 * space. It will return -ENOSPC if no space is available, and -EIO
1992 * and -EEXIST if directory entry already exists.
1994 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1996 struct inode *inode, struct ext4_dir_entry_2 *de,
1997 struct buffer_head *bh)
1999 unsigned int blocksize = dir->i_sb->s_blocksize;
2003 if (ext4_has_metadata_csum(inode->i_sb))
2004 csum_size = sizeof(struct ext4_dir_entry_tail);
2007 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2008 blocksize - csum_size, fname, &de);
2012 BUFFER_TRACE(bh, "get_write_access");
2013 err = ext4_journal_get_write_access(handle, bh);
2015 ext4_std_error(dir->i_sb, err);
2019 /* By now the buffer is marked for journaling */
2020 ext4_insert_dentry(inode, de, blocksize, fname);
2023 * XXX shouldn't update any times until successful
2024 * completion of syscall, but too many callers depend
2027 * XXX similarly, too many callers depend on
2028 * ext4_new_inode() setting the times, but error
2029 * recovery deletes the inode, so the worst that can
2030 * happen is that the times are slightly out of date
2031 * and/or different from the directory change time.
2033 dir->i_mtime = dir->i_ctime = current_time(dir);
2034 ext4_update_dx_flag(dir);
2035 inode_inc_iversion(dir);
2036 err2 = ext4_mark_inode_dirty(handle, dir);
2037 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2038 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2040 ext4_std_error(dir->i_sb, err);
2041 return err ? err : err2;
2045 * This converts a one block unindexed directory to a 3 block indexed
2046 * directory, and adds the dentry to the indexed directory.
2048 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2050 struct inode *inode, struct buffer_head *bh)
2052 struct buffer_head *bh2;
2053 struct dx_root *root;
2054 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2055 struct dx_entry *entries;
2056 struct ext4_dir_entry_2 *de, *de2;
2062 struct fake_dirent *fde;
2065 if (ext4_has_metadata_csum(inode->i_sb))
2066 csum_size = sizeof(struct ext4_dir_entry_tail);
2068 blocksize = dir->i_sb->s_blocksize;
2069 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2070 BUFFER_TRACE(bh, "get_write_access");
2071 retval = ext4_journal_get_write_access(handle, bh);
2073 ext4_std_error(dir->i_sb, retval);
2077 root = (struct dx_root *) bh->b_data;
2079 /* The 0th block becomes the root, move the dirents out */
2080 fde = &root->dotdot;
2081 de = (struct ext4_dir_entry_2 *)((char *)fde +
2082 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2083 if ((char *) de >= (((char *) root) + blocksize)) {
2084 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2086 return -EFSCORRUPTED;
2088 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2090 /* Allocate new block for the 0th block's dirents */
2091 bh2 = ext4_append(handle, dir, &block);
2094 return PTR_ERR(bh2);
2096 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2097 data2 = bh2->b_data;
2099 memcpy(data2, de, len);
2100 de = (struct ext4_dir_entry_2 *) data2;
2102 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2104 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2105 (char *) de, blocksize);
2108 ext4_initialize_dirent_tail(bh2, blocksize);
2110 /* Initialize the root; the dot dirents already exist */
2111 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2112 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2114 memset (&root->info, 0, sizeof(root->info));
2115 root->info.info_length = sizeof(root->info);
2116 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2117 entries = root->entries;
2118 dx_set_block(entries, 1);
2119 dx_set_count(entries, 1);
2120 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2122 /* Initialize as for dx_probe */
2123 fname->hinfo.hash_version = root->info.hash_version;
2124 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2125 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2126 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2127 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2129 memset(frames, 0, sizeof(frames));
2131 frame->entries = entries;
2132 frame->at = entries;
2135 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2138 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2142 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2144 retval = PTR_ERR(de);
2148 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2151 * Even if the block split failed, we have to properly write
2152 * out all the changes we did so far. Otherwise we can end up
2153 * with corrupted filesystem.
2156 ext4_mark_inode_dirty(handle, dir);
2165 * adds a file entry to the specified directory, using the same
2166 * semantics as ext4_find_entry(). It returns NULL if it failed.
2168 * NOTE!! The inode part of 'de' is left at 0 - which means you
2169 * may not sleep between calling this and putting something into
2170 * the entry, as someone else might have used it while you slept.
2172 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2173 struct inode *inode)
2175 struct inode *dir = d_inode(dentry->d_parent);
2176 struct buffer_head *bh = NULL;
2177 struct ext4_dir_entry_2 *de;
2178 struct super_block *sb;
2179 struct ext4_filename fname;
2183 ext4_lblk_t block, blocks;
2186 if (ext4_has_metadata_csum(inode->i_sb))
2187 csum_size = sizeof(struct ext4_dir_entry_tail);
2190 blocksize = sb->s_blocksize;
2191 if (!dentry->d_name.len)
2194 #ifdef CONFIG_UNICODE
2195 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2196 sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2200 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2204 if (ext4_has_inline_data(dir)) {
2205 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2215 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2216 if (!retval || (retval != ERR_BAD_DX_DIR))
2218 /* Can we just ignore htree data? */
2219 if (ext4_has_metadata_csum(sb)) {
2220 EXT4_ERROR_INODE(dir,
2221 "Directory has corrupted htree index.");
2222 retval = -EFSCORRUPTED;
2225 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2227 retval = ext4_mark_inode_dirty(handle, dir);
2228 if (unlikely(retval))
2231 blocks = dir->i_size >> sb->s_blocksize_bits;
2232 for (block = 0; block < blocks; block++) {
2233 bh = ext4_read_dirblock(dir, block, DIRENT);
2235 bh = ext4_bread(handle, dir, block,
2236 EXT4_GET_BLOCKS_CREATE);
2237 goto add_to_new_block;
2240 retval = PTR_ERR(bh);
2244 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2246 if (retval != -ENOSPC)
2249 if (blocks == 1 && !dx_fallback &&
2250 ext4_has_feature_dir_index(sb)) {
2251 retval = make_indexed_dir(handle, &fname, dir,
2253 bh = NULL; /* make_indexed_dir releases bh */
2258 bh = ext4_append(handle, dir, &block);
2261 retval = PTR_ERR(bh);
2265 de = (struct ext4_dir_entry_2 *) bh->b_data;
2267 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2270 ext4_initialize_dirent_tail(bh, blocksize);
2272 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2274 ext4_fname_free_filename(&fname);
2277 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2282 * Returns 0 for success, or a negative error value
2284 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2285 struct inode *dir, struct inode *inode)
2287 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2288 struct dx_entry *entries, *at;
2289 struct buffer_head *bh;
2290 struct super_block *sb = dir->i_sb;
2291 struct ext4_dir_entry_2 *de;
2297 frame = dx_probe(fname, dir, NULL, frames);
2299 return PTR_ERR(frame);
2300 entries = frame->entries;
2302 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2309 BUFFER_TRACE(bh, "get_write_access");
2310 err = ext4_journal_get_write_access(handle, bh);
2314 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2319 /* Block full, should compress but for now just split */
2320 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2321 dx_get_count(entries), dx_get_limit(entries)));
2322 /* Need to split index? */
2323 if (dx_get_count(entries) == dx_get_limit(entries)) {
2324 ext4_lblk_t newblock;
2325 int levels = frame - frames + 1;
2326 unsigned int icount;
2328 struct dx_entry *entries2;
2329 struct dx_node *node2;
2330 struct buffer_head *bh2;
2332 while (frame > frames) {
2333 if (dx_get_count((frame - 1)->entries) <
2334 dx_get_limit((frame - 1)->entries)) {
2338 frame--; /* split higher index block */
2340 entries = frame->entries;
2343 if (add_level && levels == ext4_dir_htree_level(sb)) {
2344 ext4_warning(sb, "Directory (ino: %lu) index full, "
2345 "reach max htree level :%d",
2346 dir->i_ino, levels);
2347 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2348 ext4_warning(sb, "Large directory feature is "
2349 "not enabled on this "
2355 icount = dx_get_count(entries);
2356 bh2 = ext4_append(handle, dir, &newblock);
2361 node2 = (struct dx_node *)(bh2->b_data);
2362 entries2 = node2->entries;
2363 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2364 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2366 BUFFER_TRACE(frame->bh, "get_write_access");
2367 err = ext4_journal_get_write_access(handle, frame->bh);
2371 unsigned icount1 = icount/2, icount2 = icount - icount1;
2372 unsigned hash2 = dx_get_hash(entries + icount1);
2373 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2376 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2377 err = ext4_journal_get_write_access(handle,
2382 memcpy((char *) entries2, (char *) (entries + icount1),
2383 icount2 * sizeof(struct dx_entry));
2384 dx_set_count(entries, icount1);
2385 dx_set_count(entries2, icount2);
2386 dx_set_limit(entries2, dx_node_limit(dir));
2388 /* Which index block gets the new entry? */
2389 if (at - entries >= icount1) {
2390 frame->at = at = at - entries - icount1 + entries2;
2391 frame->entries = entries = entries2;
2392 swap(frame->bh, bh2);
2394 dx_insert_block((frame - 1), hash2, newblock);
2395 dxtrace(dx_show_index("node", frame->entries));
2396 dxtrace(dx_show_index("node",
2397 ((struct dx_node *) bh2->b_data)->entries));
2398 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2402 err = ext4_handle_dirty_dx_node(handle, dir,
2407 err = ext4_handle_dirty_dx_node(handle, dir,
2412 struct dx_root *dxroot;
2413 memcpy((char *) entries2, (char *) entries,
2414 icount * sizeof(struct dx_entry));
2415 dx_set_limit(entries2, dx_node_limit(dir));
2418 dx_set_count(entries, 1);
2419 dx_set_block(entries + 0, newblock);
2420 dxroot = (struct dx_root *)frames[0].bh->b_data;
2421 dxroot->info.indirect_levels += 1;
2422 dxtrace(printk(KERN_DEBUG
2423 "Creating %d level index...\n",
2424 dxroot->info.indirect_levels));
2425 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2428 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2434 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2439 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2443 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2447 /* @restart is true means htree-path has been changed, we need to
2448 * repeat dx_probe() to find out valid htree-path
2450 if (restart && err == 0)
2456 * ext4_generic_delete_entry deletes a directory entry by merging it
2457 * with the previous entry
2459 int ext4_generic_delete_entry(struct inode *dir,
2460 struct ext4_dir_entry_2 *de_del,
2461 struct buffer_head *bh,
2466 struct ext4_dir_entry_2 *de, *pde;
2467 unsigned int blocksize = dir->i_sb->s_blocksize;
2472 de = (struct ext4_dir_entry_2 *)entry_buf;
2473 while (i < buf_size - csum_size) {
2474 if (ext4_check_dir_entry(dir, NULL, de, bh,
2475 entry_buf, buf_size, i))
2476 return -EFSCORRUPTED;
2479 pde->rec_len = ext4_rec_len_to_disk(
2480 ext4_rec_len_from_disk(pde->rec_len,
2482 ext4_rec_len_from_disk(de->rec_len,
2487 inode_inc_iversion(dir);
2490 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2492 de = ext4_next_entry(de, blocksize);
2497 static int ext4_delete_entry(handle_t *handle,
2499 struct ext4_dir_entry_2 *de_del,
2500 struct buffer_head *bh)
2502 int err, csum_size = 0;
2504 if (ext4_has_inline_data(dir)) {
2505 int has_inline_data = 1;
2506 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2508 if (has_inline_data)
2512 if (ext4_has_metadata_csum(dir->i_sb))
2513 csum_size = sizeof(struct ext4_dir_entry_tail);
2515 BUFFER_TRACE(bh, "get_write_access");
2516 err = ext4_journal_get_write_access(handle, bh);
2520 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2521 dir->i_sb->s_blocksize, csum_size);
2525 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2526 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2533 ext4_std_error(dir->i_sb, err);
2538 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2539 * since this indicates that nlinks count was previously 1 to avoid overflowing
2540 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2541 * that subdirectory link counts are not being maintained accurately.
2543 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2544 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2545 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2546 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2548 static void ext4_inc_count(struct inode *inode)
2552 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2553 set_nlink(inode, 1);
2557 * If a directory had nlink == 1, then we should let it be 1. This indicates
2558 * directory has >EXT4_LINK_MAX subdirs.
2560 static void ext4_dec_count(struct inode *inode)
2562 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2568 * Add non-directory inode to a directory. On success, the inode reference is
2569 * consumed by dentry is instantiation. This is also indicated by clearing of
2570 * *inodep pointer. On failure, the caller is responsible for dropping the
2571 * inode reference in the safe context.
2573 static int ext4_add_nondir(handle_t *handle,
2574 struct dentry *dentry, struct inode **inodep)
2576 struct inode *dir = d_inode(dentry->d_parent);
2577 struct inode *inode = *inodep;
2578 int err = ext4_add_entry(handle, dentry, inode);
2580 err = ext4_mark_inode_dirty(handle, inode);
2581 if (IS_DIRSYNC(dir))
2582 ext4_handle_sync(handle);
2583 d_instantiate_new(dentry, inode);
2588 ext4_orphan_add(handle, inode);
2589 unlock_new_inode(inode);
2594 * By the time this is called, we already have created
2595 * the directory cache entry for the new file, but it
2596 * is so far negative - it has no inode.
2598 * If the create succeeds, we fill in the inode information
2599 * with d_instantiate().
2601 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2605 struct inode *inode;
2606 int err, credits, retries = 0;
2608 err = dquot_initialize(dir);
2612 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2613 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2615 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2616 NULL, EXT4_HT_DIR, credits);
2617 handle = ext4_journal_current_handle();
2618 err = PTR_ERR(inode);
2619 if (!IS_ERR(inode)) {
2620 inode->i_op = &ext4_file_inode_operations;
2621 inode->i_fop = &ext4_file_operations;
2622 ext4_set_aops(inode);
2623 err = ext4_add_nondir(handle, dentry, &inode);
2625 ext4_fc_track_create(handle, dentry);
2628 ext4_journal_stop(handle);
2629 if (!IS_ERR_OR_NULL(inode))
2631 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2636 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2637 umode_t mode, dev_t rdev)
2640 struct inode *inode;
2641 int err, credits, retries = 0;
2643 err = dquot_initialize(dir);
2647 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2648 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2650 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2651 NULL, EXT4_HT_DIR, credits);
2652 handle = ext4_journal_current_handle();
2653 err = PTR_ERR(inode);
2654 if (!IS_ERR(inode)) {
2655 init_special_inode(inode, inode->i_mode, rdev);
2656 inode->i_op = &ext4_special_inode_operations;
2657 err = ext4_add_nondir(handle, dentry, &inode);
2659 ext4_fc_track_create(handle, dentry);
2662 ext4_journal_stop(handle);
2663 if (!IS_ERR_OR_NULL(inode))
2665 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2670 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2673 struct inode *inode;
2674 int err, retries = 0;
2676 err = dquot_initialize(dir);
2681 inode = ext4_new_inode_start_handle(dir, mode,
2684 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2685 4 + EXT4_XATTR_TRANS_BLOCKS);
2686 handle = ext4_journal_current_handle();
2687 err = PTR_ERR(inode);
2688 if (!IS_ERR(inode)) {
2689 inode->i_op = &ext4_file_inode_operations;
2690 inode->i_fop = &ext4_file_operations;
2691 ext4_set_aops(inode);
2692 d_tmpfile(dentry, inode);
2693 err = ext4_orphan_add(handle, inode);
2695 goto err_unlock_inode;
2696 mark_inode_dirty(inode);
2697 unlock_new_inode(inode);
2700 ext4_journal_stop(handle);
2701 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2705 ext4_journal_stop(handle);
2706 unlock_new_inode(inode);
2710 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2711 struct ext4_dir_entry_2 *de,
2712 int blocksize, int csum_size,
2713 unsigned int parent_ino, int dotdot_real_len)
2715 de->inode = cpu_to_le32(inode->i_ino);
2717 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2719 strcpy(de->name, ".");
2720 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2722 de = ext4_next_entry(de, blocksize);
2723 de->inode = cpu_to_le32(parent_ino);
2725 if (!dotdot_real_len)
2726 de->rec_len = ext4_rec_len_to_disk(blocksize -
2727 (csum_size + EXT4_DIR_REC_LEN(1)),
2730 de->rec_len = ext4_rec_len_to_disk(
2731 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2732 strcpy(de->name, "..");
2733 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2735 return ext4_next_entry(de, blocksize);
2738 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2739 struct inode *inode)
2741 struct buffer_head *dir_block = NULL;
2742 struct ext4_dir_entry_2 *de;
2743 ext4_lblk_t block = 0;
2744 unsigned int blocksize = dir->i_sb->s_blocksize;
2748 if (ext4_has_metadata_csum(dir->i_sb))
2749 csum_size = sizeof(struct ext4_dir_entry_tail);
2751 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2752 err = ext4_try_create_inline_dir(handle, dir, inode);
2753 if (err < 0 && err != -ENOSPC)
2760 dir_block = ext4_append(handle, inode, &block);
2761 if (IS_ERR(dir_block))
2762 return PTR_ERR(dir_block);
2763 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2764 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2765 set_nlink(inode, 2);
2767 ext4_initialize_dirent_tail(dir_block, blocksize);
2769 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2770 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2773 set_buffer_verified(dir_block);
2779 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2782 struct inode *inode;
2783 int err, err2 = 0, credits, retries = 0;
2785 if (EXT4_DIR_LINK_MAX(dir))
2788 err = dquot_initialize(dir);
2792 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2793 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2795 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2797 0, NULL, EXT4_HT_DIR, credits);
2798 handle = ext4_journal_current_handle();
2799 err = PTR_ERR(inode);
2803 inode->i_op = &ext4_dir_inode_operations;
2804 inode->i_fop = &ext4_dir_operations;
2805 err = ext4_init_new_dir(handle, dir, inode);
2807 goto out_clear_inode;
2808 err = ext4_mark_inode_dirty(handle, inode);
2810 err = ext4_add_entry(handle, dentry, inode);
2814 ext4_orphan_add(handle, inode);
2815 unlock_new_inode(inode);
2816 err2 = ext4_mark_inode_dirty(handle, inode);
2819 ext4_journal_stop(handle);
2823 ext4_inc_count(dir);
2825 ext4_update_dx_flag(dir);
2826 err = ext4_mark_inode_dirty(handle, dir);
2828 goto out_clear_inode;
2829 d_instantiate_new(dentry, inode);
2830 ext4_fc_track_create(handle, dentry);
2831 if (IS_DIRSYNC(dir))
2832 ext4_handle_sync(handle);
2836 ext4_journal_stop(handle);
2838 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2844 * routine to check that the specified directory is empty (for rmdir)
2846 bool ext4_empty_dir(struct inode *inode)
2848 unsigned int offset;
2849 struct buffer_head *bh;
2850 struct ext4_dir_entry_2 *de;
2851 struct super_block *sb;
2853 if (ext4_has_inline_data(inode)) {
2854 int has_inline_data = 1;
2857 ret = empty_inline_dir(inode, &has_inline_data);
2858 if (has_inline_data)
2863 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2864 EXT4_ERROR_INODE(inode, "invalid size");
2867 /* The first directory block must not be a hole,
2868 * so treat it as DIRENT_HTREE
2870 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2874 de = (struct ext4_dir_entry_2 *) bh->b_data;
2875 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2877 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2878 ext4_warning_inode(inode, "directory missing '.'");
2882 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2883 de = ext4_next_entry(de, sb->s_blocksize);
2884 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2886 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2887 ext4_warning_inode(inode, "directory missing '..'");
2891 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2892 while (offset < inode->i_size) {
2893 if (!(offset & (sb->s_blocksize - 1))) {
2894 unsigned int lblock;
2896 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2897 bh = ext4_read_dirblock(inode, lblock, EITHER);
2899 offset += sb->s_blocksize;
2905 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2906 (offset & (sb->s_blocksize - 1)));
2907 if (ext4_check_dir_entry(inode, NULL, de, bh,
2908 bh->b_data, bh->b_size, offset)) {
2909 offset = (offset | (sb->s_blocksize - 1)) + 1;
2912 if (le32_to_cpu(de->inode)) {
2916 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2923 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2924 * such inodes, starting at the superblock, in case we crash before the
2925 * file is closed/deleted, or in case the inode truncate spans multiple
2926 * transactions and the last transaction is not recovered after a crash.
2928 * At filesystem recovery time, we walk this list deleting unlinked
2929 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2931 * Orphan list manipulation functions must be called under i_mutex unless
2932 * we are just creating the inode or deleting it.
2934 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2936 struct super_block *sb = inode->i_sb;
2937 struct ext4_sb_info *sbi = EXT4_SB(sb);
2938 struct ext4_iloc iloc;
2942 if (!sbi->s_journal || is_bad_inode(inode))
2945 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2946 !inode_is_locked(inode));
2948 * Exit early if inode already is on orphan list. This is a big speedup
2949 * since we don't have to contend on the global s_orphan_lock.
2951 if (!list_empty(&EXT4_I(inode)->i_orphan))
2955 * Orphan handling is only valid for files with data blocks
2956 * being truncated, or files being unlinked. Note that we either
2957 * hold i_mutex, or the inode can not be referenced from outside,
2958 * so i_nlink should not be bumped due to race
2960 ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2961 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2963 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2964 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2968 err = ext4_reserve_inode_write(handle, inode, &iloc);
2972 mutex_lock(&sbi->s_orphan_lock);
2974 * Due to previous errors inode may be already a part of on-disk
2975 * orphan list. If so skip on-disk list modification.
2977 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2978 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2979 /* Insert this inode at the head of the on-disk orphan list */
2980 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2981 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2984 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2985 mutex_unlock(&sbi->s_orphan_lock);
2988 err = ext4_handle_dirty_super(handle, sb);
2989 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2994 * We have to remove inode from in-memory list if
2995 * addition to on disk orphan list failed. Stray orphan
2996 * list entries can cause panics at unmount time.
2998 mutex_lock(&sbi->s_orphan_lock);
2999 list_del_init(&EXT4_I(inode)->i_orphan);
3000 mutex_unlock(&sbi->s_orphan_lock);
3005 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3006 jbd_debug(4, "orphan inode %lu will point to %d\n",
3007 inode->i_ino, NEXT_ORPHAN(inode));
3009 ext4_std_error(sb, err);
3014 * ext4_orphan_del() removes an unlinked or truncated inode from the list
3015 * of such inodes stored on disk, because it is finally being cleaned up.
3017 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3019 struct list_head *prev;
3020 struct ext4_inode_info *ei = EXT4_I(inode);
3021 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3023 struct ext4_iloc iloc;
3026 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3029 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3030 !inode_is_locked(inode));
3031 /* Do this quick check before taking global s_orphan_lock. */
3032 if (list_empty(&ei->i_orphan))
3036 /* Grab inode buffer early before taking global s_orphan_lock */
3037 err = ext4_reserve_inode_write(handle, inode, &iloc);
3040 mutex_lock(&sbi->s_orphan_lock);
3041 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3043 prev = ei->i_orphan.prev;
3044 list_del_init(&ei->i_orphan);
3046 /* If we're on an error path, we may not have a valid
3047 * transaction handle with which to update the orphan list on
3048 * disk, but we still need to remove the inode from the linked
3049 * list in memory. */
3050 if (!handle || err) {
3051 mutex_unlock(&sbi->s_orphan_lock);
3055 ino_next = NEXT_ORPHAN(inode);
3056 if (prev == &sbi->s_orphan) {
3057 jbd_debug(4, "superblock will point to %u\n", ino_next);
3058 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3059 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3061 mutex_unlock(&sbi->s_orphan_lock);
3064 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3065 mutex_unlock(&sbi->s_orphan_lock);
3066 err = ext4_handle_dirty_super(handle, inode->i_sb);
3068 struct ext4_iloc iloc2;
3069 struct inode *i_prev =
3070 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3072 jbd_debug(4, "orphan inode %lu will point to %u\n",
3073 i_prev->i_ino, ino_next);
3074 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3076 mutex_unlock(&sbi->s_orphan_lock);
3079 NEXT_ORPHAN(i_prev) = ino_next;
3080 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3081 mutex_unlock(&sbi->s_orphan_lock);
3085 NEXT_ORPHAN(inode) = 0;
3086 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3088 ext4_std_error(inode->i_sb, err);
3096 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3099 struct inode *inode;
3100 struct buffer_head *bh;
3101 struct ext4_dir_entry_2 *de;
3102 handle_t *handle = NULL;
3104 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3107 /* Initialize quotas before so that eventual writes go in
3108 * separate transaction */
3109 retval = dquot_initialize(dir);
3112 retval = dquot_initialize(d_inode(dentry));
3117 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3123 inode = d_inode(dentry);
3125 retval = -EFSCORRUPTED;
3126 if (le32_to_cpu(de->inode) != inode->i_ino)
3129 retval = -ENOTEMPTY;
3130 if (!ext4_empty_dir(inode))
3133 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3134 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3135 if (IS_ERR(handle)) {
3136 retval = PTR_ERR(handle);
3141 if (IS_DIRSYNC(dir))
3142 ext4_handle_sync(handle);
3144 retval = ext4_delete_entry(handle, dir, de, bh);
3147 if (!EXT4_DIR_LINK_EMPTY(inode))
3148 ext4_warning_inode(inode,
3149 "empty directory '%.*s' has too many links (%u)",
3150 dentry->d_name.len, dentry->d_name.name,
3152 inode_inc_iversion(inode);
3154 /* There's no need to set i_disksize: the fact that i_nlink is
3155 * zero will ensure that the right thing happens during any
3158 ext4_orphan_add(handle, inode);
3159 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3160 retval = ext4_mark_inode_dirty(handle, inode);
3163 ext4_dec_count(dir);
3164 ext4_update_dx_flag(dir);
3165 ext4_fc_track_unlink(handle, dentry);
3166 retval = ext4_mark_inode_dirty(handle, dir);
3168 #ifdef CONFIG_UNICODE
3169 /* VFS negative dentries are incompatible with Encoding and
3170 * Case-insensitiveness. Eventually we'll want avoid
3171 * invalidating the dentries here, alongside with returning the
3172 * negative dentries at ext4_lookup(), when it is better
3173 * supported by the VFS for the CI case.
3175 if (IS_CASEFOLDED(dir))
3176 d_invalidate(dentry);
3182 ext4_journal_stop(handle);
3186 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3187 struct inode *inode)
3189 int retval = -ENOENT;
3190 struct buffer_head *bh;
3191 struct ext4_dir_entry_2 *de;
3192 int skip_remove_dentry = 0;
3194 bh = ext4_find_entry(dir, d_name, &de, NULL);
3201 if (le32_to_cpu(de->inode) != inode->i_ino) {
3203 * It's okay if we find dont find dentry which matches
3204 * the inode. That's because it might have gotten
3205 * renamed to a different inode number
3207 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3208 skip_remove_dentry = 1;
3213 if (IS_DIRSYNC(dir))
3214 ext4_handle_sync(handle);
3216 if (!skip_remove_dentry) {
3217 retval = ext4_delete_entry(handle, dir, de, bh);
3220 dir->i_ctime = dir->i_mtime = current_time(dir);
3221 ext4_update_dx_flag(dir);
3222 retval = ext4_mark_inode_dirty(handle, dir);
3228 if (inode->i_nlink == 0)
3229 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3230 d_name->len, d_name->name);
3233 if (!inode->i_nlink)
3234 ext4_orphan_add(handle, inode);
3235 inode->i_ctime = current_time(inode);
3236 retval = ext4_mark_inode_dirty(handle, inode);
3243 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3248 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3251 trace_ext4_unlink_enter(dir, dentry);
3253 * Initialize quotas before so that eventual writes go
3254 * in separate transaction
3256 retval = dquot_initialize(dir);
3259 retval = dquot_initialize(d_inode(dentry));
3263 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3264 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3265 if (IS_ERR(handle)) {
3266 retval = PTR_ERR(handle);
3270 retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3272 ext4_fc_track_unlink(handle, dentry);
3273 #ifdef CONFIG_UNICODE
3274 /* VFS negative dentries are incompatible with Encoding and
3275 * Case-insensitiveness. Eventually we'll want avoid
3276 * invalidating the dentries here, alongside with returning the
3277 * negative dentries at ext4_lookup(), when it is better
3278 * supported by the VFS for the CI case.
3280 if (IS_CASEFOLDED(dir))
3281 d_invalidate(dentry);
3284 ext4_journal_stop(handle);
3287 trace_ext4_unlink_exit(dentry, retval);
3291 static int ext4_symlink(struct inode *dir,
3292 struct dentry *dentry, const char *symname)
3295 struct inode *inode;
3296 int err, len = strlen(symname);
3298 struct fscrypt_str disk_link;
3300 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3303 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3308 err = dquot_initialize(dir);
3312 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3314 * For non-fast symlinks, we just allocate inode and put it on
3315 * orphan list in the first transaction => we need bitmap,
3316 * group descriptor, sb, inode block, quota blocks, and
3317 * possibly selinux xattr blocks.
3319 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3320 EXT4_XATTR_TRANS_BLOCKS;
3323 * Fast symlink. We have to add entry to directory
3324 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3325 * allocate new inode (bitmap, group descriptor, inode block,
3326 * quota blocks, sb is already counted in previous macros).
3328 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3329 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3332 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3333 &dentry->d_name, 0, NULL,
3334 EXT4_HT_DIR, credits);
3335 handle = ext4_journal_current_handle();
3336 if (IS_ERR(inode)) {
3338 ext4_journal_stop(handle);
3339 return PTR_ERR(inode);
3342 if (IS_ENCRYPTED(inode)) {
3343 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3345 goto err_drop_inode;
3346 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3349 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3350 if (!IS_ENCRYPTED(inode))
3351 inode->i_op = &ext4_symlink_inode_operations;
3352 inode_nohighmem(inode);
3353 ext4_set_aops(inode);
3355 * We cannot call page_symlink() with transaction started
3356 * because it calls into ext4_write_begin() which can wait
3357 * for transaction commit if we are running out of space
3358 * and thus we deadlock. So we have to stop transaction now
3359 * and restart it when symlink contents is written.
3361 * To keep fs consistent in case of crash, we have to put inode
3362 * to orphan list in the mean time.
3365 err = ext4_orphan_add(handle, inode);
3367 ext4_journal_stop(handle);
3370 goto err_drop_inode;
3371 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3373 goto err_drop_inode;
3375 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3376 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3378 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3379 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3380 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3381 if (IS_ERR(handle)) {
3382 err = PTR_ERR(handle);
3384 goto err_drop_inode;
3386 set_nlink(inode, 1);
3387 err = ext4_orphan_del(handle, inode);
3389 goto err_drop_inode;
3391 /* clear the extent format for fast symlink */
3392 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3393 if (!IS_ENCRYPTED(inode)) {
3394 inode->i_op = &ext4_fast_symlink_inode_operations;
3395 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3397 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3399 inode->i_size = disk_link.len - 1;
3401 EXT4_I(inode)->i_disksize = inode->i_size;
3402 err = ext4_add_nondir(handle, dentry, &inode);
3404 ext4_journal_stop(handle);
3407 goto out_free_encrypted_link;
3411 ext4_journal_stop(handle);
3413 unlock_new_inode(inode);
3415 out_free_encrypted_link:
3416 if (disk_link.name != (unsigned char *)symname)
3417 kfree(disk_link.name);
3421 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3424 int err, retries = 0;
3426 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3427 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3428 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3430 return PTR_ERR(handle);
3432 if (IS_DIRSYNC(dir))
3433 ext4_handle_sync(handle);
3435 inode->i_ctime = current_time(inode);
3436 ext4_inc_count(inode);
3439 err = ext4_add_entry(handle, dentry, inode);
3441 err = ext4_mark_inode_dirty(handle, inode);
3442 /* this can happen only for tmpfile being
3443 * linked the first time
3445 if (inode->i_nlink == 1)
3446 ext4_orphan_del(handle, inode);
3447 d_instantiate(dentry, inode);
3448 ext4_fc_track_link(handle, dentry);
3453 ext4_journal_stop(handle);
3454 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3459 static int ext4_link(struct dentry *old_dentry,
3460 struct inode *dir, struct dentry *dentry)
3462 struct inode *inode = d_inode(old_dentry);
3465 if (inode->i_nlink >= EXT4_LINK_MAX)
3468 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3472 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3473 (!projid_eq(EXT4_I(dir)->i_projid,
3474 EXT4_I(old_dentry->d_inode)->i_projid)))
3477 err = dquot_initialize(dir);
3480 return __ext4_link(dir, inode, dentry);
3484 * Try to find buffer head where contains the parent block.
3485 * It should be the inode block if it is inlined or the 1st block
3486 * if it is a normal dir.
3488 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3489 struct inode *inode,
3491 struct ext4_dir_entry_2 **parent_de,
3494 struct buffer_head *bh;
3496 if (!ext4_has_inline_data(inode)) {
3497 /* The first directory block must not be a hole, so
3498 * treat it as DIRENT_HTREE
3500 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3502 *retval = PTR_ERR(bh);
3505 *parent_de = ext4_next_entry(
3506 (struct ext4_dir_entry_2 *)bh->b_data,
3507 inode->i_sb->s_blocksize);
3512 return ext4_get_first_inline_block(inode, parent_de, retval);
3515 struct ext4_renament {
3517 struct dentry *dentry;
3518 struct inode *inode;
3520 int dir_nlink_delta;
3522 /* entry for "dentry" */
3523 struct buffer_head *bh;
3524 struct ext4_dir_entry_2 *de;
3527 /* entry for ".." in inode if it's a directory */
3528 struct buffer_head *dir_bh;
3529 struct ext4_dir_entry_2 *parent_de;
3533 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3537 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3538 &retval, &ent->parent_de,
3542 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3543 return -EFSCORRUPTED;
3544 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3545 return ext4_journal_get_write_access(handle, ent->dir_bh);
3548 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3553 ent->parent_de->inode = cpu_to_le32(dir_ino);
3554 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3555 if (!ent->dir_inlined) {
3556 if (is_dx(ent->inode)) {
3557 retval = ext4_handle_dirty_dx_node(handle,
3561 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3565 retval = ext4_mark_inode_dirty(handle, ent->inode);
3568 ext4_std_error(ent->dir->i_sb, retval);
3574 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3575 unsigned ino, unsigned file_type)
3577 int retval, retval2;
3579 BUFFER_TRACE(ent->bh, "get write access");
3580 retval = ext4_journal_get_write_access(handle, ent->bh);
3583 ent->de->inode = cpu_to_le32(ino);
3584 if (ext4_has_feature_filetype(ent->dir->i_sb))
3585 ent->de->file_type = file_type;
3586 inode_inc_iversion(ent->dir);
3587 ent->dir->i_ctime = ent->dir->i_mtime =
3588 current_time(ent->dir);
3589 retval = ext4_mark_inode_dirty(handle, ent->dir);
3590 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3591 if (!ent->inlined) {
3592 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3593 if (unlikely(retval2)) {
3594 ext4_std_error(ent->dir->i_sb, retval2);
3604 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3605 const struct qstr *d_name)
3607 int retval = -ENOENT;
3608 struct buffer_head *bh;
3609 struct ext4_dir_entry_2 *de;
3611 bh = ext4_find_entry(dir, d_name, &de, NULL);
3615 retval = ext4_delete_entry(handle, dir, de, bh);
3621 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3626 * ent->de could have moved from under us during htree split, so make
3627 * sure that we are deleting the right entry. We might also be pointing
3628 * to a stale entry in the unused part of ent->bh so just checking inum
3629 * and the name isn't enough.
3631 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3632 ent->de->name_len != ent->dentry->d_name.len ||
3633 strncmp(ent->de->name, ent->dentry->d_name.name,
3634 ent->de->name_len) ||
3636 retval = ext4_find_delete_entry(handle, ent->dir,
3637 &ent->dentry->d_name);
3639 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3640 if (retval == -ENOENT) {
3641 retval = ext4_find_delete_entry(handle, ent->dir,
3642 &ent->dentry->d_name);
3647 ext4_warning_inode(ent->dir,
3648 "Deleting old file: nlink %d, error=%d",
3649 ent->dir->i_nlink, retval);
3653 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3655 if (ent->dir_nlink_delta) {
3656 if (ent->dir_nlink_delta == -1)
3657 ext4_dec_count(ent->dir);
3659 ext4_inc_count(ent->dir);
3660 ext4_mark_inode_dirty(handle, ent->dir);
3664 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3665 int credits, handle_t **h)
3672 * for inode block, sb block, group summaries,
3675 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3676 EXT4_XATTR_TRANS_BLOCKS + 4);
3678 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3679 &ent->dentry->d_name, 0, NULL,
3680 EXT4_HT_DIR, credits);
3682 handle = ext4_journal_current_handle();
3685 ext4_journal_stop(handle);
3686 if (PTR_ERR(wh) == -ENOSPC &&
3687 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3691 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3692 wh->i_op = &ext4_special_inode_operations;
3698 * Anybody can rename anything with this: the permission checks are left to the
3699 * higher-level routines.
3701 * n.b. old_{dentry,inode) refers to the source dentry/inode
3702 * while new_{dentry,inode) refers to the destination dentry/inode
3703 * This comes from rename(const char *oldpath, const char *newpath)
3705 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3706 struct inode *new_dir, struct dentry *new_dentry,
3709 handle_t *handle = NULL;
3710 struct ext4_renament old = {
3712 .dentry = old_dentry,
3713 .inode = d_inode(old_dentry),
3715 struct ext4_renament new = {
3717 .dentry = new_dentry,
3718 .inode = d_inode(new_dentry),
3722 struct inode *whiteout = NULL;
3726 if (new.inode && new.inode->i_nlink == 0) {
3727 EXT4_ERROR_INODE(new.inode,
3728 "target of rename is already freed");
3729 return -EFSCORRUPTED;
3732 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3733 (!projid_eq(EXT4_I(new_dir)->i_projid,
3734 EXT4_I(old_dentry->d_inode)->i_projid)))
3737 retval = dquot_initialize(old.dir);
3740 retval = dquot_initialize(new.dir);
3744 /* Initialize quotas before so that eventual writes go
3745 * in separate transaction */
3747 retval = dquot_initialize(new.inode);
3752 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3754 return PTR_ERR(old.bh);
3756 * Check for inode number is _not_ due to possible IO errors.
3757 * We might rmdir the source, keep it as pwd of some process
3758 * and merrily kill the link to whatever was created under the
3759 * same name. Goodbye sticky bit ;-<
3762 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3765 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3766 &new.de, &new.inlined);
3767 if (IS_ERR(new.bh)) {
3768 retval = PTR_ERR(new.bh);
3778 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3779 ext4_alloc_da_blocks(old.inode);
3781 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3782 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3783 if (!(flags & RENAME_WHITEOUT)) {
3784 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3785 if (IS_ERR(handle)) {
3786 retval = PTR_ERR(handle);
3791 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3792 if (IS_ERR(whiteout)) {
3793 retval = PTR_ERR(whiteout);
3799 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3800 ext4_handle_sync(handle);
3802 if (S_ISDIR(old.inode->i_mode)) {
3804 retval = -ENOTEMPTY;
3805 if (!ext4_empty_dir(new.inode))
3809 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3812 retval = ext4_rename_dir_prepare(handle, &old);
3817 * If we're renaming a file within an inline_data dir and adding or
3818 * setting the new dirent causes a conversion from inline_data to
3819 * extents/blockmap, we need to force the dirent delete code to
3820 * re-read the directory, or else we end up trying to delete a dirent
3821 * from what is now the extent tree root (or a block map).
3823 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3824 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3826 old_file_type = old.de->file_type;
3829 * Do this before adding a new entry, so the old entry is sure
3830 * to be still pointing to the valid old entry.
3832 retval = ext4_setent(handle, &old, whiteout->i_ino,
3836 retval = ext4_mark_inode_dirty(handle, whiteout);
3837 if (unlikely(retval))
3841 retval = ext4_add_entry(handle, new.dentry, old.inode);
3845 retval = ext4_setent(handle, &new,
3846 old.inode->i_ino, old_file_type);
3851 force_reread = !ext4_test_inode_flag(new.dir,
3852 EXT4_INODE_INLINE_DATA);
3855 * Like most other Unix systems, set the ctime for inodes on a
3858 old.inode->i_ctime = current_time(old.inode);
3859 retval = ext4_mark_inode_dirty(handle, old.inode);
3860 if (unlikely(retval))
3867 ext4_rename_delete(handle, &old, force_reread);
3871 ext4_dec_count(new.inode);
3872 new.inode->i_ctime = current_time(new.inode);
3874 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3875 ext4_update_dx_flag(old.dir);
3877 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3881 ext4_dec_count(old.dir);
3883 /* checked ext4_empty_dir above, can't have another
3884 * parent, ext4_dec_count() won't work for many-linked
3886 clear_nlink(new.inode);
3888 ext4_inc_count(new.dir);
3889 ext4_update_dx_flag(new.dir);
3890 retval = ext4_mark_inode_dirty(handle, new.dir);
3891 if (unlikely(retval))
3895 retval = ext4_mark_inode_dirty(handle, old.dir);
3896 if (unlikely(retval))
3899 if (S_ISDIR(old.inode->i_mode)) {
3901 * We disable fast commits here that's because the
3902 * replay code is not yet capable of changing dot dot
3903 * dirents in directories.
3905 ext4_fc_mark_ineligible(old.inode->i_sb,
3906 EXT4_FC_REASON_RENAME_DIR);
3909 ext4_fc_track_unlink(handle, new.dentry);
3910 __ext4_fc_track_link(handle, old.inode, new.dentry);
3911 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3915 retval = ext4_mark_inode_dirty(handle, new.inode);
3916 if (unlikely(retval))
3918 if (!new.inode->i_nlink)
3919 ext4_orphan_add(handle, new.inode);
3929 drop_nlink(whiteout);
3930 unlock_new_inode(whiteout);
3934 ext4_journal_stop(handle);
3938 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3939 struct inode *new_dir, struct dentry *new_dentry)
3941 handle_t *handle = NULL;
3942 struct ext4_renament old = {
3944 .dentry = old_dentry,
3945 .inode = d_inode(old_dentry),
3947 struct ext4_renament new = {
3949 .dentry = new_dentry,
3950 .inode = d_inode(new_dentry),
3954 struct timespec64 ctime;
3956 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3957 !projid_eq(EXT4_I(new_dir)->i_projid,
3958 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3959 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3960 !projid_eq(EXT4_I(old_dir)->i_projid,
3961 EXT4_I(new_dentry->d_inode)->i_projid)))
3964 retval = dquot_initialize(old.dir);
3967 retval = dquot_initialize(new.dir);
3971 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3972 &old.de, &old.inlined);
3974 return PTR_ERR(old.bh);
3976 * Check for inode number is _not_ due to possible IO errors.
3977 * We might rmdir the source, keep it as pwd of some process
3978 * and merrily kill the link to whatever was created under the
3979 * same name. Goodbye sticky bit ;-<
3982 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3985 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3986 &new.de, &new.inlined);
3987 if (IS_ERR(new.bh)) {
3988 retval = PTR_ERR(new.bh);
3993 /* RENAME_EXCHANGE case: old *and* new must both exist */
3994 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3997 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3998 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3999 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4000 if (IS_ERR(handle)) {
4001 retval = PTR_ERR(handle);
4006 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4007 ext4_handle_sync(handle);
4009 if (S_ISDIR(old.inode->i_mode)) {
4011 retval = ext4_rename_dir_prepare(handle, &old);
4015 if (S_ISDIR(new.inode->i_mode)) {
4017 retval = ext4_rename_dir_prepare(handle, &new);
4023 * Other than the special case of overwriting a directory, parents'
4024 * nlink only needs to be modified if this is a cross directory rename.
4026 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4027 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4028 new.dir_nlink_delta = -old.dir_nlink_delta;
4030 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4031 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4035 new_file_type = new.de->file_type;
4036 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4040 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4045 * Like most other Unix systems, set the ctime for inodes on a
4048 ctime = current_time(old.inode);
4049 old.inode->i_ctime = ctime;
4050 new.inode->i_ctime = ctime;
4051 retval = ext4_mark_inode_dirty(handle, old.inode);
4052 if (unlikely(retval))
4054 retval = ext4_mark_inode_dirty(handle, new.inode);
4055 if (unlikely(retval))
4057 ext4_fc_mark_ineligible(new.inode->i_sb,
4058 EXT4_FC_REASON_CROSS_RENAME);
4060 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4065 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4069 ext4_update_dir_count(handle, &old);
4070 ext4_update_dir_count(handle, &new);
4079 ext4_journal_stop(handle);
4083 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4084 struct inode *new_dir, struct dentry *new_dentry,
4089 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4092 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4095 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4100 if (flags & RENAME_EXCHANGE) {
4101 return ext4_cross_rename(old_dir, old_dentry,
4102 new_dir, new_dentry);
4105 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4109 * directories can handle most operations...
4111 const struct inode_operations ext4_dir_inode_operations = {
4112 .create = ext4_create,
4113 .lookup = ext4_lookup,
4115 .unlink = ext4_unlink,
4116 .symlink = ext4_symlink,
4117 .mkdir = ext4_mkdir,
4118 .rmdir = ext4_rmdir,
4119 .mknod = ext4_mknod,
4120 .tmpfile = ext4_tmpfile,
4121 .rename = ext4_rename2,
4122 .setattr = ext4_setattr,
4123 .getattr = ext4_getattr,
4124 .listxattr = ext4_listxattr,
4125 .get_acl = ext4_get_acl,
4126 .set_acl = ext4_set_acl,
4127 .fiemap = ext4_fiemap,
4130 const struct inode_operations ext4_special_inode_operations = {
4131 .setattr = ext4_setattr,
4132 .getattr = ext4_getattr,
4133 .listxattr = ext4_listxattr,
4134 .get_acl = ext4_get_acl,
4135 .set_acl = ext4_set_acl,