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>
39 #include "ext4_jbd2.h"
44 #include <trace/events/ext4.h>
46 * define how far ahead to read directories while searching them.
48 #define NAMEI_RA_CHUNKS 2
49 #define NAMEI_RA_BLOCKS 4
50 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52 static struct buffer_head *ext4_append(handle_t *handle,
56 struct buffer_head *bh;
59 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
60 ((inode->i_size >> 10) >=
61 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
62 return ERR_PTR(-ENOSPC);
64 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
66 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
69 inode->i_size += inode->i_sb->s_blocksize;
70 EXT4_I(inode)->i_disksize = inode->i_size;
71 BUFFER_TRACE(bh, "get_write_access");
72 err = ext4_journal_get_write_access(handle, bh);
75 ext4_std_error(inode->i_sb, err);
81 static int ext4_dx_csum_verify(struct inode *inode,
82 struct ext4_dir_entry *dirent);
88 #define ext4_read_dirblock(inode, block, type) \
89 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
91 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
97 struct buffer_head *bh;
98 struct ext4_dir_entry *dirent;
101 bh = ext4_bread(NULL, inode, block, 0);
103 __ext4_warning(inode->i_sb, func, line,
104 "inode #%lu: lblock %lu: comm %s: "
105 "error %ld reading directory block",
106 inode->i_ino, (unsigned long)block,
107 current->comm, PTR_ERR(bh));
112 ext4_error_inode(inode, func, line, block,
113 "Directory hole found");
114 return ERR_PTR(-EFSCORRUPTED);
116 dirent = (struct ext4_dir_entry *) bh->b_data;
117 /* Determine whether or not we have an index block */
121 else if (ext4_rec_len_from_disk(dirent->rec_len,
122 inode->i_sb->s_blocksize) ==
123 inode->i_sb->s_blocksize)
126 if (!is_dx_block && type == INDEX) {
127 ext4_error_inode(inode, func, line, block,
128 "directory leaf block found instead of index block");
129 return ERR_PTR(-EFSCORRUPTED);
131 if (!ext4_has_metadata_csum(inode->i_sb) ||
136 * An empty leaf block can get mistaken for a index block; for
137 * this reason, we can only check the index checksum when the
138 * caller is sure it should be an index block.
140 if (is_dx_block && type == INDEX) {
141 if (ext4_dx_csum_verify(inode, dirent))
142 set_buffer_verified(bh);
144 ext4_error_inode(inode, func, line, block,
145 "Directory index failed checksum");
147 return ERR_PTR(-EFSBADCRC);
151 if (ext4_dirent_csum_verify(inode, dirent))
152 set_buffer_verified(bh);
154 ext4_error_inode(inode, func, line, block,
155 "Directory block failed checksum");
157 return ERR_PTR(-EFSBADCRC);
164 #define assert(test) J_ASSERT(test)
168 #define dxtrace(command) command
170 #define dxtrace(command)
194 * dx_root_info is laid out so that if it should somehow get overlaid by a
195 * dirent the two low bits of the hash version will be zero. Therefore, the
196 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
201 struct fake_dirent dot;
203 struct fake_dirent dotdot;
207 __le32 reserved_zero;
209 u8 info_length; /* 8 */
214 struct dx_entry entries[0];
219 struct fake_dirent fake;
220 struct dx_entry entries[0];
226 struct buffer_head *bh;
227 struct dx_entry *entries;
239 * This goes at the end of each htree block.
243 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
246 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
247 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
248 static inline unsigned dx_get_hash(struct dx_entry *entry);
249 static void dx_set_hash(struct dx_entry *entry, unsigned value);
250 static unsigned dx_get_count(struct dx_entry *entries);
251 static unsigned dx_get_limit(struct dx_entry *entries);
252 static void dx_set_count(struct dx_entry *entries, unsigned value);
253 static void dx_set_limit(struct dx_entry *entries, unsigned value);
254 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
255 static unsigned dx_node_limit(struct inode *dir);
256 static struct dx_frame *dx_probe(struct ext4_filename *fname,
258 struct dx_hash_info *hinfo,
259 struct dx_frame *frame);
260 static void dx_release(struct dx_frame *frames);
261 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
262 unsigned blocksize, struct dx_hash_info *hinfo,
263 struct dx_map_entry map[]);
264 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
265 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
266 struct dx_map_entry *offsets, int count, unsigned blocksize);
267 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
268 static void dx_insert_block(struct dx_frame *frame,
269 u32 hash, ext4_lblk_t block);
270 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
271 struct dx_frame *frame,
272 struct dx_frame *frames,
274 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
275 struct ext4_filename *fname,
276 struct ext4_dir_entry_2 **res_dir);
277 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
278 struct inode *dir, struct inode *inode);
280 /* checksumming functions */
281 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
282 unsigned int blocksize)
284 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
285 t->det_rec_len = ext4_rec_len_to_disk(
286 sizeof(struct ext4_dir_entry_tail), blocksize);
287 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
290 /* Walk through a dirent block to find a checksum "dirent" at the tail */
291 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
292 struct ext4_dir_entry *de)
294 struct ext4_dir_entry_tail *t;
297 struct ext4_dir_entry *d, *top;
300 top = (struct ext4_dir_entry *)(((void *)de) +
301 (EXT4_BLOCK_SIZE(inode->i_sb) -
302 sizeof(struct ext4_dir_entry_tail)));
303 while (d < top && d->rec_len)
304 d = (struct ext4_dir_entry *)(((void *)d) +
305 le16_to_cpu(d->rec_len));
310 t = (struct ext4_dir_entry_tail *)d;
312 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
315 if (t->det_reserved_zero1 ||
316 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
317 t->det_reserved_zero2 ||
318 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
324 static __le32 ext4_dirent_csum(struct inode *inode,
325 struct ext4_dir_entry *dirent, int size)
327 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
328 struct ext4_inode_info *ei = EXT4_I(inode);
331 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
332 return cpu_to_le32(csum);
335 #define warn_no_space_for_csum(inode) \
336 __warn_no_space_for_csum((inode), __func__, __LINE__)
338 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
341 __ext4_warning_inode(inode, func, line,
342 "No space for directory leaf checksum. Please run e2fsck -D.");
345 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
347 struct ext4_dir_entry_tail *t;
349 if (!ext4_has_metadata_csum(inode->i_sb))
352 t = get_dirent_tail(inode, dirent);
354 warn_no_space_for_csum(inode);
358 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
359 (void *)t - (void *)dirent))
365 static void ext4_dirent_csum_set(struct inode *inode,
366 struct ext4_dir_entry *dirent)
368 struct ext4_dir_entry_tail *t;
370 if (!ext4_has_metadata_csum(inode->i_sb))
373 t = get_dirent_tail(inode, dirent);
375 warn_no_space_for_csum(inode);
379 t->det_checksum = ext4_dirent_csum(inode, dirent,
380 (void *)t - (void *)dirent);
383 int ext4_handle_dirty_dirent_node(handle_t *handle,
385 struct buffer_head *bh)
387 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
388 return ext4_handle_dirty_metadata(handle, inode, bh);
391 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
392 struct ext4_dir_entry *dirent,
395 struct ext4_dir_entry *dp;
396 struct dx_root_info *root;
399 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
401 else if (le16_to_cpu(dirent->rec_len) == 12) {
402 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
403 if (le16_to_cpu(dp->rec_len) !=
404 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
406 root = (struct dx_root_info *)(((void *)dp + 12));
407 if (root->reserved_zero ||
408 root->info_length != sizeof(struct dx_root_info))
415 *offset = count_offset;
416 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
419 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
420 int count_offset, int count, struct dx_tail *t)
422 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
423 struct ext4_inode_info *ei = EXT4_I(inode);
426 __u32 dummy_csum = 0;
427 int offset = offsetof(struct dx_tail, dt_checksum);
429 size = count_offset + (count * sizeof(struct dx_entry));
430 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
431 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
432 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
434 return cpu_to_le32(csum);
437 static int ext4_dx_csum_verify(struct inode *inode,
438 struct ext4_dir_entry *dirent)
440 struct dx_countlimit *c;
442 int count_offset, limit, count;
444 if (!ext4_has_metadata_csum(inode->i_sb))
447 c = get_dx_countlimit(inode, dirent, &count_offset);
449 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
452 limit = le16_to_cpu(c->limit);
453 count = le16_to_cpu(c->count);
454 if (count_offset + (limit * sizeof(struct dx_entry)) >
455 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
456 warn_no_space_for_csum(inode);
459 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
461 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
467 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
469 struct dx_countlimit *c;
471 int count_offset, limit, count;
473 if (!ext4_has_metadata_csum(inode->i_sb))
476 c = get_dx_countlimit(inode, dirent, &count_offset);
478 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
481 limit = le16_to_cpu(c->limit);
482 count = le16_to_cpu(c->count);
483 if (count_offset + (limit * sizeof(struct dx_entry)) >
484 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
485 warn_no_space_for_csum(inode);
488 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
490 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
493 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
495 struct buffer_head *bh)
497 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
498 return ext4_handle_dirty_metadata(handle, inode, bh);
502 * p is at least 6 bytes before the end of page
504 static inline struct ext4_dir_entry_2 *
505 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
507 return (struct ext4_dir_entry_2 *)((char *)p +
508 ext4_rec_len_from_disk(p->rec_len, blocksize));
512 * Future: use high four bits of block for coalesce-on-delete flags
513 * Mask them off for now.
516 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
518 return le32_to_cpu(entry->block) & 0x0fffffff;
521 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
523 entry->block = cpu_to_le32(value);
526 static inline unsigned dx_get_hash(struct dx_entry *entry)
528 return le32_to_cpu(entry->hash);
531 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
533 entry->hash = cpu_to_le32(value);
536 static inline unsigned dx_get_count(struct dx_entry *entries)
538 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
541 static inline unsigned dx_get_limit(struct dx_entry *entries)
543 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
546 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
548 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
551 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
553 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
556 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
558 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
559 EXT4_DIR_REC_LEN(2) - infosize;
561 if (ext4_has_metadata_csum(dir->i_sb))
562 entry_space -= sizeof(struct dx_tail);
563 return entry_space / sizeof(struct dx_entry);
566 static inline unsigned dx_node_limit(struct inode *dir)
568 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
570 if (ext4_has_metadata_csum(dir->i_sb))
571 entry_space -= sizeof(struct dx_tail);
572 return entry_space / sizeof(struct dx_entry);
579 static void dx_show_index(char * label, struct dx_entry *entries)
581 int i, n = dx_get_count (entries);
582 printk(KERN_DEBUG "%s index", label);
583 for (i = 0; i < n; i++) {
584 printk(KERN_CONT " %x->%lu",
585 i ? dx_get_hash(entries + i) : 0,
586 (unsigned long)dx_get_block(entries + i));
588 printk(KERN_CONT "\n");
598 static struct stats dx_show_leaf(struct inode *dir,
599 struct dx_hash_info *hinfo,
600 struct ext4_dir_entry_2 *de,
601 int size, int show_names)
603 unsigned names = 0, space = 0;
604 char *base = (char *) de;
605 struct dx_hash_info h = *hinfo;
608 while ((char *) de < base + size)
614 #ifdef CONFIG_EXT4_FS_ENCRYPTION
617 struct fscrypt_str fname_crypto_str =
623 if (ext4_encrypted_inode(dir))
624 res = fscrypt_get_encryption_info(dir);
626 printk(KERN_WARNING "Error setting up"
627 " fname crypto: %d\n", res);
629 if (!fscrypt_has_encryption_key(dir)) {
630 /* Directory is not encrypted */
631 ext4fs_dirhash(de->name,
633 printk("%*.s:(U)%x.%u ", len,
635 (unsigned) ((char *) de
638 struct fscrypt_str de_name =
639 FSTR_INIT(name, len);
641 /* Directory is encrypted */
642 res = fscrypt_fname_alloc_buffer(
646 printk(KERN_WARNING "Error "
650 res = fscrypt_fname_disk_to_usr(dir,
654 printk(KERN_WARNING "Error "
655 "converting filename "
661 name = fname_crypto_str.name;
662 len = fname_crypto_str.len;
664 ext4fs_dirhash(de->name, de->name_len,
666 printk("%*.s:(E)%x.%u ", len, name,
667 h.hash, (unsigned) ((char *) de
669 fscrypt_fname_free_buffer(
673 int len = de->name_len;
674 char *name = de->name;
675 ext4fs_dirhash(de->name, de->name_len, &h);
676 printk("%*.s:%x.%u ", len, name, h.hash,
677 (unsigned) ((char *) de - base));
680 space += EXT4_DIR_REC_LEN(de->name_len);
683 de = ext4_next_entry(de, size);
685 printk(KERN_CONT "(%i)\n", names);
686 return (struct stats) { names, space, 1 };
689 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
690 struct dx_entry *entries, int levels)
692 unsigned blocksize = dir->i_sb->s_blocksize;
693 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
695 struct buffer_head *bh;
696 printk("%i indexed blocks...\n", count);
697 for (i = 0; i < count; i++, entries++)
699 ext4_lblk_t block = dx_get_block(entries);
700 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
701 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
703 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
704 bh = ext4_bread(NULL,dir, block, 0);
705 if (!bh || IS_ERR(bh))
708 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
709 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
710 bh->b_data, blocksize, 0);
711 names += stats.names;
712 space += stats.space;
713 bcount += stats.bcount;
717 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
718 levels ? "" : " ", names, space/bcount,
719 (space/bcount)*100/blocksize);
720 return (struct stats) { names, space, bcount};
722 #endif /* DX_DEBUG */
725 * Probe for a directory leaf block to search.
727 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
728 * error in the directory index, and the caller should fall back to
729 * searching the directory normally. The callers of dx_probe **MUST**
730 * check for this error code, and make sure it never gets reflected
733 static struct dx_frame *
734 dx_probe(struct ext4_filename *fname, struct inode *dir,
735 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
737 unsigned count, indirect;
738 struct dx_entry *at, *entries, *p, *q, *m;
739 struct dx_root *root;
740 struct dx_frame *frame = frame_in;
741 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
744 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
745 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
746 if (IS_ERR(frame->bh))
747 return (struct dx_frame *) frame->bh;
749 root = (struct dx_root *) frame->bh->b_data;
750 if (root->info.hash_version != DX_HASH_TEA &&
751 root->info.hash_version != DX_HASH_HALF_MD4 &&
752 root->info.hash_version != DX_HASH_LEGACY) {
753 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
754 root->info.hash_version);
758 hinfo = &fname->hinfo;
759 hinfo->hash_version = root->info.hash_version;
760 if (hinfo->hash_version <= DX_HASH_TEA)
761 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
762 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
763 if (fname && fname_name(fname))
764 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
767 if (root->info.unused_flags & 1) {
768 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
769 root->info.unused_flags);
773 indirect = root->info.indirect_levels;
774 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
775 ext4_warning(dir->i_sb,
776 "Directory (ino: %lu) htree depth %#06x exceed"
777 "supported value", dir->i_ino,
778 ext4_dir_htree_level(dir->i_sb));
779 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
780 ext4_warning(dir->i_sb, "Enable large directory "
781 "feature to access it");
786 entries = (struct dx_entry *)(((char *)&root->info) +
787 root->info.info_length);
789 if (dx_get_limit(entries) != dx_root_limit(dir,
790 root->info.info_length)) {
791 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
792 dx_get_limit(entries),
793 dx_root_limit(dir, root->info.info_length));
797 dxtrace(printk("Look up %x", hash));
799 count = dx_get_count(entries);
800 if (!count || count > dx_get_limit(entries)) {
801 ext4_warning_inode(dir,
802 "dx entry: count %u beyond limit %u",
803 count, dx_get_limit(entries));
808 q = entries + count - 1;
811 dxtrace(printk(KERN_CONT "."));
812 if (dx_get_hash(m) > hash)
818 if (0) { // linear search cross check
819 unsigned n = count - 1;
823 dxtrace(printk(KERN_CONT ","));
824 if (dx_get_hash(++at) > hash)
830 assert (at == p - 1);
834 dxtrace(printk(KERN_CONT " %x->%u\n",
835 at == entries ? 0 : dx_get_hash(at),
837 frame->entries = entries;
842 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
843 if (IS_ERR(frame->bh)) {
844 ret_err = (struct dx_frame *) frame->bh;
848 entries = ((struct dx_node *) frame->bh->b_data)->entries;
850 if (dx_get_limit(entries) != dx_node_limit(dir)) {
851 ext4_warning_inode(dir,
852 "dx entry: limit %u != node limit %u",
853 dx_get_limit(entries), dx_node_limit(dir));
858 while (frame >= frame_in) {
863 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
864 ext4_warning_inode(dir,
865 "Corrupt directory, running e2fsck is recommended");
869 static void dx_release(struct dx_frame *frames)
871 struct dx_root_info *info;
874 if (frames[0].bh == NULL)
877 info = &((struct dx_root *)frames[0].bh->b_data)->info;
878 for (i = 0; i <= info->indirect_levels; i++) {
879 if (frames[i].bh == NULL)
881 brelse(frames[i].bh);
887 * This function increments the frame pointer to search the next leaf
888 * block, and reads in the necessary intervening nodes if the search
889 * should be necessary. Whether or not the search is necessary is
890 * controlled by the hash parameter. If the hash value is even, then
891 * the search is only continued if the next block starts with that
892 * hash value. This is used if we are searching for a specific file.
894 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
896 * This function returns 1 if the caller should continue to search,
897 * or 0 if it should not. If there is an error reading one of the
898 * index blocks, it will a negative error code.
900 * If start_hash is non-null, it will be filled in with the starting
901 * hash of the next page.
903 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
904 struct dx_frame *frame,
905 struct dx_frame *frames,
909 struct buffer_head *bh;
915 * Find the next leaf page by incrementing the frame pointer.
916 * If we run out of entries in the interior node, loop around and
917 * increment pointer in the parent node. When we break out of
918 * this loop, num_frames indicates the number of interior
919 * nodes need to be read.
922 if (++(p->at) < p->entries + dx_get_count(p->entries))
931 * If the hash is 1, then continue only if the next page has a
932 * continuation hash of any value. This is used for readdir
933 * handling. Otherwise, check to see if the hash matches the
934 * desired contiuation hash. If it doesn't, return since
935 * there's no point to read in the successive index pages.
937 bhash = dx_get_hash(p->at);
940 if ((hash & 1) == 0) {
941 if ((bhash & ~1) != hash)
945 * If the hash is HASH_NB_ALWAYS, we always go to the next
946 * block so no check is necessary
948 while (num_frames--) {
949 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
955 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
962 * This function fills a red-black tree with information from a
963 * directory block. It returns the number directory entries loaded
964 * into the tree. If there is an error it is returned in err.
966 static int htree_dirblock_to_tree(struct file *dir_file,
967 struct inode *dir, ext4_lblk_t block,
968 struct dx_hash_info *hinfo,
969 __u32 start_hash, __u32 start_minor_hash)
971 struct buffer_head *bh;
972 struct ext4_dir_entry_2 *de, *top;
973 int err = 0, count = 0;
974 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
976 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
977 (unsigned long)block));
978 bh = ext4_read_dirblock(dir, block, DIRENT);
982 de = (struct ext4_dir_entry_2 *) bh->b_data;
983 top = (struct ext4_dir_entry_2 *) ((char *) de +
984 dir->i_sb->s_blocksize -
985 EXT4_DIR_REC_LEN(0));
986 #ifdef CONFIG_EXT4_FS_ENCRYPTION
987 /* Check if the directory is encrypted */
988 if (ext4_encrypted_inode(dir)) {
989 err = fscrypt_get_encryption_info(dir);
994 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1002 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1003 if (ext4_check_dir_entry(dir, NULL, de, bh,
1004 bh->b_data, bh->b_size,
1005 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1006 + ((char *)de - bh->b_data))) {
1007 /* silently ignore the rest of the block */
1010 ext4fs_dirhash(de->name, de->name_len, hinfo);
1011 if ((hinfo->hash < start_hash) ||
1012 ((hinfo->hash == start_hash) &&
1013 (hinfo->minor_hash < start_minor_hash)))
1017 if (!ext4_encrypted_inode(dir)) {
1018 tmp_str.name = de->name;
1019 tmp_str.len = de->name_len;
1020 err = ext4_htree_store_dirent(dir_file,
1021 hinfo->hash, hinfo->minor_hash, de,
1024 int save_len = fname_crypto_str.len;
1025 struct fscrypt_str de_name = FSTR_INIT(de->name,
1028 /* Directory is encrypted */
1029 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1030 hinfo->minor_hash, &de_name,
1036 err = ext4_htree_store_dirent(dir_file,
1037 hinfo->hash, hinfo->minor_hash, de,
1039 fname_crypto_str.len = save_len;
1049 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1050 fscrypt_fname_free_buffer(&fname_crypto_str);
1057 * This function fills a red-black tree with information from a
1058 * directory. We start scanning the directory in hash order, starting
1059 * at start_hash and start_minor_hash.
1061 * This function returns the number of entries inserted into the tree,
1062 * or a negative error code.
1064 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1065 __u32 start_minor_hash, __u32 *next_hash)
1067 struct dx_hash_info hinfo;
1068 struct ext4_dir_entry_2 *de;
1069 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1075 struct fscrypt_str tmp_str;
1077 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1078 start_hash, start_minor_hash));
1079 dir = file_inode(dir_file);
1080 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1081 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1082 if (hinfo.hash_version <= DX_HASH_TEA)
1083 hinfo.hash_version +=
1084 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1085 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1086 if (ext4_has_inline_data(dir)) {
1087 int has_inline_data = 1;
1088 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1092 if (has_inline_data) {
1097 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1098 start_hash, start_minor_hash);
1102 hinfo.hash = start_hash;
1103 hinfo.minor_hash = 0;
1104 frame = dx_probe(NULL, dir, &hinfo, frames);
1106 return PTR_ERR(frame);
1108 /* Add '.' and '..' from the htree header */
1109 if (!start_hash && !start_minor_hash) {
1110 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1111 tmp_str.name = de->name;
1112 tmp_str.len = de->name_len;
1113 err = ext4_htree_store_dirent(dir_file, 0, 0,
1119 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1120 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1121 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1122 tmp_str.name = de->name;
1123 tmp_str.len = de->name_len;
1124 err = ext4_htree_store_dirent(dir_file, 2, 0,
1132 if (fatal_signal_pending(current)) {
1137 block = dx_get_block(frame->at);
1138 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1139 start_hash, start_minor_hash);
1146 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1147 frame, frames, &hashval);
1148 *next_hash = hashval;
1154 * Stop if: (a) there are no more entries, or
1155 * (b) we have inserted at least one entry and the
1156 * next hash value is not a continuation
1159 (count && ((hashval & 1) == 0)))
1163 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1164 "next hash: %x\n", count, *next_hash));
1171 static inline int search_dirblock(struct buffer_head *bh,
1173 struct ext4_filename *fname,
1174 unsigned int offset,
1175 struct ext4_dir_entry_2 **res_dir)
1177 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1178 fname, offset, res_dir);
1182 * Directory block splitting, compacting
1186 * Create map of hash values, offsets, and sizes, stored at end of block.
1187 * Returns number of entries mapped.
1189 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1190 unsigned blocksize, struct dx_hash_info *hinfo,
1191 struct dx_map_entry *map_tail)
1194 char *base = (char *) de;
1195 struct dx_hash_info h = *hinfo;
1197 while ((char *) de < base + blocksize) {
1198 if (de->name_len && de->inode) {
1199 ext4fs_dirhash(de->name, de->name_len, &h);
1201 map_tail->hash = h.hash;
1202 map_tail->offs = ((char *) de - base)>>2;
1203 map_tail->size = le16_to_cpu(de->rec_len);
1207 /* XXX: do we need to check rec_len == 0 case? -Chris */
1208 de = ext4_next_entry(de, blocksize);
1213 /* Sort map by hash value */
1214 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1216 struct dx_map_entry *p, *q, *top = map + count - 1;
1218 /* Combsort until bubble sort doesn't suck */
1220 count = count*10/13;
1221 if (count - 9 < 2) /* 9, 10 -> 11 */
1223 for (p = top, q = p - count; q >= map; p--, q--)
1224 if (p->hash < q->hash)
1227 /* Garden variety bubble sort */
1232 if (q[1].hash >= q[0].hash)
1240 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1242 struct dx_entry *entries = frame->entries;
1243 struct dx_entry *old = frame->at, *new = old + 1;
1244 int count = dx_get_count(entries);
1246 assert(count < dx_get_limit(entries));
1247 assert(old < entries + count);
1248 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1249 dx_set_hash(new, hash);
1250 dx_set_block(new, block);
1251 dx_set_count(entries, count + 1);
1255 * Test whether a directory entry matches the filename being searched for.
1257 * Return: %true if the directory entry matches, otherwise %false.
1259 static inline bool ext4_match(const struct ext4_filename *fname,
1260 const struct ext4_dir_entry_2 *de)
1262 struct fscrypt_name f;
1267 f.usr_fname = fname->usr_fname;
1268 f.disk_name = fname->disk_name;
1269 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1270 f.crypto_buf = fname->crypto_buf;
1272 return fscrypt_match_name(&f, de->name, de->name_len);
1276 * Returns 0 if not found, -1 on failure, and 1 on success
1278 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1279 struct inode *dir, struct ext4_filename *fname,
1280 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1282 struct ext4_dir_entry_2 * de;
1286 de = (struct ext4_dir_entry_2 *)search_buf;
1287 dlimit = search_buf + buf_size;
1288 while ((char *) de < dlimit) {
1289 /* this code is executed quadratically often */
1290 /* do minimal checking `by hand' */
1291 if ((char *) de + de->name_len <= dlimit &&
1292 ext4_match(fname, de)) {
1293 /* found a match - just to be sure, do
1295 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1296 bh->b_size, offset))
1301 /* prevent looping on a bad block */
1302 de_len = ext4_rec_len_from_disk(de->rec_len,
1303 dir->i_sb->s_blocksize);
1307 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1312 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1313 struct ext4_dir_entry *de)
1315 struct super_block *sb = dir->i_sb;
1321 if (de->inode == 0 &&
1322 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1331 * finds an entry in the specified directory with the wanted name. It
1332 * returns the cache buffer in which the entry was found, and the entry
1333 * itself (as a parameter - res_dir). It does NOT read the inode of the
1334 * entry - you'll have to do that yourself if you want to.
1336 * The returned buffer_head has ->b_count elevated. The caller is expected
1337 * to brelse() it when appropriate.
1339 static struct buffer_head * ext4_find_entry (struct inode *dir,
1340 const struct qstr *d_name,
1341 struct ext4_dir_entry_2 **res_dir,
1344 struct super_block *sb;
1345 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1346 struct buffer_head *bh, *ret = NULL;
1347 ext4_lblk_t start, block;
1348 const u8 *name = d_name->name;
1349 size_t ra_max = 0; /* Number of bh's in the readahead
1351 size_t ra_ptr = 0; /* Current index into readahead
1353 ext4_lblk_t nblocks;
1354 int i, namelen, retval;
1355 struct ext4_filename fname;
1359 namelen = d_name->len;
1360 if (namelen > EXT4_NAME_LEN)
1363 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1364 if (retval == -ENOENT)
1367 return ERR_PTR(retval);
1369 if (ext4_has_inline_data(dir)) {
1370 int has_inline_data = 1;
1371 ret = ext4_find_inline_entry(dir, &fname, res_dir,
1373 if (has_inline_data) {
1376 goto cleanup_and_exit;
1380 if ((namelen <= 2) && (name[0] == '.') &&
1381 (name[1] == '.' || name[1] == '\0')) {
1383 * "." or ".." will only be in the first block
1384 * NFS may look up ".."; "." should be handled by the VFS
1391 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1393 * On success, or if the error was file not found,
1394 * return. Otherwise, fall back to doing a search the
1395 * old fashioned way.
1397 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1398 goto cleanup_and_exit;
1399 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1402 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1403 start = EXT4_I(dir)->i_dir_start_lookup;
1404 if (start >= nblocks)
1410 * We deal with the read-ahead logic here.
1412 if (ra_ptr >= ra_max) {
1413 /* Refill the readahead buffer */
1416 ra_max = start - block;
1418 ra_max = nblocks - block;
1419 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1420 retval = ext4_bread_batch(dir, block, ra_max,
1421 false /* wait */, bh_use);
1423 ret = ERR_PTR(retval);
1425 goto cleanup_and_exit;
1428 if ((bh = bh_use[ra_ptr++]) == NULL)
1431 if (!buffer_uptodate(bh)) {
1432 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1433 (unsigned long) block);
1435 ret = ERR_PTR(-EIO);
1436 goto cleanup_and_exit;
1438 if (!buffer_verified(bh) &&
1439 !is_dx_internal_node(dir, block,
1440 (struct ext4_dir_entry *)bh->b_data) &&
1441 !ext4_dirent_csum_verify(dir,
1442 (struct ext4_dir_entry *)bh->b_data)) {
1443 EXT4_ERROR_INODE(dir, "checksumming directory "
1444 "block %lu", (unsigned long)block);
1446 ret = ERR_PTR(-EFSBADCRC);
1447 goto cleanup_and_exit;
1449 set_buffer_verified(bh);
1450 i = search_dirblock(bh, dir, &fname,
1451 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1453 EXT4_I(dir)->i_dir_start_lookup = block;
1455 goto cleanup_and_exit;
1459 goto cleanup_and_exit;
1462 if (++block >= nblocks)
1464 } while (block != start);
1467 * If the directory has grown while we were searching, then
1468 * search the last part of the directory before giving up.
1471 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1472 if (block < nblocks) {
1478 /* Clean up the read-ahead blocks */
1479 for (; ra_ptr < ra_max; ra_ptr++)
1480 brelse(bh_use[ra_ptr]);
1481 ext4_fname_free_filename(&fname);
1485 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1486 struct ext4_filename *fname,
1487 struct ext4_dir_entry_2 **res_dir)
1489 struct super_block * sb = dir->i_sb;
1490 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1491 struct buffer_head *bh;
1495 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1498 frame = dx_probe(fname, dir, NULL, frames);
1500 return (struct buffer_head *) frame;
1502 block = dx_get_block(frame->at);
1503 bh = ext4_read_dirblock(dir, block, DIRENT);
1507 retval = search_dirblock(bh, dir, fname,
1508 block << EXT4_BLOCK_SIZE_BITS(sb),
1514 bh = ERR_PTR(ERR_BAD_DX_DIR);
1518 /* Check to see if we should continue to search */
1519 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1522 ext4_warning_inode(dir,
1523 "error %d reading directory index block",
1525 bh = ERR_PTR(retval);
1528 } while (retval == 1);
1532 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1538 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1540 struct inode *inode;
1541 struct ext4_dir_entry_2 *de;
1542 struct buffer_head *bh;
1545 err = fscrypt_prepare_lookup(dir, dentry, flags);
1547 return ERR_PTR(err);
1549 if (dentry->d_name.len > EXT4_NAME_LEN)
1550 return ERR_PTR(-ENAMETOOLONG);
1552 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1554 return (struct dentry *) bh;
1557 __u32 ino = le32_to_cpu(de->inode);
1559 if (!ext4_valid_inum(dir->i_sb, ino)) {
1560 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1561 return ERR_PTR(-EFSCORRUPTED);
1563 if (unlikely(ino == dir->i_ino)) {
1564 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1566 return ERR_PTR(-EFSCORRUPTED);
1568 inode = ext4_iget_normal(dir->i_sb, ino);
1569 if (inode == ERR_PTR(-ESTALE)) {
1570 EXT4_ERROR_INODE(dir,
1571 "deleted inode referenced: %u",
1573 return ERR_PTR(-EFSCORRUPTED);
1575 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1576 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1577 !fscrypt_has_permitted_context(dir, inode)) {
1578 ext4_warning(inode->i_sb,
1579 "Inconsistent encryption contexts: %lu/%lu",
1580 dir->i_ino, inode->i_ino);
1582 return ERR_PTR(-EPERM);
1585 return d_splice_alias(inode, dentry);
1589 struct dentry *ext4_get_parent(struct dentry *child)
1592 static const struct qstr dotdot = QSTR_INIT("..", 2);
1593 struct ext4_dir_entry_2 * de;
1594 struct buffer_head *bh;
1596 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1598 return (struct dentry *) bh;
1600 return ERR_PTR(-ENOENT);
1601 ino = le32_to_cpu(de->inode);
1604 if (!ext4_valid_inum(child->d_sb, ino)) {
1605 EXT4_ERROR_INODE(d_inode(child),
1606 "bad parent inode number: %u", ino);
1607 return ERR_PTR(-EFSCORRUPTED);
1610 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1614 * Move count entries from end of map between two memory locations.
1615 * Returns pointer to last entry moved.
1617 static struct ext4_dir_entry_2 *
1618 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1621 unsigned rec_len = 0;
1624 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1625 (from + (map->offs<<2));
1626 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1627 memcpy (to, de, rec_len);
1628 ((struct ext4_dir_entry_2 *) to)->rec_len =
1629 ext4_rec_len_to_disk(rec_len, blocksize);
1634 return (struct ext4_dir_entry_2 *) (to - rec_len);
1638 * Compact each dir entry in the range to the minimal rec_len.
1639 * Returns pointer to last entry in range.
1641 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1643 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1644 unsigned rec_len = 0;
1647 while ((char*)de < base + blocksize) {
1648 next = ext4_next_entry(de, blocksize);
1649 if (de->inode && de->name_len) {
1650 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1652 memmove(to, de, rec_len);
1653 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1655 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1663 * Split a full leaf block to make room for a new dir entry.
1664 * Allocate a new block, and move entries so that they are approx. equally full.
1665 * Returns pointer to de in block into which the new entry will be inserted.
1667 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1668 struct buffer_head **bh,struct dx_frame *frame,
1669 struct dx_hash_info *hinfo)
1671 unsigned blocksize = dir->i_sb->s_blocksize;
1672 unsigned count, continued;
1673 struct buffer_head *bh2;
1674 ext4_lblk_t newblock;
1676 struct dx_map_entry *map;
1677 char *data1 = (*bh)->b_data, *data2;
1678 unsigned split, move, size;
1679 struct ext4_dir_entry_2 *de = NULL, *de2;
1680 struct ext4_dir_entry_tail *t;
1684 if (ext4_has_metadata_csum(dir->i_sb))
1685 csum_size = sizeof(struct ext4_dir_entry_tail);
1687 bh2 = ext4_append(handle, dir, &newblock);
1691 return (struct ext4_dir_entry_2 *) bh2;
1694 BUFFER_TRACE(*bh, "get_write_access");
1695 err = ext4_journal_get_write_access(handle, *bh);
1699 BUFFER_TRACE(frame->bh, "get_write_access");
1700 err = ext4_journal_get_write_access(handle, frame->bh);
1704 data2 = bh2->b_data;
1706 /* create map in the end of data2 block */
1707 map = (struct dx_map_entry *) (data2 + blocksize);
1708 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1709 blocksize, hinfo, map);
1711 dx_sort_map(map, count);
1712 /* Split the existing block in the middle, size-wise */
1715 for (i = count-1; i >= 0; i--) {
1716 /* is more than half of this entry in 2nd half of the block? */
1717 if (size + map[i].size/2 > blocksize/2)
1719 size += map[i].size;
1722 /* map index at which we will split */
1723 split = count - move;
1724 hash2 = map[split].hash;
1725 continued = hash2 == map[split - 1].hash;
1726 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1727 (unsigned long)dx_get_block(frame->at),
1728 hash2, split, count-split));
1730 /* Fancy dance to stay within two buffers */
1731 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1733 de = dx_pack_dirents(data1, blocksize);
1734 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1737 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1741 t = EXT4_DIRENT_TAIL(data2, blocksize);
1742 initialize_dirent_tail(t, blocksize);
1744 t = EXT4_DIRENT_TAIL(data1, blocksize);
1745 initialize_dirent_tail(t, blocksize);
1748 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1750 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1753 /* Which block gets the new entry? */
1754 if (hinfo->hash >= hash2) {
1758 dx_insert_block(frame, hash2 + continued, newblock);
1759 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1762 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1766 dxtrace(dx_show_index("frame", frame->entries));
1773 ext4_std_error(dir->i_sb, err);
1774 return ERR_PTR(err);
1777 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1778 struct buffer_head *bh,
1779 void *buf, int buf_size,
1780 struct ext4_filename *fname,
1781 struct ext4_dir_entry_2 **dest_de)
1783 struct ext4_dir_entry_2 *de;
1784 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1786 unsigned int offset = 0;
1789 de = (struct ext4_dir_entry_2 *)buf;
1790 top = buf + buf_size - reclen;
1791 while ((char *) de <= top) {
1792 if (ext4_check_dir_entry(dir, NULL, de, bh,
1793 buf, buf_size, offset))
1794 return -EFSCORRUPTED;
1795 if (ext4_match(fname, de))
1797 nlen = EXT4_DIR_REC_LEN(de->name_len);
1798 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1799 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1801 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1804 if ((char *) de > top)
1811 void ext4_insert_dentry(struct inode *inode,
1812 struct ext4_dir_entry_2 *de,
1814 struct ext4_filename *fname)
1819 nlen = EXT4_DIR_REC_LEN(de->name_len);
1820 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1822 struct ext4_dir_entry_2 *de1 =
1823 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1824 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1825 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1828 de->file_type = EXT4_FT_UNKNOWN;
1829 de->inode = cpu_to_le32(inode->i_ino);
1830 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1831 de->name_len = fname_len(fname);
1832 memcpy(de->name, fname_name(fname), fname_len(fname));
1836 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1837 * it points to a directory entry which is guaranteed to be large
1838 * enough for new directory entry. If de is NULL, then
1839 * add_dirent_to_buf will attempt search the directory block for
1840 * space. It will return -ENOSPC if no space is available, and -EIO
1841 * and -EEXIST if directory entry already exists.
1843 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1845 struct inode *inode, struct ext4_dir_entry_2 *de,
1846 struct buffer_head *bh)
1848 unsigned int blocksize = dir->i_sb->s_blocksize;
1852 if (ext4_has_metadata_csum(inode->i_sb))
1853 csum_size = sizeof(struct ext4_dir_entry_tail);
1856 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1857 blocksize - csum_size, fname, &de);
1861 BUFFER_TRACE(bh, "get_write_access");
1862 err = ext4_journal_get_write_access(handle, bh);
1864 ext4_std_error(dir->i_sb, err);
1868 /* By now the buffer is marked for journaling */
1869 ext4_insert_dentry(inode, de, blocksize, fname);
1872 * XXX shouldn't update any times until successful
1873 * completion of syscall, but too many callers depend
1876 * XXX similarly, too many callers depend on
1877 * ext4_new_inode() setting the times, but error
1878 * recovery deletes the inode, so the worst that can
1879 * happen is that the times are slightly out of date
1880 * and/or different from the directory change time.
1882 dir->i_mtime = dir->i_ctime = current_time(dir);
1883 ext4_update_dx_flag(dir);
1884 inode_inc_iversion(dir);
1885 ext4_mark_inode_dirty(handle, dir);
1886 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1887 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1889 ext4_std_error(dir->i_sb, err);
1894 * This converts a one block unindexed directory to a 3 block indexed
1895 * directory, and adds the dentry to the indexed directory.
1897 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1899 struct inode *inode, struct buffer_head *bh)
1901 struct buffer_head *bh2;
1902 struct dx_root *root;
1903 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1904 struct dx_entry *entries;
1905 struct ext4_dir_entry_2 *de, *de2;
1906 struct ext4_dir_entry_tail *t;
1912 struct fake_dirent *fde;
1915 if (ext4_has_metadata_csum(inode->i_sb))
1916 csum_size = sizeof(struct ext4_dir_entry_tail);
1918 blocksize = dir->i_sb->s_blocksize;
1919 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1920 BUFFER_TRACE(bh, "get_write_access");
1921 retval = ext4_journal_get_write_access(handle, bh);
1923 ext4_std_error(dir->i_sb, retval);
1927 root = (struct dx_root *) bh->b_data;
1929 /* The 0th block becomes the root, move the dirents out */
1930 fde = &root->dotdot;
1931 de = (struct ext4_dir_entry_2 *)((char *)fde +
1932 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1933 if ((char *) de >= (((char *) root) + blocksize)) {
1934 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1936 return -EFSCORRUPTED;
1938 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1940 /* Allocate new block for the 0th block's dirents */
1941 bh2 = ext4_append(handle, dir, &block);
1944 return PTR_ERR(bh2);
1946 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1947 data1 = bh2->b_data;
1949 memcpy (data1, de, len);
1950 de = (struct ext4_dir_entry_2 *) data1;
1952 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1954 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1959 t = EXT4_DIRENT_TAIL(data1, blocksize);
1960 initialize_dirent_tail(t, blocksize);
1963 /* Initialize the root; the dot dirents already exist */
1964 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1965 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1967 memset (&root->info, 0, sizeof(root->info));
1968 root->info.info_length = sizeof(root->info);
1969 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1970 entries = root->entries;
1971 dx_set_block(entries, 1);
1972 dx_set_count(entries, 1);
1973 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1975 /* Initialize as for dx_probe */
1976 fname->hinfo.hash_version = root->info.hash_version;
1977 if (fname->hinfo.hash_version <= DX_HASH_TEA)
1978 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1979 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1980 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1982 memset(frames, 0, sizeof(frames));
1984 frame->entries = entries;
1985 frame->at = entries;
1988 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1991 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1995 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
1997 retval = PTR_ERR(de);
2001 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2004 * Even if the block split failed, we have to properly write
2005 * out all the changes we did so far. Otherwise we can end up
2006 * with corrupted filesystem.
2009 ext4_mark_inode_dirty(handle, dir);
2018 * adds a file entry to the specified directory, using the same
2019 * semantics as ext4_find_entry(). It returns NULL if it failed.
2021 * NOTE!! The inode part of 'de' is left at 0 - which means you
2022 * may not sleep between calling this and putting something into
2023 * the entry, as someone else might have used it while you slept.
2025 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2026 struct inode *inode)
2028 struct inode *dir = d_inode(dentry->d_parent);
2029 struct buffer_head *bh = NULL;
2030 struct ext4_dir_entry_2 *de;
2031 struct ext4_dir_entry_tail *t;
2032 struct super_block *sb;
2033 struct ext4_filename fname;
2037 ext4_lblk_t block, blocks;
2040 if (ext4_has_metadata_csum(inode->i_sb))
2041 csum_size = sizeof(struct ext4_dir_entry_tail);
2044 blocksize = sb->s_blocksize;
2045 if (!dentry->d_name.len)
2048 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2052 if (ext4_has_inline_data(dir)) {
2053 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2063 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2064 if (!retval || (retval != ERR_BAD_DX_DIR))
2066 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2068 ext4_mark_inode_dirty(handle, dir);
2070 blocks = dir->i_size >> sb->s_blocksize_bits;
2071 for (block = 0; block < blocks; block++) {
2072 bh = ext4_read_dirblock(dir, block, DIRENT);
2074 retval = PTR_ERR(bh);
2078 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2080 if (retval != -ENOSPC)
2083 if (blocks == 1 && !dx_fallback &&
2084 ext4_has_feature_dir_index(sb)) {
2085 retval = make_indexed_dir(handle, &fname, dir,
2087 bh = NULL; /* make_indexed_dir releases bh */
2092 bh = ext4_append(handle, dir, &block);
2094 retval = PTR_ERR(bh);
2098 de = (struct ext4_dir_entry_2 *) bh->b_data;
2100 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2103 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2104 initialize_dirent_tail(t, blocksize);
2107 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2109 ext4_fname_free_filename(&fname);
2112 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2117 * Returns 0 for success, or a negative error value
2119 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2120 struct inode *dir, struct inode *inode)
2122 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2123 struct dx_entry *entries, *at;
2124 struct buffer_head *bh;
2125 struct super_block *sb = dir->i_sb;
2126 struct ext4_dir_entry_2 *de;
2132 frame = dx_probe(fname, dir, NULL, frames);
2134 return PTR_ERR(frame);
2135 entries = frame->entries;
2137 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2144 BUFFER_TRACE(bh, "get_write_access");
2145 err = ext4_journal_get_write_access(handle, bh);
2149 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2154 /* Block full, should compress but for now just split */
2155 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2156 dx_get_count(entries), dx_get_limit(entries)));
2157 /* Need to split index? */
2158 if (dx_get_count(entries) == dx_get_limit(entries)) {
2159 ext4_lblk_t newblock;
2160 int levels = frame - frames + 1;
2161 unsigned int icount;
2163 struct dx_entry *entries2;
2164 struct dx_node *node2;
2165 struct buffer_head *bh2;
2167 while (frame > frames) {
2168 if (dx_get_count((frame - 1)->entries) <
2169 dx_get_limit((frame - 1)->entries)) {
2173 frame--; /* split higher index block */
2175 entries = frame->entries;
2178 if (add_level && levels == ext4_dir_htree_level(sb)) {
2179 ext4_warning(sb, "Directory (ino: %lu) index full, "
2180 "reach max htree level :%d",
2181 dir->i_ino, levels);
2182 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2183 ext4_warning(sb, "Large directory feature is "
2184 "not enabled on this "
2190 icount = dx_get_count(entries);
2191 bh2 = ext4_append(handle, dir, &newblock);
2196 node2 = (struct dx_node *)(bh2->b_data);
2197 entries2 = node2->entries;
2198 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2199 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2201 BUFFER_TRACE(frame->bh, "get_write_access");
2202 err = ext4_journal_get_write_access(handle, frame->bh);
2206 unsigned icount1 = icount/2, icount2 = icount - icount1;
2207 unsigned hash2 = dx_get_hash(entries + icount1);
2208 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2211 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2212 err = ext4_journal_get_write_access(handle,
2217 memcpy((char *) entries2, (char *) (entries + icount1),
2218 icount2 * sizeof(struct dx_entry));
2219 dx_set_count(entries, icount1);
2220 dx_set_count(entries2, icount2);
2221 dx_set_limit(entries2, dx_node_limit(dir));
2223 /* Which index block gets the new entry? */
2224 if (at - entries >= icount1) {
2225 frame->at = at = at - entries - icount1 + entries2;
2226 frame->entries = entries = entries2;
2227 swap(frame->bh, bh2);
2229 dx_insert_block((frame - 1), hash2, newblock);
2230 dxtrace(dx_show_index("node", frame->entries));
2231 dxtrace(dx_show_index("node",
2232 ((struct dx_node *) bh2->b_data)->entries));
2233 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2237 err = ext4_handle_dirty_dx_node(handle, dir,
2242 err = ext4_handle_dirty_dx_node(handle, dir,
2247 struct dx_root *dxroot;
2248 memcpy((char *) entries2, (char *) entries,
2249 icount * sizeof(struct dx_entry));
2250 dx_set_limit(entries2, dx_node_limit(dir));
2253 dx_set_count(entries, 1);
2254 dx_set_block(entries + 0, newblock);
2255 dxroot = (struct dx_root *)frames[0].bh->b_data;
2256 dxroot->info.indirect_levels += 1;
2257 dxtrace(printk(KERN_DEBUG
2258 "Creating %d level index...\n",
2259 info->indirect_levels));
2260 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2263 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2269 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2274 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2278 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2282 /* @restart is true means htree-path has been changed, we need to
2283 * repeat dx_probe() to find out valid htree-path
2285 if (restart && err == 0)
2291 * ext4_generic_delete_entry deletes a directory entry by merging it
2292 * with the previous entry
2294 int ext4_generic_delete_entry(handle_t *handle,
2296 struct ext4_dir_entry_2 *de_del,
2297 struct buffer_head *bh,
2302 struct ext4_dir_entry_2 *de, *pde;
2303 unsigned int blocksize = dir->i_sb->s_blocksize;
2308 de = (struct ext4_dir_entry_2 *)entry_buf;
2309 while (i < buf_size - csum_size) {
2310 if (ext4_check_dir_entry(dir, NULL, de, bh,
2311 bh->b_data, bh->b_size, i))
2312 return -EFSCORRUPTED;
2315 pde->rec_len = ext4_rec_len_to_disk(
2316 ext4_rec_len_from_disk(pde->rec_len,
2318 ext4_rec_len_from_disk(de->rec_len,
2323 inode_inc_iversion(dir);
2326 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2328 de = ext4_next_entry(de, blocksize);
2333 static int ext4_delete_entry(handle_t *handle,
2335 struct ext4_dir_entry_2 *de_del,
2336 struct buffer_head *bh)
2338 int err, csum_size = 0;
2340 if (ext4_has_inline_data(dir)) {
2341 int has_inline_data = 1;
2342 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2344 if (has_inline_data)
2348 if (ext4_has_metadata_csum(dir->i_sb))
2349 csum_size = sizeof(struct ext4_dir_entry_tail);
2351 BUFFER_TRACE(bh, "get_write_access");
2352 err = ext4_journal_get_write_access(handle, bh);
2356 err = ext4_generic_delete_entry(handle, dir, de_del,
2358 dir->i_sb->s_blocksize, csum_size);
2362 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2363 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2370 ext4_std_error(dir->i_sb, err);
2375 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2376 * since this indicates that nlinks count was previously 1 to avoid overflowing
2377 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2378 * that subdirectory link counts are not being maintained accurately.
2380 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2381 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2382 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2383 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2385 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2389 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2390 set_nlink(inode, 1);
2394 * If a directory had nlink == 1, then we should let it be 1. This indicates
2395 * directory has >EXT4_LINK_MAX subdirs.
2397 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2399 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2404 static int ext4_add_nondir(handle_t *handle,
2405 struct dentry *dentry, struct inode *inode)
2407 int err = ext4_add_entry(handle, dentry, inode);
2409 ext4_mark_inode_dirty(handle, inode);
2410 unlock_new_inode(inode);
2411 d_instantiate(dentry, inode);
2415 unlock_new_inode(inode);
2421 * By the time this is called, we already have created
2422 * the directory cache entry for the new file, but it
2423 * is so far negative - it has no inode.
2425 * If the create succeeds, we fill in the inode information
2426 * with d_instantiate().
2428 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2432 struct inode *inode;
2433 int err, credits, retries = 0;
2435 err = dquot_initialize(dir);
2439 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2440 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2442 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2443 NULL, EXT4_HT_DIR, credits);
2444 handle = ext4_journal_current_handle();
2445 err = PTR_ERR(inode);
2446 if (!IS_ERR(inode)) {
2447 inode->i_op = &ext4_file_inode_operations;
2448 inode->i_fop = &ext4_file_operations;
2449 ext4_set_aops(inode);
2450 err = ext4_add_nondir(handle, dentry, inode);
2451 if (!err && IS_DIRSYNC(dir))
2452 ext4_handle_sync(handle);
2455 ext4_journal_stop(handle);
2456 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2461 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2462 umode_t mode, dev_t rdev)
2465 struct inode *inode;
2466 int err, credits, retries = 0;
2468 err = dquot_initialize(dir);
2472 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2473 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2475 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2476 NULL, EXT4_HT_DIR, credits);
2477 handle = ext4_journal_current_handle();
2478 err = PTR_ERR(inode);
2479 if (!IS_ERR(inode)) {
2480 init_special_inode(inode, inode->i_mode, rdev);
2481 inode->i_op = &ext4_special_inode_operations;
2482 err = ext4_add_nondir(handle, dentry, inode);
2483 if (!err && IS_DIRSYNC(dir))
2484 ext4_handle_sync(handle);
2487 ext4_journal_stop(handle);
2488 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2493 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2496 struct inode *inode;
2497 int err, retries = 0;
2499 err = dquot_initialize(dir);
2504 inode = ext4_new_inode_start_handle(dir, mode,
2507 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2508 4 + EXT4_XATTR_TRANS_BLOCKS);
2509 handle = ext4_journal_current_handle();
2510 err = PTR_ERR(inode);
2511 if (!IS_ERR(inode)) {
2512 inode->i_op = &ext4_file_inode_operations;
2513 inode->i_fop = &ext4_file_operations;
2514 ext4_set_aops(inode);
2515 d_tmpfile(dentry, inode);
2516 err = ext4_orphan_add(handle, inode);
2518 goto err_unlock_inode;
2519 mark_inode_dirty(inode);
2520 unlock_new_inode(inode);
2523 ext4_journal_stop(handle);
2524 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2528 ext4_journal_stop(handle);
2529 unlock_new_inode(inode);
2533 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2534 struct ext4_dir_entry_2 *de,
2535 int blocksize, int csum_size,
2536 unsigned int parent_ino, int dotdot_real_len)
2538 de->inode = cpu_to_le32(inode->i_ino);
2540 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2542 strcpy(de->name, ".");
2543 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2545 de = ext4_next_entry(de, blocksize);
2546 de->inode = cpu_to_le32(parent_ino);
2548 if (!dotdot_real_len)
2549 de->rec_len = ext4_rec_len_to_disk(blocksize -
2550 (csum_size + EXT4_DIR_REC_LEN(1)),
2553 de->rec_len = ext4_rec_len_to_disk(
2554 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2555 strcpy(de->name, "..");
2556 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2558 return ext4_next_entry(de, blocksize);
2561 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2562 struct inode *inode)
2564 struct buffer_head *dir_block = NULL;
2565 struct ext4_dir_entry_2 *de;
2566 struct ext4_dir_entry_tail *t;
2567 ext4_lblk_t block = 0;
2568 unsigned int blocksize = dir->i_sb->s_blocksize;
2572 if (ext4_has_metadata_csum(dir->i_sb))
2573 csum_size = sizeof(struct ext4_dir_entry_tail);
2575 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2576 err = ext4_try_create_inline_dir(handle, dir, inode);
2577 if (err < 0 && err != -ENOSPC)
2584 dir_block = ext4_append(handle, inode, &block);
2585 if (IS_ERR(dir_block))
2586 return PTR_ERR(dir_block);
2587 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2588 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2589 set_nlink(inode, 2);
2591 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2592 initialize_dirent_tail(t, blocksize);
2595 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2596 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2599 set_buffer_verified(dir_block);
2605 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2608 struct inode *inode;
2609 int err, credits, retries = 0;
2611 if (EXT4_DIR_LINK_MAX(dir))
2614 err = dquot_initialize(dir);
2618 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2619 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2621 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2623 0, NULL, EXT4_HT_DIR, credits);
2624 handle = ext4_journal_current_handle();
2625 err = PTR_ERR(inode);
2629 inode->i_op = &ext4_dir_inode_operations;
2630 inode->i_fop = &ext4_dir_operations;
2631 err = ext4_init_new_dir(handle, dir, inode);
2633 goto out_clear_inode;
2634 err = ext4_mark_inode_dirty(handle, inode);
2636 err = ext4_add_entry(handle, dentry, inode);
2640 unlock_new_inode(inode);
2641 ext4_mark_inode_dirty(handle, inode);
2645 ext4_inc_count(handle, dir);
2646 ext4_update_dx_flag(dir);
2647 err = ext4_mark_inode_dirty(handle, dir);
2649 goto out_clear_inode;
2650 unlock_new_inode(inode);
2651 d_instantiate(dentry, inode);
2652 if (IS_DIRSYNC(dir))
2653 ext4_handle_sync(handle);
2657 ext4_journal_stop(handle);
2658 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2664 * routine to check that the specified directory is empty (for rmdir)
2666 bool ext4_empty_dir(struct inode *inode)
2668 unsigned int offset;
2669 struct buffer_head *bh;
2670 struct ext4_dir_entry_2 *de, *de1;
2671 struct super_block *sb;
2673 if (ext4_has_inline_data(inode)) {
2674 int has_inline_data = 1;
2677 ret = empty_inline_dir(inode, &has_inline_data);
2678 if (has_inline_data)
2683 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2684 EXT4_ERROR_INODE(inode, "invalid size");
2687 bh = ext4_read_dirblock(inode, 0, EITHER);
2691 de = (struct ext4_dir_entry_2 *) bh->b_data;
2692 de1 = ext4_next_entry(de, sb->s_blocksize);
2693 if (le32_to_cpu(de->inode) != inode->i_ino ||
2694 le32_to_cpu(de1->inode) == 0 ||
2695 strcmp(".", de->name) || strcmp("..", de1->name)) {
2696 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2700 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2701 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2702 de = ext4_next_entry(de1, sb->s_blocksize);
2703 while (offset < inode->i_size) {
2704 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2705 unsigned int lblock;
2707 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2708 bh = ext4_read_dirblock(inode, lblock, EITHER);
2711 de = (struct ext4_dir_entry_2 *) bh->b_data;
2713 if (ext4_check_dir_entry(inode, NULL, de, bh,
2714 bh->b_data, bh->b_size, offset)) {
2715 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2717 offset = (offset | (sb->s_blocksize - 1)) + 1;
2720 if (le32_to_cpu(de->inode)) {
2724 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2725 de = ext4_next_entry(de, sb->s_blocksize);
2732 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2733 * such inodes, starting at the superblock, in case we crash before the
2734 * file is closed/deleted, or in case the inode truncate spans multiple
2735 * transactions and the last transaction is not recovered after a crash.
2737 * At filesystem recovery time, we walk this list deleting unlinked
2738 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2740 * Orphan list manipulation functions must be called under i_mutex unless
2741 * we are just creating the inode or deleting it.
2743 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2745 struct super_block *sb = inode->i_sb;
2746 struct ext4_sb_info *sbi = EXT4_SB(sb);
2747 struct ext4_iloc iloc;
2751 if (!sbi->s_journal || is_bad_inode(inode))
2754 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2755 !inode_is_locked(inode));
2757 * Exit early if inode already is on orphan list. This is a big speedup
2758 * since we don't have to contend on the global s_orphan_lock.
2760 if (!list_empty(&EXT4_I(inode)->i_orphan))
2764 * Orphan handling is only valid for files with data blocks
2765 * being truncated, or files being unlinked. Note that we either
2766 * hold i_mutex, or the inode can not be referenced from outside,
2767 * so i_nlink should not be bumped due to race
2769 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2770 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2772 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2773 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2777 err = ext4_reserve_inode_write(handle, inode, &iloc);
2781 mutex_lock(&sbi->s_orphan_lock);
2783 * Due to previous errors inode may be already a part of on-disk
2784 * orphan list. If so skip on-disk list modification.
2786 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2787 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2788 /* Insert this inode at the head of the on-disk orphan list */
2789 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2790 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2793 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2794 mutex_unlock(&sbi->s_orphan_lock);
2797 err = ext4_handle_dirty_super(handle, sb);
2798 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2803 * We have to remove inode from in-memory list if
2804 * addition to on disk orphan list failed. Stray orphan
2805 * list entries can cause panics at unmount time.
2807 mutex_lock(&sbi->s_orphan_lock);
2808 list_del_init(&EXT4_I(inode)->i_orphan);
2809 mutex_unlock(&sbi->s_orphan_lock);
2812 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2813 jbd_debug(4, "orphan inode %lu will point to %d\n",
2814 inode->i_ino, NEXT_ORPHAN(inode));
2816 ext4_std_error(sb, err);
2821 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2822 * of such inodes stored on disk, because it is finally being cleaned up.
2824 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2826 struct list_head *prev;
2827 struct ext4_inode_info *ei = EXT4_I(inode);
2828 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2830 struct ext4_iloc iloc;
2833 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2836 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2837 !inode_is_locked(inode));
2838 /* Do this quick check before taking global s_orphan_lock. */
2839 if (list_empty(&ei->i_orphan))
2843 /* Grab inode buffer early before taking global s_orphan_lock */
2844 err = ext4_reserve_inode_write(handle, inode, &iloc);
2847 mutex_lock(&sbi->s_orphan_lock);
2848 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2850 prev = ei->i_orphan.prev;
2851 list_del_init(&ei->i_orphan);
2853 /* If we're on an error path, we may not have a valid
2854 * transaction handle with which to update the orphan list on
2855 * disk, but we still need to remove the inode from the linked
2856 * list in memory. */
2857 if (!handle || err) {
2858 mutex_unlock(&sbi->s_orphan_lock);
2862 ino_next = NEXT_ORPHAN(inode);
2863 if (prev == &sbi->s_orphan) {
2864 jbd_debug(4, "superblock will point to %u\n", ino_next);
2865 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2866 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2868 mutex_unlock(&sbi->s_orphan_lock);
2871 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2872 mutex_unlock(&sbi->s_orphan_lock);
2873 err = ext4_handle_dirty_super(handle, inode->i_sb);
2875 struct ext4_iloc iloc2;
2876 struct inode *i_prev =
2877 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2879 jbd_debug(4, "orphan inode %lu will point to %u\n",
2880 i_prev->i_ino, ino_next);
2881 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2883 mutex_unlock(&sbi->s_orphan_lock);
2886 NEXT_ORPHAN(i_prev) = ino_next;
2887 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2888 mutex_unlock(&sbi->s_orphan_lock);
2892 NEXT_ORPHAN(inode) = 0;
2893 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2895 ext4_std_error(inode->i_sb, err);
2903 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2906 struct inode *inode;
2907 struct buffer_head *bh;
2908 struct ext4_dir_entry_2 *de;
2909 handle_t *handle = NULL;
2911 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2914 /* Initialize quotas before so that eventual writes go in
2915 * separate transaction */
2916 retval = dquot_initialize(dir);
2919 retval = dquot_initialize(d_inode(dentry));
2924 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2930 inode = d_inode(dentry);
2932 retval = -EFSCORRUPTED;
2933 if (le32_to_cpu(de->inode) != inode->i_ino)
2936 retval = -ENOTEMPTY;
2937 if (!ext4_empty_dir(inode))
2940 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2941 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2942 if (IS_ERR(handle)) {
2943 retval = PTR_ERR(handle);
2948 if (IS_DIRSYNC(dir))
2949 ext4_handle_sync(handle);
2951 retval = ext4_delete_entry(handle, dir, de, bh);
2954 if (!EXT4_DIR_LINK_EMPTY(inode))
2955 ext4_warning_inode(inode,
2956 "empty directory '%.*s' has too many links (%u)",
2957 dentry->d_name.len, dentry->d_name.name,
2961 /* There's no need to set i_disksize: the fact that i_nlink is
2962 * zero will ensure that the right thing happens during any
2965 ext4_orphan_add(handle, inode);
2966 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2967 ext4_mark_inode_dirty(handle, inode);
2968 ext4_dec_count(handle, dir);
2969 ext4_update_dx_flag(dir);
2970 ext4_mark_inode_dirty(handle, dir);
2975 ext4_journal_stop(handle);
2979 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2982 struct inode *inode;
2983 struct buffer_head *bh;
2984 struct ext4_dir_entry_2 *de;
2985 handle_t *handle = NULL;
2987 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2990 trace_ext4_unlink_enter(dir, dentry);
2991 /* Initialize quotas before so that eventual writes go
2992 * in separate transaction */
2993 retval = dquot_initialize(dir);
2996 retval = dquot_initialize(d_inode(dentry));
3001 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3007 inode = d_inode(dentry);
3009 retval = -EFSCORRUPTED;
3010 if (le32_to_cpu(de->inode) != inode->i_ino)
3013 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3014 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3015 if (IS_ERR(handle)) {
3016 retval = PTR_ERR(handle);
3021 if (IS_DIRSYNC(dir))
3022 ext4_handle_sync(handle);
3024 if (inode->i_nlink == 0) {
3025 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3026 dentry->d_name.len, dentry->d_name.name);
3027 set_nlink(inode, 1);
3029 retval = ext4_delete_entry(handle, dir, de, bh);
3032 dir->i_ctime = dir->i_mtime = current_time(dir);
3033 ext4_update_dx_flag(dir);
3034 ext4_mark_inode_dirty(handle, dir);
3036 if (!inode->i_nlink)
3037 ext4_orphan_add(handle, inode);
3038 inode->i_ctime = current_time(inode);
3039 ext4_mark_inode_dirty(handle, inode);
3044 ext4_journal_stop(handle);
3045 trace_ext4_unlink_exit(dentry, retval);
3049 static int ext4_symlink(struct inode *dir,
3050 struct dentry *dentry, const char *symname)
3053 struct inode *inode;
3054 int err, len = strlen(symname);
3056 bool encryption_required;
3057 struct fscrypt_str disk_link;
3058 struct fscrypt_symlink_data *sd = NULL;
3060 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3063 disk_link.len = len + 1;
3064 disk_link.name = (char *) symname;
3066 encryption_required = (ext4_encrypted_inode(dir) ||
3067 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3068 if (encryption_required) {
3069 err = fscrypt_get_encryption_info(dir);
3072 if (!fscrypt_has_encryption_key(dir))
3074 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3075 sizeof(struct fscrypt_symlink_data));
3076 sd = kzalloc(disk_link.len, GFP_KERNEL);
3081 if (disk_link.len > dir->i_sb->s_blocksize) {
3082 err = -ENAMETOOLONG;
3086 err = dquot_initialize(dir);
3090 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3092 * For non-fast symlinks, we just allocate inode and put it on
3093 * orphan list in the first transaction => we need bitmap,
3094 * group descriptor, sb, inode block, quota blocks, and
3095 * possibly selinux xattr blocks.
3097 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3098 EXT4_XATTR_TRANS_BLOCKS;
3101 * Fast symlink. We have to add entry to directory
3102 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3103 * allocate new inode (bitmap, group descriptor, inode block,
3104 * quota blocks, sb is already counted in previous macros).
3106 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3107 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3110 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3111 &dentry->d_name, 0, NULL,
3112 EXT4_HT_DIR, credits);
3113 handle = ext4_journal_current_handle();
3114 if (IS_ERR(inode)) {
3116 ext4_journal_stop(handle);
3117 err = PTR_ERR(inode);
3121 if (encryption_required) {
3123 struct fscrypt_str ostr =
3124 FSTR_INIT(sd->encrypted_path, disk_link.len);
3126 istr.name = (const unsigned char *) symname;
3128 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3130 goto err_drop_inode;
3131 sd->len = cpu_to_le16(ostr.len);
3132 disk_link.name = (char *) sd;
3133 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3136 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3137 if (!encryption_required)
3138 inode->i_op = &ext4_symlink_inode_operations;
3139 inode_nohighmem(inode);
3140 ext4_set_aops(inode);
3142 * We cannot call page_symlink() with transaction started
3143 * because it calls into ext4_write_begin() which can wait
3144 * for transaction commit if we are running out of space
3145 * and thus we deadlock. So we have to stop transaction now
3146 * and restart it when symlink contents is written.
3148 * To keep fs consistent in case of crash, we have to put inode
3149 * to orphan list in the mean time.
3152 err = ext4_orphan_add(handle, inode);
3153 ext4_journal_stop(handle);
3156 goto err_drop_inode;
3157 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3159 goto err_drop_inode;
3161 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3162 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3164 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3165 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3166 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3167 if (IS_ERR(handle)) {
3168 err = PTR_ERR(handle);
3170 goto err_drop_inode;
3172 set_nlink(inode, 1);
3173 err = ext4_orphan_del(handle, inode);
3175 goto err_drop_inode;
3177 /* clear the extent format for fast symlink */
3178 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3179 if (!encryption_required) {
3180 inode->i_op = &ext4_fast_symlink_inode_operations;
3181 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3183 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3185 inode->i_size = disk_link.len - 1;
3187 EXT4_I(inode)->i_disksize = inode->i_size;
3188 err = ext4_add_nondir(handle, dentry, inode);
3189 if (!err && IS_DIRSYNC(dir))
3190 ext4_handle_sync(handle);
3193 ext4_journal_stop(handle);
3198 ext4_journal_stop(handle);
3200 unlock_new_inode(inode);
3207 static int ext4_link(struct dentry *old_dentry,
3208 struct inode *dir, struct dentry *dentry)
3211 struct inode *inode = d_inode(old_dentry);
3212 int err, retries = 0;
3214 if (inode->i_nlink >= EXT4_LINK_MAX)
3217 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3221 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3222 (!projid_eq(EXT4_I(dir)->i_projid,
3223 EXT4_I(old_dentry->d_inode)->i_projid)))
3226 err = dquot_initialize(dir);
3231 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3232 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3233 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3235 return PTR_ERR(handle);
3237 if (IS_DIRSYNC(dir))
3238 ext4_handle_sync(handle);
3240 inode->i_ctime = current_time(inode);
3241 ext4_inc_count(handle, inode);
3244 err = ext4_add_entry(handle, dentry, inode);
3246 ext4_mark_inode_dirty(handle, inode);
3247 /* this can happen only for tmpfile being
3248 * linked the first time
3250 if (inode->i_nlink == 1)
3251 ext4_orphan_del(handle, inode);
3252 d_instantiate(dentry, inode);
3257 ext4_journal_stop(handle);
3258 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3265 * Try to find buffer head where contains the parent block.
3266 * It should be the inode block if it is inlined or the 1st block
3267 * if it is a normal dir.
3269 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3270 struct inode *inode,
3272 struct ext4_dir_entry_2 **parent_de,
3275 struct buffer_head *bh;
3277 if (!ext4_has_inline_data(inode)) {
3278 bh = ext4_read_dirblock(inode, 0, EITHER);
3280 *retval = PTR_ERR(bh);
3283 *parent_de = ext4_next_entry(
3284 (struct ext4_dir_entry_2 *)bh->b_data,
3285 inode->i_sb->s_blocksize);
3290 return ext4_get_first_inline_block(inode, parent_de, retval);
3293 struct ext4_renament {
3295 struct dentry *dentry;
3296 struct inode *inode;
3298 int dir_nlink_delta;
3300 /* entry for "dentry" */
3301 struct buffer_head *bh;
3302 struct ext4_dir_entry_2 *de;
3305 /* entry for ".." in inode if it's a directory */
3306 struct buffer_head *dir_bh;
3307 struct ext4_dir_entry_2 *parent_de;
3311 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3315 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3316 &retval, &ent->parent_de,
3320 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3321 return -EFSCORRUPTED;
3322 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3323 return ext4_journal_get_write_access(handle, ent->dir_bh);
3326 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3331 ent->parent_de->inode = cpu_to_le32(dir_ino);
3332 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3333 if (!ent->dir_inlined) {
3334 if (is_dx(ent->inode)) {
3335 retval = ext4_handle_dirty_dx_node(handle,
3339 retval = ext4_handle_dirty_dirent_node(handle,
3344 retval = ext4_mark_inode_dirty(handle, ent->inode);
3347 ext4_std_error(ent->dir->i_sb, retval);
3353 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3354 unsigned ino, unsigned file_type)
3358 BUFFER_TRACE(ent->bh, "get write access");
3359 retval = ext4_journal_get_write_access(handle, ent->bh);
3362 ent->de->inode = cpu_to_le32(ino);
3363 if (ext4_has_feature_filetype(ent->dir->i_sb))
3364 ent->de->file_type = file_type;
3365 ent->dir->i_version++;
3366 ent->dir->i_ctime = ent->dir->i_mtime =
3367 current_time(ent->dir);
3368 ext4_mark_inode_dirty(handle, ent->dir);
3369 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3370 if (!ent->inlined) {
3371 retval = ext4_handle_dirty_dirent_node(handle,
3373 if (unlikely(retval)) {
3374 ext4_std_error(ent->dir->i_sb, retval);
3384 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3385 const struct qstr *d_name)
3387 int retval = -ENOENT;
3388 struct buffer_head *bh;
3389 struct ext4_dir_entry_2 *de;
3391 bh = ext4_find_entry(dir, d_name, &de, NULL);
3395 retval = ext4_delete_entry(handle, dir, de, bh);
3401 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3406 * ent->de could have moved from under us during htree split, so make
3407 * sure that we are deleting the right entry. We might also be pointing
3408 * to a stale entry in the unused part of ent->bh so just checking inum
3409 * and the name isn't enough.
3411 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3412 ent->de->name_len != ent->dentry->d_name.len ||
3413 strncmp(ent->de->name, ent->dentry->d_name.name,
3414 ent->de->name_len) ||
3416 retval = ext4_find_delete_entry(handle, ent->dir,
3417 &ent->dentry->d_name);
3419 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3420 if (retval == -ENOENT) {
3421 retval = ext4_find_delete_entry(handle, ent->dir,
3422 &ent->dentry->d_name);
3427 ext4_warning_inode(ent->dir,
3428 "Deleting old file: nlink %d, error=%d",
3429 ent->dir->i_nlink, retval);
3433 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3435 if (ent->dir_nlink_delta) {
3436 if (ent->dir_nlink_delta == -1)
3437 ext4_dec_count(handle, ent->dir);
3439 ext4_inc_count(handle, ent->dir);
3440 ext4_mark_inode_dirty(handle, ent->dir);
3444 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3445 int credits, handle_t **h)
3452 * for inode block, sb block, group summaries,
3455 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3456 EXT4_XATTR_TRANS_BLOCKS + 4);
3458 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3459 &ent->dentry->d_name, 0, NULL,
3460 EXT4_HT_DIR, credits);
3462 handle = ext4_journal_current_handle();
3465 ext4_journal_stop(handle);
3466 if (PTR_ERR(wh) == -ENOSPC &&
3467 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3471 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3472 wh->i_op = &ext4_special_inode_operations;
3478 * Anybody can rename anything with this: the permission checks are left to the
3479 * higher-level routines.
3481 * n.b. old_{dentry,inode) refers to the source dentry/inode
3482 * while new_{dentry,inode) refers to the destination dentry/inode
3483 * This comes from rename(const char *oldpath, const char *newpath)
3485 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3486 struct inode *new_dir, struct dentry *new_dentry,
3489 handle_t *handle = NULL;
3490 struct ext4_renament old = {
3492 .dentry = old_dentry,
3493 .inode = d_inode(old_dentry),
3495 struct ext4_renament new = {
3497 .dentry = new_dentry,
3498 .inode = d_inode(new_dentry),
3502 struct inode *whiteout = NULL;
3506 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3507 (!projid_eq(EXT4_I(new_dir)->i_projid,
3508 EXT4_I(old_dentry->d_inode)->i_projid)))
3511 retval = dquot_initialize(old.dir);
3514 retval = dquot_initialize(new.dir);
3518 /* Initialize quotas before so that eventual writes go
3519 * in separate transaction */
3521 retval = dquot_initialize(new.inode);
3526 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3528 return PTR_ERR(old.bh);
3530 * Check for inode number is _not_ due to possible IO errors.
3531 * We might rmdir the source, keep it as pwd of some process
3532 * and merrily kill the link to whatever was created under the
3533 * same name. Goodbye sticky bit ;-<
3536 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3539 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3540 &new.de, &new.inlined);
3541 if (IS_ERR(new.bh)) {
3542 retval = PTR_ERR(new.bh);
3552 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3553 ext4_alloc_da_blocks(old.inode);
3555 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3556 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3557 if (!(flags & RENAME_WHITEOUT)) {
3558 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3559 if (IS_ERR(handle)) {
3560 retval = PTR_ERR(handle);
3565 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3566 if (IS_ERR(whiteout)) {
3567 retval = PTR_ERR(whiteout);
3573 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3574 ext4_handle_sync(handle);
3576 if (S_ISDIR(old.inode->i_mode)) {
3578 retval = -ENOTEMPTY;
3579 if (!ext4_empty_dir(new.inode))
3583 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3586 retval = ext4_rename_dir_prepare(handle, &old);
3591 * If we're renaming a file within an inline_data dir and adding or
3592 * setting the new dirent causes a conversion from inline_data to
3593 * extents/blockmap, we need to force the dirent delete code to
3594 * re-read the directory, or else we end up trying to delete a dirent
3595 * from what is now the extent tree root (or a block map).
3597 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3598 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3600 old_file_type = old.de->file_type;
3603 * Do this before adding a new entry, so the old entry is sure
3604 * to be still pointing to the valid old entry.
3606 retval = ext4_setent(handle, &old, whiteout->i_ino,
3610 ext4_mark_inode_dirty(handle, whiteout);
3613 retval = ext4_add_entry(handle, new.dentry, old.inode);
3617 retval = ext4_setent(handle, &new,
3618 old.inode->i_ino, old_file_type);
3623 force_reread = !ext4_test_inode_flag(new.dir,
3624 EXT4_INODE_INLINE_DATA);
3627 * Like most other Unix systems, set the ctime for inodes on a
3630 old.inode->i_ctime = current_time(old.inode);
3631 ext4_mark_inode_dirty(handle, old.inode);
3637 ext4_rename_delete(handle, &old, force_reread);
3641 ext4_dec_count(handle, new.inode);
3642 new.inode->i_ctime = current_time(new.inode);
3644 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3645 ext4_update_dx_flag(old.dir);
3647 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3651 ext4_dec_count(handle, old.dir);
3653 /* checked ext4_empty_dir above, can't have another
3654 * parent, ext4_dec_count() won't work for many-linked
3656 clear_nlink(new.inode);
3658 ext4_inc_count(handle, new.dir);
3659 ext4_update_dx_flag(new.dir);
3660 ext4_mark_inode_dirty(handle, new.dir);
3663 ext4_mark_inode_dirty(handle, old.dir);
3665 ext4_mark_inode_dirty(handle, new.inode);
3666 if (!new.inode->i_nlink)
3667 ext4_orphan_add(handle, new.inode);
3677 drop_nlink(whiteout);
3678 unlock_new_inode(whiteout);
3682 ext4_journal_stop(handle);
3686 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3687 struct inode *new_dir, struct dentry *new_dentry)
3689 handle_t *handle = NULL;
3690 struct ext4_renament old = {
3692 .dentry = old_dentry,
3693 .inode = d_inode(old_dentry),
3695 struct ext4_renament new = {
3697 .dentry = new_dentry,
3698 .inode = d_inode(new_dentry),
3702 struct timespec ctime;
3704 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3705 !projid_eq(EXT4_I(new_dir)->i_projid,
3706 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3707 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3708 !projid_eq(EXT4_I(old_dir)->i_projid,
3709 EXT4_I(new_dentry->d_inode)->i_projid)))
3712 retval = dquot_initialize(old.dir);
3715 retval = dquot_initialize(new.dir);
3719 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3720 &old.de, &old.inlined);
3722 return PTR_ERR(old.bh);
3724 * Check for inode number is _not_ due to possible IO errors.
3725 * We might rmdir the source, keep it as pwd of some process
3726 * and merrily kill the link to whatever was created under the
3727 * same name. Goodbye sticky bit ;-<
3730 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3733 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3734 &new.de, &new.inlined);
3735 if (IS_ERR(new.bh)) {
3736 retval = PTR_ERR(new.bh);
3741 /* RENAME_EXCHANGE case: old *and* new must both exist */
3742 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3745 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3746 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3747 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3748 if (IS_ERR(handle)) {
3749 retval = PTR_ERR(handle);
3754 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3755 ext4_handle_sync(handle);
3757 if (S_ISDIR(old.inode->i_mode)) {
3759 retval = ext4_rename_dir_prepare(handle, &old);
3763 if (S_ISDIR(new.inode->i_mode)) {
3765 retval = ext4_rename_dir_prepare(handle, &new);
3771 * Other than the special case of overwriting a directory, parents'
3772 * nlink only needs to be modified if this is a cross directory rename.
3774 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3775 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3776 new.dir_nlink_delta = -old.dir_nlink_delta;
3778 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3779 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3783 new_file_type = new.de->file_type;
3784 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3788 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3793 * Like most other Unix systems, set the ctime for inodes on a
3796 ctime = current_time(old.inode);
3797 old.inode->i_ctime = ctime;
3798 new.inode->i_ctime = ctime;
3799 ext4_mark_inode_dirty(handle, old.inode);
3800 ext4_mark_inode_dirty(handle, new.inode);
3803 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3808 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3812 ext4_update_dir_count(handle, &old);
3813 ext4_update_dir_count(handle, &new);
3822 ext4_journal_stop(handle);
3826 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3827 struct inode *new_dir, struct dentry *new_dentry,
3832 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3835 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3838 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
3843 if (flags & RENAME_EXCHANGE) {
3844 return ext4_cross_rename(old_dir, old_dentry,
3845 new_dir, new_dentry);
3848 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3852 * directories can handle most operations...
3854 const struct inode_operations ext4_dir_inode_operations = {
3855 .create = ext4_create,
3856 .lookup = ext4_lookup,
3858 .unlink = ext4_unlink,
3859 .symlink = ext4_symlink,
3860 .mkdir = ext4_mkdir,
3861 .rmdir = ext4_rmdir,
3862 .mknod = ext4_mknod,
3863 .tmpfile = ext4_tmpfile,
3864 .rename = ext4_rename2,
3865 .setattr = ext4_setattr,
3866 .getattr = ext4_getattr,
3867 .listxattr = ext4_listxattr,
3868 .get_acl = ext4_get_acl,
3869 .set_acl = ext4_set_acl,
3870 .fiemap = ext4_fiemap,
3873 const struct inode_operations ext4_special_inode_operations = {
3874 .setattr = ext4_setattr,
3875 .getattr = ext4_getattr,
3876 .listxattr = ext4_listxattr,
3877 .get_acl = ext4_get_acl,
3878 .set_acl = ext4_set_acl,