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 bh = ext4_bread(NULL, inode, block, 0);
114 __ext4_warning(inode->i_sb, func, line,
115 "inode #%lu: lblock %lu: comm %s: "
116 "error %ld reading directory block",
117 inode->i_ino, (unsigned long)block,
118 current->comm, PTR_ERR(bh));
122 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
123 ext4_error_inode(inode, func, line, block,
124 "Directory hole found for htree %s block",
125 (type == INDEX) ? "index" : "leaf");
126 return ERR_PTR(-EFSCORRUPTED);
130 dirent = (struct ext4_dir_entry *) bh->b_data;
131 /* Determine whether or not we have an index block */
135 else if (ext4_rec_len_from_disk(dirent->rec_len,
136 inode->i_sb->s_blocksize) ==
137 inode->i_sb->s_blocksize)
140 if (!is_dx_block && type == INDEX) {
141 ext4_error_inode(inode, func, line, block,
142 "directory leaf block found instead of index block");
144 return ERR_PTR(-EFSCORRUPTED);
146 if (!ext4_has_metadata_csum(inode->i_sb) ||
151 * An empty leaf block can get mistaken for a index block; for
152 * this reason, we can only check the index checksum when the
153 * caller is sure it should be an index block.
155 if (is_dx_block && type == INDEX) {
156 if (ext4_dx_csum_verify(inode, dirent))
157 set_buffer_verified(bh);
159 ext4_set_errno(inode->i_sb, EFSBADCRC);
160 ext4_error_inode(inode, func, line, block,
161 "Directory index failed checksum");
163 return ERR_PTR(-EFSBADCRC);
167 if (ext4_dirblock_csum_verify(inode, bh))
168 set_buffer_verified(bh);
170 ext4_set_errno(inode->i_sb, EFSBADCRC);
171 ext4_error_inode(inode, func, line, block,
172 "Directory block failed checksum");
174 return ERR_PTR(-EFSBADCRC);
181 #define assert(test) J_ASSERT(test)
185 #define dxtrace(command) command
187 #define dxtrace(command)
211 * dx_root_info is laid out so that if it should somehow get overlaid by a
212 * dirent the two low bits of the hash version will be zero. Therefore, the
213 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
218 struct fake_dirent dot;
220 struct fake_dirent dotdot;
224 __le32 reserved_zero;
226 u8 info_length; /* 8 */
231 struct dx_entry entries[0];
236 struct fake_dirent fake;
237 struct dx_entry entries[0];
243 struct buffer_head *bh;
244 struct dx_entry *entries;
256 * This goes at the end of each htree block.
260 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
263 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
264 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
265 static inline unsigned dx_get_hash(struct dx_entry *entry);
266 static void dx_set_hash(struct dx_entry *entry, unsigned value);
267 static unsigned dx_get_count(struct dx_entry *entries);
268 static unsigned dx_get_limit(struct dx_entry *entries);
269 static void dx_set_count(struct dx_entry *entries, unsigned value);
270 static void dx_set_limit(struct dx_entry *entries, unsigned value);
271 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
272 static unsigned dx_node_limit(struct inode *dir);
273 static struct dx_frame *dx_probe(struct ext4_filename *fname,
275 struct dx_hash_info *hinfo,
276 struct dx_frame *frame);
277 static void dx_release(struct dx_frame *frames);
278 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
279 unsigned blocksize, struct dx_hash_info *hinfo,
280 struct dx_map_entry map[]);
281 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
282 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
283 struct dx_map_entry *offsets, int count, unsigned blocksize);
284 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
285 static void dx_insert_block(struct dx_frame *frame,
286 u32 hash, ext4_lblk_t block);
287 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
288 struct dx_frame *frame,
289 struct dx_frame *frames,
291 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
292 struct ext4_filename *fname,
293 struct ext4_dir_entry_2 **res_dir);
294 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
295 struct inode *dir, struct inode *inode);
297 /* checksumming functions */
298 void ext4_initialize_dirent_tail(struct buffer_head *bh,
299 unsigned int blocksize)
301 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
303 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
304 t->det_rec_len = ext4_rec_len_to_disk(
305 sizeof(struct ext4_dir_entry_tail), blocksize);
306 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
309 /* Walk through a dirent block to find a checksum "dirent" at the tail */
310 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
311 struct buffer_head *bh)
313 struct ext4_dir_entry_tail *t;
316 struct ext4_dir_entry *d, *top;
318 d = (struct ext4_dir_entry *)bh->b_data;
319 top = (struct ext4_dir_entry *)(bh->b_data +
320 (EXT4_BLOCK_SIZE(inode->i_sb) -
321 sizeof(struct ext4_dir_entry_tail)));
322 while (d < top && d->rec_len)
323 d = (struct ext4_dir_entry *)(((void *)d) +
324 le16_to_cpu(d->rec_len));
329 t = (struct ext4_dir_entry_tail *)d;
331 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
334 if (t->det_reserved_zero1 ||
335 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
336 t->det_reserved_zero2 ||
337 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
343 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
345 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
346 struct ext4_inode_info *ei = EXT4_I(inode);
349 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
350 return cpu_to_le32(csum);
353 #define warn_no_space_for_csum(inode) \
354 __warn_no_space_for_csum((inode), __func__, __LINE__)
356 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
359 __ext4_warning_inode(inode, func, line,
360 "No space for directory leaf checksum. Please run e2fsck -D.");
363 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
365 struct ext4_dir_entry_tail *t;
367 if (!ext4_has_metadata_csum(inode->i_sb))
370 t = get_dirent_tail(inode, bh);
372 warn_no_space_for_csum(inode);
376 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
377 (char *)t - bh->b_data))
383 static void ext4_dirblock_csum_set(struct inode *inode,
384 struct buffer_head *bh)
386 struct ext4_dir_entry_tail *t;
388 if (!ext4_has_metadata_csum(inode->i_sb))
391 t = get_dirent_tail(inode, bh);
393 warn_no_space_for_csum(inode);
397 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
398 (char *)t - bh->b_data);
401 int ext4_handle_dirty_dirblock(handle_t *handle,
403 struct buffer_head *bh)
405 ext4_dirblock_csum_set(inode, bh);
406 return ext4_handle_dirty_metadata(handle, inode, bh);
409 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
410 struct ext4_dir_entry *dirent,
413 struct ext4_dir_entry *dp;
414 struct dx_root_info *root;
417 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
419 else if (le16_to_cpu(dirent->rec_len) == 12) {
420 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
421 if (le16_to_cpu(dp->rec_len) !=
422 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
424 root = (struct dx_root_info *)(((void *)dp + 12));
425 if (root->reserved_zero ||
426 root->info_length != sizeof(struct dx_root_info))
433 *offset = count_offset;
434 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
437 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
438 int count_offset, int count, struct dx_tail *t)
440 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
441 struct ext4_inode_info *ei = EXT4_I(inode);
444 __u32 dummy_csum = 0;
445 int offset = offsetof(struct dx_tail, dt_checksum);
447 size = count_offset + (count * sizeof(struct dx_entry));
448 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
449 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
450 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
452 return cpu_to_le32(csum);
455 static int ext4_dx_csum_verify(struct inode *inode,
456 struct ext4_dir_entry *dirent)
458 struct dx_countlimit *c;
460 int count_offset, limit, count;
462 if (!ext4_has_metadata_csum(inode->i_sb))
465 c = get_dx_countlimit(inode, dirent, &count_offset);
467 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
470 limit = le16_to_cpu(c->limit);
471 count = le16_to_cpu(c->count);
472 if (count_offset + (limit * sizeof(struct dx_entry)) >
473 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
474 warn_no_space_for_csum(inode);
477 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
479 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
485 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
487 struct dx_countlimit *c;
489 int count_offset, limit, count;
491 if (!ext4_has_metadata_csum(inode->i_sb))
494 c = get_dx_countlimit(inode, dirent, &count_offset);
496 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
499 limit = le16_to_cpu(c->limit);
500 count = le16_to_cpu(c->count);
501 if (count_offset + (limit * sizeof(struct dx_entry)) >
502 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
503 warn_no_space_for_csum(inode);
506 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
508 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
511 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
513 struct buffer_head *bh)
515 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
516 return ext4_handle_dirty_metadata(handle, inode, bh);
520 * p is at least 6 bytes before the end of page
522 static inline struct ext4_dir_entry_2 *
523 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
525 return (struct ext4_dir_entry_2 *)((char *)p +
526 ext4_rec_len_from_disk(p->rec_len, blocksize));
530 * Future: use high four bits of block for coalesce-on-delete flags
531 * Mask them off for now.
534 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
536 return le32_to_cpu(entry->block) & 0x0fffffff;
539 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
541 entry->block = cpu_to_le32(value);
544 static inline unsigned dx_get_hash(struct dx_entry *entry)
546 return le32_to_cpu(entry->hash);
549 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
551 entry->hash = cpu_to_le32(value);
554 static inline unsigned dx_get_count(struct dx_entry *entries)
556 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
559 static inline unsigned dx_get_limit(struct dx_entry *entries)
561 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
564 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
566 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
569 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
571 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
574 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
576 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
577 EXT4_DIR_REC_LEN(2) - infosize;
579 if (ext4_has_metadata_csum(dir->i_sb))
580 entry_space -= sizeof(struct dx_tail);
581 return entry_space / sizeof(struct dx_entry);
584 static inline unsigned dx_node_limit(struct inode *dir)
586 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
588 if (ext4_has_metadata_csum(dir->i_sb))
589 entry_space -= sizeof(struct dx_tail);
590 return entry_space / sizeof(struct dx_entry);
597 static void dx_show_index(char * label, struct dx_entry *entries)
599 int i, n = dx_get_count (entries);
600 printk(KERN_DEBUG "%s index", label);
601 for (i = 0; i < n; i++) {
602 printk(KERN_CONT " %x->%lu",
603 i ? dx_get_hash(entries + i) : 0,
604 (unsigned long)dx_get_block(entries + i));
606 printk(KERN_CONT "\n");
616 static struct stats dx_show_leaf(struct inode *dir,
617 struct dx_hash_info *hinfo,
618 struct ext4_dir_entry_2 *de,
619 int size, int show_names)
621 unsigned names = 0, space = 0;
622 char *base = (char *) de;
623 struct dx_hash_info h = *hinfo;
626 while ((char *) de < base + size)
632 #ifdef CONFIG_FS_ENCRYPTION
635 struct fscrypt_str fname_crypto_str =
641 if (IS_ENCRYPTED(dir))
642 res = fscrypt_get_encryption_info(dir);
644 printk(KERN_WARNING "Error setting up"
645 " fname crypto: %d\n", res);
647 if (!fscrypt_has_encryption_key(dir)) {
648 /* Directory is not encrypted */
649 ext4fs_dirhash(dir, de->name,
651 printk("%*.s:(U)%x.%u ", len,
653 (unsigned) ((char *) de
656 struct fscrypt_str de_name =
657 FSTR_INIT(name, len);
659 /* Directory is encrypted */
660 res = fscrypt_fname_alloc_buffer(
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)
848 assert (at == p - 1);
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 #ifdef CONFIG_FS_ENCRYPTION
1008 /* Check if the directory is encrypted */
1009 if (IS_ENCRYPTED(dir)) {
1010 err = fscrypt_get_encryption_info(dir);
1015 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1023 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1024 if (ext4_check_dir_entry(dir, NULL, de, bh,
1025 bh->b_data, bh->b_size,
1026 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1027 + ((char *)de - bh->b_data))) {
1028 /* silently ignore the rest of the block */
1031 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1032 if ((hinfo->hash < start_hash) ||
1033 ((hinfo->hash == start_hash) &&
1034 (hinfo->minor_hash < start_minor_hash)))
1038 if (!IS_ENCRYPTED(dir)) {
1039 tmp_str.name = de->name;
1040 tmp_str.len = de->name_len;
1041 err = ext4_htree_store_dirent(dir_file,
1042 hinfo->hash, hinfo->minor_hash, de,
1045 int save_len = fname_crypto_str.len;
1046 struct fscrypt_str de_name = FSTR_INIT(de->name,
1049 /* Directory is encrypted */
1050 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1051 hinfo->minor_hash, &de_name,
1057 err = ext4_htree_store_dirent(dir_file,
1058 hinfo->hash, hinfo->minor_hash, de,
1060 fname_crypto_str.len = save_len;
1070 #ifdef CONFIG_FS_ENCRYPTION
1071 fscrypt_fname_free_buffer(&fname_crypto_str);
1078 * This function fills a red-black tree with information from a
1079 * directory. We start scanning the directory in hash order, starting
1080 * at start_hash and start_minor_hash.
1082 * This function returns the number of entries inserted into the tree,
1083 * or a negative error code.
1085 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1086 __u32 start_minor_hash, __u32 *next_hash)
1088 struct dx_hash_info hinfo;
1089 struct ext4_dir_entry_2 *de;
1090 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1096 struct fscrypt_str tmp_str;
1098 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1099 start_hash, start_minor_hash));
1100 dir = file_inode(dir_file);
1101 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1102 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1103 if (hinfo.hash_version <= DX_HASH_TEA)
1104 hinfo.hash_version +=
1105 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1106 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1107 if (ext4_has_inline_data(dir)) {
1108 int has_inline_data = 1;
1109 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1113 if (has_inline_data) {
1118 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1119 start_hash, start_minor_hash);
1123 hinfo.hash = start_hash;
1124 hinfo.minor_hash = 0;
1125 frame = dx_probe(NULL, dir, &hinfo, frames);
1127 return PTR_ERR(frame);
1129 /* Add '.' and '..' from the htree header */
1130 if (!start_hash && !start_minor_hash) {
1131 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1132 tmp_str.name = de->name;
1133 tmp_str.len = de->name_len;
1134 err = ext4_htree_store_dirent(dir_file, 0, 0,
1140 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1141 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1142 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1143 tmp_str.name = de->name;
1144 tmp_str.len = de->name_len;
1145 err = ext4_htree_store_dirent(dir_file, 2, 0,
1153 if (fatal_signal_pending(current)) {
1158 block = dx_get_block(frame->at);
1159 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1160 start_hash, start_minor_hash);
1167 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1168 frame, frames, &hashval);
1169 *next_hash = hashval;
1175 * Stop if: (a) there are no more entries, or
1176 * (b) we have inserted at least one entry and the
1177 * next hash value is not a continuation
1180 (count && ((hashval & 1) == 0)))
1184 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1185 "next hash: %x\n", count, *next_hash));
1192 static inline int search_dirblock(struct buffer_head *bh,
1194 struct ext4_filename *fname,
1195 unsigned int offset,
1196 struct ext4_dir_entry_2 **res_dir)
1198 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1199 fname, offset, res_dir);
1203 * Directory block splitting, compacting
1207 * Create map of hash values, offsets, and sizes, stored at end of block.
1208 * Returns number of entries mapped.
1210 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1211 unsigned blocksize, struct dx_hash_info *hinfo,
1212 struct dx_map_entry *map_tail)
1215 char *base = (char *) de;
1216 struct dx_hash_info h = *hinfo;
1218 while ((char *) de < base + blocksize) {
1219 if (de->name_len && de->inode) {
1220 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1222 map_tail->hash = h.hash;
1223 map_tail->offs = ((char *) de - base)>>2;
1224 map_tail->size = le16_to_cpu(de->rec_len);
1228 /* XXX: do we need to check rec_len == 0 case? -Chris */
1229 de = ext4_next_entry(de, blocksize);
1234 /* Sort map by hash value */
1235 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1237 struct dx_map_entry *p, *q, *top = map + count - 1;
1239 /* Combsort until bubble sort doesn't suck */
1241 count = count*10/13;
1242 if (count - 9 < 2) /* 9, 10 -> 11 */
1244 for (p = top, q = p - count; q >= map; p--, q--)
1245 if (p->hash < q->hash)
1248 /* Garden variety bubble sort */
1253 if (q[1].hash >= q[0].hash)
1261 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1263 struct dx_entry *entries = frame->entries;
1264 struct dx_entry *old = frame->at, *new = old + 1;
1265 int count = dx_get_count(entries);
1267 assert(count < dx_get_limit(entries));
1268 assert(old < entries + count);
1269 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1270 dx_set_hash(new, hash);
1271 dx_set_block(new, block);
1272 dx_set_count(entries, count + 1);
1275 #ifdef CONFIG_UNICODE
1277 * Test whether a case-insensitive directory entry matches the filename
1278 * being searched for. If quick is set, assume the name being looked up
1279 * is already in the casefolded form.
1281 * Returns: 0 if the directory entry matches, more than 0 if it
1282 * doesn't match or less than zero on error.
1284 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1285 const struct qstr *entry, bool quick)
1287 const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1288 const struct unicode_map *um = sbi->s_encoding;
1292 ret = utf8_strncasecmp_folded(um, name, entry);
1294 ret = utf8_strncasecmp(um, name, entry);
1297 /* Handle invalid character sequence as either an error
1298 * or as an opaque byte sequence.
1300 if (ext4_has_strict_mode(sbi))
1303 if (name->len != entry->len)
1306 return !!memcmp(name->name, entry->name, name->len);
1312 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1313 struct fscrypt_str *cf_name)
1317 if (!IS_CASEFOLDED(dir) || !EXT4_SB(dir->i_sb)->s_encoding) {
1318 cf_name->name = NULL;
1322 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1326 len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1327 iname, cf_name->name,
1330 kfree(cf_name->name);
1331 cf_name->name = NULL;
1334 cf_name->len = (unsigned) len;
1340 * Test whether a directory entry matches the filename being searched for.
1342 * Return: %true if the directory entry matches, otherwise %false.
1344 static inline bool ext4_match(const struct inode *parent,
1345 const struct ext4_filename *fname,
1346 const struct ext4_dir_entry_2 *de)
1348 struct fscrypt_name f;
1349 #ifdef CONFIG_UNICODE
1350 const struct qstr entry = {.name = de->name, .len = de->name_len};
1356 f.usr_fname = fname->usr_fname;
1357 f.disk_name = fname->disk_name;
1358 #ifdef CONFIG_FS_ENCRYPTION
1359 f.crypto_buf = fname->crypto_buf;
1362 #ifdef CONFIG_UNICODE
1363 if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1364 if (fname->cf_name.name) {
1365 struct qstr cf = {.name = fname->cf_name.name,
1366 .len = fname->cf_name.len};
1367 return !ext4_ci_compare(parent, &cf, &entry, true);
1369 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1374 return fscrypt_match_name(&f, de->name, de->name_len);
1378 * Returns 0 if not found, -1 on failure, and 1 on success
1380 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1381 struct inode *dir, struct ext4_filename *fname,
1382 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1384 struct ext4_dir_entry_2 * de;
1388 de = (struct ext4_dir_entry_2 *)search_buf;
1389 dlimit = search_buf + buf_size;
1390 while ((char *) de < dlimit) {
1391 /* this code is executed quadratically often */
1392 /* do minimal checking `by hand' */
1393 if ((char *) de + de->name_len <= dlimit &&
1394 ext4_match(dir, fname, de)) {
1395 /* found a match - just to be sure, do
1397 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1398 bh->b_size, offset))
1403 /* prevent looping on a bad block */
1404 de_len = ext4_rec_len_from_disk(de->rec_len,
1405 dir->i_sb->s_blocksize);
1409 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1414 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1415 struct ext4_dir_entry *de)
1417 struct super_block *sb = dir->i_sb;
1423 if (de->inode == 0 &&
1424 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1431 * __ext4_find_entry()
1433 * finds an entry in the specified directory with the wanted name. It
1434 * returns the cache buffer in which the entry was found, and the entry
1435 * itself (as a parameter - res_dir). It does NOT read the inode of the
1436 * entry - you'll have to do that yourself if you want to.
1438 * The returned buffer_head has ->b_count elevated. The caller is expected
1439 * to brelse() it when appropriate.
1441 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1442 struct ext4_filename *fname,
1443 struct ext4_dir_entry_2 **res_dir,
1446 struct super_block *sb;
1447 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1448 struct buffer_head *bh, *ret = NULL;
1449 ext4_lblk_t start, block;
1450 const u8 *name = fname->usr_fname->name;
1451 size_t ra_max = 0; /* Number of bh's in the readahead
1453 size_t ra_ptr = 0; /* Current index into readahead
1455 ext4_lblk_t nblocks;
1456 int i, namelen, retval;
1460 namelen = fname->usr_fname->len;
1461 if (namelen > EXT4_NAME_LEN)
1464 if (ext4_has_inline_data(dir)) {
1465 int has_inline_data = 1;
1466 ret = ext4_find_inline_entry(dir, fname, res_dir,
1468 if (has_inline_data) {
1471 goto cleanup_and_exit;
1475 if ((namelen <= 2) && (name[0] == '.') &&
1476 (name[1] == '.' || name[1] == '\0')) {
1478 * "." or ".." will only be in the first block
1479 * NFS may look up ".."; "." should be handled by the VFS
1486 ret = ext4_dx_find_entry(dir, fname, res_dir);
1488 * On success, or if the error was file not found,
1489 * return. Otherwise, fall back to doing a search the
1490 * old fashioned way.
1492 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1493 goto cleanup_and_exit;
1494 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1498 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1501 goto cleanup_and_exit;
1503 start = EXT4_I(dir)->i_dir_start_lookup;
1504 if (start >= nblocks)
1510 * We deal with the read-ahead logic here.
1512 if (ra_ptr >= ra_max) {
1513 /* Refill the readahead buffer */
1516 ra_max = start - block;
1518 ra_max = nblocks - block;
1519 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1520 retval = ext4_bread_batch(dir, block, ra_max,
1521 false /* wait */, bh_use);
1523 ret = ERR_PTR(retval);
1525 goto cleanup_and_exit;
1528 if ((bh = bh_use[ra_ptr++]) == NULL)
1531 if (!buffer_uptodate(bh)) {
1532 ext4_set_errno(sb, EIO);
1533 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1534 (unsigned long) block);
1536 ret = ERR_PTR(-EIO);
1537 goto cleanup_and_exit;
1539 if (!buffer_verified(bh) &&
1540 !is_dx_internal_node(dir, block,
1541 (struct ext4_dir_entry *)bh->b_data) &&
1542 !ext4_dirblock_csum_verify(dir, bh)) {
1543 ext4_set_errno(sb, EFSBADCRC);
1544 EXT4_ERROR_INODE(dir, "checksumming directory "
1545 "block %lu", (unsigned long)block);
1547 ret = ERR_PTR(-EFSBADCRC);
1548 goto cleanup_and_exit;
1550 set_buffer_verified(bh);
1551 i = search_dirblock(bh, dir, fname,
1552 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1554 EXT4_I(dir)->i_dir_start_lookup = block;
1556 goto cleanup_and_exit;
1560 goto cleanup_and_exit;
1563 if (++block >= nblocks)
1565 } while (block != start);
1568 * If the directory has grown while we were searching, then
1569 * search the last part of the directory before giving up.
1572 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1573 if (block < nblocks) {
1579 /* Clean up the read-ahead blocks */
1580 for (; ra_ptr < ra_max; ra_ptr++)
1581 brelse(bh_use[ra_ptr]);
1585 static struct buffer_head *ext4_find_entry(struct inode *dir,
1586 const struct qstr *d_name,
1587 struct ext4_dir_entry_2 **res_dir,
1591 struct ext4_filename fname;
1592 struct buffer_head *bh;
1594 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1598 return ERR_PTR(err);
1600 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1602 ext4_fname_free_filename(&fname);
1606 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1607 struct dentry *dentry,
1608 struct ext4_dir_entry_2 **res_dir)
1611 struct ext4_filename fname;
1612 struct buffer_head *bh;
1614 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1618 return ERR_PTR(err);
1620 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1622 ext4_fname_free_filename(&fname);
1626 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1627 struct ext4_filename *fname,
1628 struct ext4_dir_entry_2 **res_dir)
1630 struct super_block * sb = dir->i_sb;
1631 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1632 struct buffer_head *bh;
1636 #ifdef CONFIG_FS_ENCRYPTION
1639 frame = dx_probe(fname, dir, NULL, frames);
1641 return (struct buffer_head *) frame;
1643 block = dx_get_block(frame->at);
1644 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1648 retval = search_dirblock(bh, dir, fname,
1649 block << EXT4_BLOCK_SIZE_BITS(sb),
1655 bh = ERR_PTR(ERR_BAD_DX_DIR);
1659 /* Check to see if we should continue to search */
1660 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1663 ext4_warning_inode(dir,
1664 "error %d reading directory index block",
1666 bh = ERR_PTR(retval);
1669 } while (retval == 1);
1673 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1679 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1681 struct inode *inode;
1682 struct ext4_dir_entry_2 *de;
1683 struct buffer_head *bh;
1685 if (dentry->d_name.len > EXT4_NAME_LEN)
1686 return ERR_PTR(-ENAMETOOLONG);
1688 bh = ext4_lookup_entry(dir, dentry, &de);
1690 return ERR_CAST(bh);
1693 __u32 ino = le32_to_cpu(de->inode);
1695 if (!ext4_valid_inum(dir->i_sb, ino)) {
1696 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1697 return ERR_PTR(-EFSCORRUPTED);
1699 if (unlikely(ino == dir->i_ino)) {
1700 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1702 return ERR_PTR(-EFSCORRUPTED);
1704 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1705 if (inode == ERR_PTR(-ESTALE)) {
1706 EXT4_ERROR_INODE(dir,
1707 "deleted inode referenced: %u",
1709 return ERR_PTR(-EFSCORRUPTED);
1711 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1712 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1713 !fscrypt_has_permitted_context(dir, inode)) {
1714 ext4_warning(inode->i_sb,
1715 "Inconsistent encryption contexts: %lu/%lu",
1716 dir->i_ino, inode->i_ino);
1718 return ERR_PTR(-EPERM);
1722 #ifdef CONFIG_UNICODE
1723 if (!inode && IS_CASEFOLDED(dir)) {
1724 /* Eventually we want to call d_add_ci(dentry, NULL)
1725 * for negative dentries in the encoding case as
1726 * well. For now, prevent the negative dentry
1727 * from being cached.
1732 return d_splice_alias(inode, dentry);
1736 struct dentry *ext4_get_parent(struct dentry *child)
1739 static const struct qstr dotdot = QSTR_INIT("..", 2);
1740 struct ext4_dir_entry_2 * de;
1741 struct buffer_head *bh;
1743 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1745 return ERR_CAST(bh);
1747 return ERR_PTR(-ENOENT);
1748 ino = le32_to_cpu(de->inode);
1751 if (!ext4_valid_inum(child->d_sb, ino)) {
1752 EXT4_ERROR_INODE(d_inode(child),
1753 "bad parent inode number: %u", ino);
1754 return ERR_PTR(-EFSCORRUPTED);
1757 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1761 * Move count entries from end of map between two memory locations.
1762 * Returns pointer to last entry moved.
1764 static struct ext4_dir_entry_2 *
1765 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1768 unsigned rec_len = 0;
1771 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1772 (from + (map->offs<<2));
1773 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1774 memcpy (to, de, rec_len);
1775 ((struct ext4_dir_entry_2 *) to)->rec_len =
1776 ext4_rec_len_to_disk(rec_len, blocksize);
1781 return (struct ext4_dir_entry_2 *) (to - rec_len);
1785 * Compact each dir entry in the range to the minimal rec_len.
1786 * Returns pointer to last entry in range.
1788 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1790 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1791 unsigned rec_len = 0;
1794 while ((char*)de < base + blocksize) {
1795 next = ext4_next_entry(de, blocksize);
1796 if (de->inode && de->name_len) {
1797 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1799 memmove(to, de, rec_len);
1800 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1802 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1810 * Split a full leaf block to make room for a new dir entry.
1811 * Allocate a new block, and move entries so that they are approx. equally full.
1812 * Returns pointer to de in block into which the new entry will be inserted.
1814 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1815 struct buffer_head **bh,struct dx_frame *frame,
1816 struct dx_hash_info *hinfo)
1818 unsigned blocksize = dir->i_sb->s_blocksize;
1819 unsigned count, continued;
1820 struct buffer_head *bh2;
1821 ext4_lblk_t newblock;
1823 struct dx_map_entry *map;
1824 char *data1 = (*bh)->b_data, *data2;
1825 unsigned split, move, size;
1826 struct ext4_dir_entry_2 *de = NULL, *de2;
1830 if (ext4_has_metadata_csum(dir->i_sb))
1831 csum_size = sizeof(struct ext4_dir_entry_tail);
1833 bh2 = ext4_append(handle, dir, &newblock);
1837 return (struct ext4_dir_entry_2 *) bh2;
1840 BUFFER_TRACE(*bh, "get_write_access");
1841 err = ext4_journal_get_write_access(handle, *bh);
1845 BUFFER_TRACE(frame->bh, "get_write_access");
1846 err = ext4_journal_get_write_access(handle, frame->bh);
1850 data2 = bh2->b_data;
1852 /* create map in the end of data2 block */
1853 map = (struct dx_map_entry *) (data2 + blocksize);
1854 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1855 blocksize, hinfo, map);
1857 dx_sort_map(map, count);
1858 /* Split the existing block in the middle, size-wise */
1861 for (i = count-1; i >= 0; i--) {
1862 /* is more than half of this entry in 2nd half of the block? */
1863 if (size + map[i].size/2 > blocksize/2)
1865 size += map[i].size;
1868 /* map index at which we will split */
1869 split = count - move;
1870 hash2 = map[split].hash;
1871 continued = hash2 == map[split - 1].hash;
1872 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1873 (unsigned long)dx_get_block(frame->at),
1874 hash2, split, count-split));
1876 /* Fancy dance to stay within two buffers */
1877 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1879 de = dx_pack_dirents(data1, blocksize);
1880 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1883 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1887 ext4_initialize_dirent_tail(*bh, blocksize);
1888 ext4_initialize_dirent_tail(bh2, blocksize);
1891 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1893 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1896 /* Which block gets the new entry? */
1897 if (hinfo->hash >= hash2) {
1901 dx_insert_block(frame, hash2 + continued, newblock);
1902 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1905 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1909 dxtrace(dx_show_index("frame", frame->entries));
1916 ext4_std_error(dir->i_sb, err);
1917 return ERR_PTR(err);
1920 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1921 struct buffer_head *bh,
1922 void *buf, int buf_size,
1923 struct ext4_filename *fname,
1924 struct ext4_dir_entry_2 **dest_de)
1926 struct ext4_dir_entry_2 *de;
1927 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1929 unsigned int offset = 0;
1932 de = (struct ext4_dir_entry_2 *)buf;
1933 top = buf + buf_size - reclen;
1934 while ((char *) de <= top) {
1935 if (ext4_check_dir_entry(dir, NULL, de, bh,
1936 buf, buf_size, offset))
1937 return -EFSCORRUPTED;
1938 if (ext4_match(dir, fname, de))
1940 nlen = EXT4_DIR_REC_LEN(de->name_len);
1941 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1942 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1944 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1947 if ((char *) de > top)
1954 void ext4_insert_dentry(struct inode *inode,
1955 struct ext4_dir_entry_2 *de,
1957 struct ext4_filename *fname)
1962 nlen = EXT4_DIR_REC_LEN(de->name_len);
1963 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1965 struct ext4_dir_entry_2 *de1 =
1966 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1967 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1968 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1971 de->file_type = EXT4_FT_UNKNOWN;
1972 de->inode = cpu_to_le32(inode->i_ino);
1973 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1974 de->name_len = fname_len(fname);
1975 memcpy(de->name, fname_name(fname), fname_len(fname));
1979 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1980 * it points to a directory entry which is guaranteed to be large
1981 * enough for new directory entry. If de is NULL, then
1982 * add_dirent_to_buf will attempt search the directory block for
1983 * space. It will return -ENOSPC if no space is available, and -EIO
1984 * and -EEXIST if directory entry already exists.
1986 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1988 struct inode *inode, struct ext4_dir_entry_2 *de,
1989 struct buffer_head *bh)
1991 unsigned int blocksize = dir->i_sb->s_blocksize;
1995 if (ext4_has_metadata_csum(inode->i_sb))
1996 csum_size = sizeof(struct ext4_dir_entry_tail);
1999 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2000 blocksize - csum_size, fname, &de);
2004 BUFFER_TRACE(bh, "get_write_access");
2005 err = ext4_journal_get_write_access(handle, bh);
2007 ext4_std_error(dir->i_sb, err);
2011 /* By now the buffer is marked for journaling */
2012 ext4_insert_dentry(inode, de, blocksize, fname);
2015 * XXX shouldn't update any times until successful
2016 * completion of syscall, but too many callers depend
2019 * XXX similarly, too many callers depend on
2020 * ext4_new_inode() setting the times, but error
2021 * recovery deletes the inode, so the worst that can
2022 * happen is that the times are slightly out of date
2023 * and/or different from the directory change time.
2025 dir->i_mtime = dir->i_ctime = current_time(dir);
2026 ext4_update_dx_flag(dir);
2027 inode_inc_iversion(dir);
2028 ext4_mark_inode_dirty(handle, dir);
2029 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2030 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2032 ext4_std_error(dir->i_sb, err);
2037 * This converts a one block unindexed directory to a 3 block indexed
2038 * directory, and adds the dentry to the indexed directory.
2040 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2042 struct inode *inode, struct buffer_head *bh)
2044 struct buffer_head *bh2;
2045 struct dx_root *root;
2046 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2047 struct dx_entry *entries;
2048 struct ext4_dir_entry_2 *de, *de2;
2054 struct fake_dirent *fde;
2057 if (ext4_has_metadata_csum(inode->i_sb))
2058 csum_size = sizeof(struct ext4_dir_entry_tail);
2060 blocksize = dir->i_sb->s_blocksize;
2061 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2062 BUFFER_TRACE(bh, "get_write_access");
2063 retval = ext4_journal_get_write_access(handle, bh);
2065 ext4_std_error(dir->i_sb, retval);
2069 root = (struct dx_root *) bh->b_data;
2071 /* The 0th block becomes the root, move the dirents out */
2072 fde = &root->dotdot;
2073 de = (struct ext4_dir_entry_2 *)((char *)fde +
2074 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2075 if ((char *) de >= (((char *) root) + blocksize)) {
2076 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2078 return -EFSCORRUPTED;
2080 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2082 /* Allocate new block for the 0th block's dirents */
2083 bh2 = ext4_append(handle, dir, &block);
2086 return PTR_ERR(bh2);
2088 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2089 data2 = bh2->b_data;
2091 memcpy(data2, de, len);
2092 de = (struct ext4_dir_entry_2 *) data2;
2094 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2096 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2097 (char *) de, blocksize);
2100 ext4_initialize_dirent_tail(bh2, blocksize);
2102 /* Initialize the root; the dot dirents already exist */
2103 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2104 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2106 memset (&root->info, 0, sizeof(root->info));
2107 root->info.info_length = sizeof(root->info);
2108 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2109 entries = root->entries;
2110 dx_set_block(entries, 1);
2111 dx_set_count(entries, 1);
2112 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2114 /* Initialize as for dx_probe */
2115 fname->hinfo.hash_version = root->info.hash_version;
2116 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2117 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2118 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2119 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2121 memset(frames, 0, sizeof(frames));
2123 frame->entries = entries;
2124 frame->at = entries;
2127 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2130 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2134 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2136 retval = PTR_ERR(de);
2140 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2143 * Even if the block split failed, we have to properly write
2144 * out all the changes we did so far. Otherwise we can end up
2145 * with corrupted filesystem.
2148 ext4_mark_inode_dirty(handle, dir);
2157 * adds a file entry to the specified directory, using the same
2158 * semantics as ext4_find_entry(). It returns NULL if it failed.
2160 * NOTE!! The inode part of 'de' is left at 0 - which means you
2161 * may not sleep between calling this and putting something into
2162 * the entry, as someone else might have used it while you slept.
2164 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2165 struct inode *inode)
2167 struct inode *dir = d_inode(dentry->d_parent);
2168 struct buffer_head *bh = NULL;
2169 struct ext4_dir_entry_2 *de;
2170 struct super_block *sb;
2171 #ifdef CONFIG_UNICODE
2172 struct ext4_sb_info *sbi;
2174 struct ext4_filename fname;
2178 ext4_lblk_t block, blocks;
2181 if (ext4_has_metadata_csum(inode->i_sb))
2182 csum_size = sizeof(struct ext4_dir_entry_tail);
2185 blocksize = sb->s_blocksize;
2186 if (!dentry->d_name.len)
2189 #ifdef CONFIG_UNICODE
2191 if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2192 sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
2196 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2200 if (ext4_has_inline_data(dir)) {
2201 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2211 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2212 if (!retval || (retval != ERR_BAD_DX_DIR))
2214 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2216 ext4_mark_inode_dirty(handle, dir);
2218 blocks = dir->i_size >> sb->s_blocksize_bits;
2219 for (block = 0; block < blocks; block++) {
2220 bh = ext4_read_dirblock(dir, block, DIRENT);
2222 bh = ext4_bread(handle, dir, block,
2223 EXT4_GET_BLOCKS_CREATE);
2224 goto add_to_new_block;
2227 retval = PTR_ERR(bh);
2231 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2233 if (retval != -ENOSPC)
2236 if (blocks == 1 && !dx_fallback &&
2237 ext4_has_feature_dir_index(sb)) {
2238 retval = make_indexed_dir(handle, &fname, dir,
2240 bh = NULL; /* make_indexed_dir releases bh */
2245 bh = ext4_append(handle, dir, &block);
2248 retval = PTR_ERR(bh);
2252 de = (struct ext4_dir_entry_2 *) bh->b_data;
2254 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2257 ext4_initialize_dirent_tail(bh, blocksize);
2259 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2261 ext4_fname_free_filename(&fname);
2264 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2269 * Returns 0 for success, or a negative error value
2271 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2272 struct inode *dir, struct inode *inode)
2274 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2275 struct dx_entry *entries, *at;
2276 struct buffer_head *bh;
2277 struct super_block *sb = dir->i_sb;
2278 struct ext4_dir_entry_2 *de;
2284 frame = dx_probe(fname, dir, NULL, frames);
2286 return PTR_ERR(frame);
2287 entries = frame->entries;
2289 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2296 BUFFER_TRACE(bh, "get_write_access");
2297 err = ext4_journal_get_write_access(handle, bh);
2301 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2306 /* Block full, should compress but for now just split */
2307 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2308 dx_get_count(entries), dx_get_limit(entries)));
2309 /* Need to split index? */
2310 if (dx_get_count(entries) == dx_get_limit(entries)) {
2311 ext4_lblk_t newblock;
2312 int levels = frame - frames + 1;
2313 unsigned int icount;
2315 struct dx_entry *entries2;
2316 struct dx_node *node2;
2317 struct buffer_head *bh2;
2319 while (frame > frames) {
2320 if (dx_get_count((frame - 1)->entries) <
2321 dx_get_limit((frame - 1)->entries)) {
2325 frame--; /* split higher index block */
2327 entries = frame->entries;
2330 if (add_level && levels == ext4_dir_htree_level(sb)) {
2331 ext4_warning(sb, "Directory (ino: %lu) index full, "
2332 "reach max htree level :%d",
2333 dir->i_ino, levels);
2334 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2335 ext4_warning(sb, "Large directory feature is "
2336 "not enabled on this "
2342 icount = dx_get_count(entries);
2343 bh2 = ext4_append(handle, dir, &newblock);
2348 node2 = (struct dx_node *)(bh2->b_data);
2349 entries2 = node2->entries;
2350 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2351 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2353 BUFFER_TRACE(frame->bh, "get_write_access");
2354 err = ext4_journal_get_write_access(handle, frame->bh);
2358 unsigned icount1 = icount/2, icount2 = icount - icount1;
2359 unsigned hash2 = dx_get_hash(entries + icount1);
2360 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2363 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2364 err = ext4_journal_get_write_access(handle,
2369 memcpy((char *) entries2, (char *) (entries + icount1),
2370 icount2 * sizeof(struct dx_entry));
2371 dx_set_count(entries, icount1);
2372 dx_set_count(entries2, icount2);
2373 dx_set_limit(entries2, dx_node_limit(dir));
2375 /* Which index block gets the new entry? */
2376 if (at - entries >= icount1) {
2377 frame->at = at = at - entries - icount1 + entries2;
2378 frame->entries = entries = entries2;
2379 swap(frame->bh, bh2);
2381 dx_insert_block((frame - 1), hash2, newblock);
2382 dxtrace(dx_show_index("node", frame->entries));
2383 dxtrace(dx_show_index("node",
2384 ((struct dx_node *) bh2->b_data)->entries));
2385 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2389 err = ext4_handle_dirty_dx_node(handle, dir,
2394 err = ext4_handle_dirty_dx_node(handle, dir,
2399 struct dx_root *dxroot;
2400 memcpy((char *) entries2, (char *) entries,
2401 icount * sizeof(struct dx_entry));
2402 dx_set_limit(entries2, dx_node_limit(dir));
2405 dx_set_count(entries, 1);
2406 dx_set_block(entries + 0, newblock);
2407 dxroot = (struct dx_root *)frames[0].bh->b_data;
2408 dxroot->info.indirect_levels += 1;
2409 dxtrace(printk(KERN_DEBUG
2410 "Creating %d level index...\n",
2411 dxroot->info.indirect_levels));
2412 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2415 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2421 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2426 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2430 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2434 /* @restart is true means htree-path has been changed, we need to
2435 * repeat dx_probe() to find out valid htree-path
2437 if (restart && err == 0)
2443 * ext4_generic_delete_entry deletes a directory entry by merging it
2444 * with the previous entry
2446 int ext4_generic_delete_entry(handle_t *handle,
2448 struct ext4_dir_entry_2 *de_del,
2449 struct buffer_head *bh,
2454 struct ext4_dir_entry_2 *de, *pde;
2455 unsigned int blocksize = dir->i_sb->s_blocksize;
2460 de = (struct ext4_dir_entry_2 *)entry_buf;
2461 while (i < buf_size - csum_size) {
2462 if (ext4_check_dir_entry(dir, NULL, de, bh,
2463 bh->b_data, bh->b_size, i))
2464 return -EFSCORRUPTED;
2467 pde->rec_len = ext4_rec_len_to_disk(
2468 ext4_rec_len_from_disk(pde->rec_len,
2470 ext4_rec_len_from_disk(de->rec_len,
2475 inode_inc_iversion(dir);
2478 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2480 de = ext4_next_entry(de, blocksize);
2485 static int ext4_delete_entry(handle_t *handle,
2487 struct ext4_dir_entry_2 *de_del,
2488 struct buffer_head *bh)
2490 int err, csum_size = 0;
2492 if (ext4_has_inline_data(dir)) {
2493 int has_inline_data = 1;
2494 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2496 if (has_inline_data)
2500 if (ext4_has_metadata_csum(dir->i_sb))
2501 csum_size = sizeof(struct ext4_dir_entry_tail);
2503 BUFFER_TRACE(bh, "get_write_access");
2504 err = ext4_journal_get_write_access(handle, bh);
2508 err = ext4_generic_delete_entry(handle, dir, de_del,
2510 dir->i_sb->s_blocksize, csum_size);
2514 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2515 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2522 ext4_std_error(dir->i_sb, err);
2527 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2528 * since this indicates that nlinks count was previously 1 to avoid overflowing
2529 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2530 * that subdirectory link counts are not being maintained accurately.
2532 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2533 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2534 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2535 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2537 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2541 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2542 set_nlink(inode, 1);
2546 * If a directory had nlink == 1, then we should let it be 1. This indicates
2547 * directory has >EXT4_LINK_MAX subdirs.
2549 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2551 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2557 * Add non-directory inode to a directory. On success, the inode reference is
2558 * consumed by dentry is instantiation. This is also indicated by clearing of
2559 * *inodep pointer. On failure, the caller is responsible for dropping the
2560 * inode reference in the safe context.
2562 static int ext4_add_nondir(handle_t *handle,
2563 struct dentry *dentry, struct inode **inodep)
2565 struct inode *dir = d_inode(dentry->d_parent);
2566 struct inode *inode = *inodep;
2567 int err = ext4_add_entry(handle, dentry, inode);
2569 ext4_mark_inode_dirty(handle, inode);
2570 if (IS_DIRSYNC(dir))
2571 ext4_handle_sync(handle);
2572 d_instantiate_new(dentry, inode);
2577 ext4_orphan_add(handle, inode);
2578 unlock_new_inode(inode);
2583 * By the time this is called, we already have created
2584 * the directory cache entry for the new file, but it
2585 * is so far negative - it has no inode.
2587 * If the create succeeds, we fill in the inode information
2588 * with d_instantiate().
2590 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2594 struct inode *inode;
2595 int err, credits, retries = 0;
2597 err = dquot_initialize(dir);
2601 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2602 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2604 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2605 NULL, EXT4_HT_DIR, credits);
2606 handle = ext4_journal_current_handle();
2607 err = PTR_ERR(inode);
2608 if (!IS_ERR(inode)) {
2609 inode->i_op = &ext4_file_inode_operations;
2610 inode->i_fop = &ext4_file_operations;
2611 ext4_set_aops(inode);
2612 err = ext4_add_nondir(handle, dentry, &inode);
2615 ext4_journal_stop(handle);
2616 if (!IS_ERR_OR_NULL(inode))
2618 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2623 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2624 umode_t mode, dev_t rdev)
2627 struct inode *inode;
2628 int err, credits, retries = 0;
2630 err = dquot_initialize(dir);
2634 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2635 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2637 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2638 NULL, EXT4_HT_DIR, credits);
2639 handle = ext4_journal_current_handle();
2640 err = PTR_ERR(inode);
2641 if (!IS_ERR(inode)) {
2642 init_special_inode(inode, inode->i_mode, rdev);
2643 inode->i_op = &ext4_special_inode_operations;
2644 err = ext4_add_nondir(handle, dentry, &inode);
2647 ext4_journal_stop(handle);
2648 if (!IS_ERR_OR_NULL(inode))
2650 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2655 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2658 struct inode *inode;
2659 int err, retries = 0;
2661 err = dquot_initialize(dir);
2666 inode = ext4_new_inode_start_handle(dir, mode,
2669 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2670 4 + EXT4_XATTR_TRANS_BLOCKS);
2671 handle = ext4_journal_current_handle();
2672 err = PTR_ERR(inode);
2673 if (!IS_ERR(inode)) {
2674 inode->i_op = &ext4_file_inode_operations;
2675 inode->i_fop = &ext4_file_operations;
2676 ext4_set_aops(inode);
2677 d_tmpfile(dentry, inode);
2678 err = ext4_orphan_add(handle, inode);
2680 goto err_unlock_inode;
2681 mark_inode_dirty(inode);
2682 unlock_new_inode(inode);
2685 ext4_journal_stop(handle);
2686 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2690 ext4_journal_stop(handle);
2691 unlock_new_inode(inode);
2695 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2696 struct ext4_dir_entry_2 *de,
2697 int blocksize, int csum_size,
2698 unsigned int parent_ino, int dotdot_real_len)
2700 de->inode = cpu_to_le32(inode->i_ino);
2702 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2704 strcpy(de->name, ".");
2705 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2707 de = ext4_next_entry(de, blocksize);
2708 de->inode = cpu_to_le32(parent_ino);
2710 if (!dotdot_real_len)
2711 de->rec_len = ext4_rec_len_to_disk(blocksize -
2712 (csum_size + EXT4_DIR_REC_LEN(1)),
2715 de->rec_len = ext4_rec_len_to_disk(
2716 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2717 strcpy(de->name, "..");
2718 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2720 return ext4_next_entry(de, blocksize);
2723 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2724 struct inode *inode)
2726 struct buffer_head *dir_block = NULL;
2727 struct ext4_dir_entry_2 *de;
2728 ext4_lblk_t block = 0;
2729 unsigned int blocksize = dir->i_sb->s_blocksize;
2733 if (ext4_has_metadata_csum(dir->i_sb))
2734 csum_size = sizeof(struct ext4_dir_entry_tail);
2736 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2737 err = ext4_try_create_inline_dir(handle, dir, inode);
2738 if (err < 0 && err != -ENOSPC)
2745 dir_block = ext4_append(handle, inode, &block);
2746 if (IS_ERR(dir_block))
2747 return PTR_ERR(dir_block);
2748 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2749 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2750 set_nlink(inode, 2);
2752 ext4_initialize_dirent_tail(dir_block, blocksize);
2754 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2755 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2758 set_buffer_verified(dir_block);
2764 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2767 struct inode *inode;
2768 int err, credits, retries = 0;
2770 if (EXT4_DIR_LINK_MAX(dir))
2773 err = dquot_initialize(dir);
2777 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2778 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2780 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2782 0, NULL, EXT4_HT_DIR, credits);
2783 handle = ext4_journal_current_handle();
2784 err = PTR_ERR(inode);
2788 inode->i_op = &ext4_dir_inode_operations;
2789 inode->i_fop = &ext4_dir_operations;
2790 err = ext4_init_new_dir(handle, dir, inode);
2792 goto out_clear_inode;
2793 err = ext4_mark_inode_dirty(handle, inode);
2795 err = ext4_add_entry(handle, dentry, inode);
2799 ext4_orphan_add(handle, inode);
2800 unlock_new_inode(inode);
2801 ext4_mark_inode_dirty(handle, inode);
2802 ext4_journal_stop(handle);
2806 ext4_inc_count(handle, dir);
2807 ext4_update_dx_flag(dir);
2808 err = ext4_mark_inode_dirty(handle, dir);
2810 goto out_clear_inode;
2811 d_instantiate_new(dentry, inode);
2812 if (IS_DIRSYNC(dir))
2813 ext4_handle_sync(handle);
2817 ext4_journal_stop(handle);
2819 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2825 * routine to check that the specified directory is empty (for rmdir)
2827 bool ext4_empty_dir(struct inode *inode)
2829 unsigned int offset;
2830 struct buffer_head *bh;
2831 struct ext4_dir_entry_2 *de;
2832 struct super_block *sb;
2834 if (ext4_has_inline_data(inode)) {
2835 int has_inline_data = 1;
2838 ret = empty_inline_dir(inode, &has_inline_data);
2839 if (has_inline_data)
2844 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2845 EXT4_ERROR_INODE(inode, "invalid size");
2848 /* The first directory block must not be a hole,
2849 * so treat it as DIRENT_HTREE
2851 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2855 de = (struct ext4_dir_entry_2 *) bh->b_data;
2856 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2858 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2859 ext4_warning_inode(inode, "directory missing '.'");
2863 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2864 de = ext4_next_entry(de, sb->s_blocksize);
2865 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2867 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2868 ext4_warning_inode(inode, "directory missing '..'");
2872 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2873 while (offset < inode->i_size) {
2874 if (!(offset & (sb->s_blocksize - 1))) {
2875 unsigned int lblock;
2877 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2878 bh = ext4_read_dirblock(inode, lblock, EITHER);
2880 offset += sb->s_blocksize;
2886 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2887 (offset & (sb->s_blocksize - 1)));
2888 if (ext4_check_dir_entry(inode, NULL, de, bh,
2889 bh->b_data, bh->b_size, offset)) {
2890 offset = (offset | (sb->s_blocksize - 1)) + 1;
2893 if (le32_to_cpu(de->inode)) {
2897 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2904 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2905 * such inodes, starting at the superblock, in case we crash before the
2906 * file is closed/deleted, or in case the inode truncate spans multiple
2907 * transactions and the last transaction is not recovered after a crash.
2909 * At filesystem recovery time, we walk this list deleting unlinked
2910 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2912 * Orphan list manipulation functions must be called under i_mutex unless
2913 * we are just creating the inode or deleting it.
2915 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2917 struct super_block *sb = inode->i_sb;
2918 struct ext4_sb_info *sbi = EXT4_SB(sb);
2919 struct ext4_iloc iloc;
2923 if (!sbi->s_journal || is_bad_inode(inode))
2926 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2927 !inode_is_locked(inode));
2929 * Exit early if inode already is on orphan list. This is a big speedup
2930 * since we don't have to contend on the global s_orphan_lock.
2932 if (!list_empty(&EXT4_I(inode)->i_orphan))
2936 * Orphan handling is only valid for files with data blocks
2937 * being truncated, or files being unlinked. Note that we either
2938 * hold i_mutex, or the inode can not be referenced from outside,
2939 * so i_nlink should not be bumped due to race
2941 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2942 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2944 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2945 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2949 err = ext4_reserve_inode_write(handle, inode, &iloc);
2953 mutex_lock(&sbi->s_orphan_lock);
2955 * Due to previous errors inode may be already a part of on-disk
2956 * orphan list. If so skip on-disk list modification.
2958 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2959 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2960 /* Insert this inode at the head of the on-disk orphan list */
2961 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2962 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2965 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2966 mutex_unlock(&sbi->s_orphan_lock);
2969 err = ext4_handle_dirty_super(handle, sb);
2970 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2975 * We have to remove inode from in-memory list if
2976 * addition to on disk orphan list failed. Stray orphan
2977 * list entries can cause panics at unmount time.
2979 mutex_lock(&sbi->s_orphan_lock);
2980 list_del_init(&EXT4_I(inode)->i_orphan);
2981 mutex_unlock(&sbi->s_orphan_lock);
2986 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2987 jbd_debug(4, "orphan inode %lu will point to %d\n",
2988 inode->i_ino, NEXT_ORPHAN(inode));
2990 ext4_std_error(sb, err);
2995 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2996 * of such inodes stored on disk, because it is finally being cleaned up.
2998 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3000 struct list_head *prev;
3001 struct ext4_inode_info *ei = EXT4_I(inode);
3002 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3004 struct ext4_iloc iloc;
3007 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3010 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3011 !inode_is_locked(inode));
3012 /* Do this quick check before taking global s_orphan_lock. */
3013 if (list_empty(&ei->i_orphan))
3017 /* Grab inode buffer early before taking global s_orphan_lock */
3018 err = ext4_reserve_inode_write(handle, inode, &iloc);
3021 mutex_lock(&sbi->s_orphan_lock);
3022 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3024 prev = ei->i_orphan.prev;
3025 list_del_init(&ei->i_orphan);
3027 /* If we're on an error path, we may not have a valid
3028 * transaction handle with which to update the orphan list on
3029 * disk, but we still need to remove the inode from the linked
3030 * list in memory. */
3031 if (!handle || err) {
3032 mutex_unlock(&sbi->s_orphan_lock);
3036 ino_next = NEXT_ORPHAN(inode);
3037 if (prev == &sbi->s_orphan) {
3038 jbd_debug(4, "superblock will point to %u\n", ino_next);
3039 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3040 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3042 mutex_unlock(&sbi->s_orphan_lock);
3045 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3046 mutex_unlock(&sbi->s_orphan_lock);
3047 err = ext4_handle_dirty_super(handle, inode->i_sb);
3049 struct ext4_iloc iloc2;
3050 struct inode *i_prev =
3051 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3053 jbd_debug(4, "orphan inode %lu will point to %u\n",
3054 i_prev->i_ino, ino_next);
3055 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3057 mutex_unlock(&sbi->s_orphan_lock);
3060 NEXT_ORPHAN(i_prev) = ino_next;
3061 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3062 mutex_unlock(&sbi->s_orphan_lock);
3066 NEXT_ORPHAN(inode) = 0;
3067 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3069 ext4_std_error(inode->i_sb, err);
3077 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3080 struct inode *inode;
3081 struct buffer_head *bh;
3082 struct ext4_dir_entry_2 *de;
3083 handle_t *handle = NULL;
3085 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3088 /* Initialize quotas before so that eventual writes go in
3089 * separate transaction */
3090 retval = dquot_initialize(dir);
3093 retval = dquot_initialize(d_inode(dentry));
3098 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3104 inode = d_inode(dentry);
3106 retval = -EFSCORRUPTED;
3107 if (le32_to_cpu(de->inode) != inode->i_ino)
3110 retval = -ENOTEMPTY;
3111 if (!ext4_empty_dir(inode))
3114 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3115 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3116 if (IS_ERR(handle)) {
3117 retval = PTR_ERR(handle);
3122 if (IS_DIRSYNC(dir))
3123 ext4_handle_sync(handle);
3125 retval = ext4_delete_entry(handle, dir, de, bh);
3128 if (!EXT4_DIR_LINK_EMPTY(inode))
3129 ext4_warning_inode(inode,
3130 "empty directory '%.*s' has too many links (%u)",
3131 dentry->d_name.len, dentry->d_name.name,
3133 inode_inc_iversion(inode);
3135 /* There's no need to set i_disksize: the fact that i_nlink is
3136 * zero will ensure that the right thing happens during any
3139 ext4_orphan_add(handle, inode);
3140 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3141 ext4_mark_inode_dirty(handle, inode);
3142 ext4_dec_count(handle, dir);
3143 ext4_update_dx_flag(dir);
3144 ext4_mark_inode_dirty(handle, dir);
3146 #ifdef CONFIG_UNICODE
3147 /* VFS negative dentries are incompatible with Encoding and
3148 * Case-insensitiveness. Eventually we'll want avoid
3149 * invalidating the dentries here, alongside with returning the
3150 * negative dentries at ext4_lookup(), when it is better
3151 * supported by the VFS for the CI case.
3153 if (IS_CASEFOLDED(dir))
3154 d_invalidate(dentry);
3160 ext4_journal_stop(handle);
3164 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3167 struct inode *inode;
3168 struct buffer_head *bh;
3169 struct ext4_dir_entry_2 *de;
3170 handle_t *handle = NULL;
3172 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3175 trace_ext4_unlink_enter(dir, dentry);
3176 /* Initialize quotas before so that eventual writes go
3177 * in separate transaction */
3178 retval = dquot_initialize(dir);
3181 retval = dquot_initialize(d_inode(dentry));
3186 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3192 inode = d_inode(dentry);
3194 retval = -EFSCORRUPTED;
3195 if (le32_to_cpu(de->inode) != inode->i_ino)
3198 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3199 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3200 if (IS_ERR(handle)) {
3201 retval = PTR_ERR(handle);
3206 if (IS_DIRSYNC(dir))
3207 ext4_handle_sync(handle);
3209 retval = ext4_delete_entry(handle, dir, de, bh);
3212 dir->i_ctime = dir->i_mtime = current_time(dir);
3213 ext4_update_dx_flag(dir);
3214 ext4_mark_inode_dirty(handle, dir);
3215 if (inode->i_nlink == 0)
3216 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3217 dentry->d_name.len, dentry->d_name.name);
3220 if (!inode->i_nlink)
3221 ext4_orphan_add(handle, inode);
3222 inode->i_ctime = current_time(inode);
3223 ext4_mark_inode_dirty(handle, inode);
3225 #ifdef CONFIG_UNICODE
3226 /* VFS negative dentries are incompatible with Encoding and
3227 * Case-insensitiveness. Eventually we'll want avoid
3228 * invalidating the dentries here, alongside with returning the
3229 * negative dentries at ext4_lookup(), when it is better
3230 * supported by the VFS for the CI case.
3232 if (IS_CASEFOLDED(dir))
3233 d_invalidate(dentry);
3239 ext4_journal_stop(handle);
3240 trace_ext4_unlink_exit(dentry, retval);
3244 static int ext4_symlink(struct inode *dir,
3245 struct dentry *dentry, const char *symname)
3248 struct inode *inode;
3249 int err, len = strlen(symname);
3251 struct fscrypt_str disk_link;
3253 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3256 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3261 err = dquot_initialize(dir);
3265 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3267 * For non-fast symlinks, we just allocate inode and put it on
3268 * orphan list in the first transaction => we need bitmap,
3269 * group descriptor, sb, inode block, quota blocks, and
3270 * possibly selinux xattr blocks.
3272 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3273 EXT4_XATTR_TRANS_BLOCKS;
3276 * Fast symlink. We have to add entry to directory
3277 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3278 * allocate new inode (bitmap, group descriptor, inode block,
3279 * quota blocks, sb is already counted in previous macros).
3281 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3282 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3285 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3286 &dentry->d_name, 0, NULL,
3287 EXT4_HT_DIR, credits);
3288 handle = ext4_journal_current_handle();
3289 if (IS_ERR(inode)) {
3291 ext4_journal_stop(handle);
3292 return PTR_ERR(inode);
3295 if (IS_ENCRYPTED(inode)) {
3296 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3298 goto err_drop_inode;
3299 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3302 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3303 if (!IS_ENCRYPTED(inode))
3304 inode->i_op = &ext4_symlink_inode_operations;
3305 inode_nohighmem(inode);
3306 ext4_set_aops(inode);
3308 * We cannot call page_symlink() with transaction started
3309 * because it calls into ext4_write_begin() which can wait
3310 * for transaction commit if we are running out of space
3311 * and thus we deadlock. So we have to stop transaction now
3312 * and restart it when symlink contents is written.
3314 * To keep fs consistent in case of crash, we have to put inode
3315 * to orphan list in the mean time.
3318 err = ext4_orphan_add(handle, inode);
3319 ext4_journal_stop(handle);
3322 goto err_drop_inode;
3323 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3325 goto err_drop_inode;
3327 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3328 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3330 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3331 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3332 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3333 if (IS_ERR(handle)) {
3334 err = PTR_ERR(handle);
3336 goto err_drop_inode;
3338 set_nlink(inode, 1);
3339 err = ext4_orphan_del(handle, inode);
3341 goto err_drop_inode;
3343 /* clear the extent format for fast symlink */
3344 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3345 if (!IS_ENCRYPTED(inode)) {
3346 inode->i_op = &ext4_fast_symlink_inode_operations;
3347 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3349 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3351 inode->i_size = disk_link.len - 1;
3353 EXT4_I(inode)->i_disksize = inode->i_size;
3354 err = ext4_add_nondir(handle, dentry, &inode);
3356 ext4_journal_stop(handle);
3359 goto out_free_encrypted_link;
3363 ext4_journal_stop(handle);
3365 unlock_new_inode(inode);
3367 out_free_encrypted_link:
3368 if (disk_link.name != (unsigned char *)symname)
3369 kfree(disk_link.name);
3373 static int ext4_link(struct dentry *old_dentry,
3374 struct inode *dir, struct dentry *dentry)
3377 struct inode *inode = d_inode(old_dentry);
3378 int err, retries = 0;
3380 if (inode->i_nlink >= EXT4_LINK_MAX)
3383 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3387 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3388 (!projid_eq(EXT4_I(dir)->i_projid,
3389 EXT4_I(old_dentry->d_inode)->i_projid)))
3392 err = dquot_initialize(dir);
3397 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3398 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3399 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3401 return PTR_ERR(handle);
3403 if (IS_DIRSYNC(dir))
3404 ext4_handle_sync(handle);
3406 inode->i_ctime = current_time(inode);
3407 ext4_inc_count(handle, inode);
3410 err = ext4_add_entry(handle, dentry, inode);
3412 ext4_mark_inode_dirty(handle, inode);
3413 /* this can happen only for tmpfile being
3414 * linked the first time
3416 if (inode->i_nlink == 1)
3417 ext4_orphan_del(handle, inode);
3418 d_instantiate(dentry, inode);
3423 ext4_journal_stop(handle);
3424 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3431 * Try to find buffer head where contains the parent block.
3432 * It should be the inode block if it is inlined or the 1st block
3433 * if it is a normal dir.
3435 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3436 struct inode *inode,
3438 struct ext4_dir_entry_2 **parent_de,
3441 struct buffer_head *bh;
3443 if (!ext4_has_inline_data(inode)) {
3444 /* The first directory block must not be a hole, so
3445 * treat it as DIRENT_HTREE
3447 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3449 *retval = PTR_ERR(bh);
3452 *parent_de = ext4_next_entry(
3453 (struct ext4_dir_entry_2 *)bh->b_data,
3454 inode->i_sb->s_blocksize);
3459 return ext4_get_first_inline_block(inode, parent_de, retval);
3462 struct ext4_renament {
3464 struct dentry *dentry;
3465 struct inode *inode;
3467 int dir_nlink_delta;
3469 /* entry for "dentry" */
3470 struct buffer_head *bh;
3471 struct ext4_dir_entry_2 *de;
3474 /* entry for ".." in inode if it's a directory */
3475 struct buffer_head *dir_bh;
3476 struct ext4_dir_entry_2 *parent_de;
3480 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3484 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3485 &retval, &ent->parent_de,
3489 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3490 return -EFSCORRUPTED;
3491 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3492 return ext4_journal_get_write_access(handle, ent->dir_bh);
3495 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3500 ent->parent_de->inode = cpu_to_le32(dir_ino);
3501 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3502 if (!ent->dir_inlined) {
3503 if (is_dx(ent->inode)) {
3504 retval = ext4_handle_dirty_dx_node(handle,
3508 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3512 retval = ext4_mark_inode_dirty(handle, ent->inode);
3515 ext4_std_error(ent->dir->i_sb, retval);
3521 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3522 unsigned ino, unsigned file_type)
3526 BUFFER_TRACE(ent->bh, "get write access");
3527 retval = ext4_journal_get_write_access(handle, ent->bh);
3530 ent->de->inode = cpu_to_le32(ino);
3531 if (ext4_has_feature_filetype(ent->dir->i_sb))
3532 ent->de->file_type = file_type;
3533 inode_inc_iversion(ent->dir);
3534 ent->dir->i_ctime = ent->dir->i_mtime =
3535 current_time(ent->dir);
3536 ext4_mark_inode_dirty(handle, ent->dir);
3537 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3538 if (!ent->inlined) {
3539 retval = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3540 if (unlikely(retval)) {
3541 ext4_std_error(ent->dir->i_sb, retval);
3551 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3552 const struct qstr *d_name)
3554 int retval = -ENOENT;
3555 struct buffer_head *bh;
3556 struct ext4_dir_entry_2 *de;
3558 bh = ext4_find_entry(dir, d_name, &de, NULL);
3562 retval = ext4_delete_entry(handle, dir, de, bh);
3568 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3573 * ent->de could have moved from under us during htree split, so make
3574 * sure that we are deleting the right entry. We might also be pointing
3575 * to a stale entry in the unused part of ent->bh so just checking inum
3576 * and the name isn't enough.
3578 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3579 ent->de->name_len != ent->dentry->d_name.len ||
3580 strncmp(ent->de->name, ent->dentry->d_name.name,
3581 ent->de->name_len) ||
3583 retval = ext4_find_delete_entry(handle, ent->dir,
3584 &ent->dentry->d_name);
3586 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3587 if (retval == -ENOENT) {
3588 retval = ext4_find_delete_entry(handle, ent->dir,
3589 &ent->dentry->d_name);
3594 ext4_warning_inode(ent->dir,
3595 "Deleting old file: nlink %d, error=%d",
3596 ent->dir->i_nlink, retval);
3600 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3602 if (ent->dir_nlink_delta) {
3603 if (ent->dir_nlink_delta == -1)
3604 ext4_dec_count(handle, ent->dir);
3606 ext4_inc_count(handle, ent->dir);
3607 ext4_mark_inode_dirty(handle, ent->dir);
3611 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3612 int credits, handle_t **h)
3619 * for inode block, sb block, group summaries,
3622 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3623 EXT4_XATTR_TRANS_BLOCKS + 4);
3625 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3626 &ent->dentry->d_name, 0, NULL,
3627 EXT4_HT_DIR, credits);
3629 handle = ext4_journal_current_handle();
3632 ext4_journal_stop(handle);
3633 if (PTR_ERR(wh) == -ENOSPC &&
3634 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3638 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3639 wh->i_op = &ext4_special_inode_operations;
3645 * Anybody can rename anything with this: the permission checks are left to the
3646 * higher-level routines.
3648 * n.b. old_{dentry,inode) refers to the source dentry/inode
3649 * while new_{dentry,inode) refers to the destination dentry/inode
3650 * This comes from rename(const char *oldpath, const char *newpath)
3652 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3653 struct inode *new_dir, struct dentry *new_dentry,
3656 handle_t *handle = NULL;
3657 struct ext4_renament old = {
3659 .dentry = old_dentry,
3660 .inode = d_inode(old_dentry),
3662 struct ext4_renament new = {
3664 .dentry = new_dentry,
3665 .inode = d_inode(new_dentry),
3669 struct inode *whiteout = NULL;
3673 if (new.inode && new.inode->i_nlink == 0) {
3674 EXT4_ERROR_INODE(new.inode,
3675 "target of rename is already freed");
3676 return -EFSCORRUPTED;
3679 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3680 (!projid_eq(EXT4_I(new_dir)->i_projid,
3681 EXT4_I(old_dentry->d_inode)->i_projid)))
3684 retval = dquot_initialize(old.dir);
3687 retval = dquot_initialize(new.dir);
3691 /* Initialize quotas before so that eventual writes go
3692 * in separate transaction */
3694 retval = dquot_initialize(new.inode);
3699 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3701 return PTR_ERR(old.bh);
3703 * Check for inode number is _not_ due to possible IO errors.
3704 * We might rmdir the source, keep it as pwd of some process
3705 * and merrily kill the link to whatever was created under the
3706 * same name. Goodbye sticky bit ;-<
3709 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3712 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3713 &new.de, &new.inlined);
3714 if (IS_ERR(new.bh)) {
3715 retval = PTR_ERR(new.bh);
3725 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3726 ext4_alloc_da_blocks(old.inode);
3728 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3729 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3730 if (!(flags & RENAME_WHITEOUT)) {
3731 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3732 if (IS_ERR(handle)) {
3733 retval = PTR_ERR(handle);
3738 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3739 if (IS_ERR(whiteout)) {
3740 retval = PTR_ERR(whiteout);
3746 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3747 ext4_handle_sync(handle);
3749 if (S_ISDIR(old.inode->i_mode)) {
3751 retval = -ENOTEMPTY;
3752 if (!ext4_empty_dir(new.inode))
3756 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3759 retval = ext4_rename_dir_prepare(handle, &old);
3764 * If we're renaming a file within an inline_data dir and adding or
3765 * setting the new dirent causes a conversion from inline_data to
3766 * extents/blockmap, we need to force the dirent delete code to
3767 * re-read the directory, or else we end up trying to delete a dirent
3768 * from what is now the extent tree root (or a block map).
3770 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3771 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3773 old_file_type = old.de->file_type;
3776 * Do this before adding a new entry, so the old entry is sure
3777 * to be still pointing to the valid old entry.
3779 retval = ext4_setent(handle, &old, whiteout->i_ino,
3783 ext4_mark_inode_dirty(handle, whiteout);
3786 retval = ext4_add_entry(handle, new.dentry, old.inode);
3790 retval = ext4_setent(handle, &new,
3791 old.inode->i_ino, old_file_type);
3796 force_reread = !ext4_test_inode_flag(new.dir,
3797 EXT4_INODE_INLINE_DATA);
3800 * Like most other Unix systems, set the ctime for inodes on a
3803 old.inode->i_ctime = current_time(old.inode);
3804 ext4_mark_inode_dirty(handle, old.inode);
3810 ext4_rename_delete(handle, &old, force_reread);
3814 ext4_dec_count(handle, new.inode);
3815 new.inode->i_ctime = current_time(new.inode);
3817 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3818 ext4_update_dx_flag(old.dir);
3820 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3824 ext4_dec_count(handle, old.dir);
3826 /* checked ext4_empty_dir above, can't have another
3827 * parent, ext4_dec_count() won't work for many-linked
3829 clear_nlink(new.inode);
3831 ext4_inc_count(handle, new.dir);
3832 ext4_update_dx_flag(new.dir);
3833 ext4_mark_inode_dirty(handle, new.dir);
3836 ext4_mark_inode_dirty(handle, old.dir);
3838 ext4_mark_inode_dirty(handle, new.inode);
3839 if (!new.inode->i_nlink)
3840 ext4_orphan_add(handle, new.inode);
3850 drop_nlink(whiteout);
3851 unlock_new_inode(whiteout);
3855 ext4_journal_stop(handle);
3859 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3860 struct inode *new_dir, struct dentry *new_dentry)
3862 handle_t *handle = NULL;
3863 struct ext4_renament old = {
3865 .dentry = old_dentry,
3866 .inode = d_inode(old_dentry),
3868 struct ext4_renament new = {
3870 .dentry = new_dentry,
3871 .inode = d_inode(new_dentry),
3875 struct timespec64 ctime;
3877 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3878 !projid_eq(EXT4_I(new_dir)->i_projid,
3879 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3880 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3881 !projid_eq(EXT4_I(old_dir)->i_projid,
3882 EXT4_I(new_dentry->d_inode)->i_projid)))
3885 retval = dquot_initialize(old.dir);
3888 retval = dquot_initialize(new.dir);
3892 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3893 &old.de, &old.inlined);
3895 return PTR_ERR(old.bh);
3897 * Check for inode number is _not_ due to possible IO errors.
3898 * We might rmdir the source, keep it as pwd of some process
3899 * and merrily kill the link to whatever was created under the
3900 * same name. Goodbye sticky bit ;-<
3903 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3906 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3907 &new.de, &new.inlined);
3908 if (IS_ERR(new.bh)) {
3909 retval = PTR_ERR(new.bh);
3914 /* RENAME_EXCHANGE case: old *and* new must both exist */
3915 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3918 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3919 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3920 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3921 if (IS_ERR(handle)) {
3922 retval = PTR_ERR(handle);
3927 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3928 ext4_handle_sync(handle);
3930 if (S_ISDIR(old.inode->i_mode)) {
3932 retval = ext4_rename_dir_prepare(handle, &old);
3936 if (S_ISDIR(new.inode->i_mode)) {
3938 retval = ext4_rename_dir_prepare(handle, &new);
3944 * Other than the special case of overwriting a directory, parents'
3945 * nlink only needs to be modified if this is a cross directory rename.
3947 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3948 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3949 new.dir_nlink_delta = -old.dir_nlink_delta;
3951 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3952 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3956 new_file_type = new.de->file_type;
3957 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3961 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3966 * Like most other Unix systems, set the ctime for inodes on a
3969 ctime = current_time(old.inode);
3970 old.inode->i_ctime = ctime;
3971 new.inode->i_ctime = ctime;
3972 ext4_mark_inode_dirty(handle, old.inode);
3973 ext4_mark_inode_dirty(handle, new.inode);
3976 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3981 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3985 ext4_update_dir_count(handle, &old);
3986 ext4_update_dir_count(handle, &new);
3995 ext4_journal_stop(handle);
3999 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4000 struct inode *new_dir, struct dentry *new_dentry,
4005 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4008 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4011 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4016 if (flags & RENAME_EXCHANGE) {
4017 return ext4_cross_rename(old_dir, old_dentry,
4018 new_dir, new_dentry);
4021 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4025 * directories can handle most operations...
4027 const struct inode_operations ext4_dir_inode_operations = {
4028 .create = ext4_create,
4029 .lookup = ext4_lookup,
4031 .unlink = ext4_unlink,
4032 .symlink = ext4_symlink,
4033 .mkdir = ext4_mkdir,
4034 .rmdir = ext4_rmdir,
4035 .mknod = ext4_mknod,
4036 .tmpfile = ext4_tmpfile,
4037 .rename = ext4_rename2,
4038 .setattr = ext4_setattr,
4039 .getattr = ext4_getattr,
4040 .listxattr = ext4_listxattr,
4041 .get_acl = ext4_get_acl,
4042 .set_acl = ext4_set_acl,
4043 .fiemap = ext4_fiemap,
4046 const struct inode_operations ext4_special_inode_operations = {
4047 .setattr = ext4_setattr,
4048 .getattr = ext4_getattr,
4049 .listxattr = ext4_listxattr,
4050 .get_acl = ext4_get_acl,
4051 .set_acl = ext4_set_acl,