2 * linux/fs/ext4/namei.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/namei.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
28 #include <linux/pagemap.h>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
37 #include "ext4_jbd2.h"
42 #include <trace/events/ext4.h>
44 * define how far ahead to read directories while searching them.
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 static struct buffer_head *ext4_append(handle_t *handle,
54 struct buffer_head *bh;
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
69 BUFFER_TRACE(bh, "get_write_access");
70 err = ext4_journal_get_write_access(handle, bh);
73 ext4_std_error(inode->i_sb, err);
79 static int ext4_dx_csum_verify(struct inode *inode,
80 struct ext4_dir_entry *dirent);
86 #define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
99 bh = ext4_bread(NULL, inode, block, 0);
101 __ext4_warning(inode->i_sb, func, line,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode->i_ino, (unsigned long)block,
105 current->comm, PTR_ERR(bh));
110 ext4_error_inode(inode, func, line, block,
111 "Directory hole found");
112 return ERR_PTR(-EFSCORRUPTED);
114 dirent = (struct ext4_dir_entry *) bh->b_data;
115 /* Determine whether or not we have an index block */
119 else if (ext4_rec_len_from_disk(dirent->rec_len,
120 inode->i_sb->s_blocksize) ==
121 inode->i_sb->s_blocksize)
124 if (!is_dx_block && type == INDEX) {
125 ext4_error_inode(inode, func, line, block,
126 "directory leaf block found instead of index block");
127 return ERR_PTR(-EFSCORRUPTED);
129 if (!ext4_has_metadata_csum(inode->i_sb) ||
134 * An empty leaf block can get mistaken for a index block; for
135 * this reason, we can only check the index checksum when the
136 * caller is sure it should be an index block.
138 if (is_dx_block && type == INDEX) {
139 if (ext4_dx_csum_verify(inode, dirent))
140 set_buffer_verified(bh);
142 ext4_error_inode(inode, func, line, block,
143 "Directory index failed checksum");
145 return ERR_PTR(-EFSBADCRC);
149 if (ext4_dirent_csum_verify(inode, dirent))
150 set_buffer_verified(bh);
152 ext4_error_inode(inode, func, line, block,
153 "Directory block failed checksum");
155 return ERR_PTR(-EFSBADCRC);
162 #define assert(test) J_ASSERT(test)
166 #define dxtrace(command) command
168 #define dxtrace(command)
192 * dx_root_info is laid out so that if it should somehow get overlaid by a
193 * dirent the two low bits of the hash version will be zero. Therefore, the
194 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
199 struct fake_dirent dot;
201 struct fake_dirent dotdot;
205 __le32 reserved_zero;
207 u8 info_length; /* 8 */
212 struct dx_entry entries[0];
217 struct fake_dirent fake;
218 struct dx_entry entries[0];
224 struct buffer_head *bh;
225 struct dx_entry *entries;
237 * This goes at the end of each htree block.
241 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246 static inline unsigned dx_get_hash(struct dx_entry *entry);
247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
248 static unsigned dx_get_count(struct dx_entry *entries);
249 static unsigned dx_get_limit(struct dx_entry *entries);
250 static void dx_set_count(struct dx_entry *entries, unsigned value);
251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253 static unsigned dx_node_limit(struct inode *dir);
254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
256 struct dx_hash_info *hinfo,
257 struct dx_frame *frame);
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 unsigned blocksize, struct dx_hash_info *hinfo,
261 struct dx_map_entry map[]);
262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264 struct dx_map_entry *offsets, int count, unsigned blocksize);
265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266 static void dx_insert_block(struct dx_frame *frame,
267 u32 hash, ext4_lblk_t block);
268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269 struct dx_frame *frame,
270 struct dx_frame *frames,
272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273 struct ext4_filename *fname,
274 struct ext4_dir_entry_2 **res_dir);
275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276 struct inode *dir, struct inode *inode);
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 unsigned int blocksize)
282 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 t->det_rec_len = ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail), blocksize);
285 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 struct ext4_dir_entry *de)
292 struct ext4_dir_entry_tail *t;
295 struct ext4_dir_entry *d, *top;
298 top = (struct ext4_dir_entry *)(((void *)de) +
299 (EXT4_BLOCK_SIZE(inode->i_sb) -
300 sizeof(struct ext4_dir_entry_tail)));
301 while (d < top && d->rec_len)
302 d = (struct ext4_dir_entry *)(((void *)d) +
303 le16_to_cpu(d->rec_len));
308 t = (struct ext4_dir_entry_tail *)d;
310 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
313 if (t->det_reserved_zero1 ||
314 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 t->det_reserved_zero2 ||
316 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
322 static __le32 ext4_dirent_csum(struct inode *inode,
323 struct ext4_dir_entry *dirent, int size)
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 struct ext4_inode_info *ei = EXT4_I(inode);
329 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 return cpu_to_le32(csum);
333 #define warn_no_space_for_csum(inode) \
334 __warn_no_space_for_csum((inode), __func__, __LINE__)
336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
339 __ext4_warning_inode(inode, func, line,
340 "No space for directory leaf checksum. Please run e2fsck -D.");
343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
345 struct ext4_dir_entry_tail *t;
347 if (!ext4_has_metadata_csum(inode->i_sb))
350 t = get_dirent_tail(inode, dirent);
352 warn_no_space_for_csum(inode);
356 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 (void *)t - (void *)dirent))
363 static void ext4_dirent_csum_set(struct inode *inode,
364 struct ext4_dir_entry *dirent)
366 struct ext4_dir_entry_tail *t;
368 if (!ext4_has_metadata_csum(inode->i_sb))
371 t = get_dirent_tail(inode, dirent);
373 warn_no_space_for_csum(inode);
377 t->det_checksum = ext4_dirent_csum(inode, dirent,
378 (void *)t - (void *)dirent);
381 int ext4_handle_dirty_dirent_node(handle_t *handle,
383 struct buffer_head *bh)
385 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 return ext4_handle_dirty_metadata(handle, inode, bh);
389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 struct ext4_dir_entry *dirent,
393 struct ext4_dir_entry *dp;
394 struct dx_root_info *root;
397 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
399 else if (le16_to_cpu(dirent->rec_len) == 12) {
400 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 if (le16_to_cpu(dp->rec_len) !=
402 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
404 root = (struct dx_root_info *)(((void *)dp + 12));
405 if (root->reserved_zero ||
406 root->info_length != sizeof(struct dx_root_info))
413 *offset = count_offset;
414 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 int count_offset, int count, struct dx_tail *t)
420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 struct ext4_inode_info *ei = EXT4_I(inode);
424 __u32 dummy_csum = 0;
425 int offset = offsetof(struct dx_tail, dt_checksum);
427 size = count_offset + (count * sizeof(struct dx_entry));
428 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
429 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
430 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
432 return cpu_to_le32(csum);
435 static int ext4_dx_csum_verify(struct inode *inode,
436 struct ext4_dir_entry *dirent)
438 struct dx_countlimit *c;
440 int count_offset, limit, count;
442 if (!ext4_has_metadata_csum(inode->i_sb))
445 c = get_dx_countlimit(inode, dirent, &count_offset);
447 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
450 limit = le16_to_cpu(c->limit);
451 count = le16_to_cpu(c->count);
452 if (count_offset + (limit * sizeof(struct dx_entry)) >
453 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
454 warn_no_space_for_csum(inode);
457 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
459 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
465 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
467 struct dx_countlimit *c;
469 int count_offset, limit, count;
471 if (!ext4_has_metadata_csum(inode->i_sb))
474 c = get_dx_countlimit(inode, dirent, &count_offset);
476 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
479 limit = le16_to_cpu(c->limit);
480 count = le16_to_cpu(c->count);
481 if (count_offset + (limit * sizeof(struct dx_entry)) >
482 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
483 warn_no_space_for_csum(inode);
486 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
488 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
493 struct buffer_head *bh)
495 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496 return ext4_handle_dirty_metadata(handle, inode, bh);
500 * p is at least 6 bytes before the end of page
502 static inline struct ext4_dir_entry_2 *
503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
505 return (struct ext4_dir_entry_2 *)((char *)p +
506 ext4_rec_len_from_disk(p->rec_len, blocksize));
510 * Future: use high four bits of block for coalesce-on-delete flags
511 * Mask them off for now.
514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
516 return le32_to_cpu(entry->block) & 0x00ffffff;
519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
521 entry->block = cpu_to_le32(value);
524 static inline unsigned dx_get_hash(struct dx_entry *entry)
526 return le32_to_cpu(entry->hash);
529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
531 entry->hash = cpu_to_le32(value);
534 static inline unsigned dx_get_count(struct dx_entry *entries)
536 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
539 static inline unsigned dx_get_limit(struct dx_entry *entries)
541 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
546 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
551 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
556 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 EXT4_DIR_REC_LEN(2) - infosize;
559 if (ext4_has_metadata_csum(dir->i_sb))
560 entry_space -= sizeof(struct dx_tail);
561 return entry_space / sizeof(struct dx_entry);
564 static inline unsigned dx_node_limit(struct inode *dir)
566 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568 if (ext4_has_metadata_csum(dir->i_sb))
569 entry_space -= sizeof(struct dx_tail);
570 return entry_space / sizeof(struct dx_entry);
577 static void dx_show_index(char * label, struct dx_entry *entries)
579 int i, n = dx_get_count (entries);
580 printk(KERN_DEBUG "%s index", label);
581 for (i = 0; i < n; i++) {
582 printk(KERN_CONT " %x->%lu",
583 i ? dx_get_hash(entries + i) : 0,
584 (unsigned long)dx_get_block(entries + i));
586 printk(KERN_CONT "\n");
596 static struct stats dx_show_leaf(struct inode *dir,
597 struct dx_hash_info *hinfo,
598 struct ext4_dir_entry_2 *de,
599 int size, int show_names)
601 unsigned names = 0, space = 0;
602 char *base = (char *) de;
603 struct dx_hash_info h = *hinfo;
606 while ((char *) de < base + size)
612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
615 struct fscrypt_str fname_crypto_str =
621 if (ext4_encrypted_inode(dir))
622 res = fscrypt_get_encryption_info(dir);
624 printk(KERN_WARNING "Error setting up"
625 " fname crypto: %d\n", res);
627 if (!fscrypt_has_encryption_key(dir)) {
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de->name,
631 printk("%*.s:(U)%x.%u ", len,
633 (unsigned) ((char *) de
636 struct fscrypt_str de_name =
637 FSTR_INIT(name, len);
639 /* Directory is encrypted */
640 res = fscrypt_fname_alloc_buffer(
644 printk(KERN_WARNING "Error "
648 res = fscrypt_fname_disk_to_usr(dir,
652 printk(KERN_WARNING "Error "
653 "converting filename "
659 name = fname_crypto_str.name;
660 len = fname_crypto_str.len;
662 ext4fs_dirhash(de->name, de->name_len,
664 printk("%*.s:(E)%x.%u ", len, name,
665 h.hash, (unsigned) ((char *) de
667 fscrypt_fname_free_buffer(
671 int len = de->name_len;
672 char *name = de->name;
673 ext4fs_dirhash(de->name, de->name_len, &h);
674 printk("%*.s:%x.%u ", len, name, h.hash,
675 (unsigned) ((char *) de - base));
678 space += EXT4_DIR_REC_LEN(de->name_len);
681 de = ext4_next_entry(de, size);
683 printk(KERN_CONT "(%i)\n", names);
684 return (struct stats) { names, space, 1 };
687 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
688 struct dx_entry *entries, int levels)
690 unsigned blocksize = dir->i_sb->s_blocksize;
691 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
693 struct buffer_head *bh;
694 printk("%i indexed blocks...\n", count);
695 for (i = 0; i < count; i++, entries++)
697 ext4_lblk_t block = dx_get_block(entries);
698 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
699 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
701 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
702 bh = ext4_bread(NULL,dir, block, 0);
703 if (!bh || IS_ERR(bh))
706 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
707 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
708 bh->b_data, blocksize, 0);
709 names += stats.names;
710 space += stats.space;
711 bcount += stats.bcount;
715 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
716 levels ? "" : " ", names, space/bcount,
717 (space/bcount)*100/blocksize);
718 return (struct stats) { names, space, bcount};
720 #endif /* DX_DEBUG */
723 * Probe for a directory leaf block to search.
725 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
726 * error in the directory index, and the caller should fall back to
727 * searching the directory normally. The callers of dx_probe **MUST**
728 * check for this error code, and make sure it never gets reflected
731 static struct dx_frame *
732 dx_probe(struct ext4_filename *fname, struct inode *dir,
733 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
735 unsigned count, indirect;
736 struct dx_entry *at, *entries, *p, *q, *m;
737 struct dx_root *root;
738 struct dx_frame *frame = frame_in;
739 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
742 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
743 if (IS_ERR(frame->bh))
744 return (struct dx_frame *) frame->bh;
746 root = (struct dx_root *) frame->bh->b_data;
747 if (root->info.hash_version != DX_HASH_TEA &&
748 root->info.hash_version != DX_HASH_HALF_MD4 &&
749 root->info.hash_version != DX_HASH_LEGACY) {
750 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
751 root->info.hash_version);
755 hinfo = &fname->hinfo;
756 hinfo->hash_version = root->info.hash_version;
757 if (hinfo->hash_version <= DX_HASH_TEA)
758 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
759 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
760 if (fname && fname_name(fname))
761 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
764 if (root->info.unused_flags & 1) {
765 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
766 root->info.unused_flags);
770 indirect = root->info.indirect_levels;
772 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
773 root->info.indirect_levels);
777 entries = (struct dx_entry *)(((char *)&root->info) +
778 root->info.info_length);
780 if (dx_get_limit(entries) != dx_root_limit(dir,
781 root->info.info_length)) {
782 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
783 dx_get_limit(entries),
784 dx_root_limit(dir, root->info.info_length));
788 dxtrace(printk("Look up %x", hash));
790 count = dx_get_count(entries);
791 if (!count || count > dx_get_limit(entries)) {
792 ext4_warning_inode(dir,
793 "dx entry: count %u beyond limit %u",
794 count, dx_get_limit(entries));
799 q = entries + count - 1;
802 dxtrace(printk(KERN_CONT "."));
803 if (dx_get_hash(m) > hash)
809 if (0) { // linear search cross check
810 unsigned n = count - 1;
814 dxtrace(printk(KERN_CONT ","));
815 if (dx_get_hash(++at) > hash)
821 assert (at == p - 1);
825 dxtrace(printk(KERN_CONT " %x->%u\n",
826 at == entries ? 0 : dx_get_hash(at),
828 frame->entries = entries;
833 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
834 if (IS_ERR(frame->bh)) {
835 ret_err = (struct dx_frame *) frame->bh;
839 entries = ((struct dx_node *) frame->bh->b_data)->entries;
841 if (dx_get_limit(entries) != dx_node_limit(dir)) {
842 ext4_warning_inode(dir,
843 "dx entry: limit %u != node limit %u",
844 dx_get_limit(entries), dx_node_limit(dir));
849 while (frame >= frame_in) {
854 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
855 ext4_warning_inode(dir,
856 "Corrupt directory, running e2fsck is recommended");
860 static void dx_release(struct dx_frame *frames)
862 if (frames[0].bh == NULL)
865 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
866 brelse(frames[1].bh);
867 brelse(frames[0].bh);
871 * This function increments the frame pointer to search the next leaf
872 * block, and reads in the necessary intervening nodes if the search
873 * should be necessary. Whether or not the search is necessary is
874 * controlled by the hash parameter. If the hash value is even, then
875 * the search is only continued if the next block starts with that
876 * hash value. This is used if we are searching for a specific file.
878 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
880 * This function returns 1 if the caller should continue to search,
881 * or 0 if it should not. If there is an error reading one of the
882 * index blocks, it will a negative error code.
884 * If start_hash is non-null, it will be filled in with the starting
885 * hash of the next page.
887 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
888 struct dx_frame *frame,
889 struct dx_frame *frames,
893 struct buffer_head *bh;
899 * Find the next leaf page by incrementing the frame pointer.
900 * If we run out of entries in the interior node, loop around and
901 * increment pointer in the parent node. When we break out of
902 * this loop, num_frames indicates the number of interior
903 * nodes need to be read.
906 if (++(p->at) < p->entries + dx_get_count(p->entries))
915 * If the hash is 1, then continue only if the next page has a
916 * continuation hash of any value. This is used for readdir
917 * handling. Otherwise, check to see if the hash matches the
918 * desired contiuation hash. If it doesn't, return since
919 * there's no point to read in the successive index pages.
921 bhash = dx_get_hash(p->at);
924 if ((hash & 1) == 0) {
925 if ((bhash & ~1) != hash)
929 * If the hash is HASH_NB_ALWAYS, we always go to the next
930 * block so no check is necessary
932 while (num_frames--) {
933 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
939 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
946 * This function fills a red-black tree with information from a
947 * directory block. It returns the number directory entries loaded
948 * into the tree. If there is an error it is returned in err.
950 static int htree_dirblock_to_tree(struct file *dir_file,
951 struct inode *dir, ext4_lblk_t block,
952 struct dx_hash_info *hinfo,
953 __u32 start_hash, __u32 start_minor_hash)
955 struct buffer_head *bh;
956 struct ext4_dir_entry_2 *de, *top;
957 int err = 0, count = 0;
958 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
960 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
961 (unsigned long)block));
962 bh = ext4_read_dirblock(dir, block, DIRENT);
966 de = (struct ext4_dir_entry_2 *) bh->b_data;
967 top = (struct ext4_dir_entry_2 *) ((char *) de +
968 dir->i_sb->s_blocksize -
969 EXT4_DIR_REC_LEN(0));
970 #ifdef CONFIG_EXT4_FS_ENCRYPTION
971 /* Check if the directory is encrypted */
972 if (ext4_encrypted_inode(dir)) {
973 err = fscrypt_get_encryption_info(dir);
978 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
986 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
987 if (ext4_check_dir_entry(dir, NULL, de, bh,
988 bh->b_data, bh->b_size,
989 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
990 + ((char *)de - bh->b_data))) {
991 /* silently ignore the rest of the block */
994 ext4fs_dirhash(de->name, de->name_len, hinfo);
995 if ((hinfo->hash < start_hash) ||
996 ((hinfo->hash == start_hash) &&
997 (hinfo->minor_hash < start_minor_hash)))
1001 if (!ext4_encrypted_inode(dir)) {
1002 tmp_str.name = de->name;
1003 tmp_str.len = de->name_len;
1004 err = ext4_htree_store_dirent(dir_file,
1005 hinfo->hash, hinfo->minor_hash, de,
1008 int save_len = fname_crypto_str.len;
1009 struct fscrypt_str de_name = FSTR_INIT(de->name,
1012 /* Directory is encrypted */
1013 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1014 hinfo->minor_hash, &de_name,
1020 err = ext4_htree_store_dirent(dir_file,
1021 hinfo->hash, hinfo->minor_hash, de,
1023 fname_crypto_str.len = save_len;
1033 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1034 fscrypt_fname_free_buffer(&fname_crypto_str);
1041 * This function fills a red-black tree with information from a
1042 * directory. We start scanning the directory in hash order, starting
1043 * at start_hash and start_minor_hash.
1045 * This function returns the number of entries inserted into the tree,
1046 * or a negative error code.
1048 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1049 __u32 start_minor_hash, __u32 *next_hash)
1051 struct dx_hash_info hinfo;
1052 struct ext4_dir_entry_2 *de;
1053 struct dx_frame frames[2], *frame;
1059 struct fscrypt_str tmp_str;
1061 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1062 start_hash, start_minor_hash));
1063 dir = file_inode(dir_file);
1064 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1065 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1066 if (hinfo.hash_version <= DX_HASH_TEA)
1067 hinfo.hash_version +=
1068 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1069 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1070 if (ext4_has_inline_data(dir)) {
1071 int has_inline_data = 1;
1072 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1076 if (has_inline_data) {
1081 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1082 start_hash, start_minor_hash);
1086 hinfo.hash = start_hash;
1087 hinfo.minor_hash = 0;
1088 frame = dx_probe(NULL, dir, &hinfo, frames);
1090 return PTR_ERR(frame);
1092 /* Add '.' and '..' from the htree header */
1093 if (!start_hash && !start_minor_hash) {
1094 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1095 tmp_str.name = de->name;
1096 tmp_str.len = de->name_len;
1097 err = ext4_htree_store_dirent(dir_file, 0, 0,
1103 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1104 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1105 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1106 tmp_str.name = de->name;
1107 tmp_str.len = de->name_len;
1108 err = ext4_htree_store_dirent(dir_file, 2, 0,
1116 if (fatal_signal_pending(current)) {
1121 block = dx_get_block(frame->at);
1122 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1123 start_hash, start_minor_hash);
1130 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1131 frame, frames, &hashval);
1132 *next_hash = hashval;
1138 * Stop if: (a) there are no more entries, or
1139 * (b) we have inserted at least one entry and the
1140 * next hash value is not a continuation
1143 (count && ((hashval & 1) == 0)))
1147 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1148 "next hash: %x\n", count, *next_hash));
1155 static inline int search_dirblock(struct buffer_head *bh,
1157 struct ext4_filename *fname,
1158 unsigned int offset,
1159 struct ext4_dir_entry_2 **res_dir)
1161 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1162 fname, offset, res_dir);
1166 * Directory block splitting, compacting
1170 * Create map of hash values, offsets, and sizes, stored at end of block.
1171 * Returns number of entries mapped.
1173 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1174 unsigned blocksize, struct dx_hash_info *hinfo,
1175 struct dx_map_entry *map_tail)
1178 char *base = (char *) de;
1179 struct dx_hash_info h = *hinfo;
1181 while ((char *) de < base + blocksize) {
1182 if (de->name_len && de->inode) {
1183 ext4fs_dirhash(de->name, de->name_len, &h);
1185 map_tail->hash = h.hash;
1186 map_tail->offs = ((char *) de - base)>>2;
1187 map_tail->size = le16_to_cpu(de->rec_len);
1191 /* XXX: do we need to check rec_len == 0 case? -Chris */
1192 de = ext4_next_entry(de, blocksize);
1197 /* Sort map by hash value */
1198 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1200 struct dx_map_entry *p, *q, *top = map + count - 1;
1202 /* Combsort until bubble sort doesn't suck */
1204 count = count*10/13;
1205 if (count - 9 < 2) /* 9, 10 -> 11 */
1207 for (p = top, q = p - count; q >= map; p--, q--)
1208 if (p->hash < q->hash)
1211 /* Garden variety bubble sort */
1216 if (q[1].hash >= q[0].hash)
1224 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1226 struct dx_entry *entries = frame->entries;
1227 struct dx_entry *old = frame->at, *new = old + 1;
1228 int count = dx_get_count(entries);
1230 assert(count < dx_get_limit(entries));
1231 assert(old < entries + count);
1232 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1233 dx_set_hash(new, hash);
1234 dx_set_block(new, block);
1235 dx_set_count(entries, count + 1);
1239 * Test whether a directory entry matches the filename being searched for.
1241 * Return: %true if the directory entry matches, otherwise %false.
1243 static inline bool ext4_match(const struct ext4_filename *fname,
1244 const struct ext4_dir_entry_2 *de)
1246 struct fscrypt_name f;
1251 f.usr_fname = fname->usr_fname;
1252 f.disk_name = fname->disk_name;
1253 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1254 f.crypto_buf = fname->crypto_buf;
1256 return fscrypt_match_name(&f, de->name, de->name_len);
1260 * Returns 0 if not found, -1 on failure, and 1 on success
1262 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1263 struct inode *dir, struct ext4_filename *fname,
1264 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1266 struct ext4_dir_entry_2 * de;
1270 de = (struct ext4_dir_entry_2 *)search_buf;
1271 dlimit = search_buf + buf_size;
1272 while ((char *) de < dlimit) {
1273 /* this code is executed quadratically often */
1274 /* do minimal checking `by hand' */
1275 if ((char *) de + de->name_len <= dlimit &&
1276 ext4_match(fname, de)) {
1277 /* found a match - just to be sure, do
1279 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1280 bh->b_size, offset))
1285 /* prevent looping on a bad block */
1286 de_len = ext4_rec_len_from_disk(de->rec_len,
1287 dir->i_sb->s_blocksize);
1291 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1296 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1297 struct ext4_dir_entry *de)
1299 struct super_block *sb = dir->i_sb;
1305 if (de->inode == 0 &&
1306 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1315 * finds an entry in the specified directory with the wanted name. It
1316 * returns the cache buffer in which the entry was found, and the entry
1317 * itself (as a parameter - res_dir). It does NOT read the inode of the
1318 * entry - you'll have to do that yourself if you want to.
1320 * The returned buffer_head has ->b_count elevated. The caller is expected
1321 * to brelse() it when appropriate.
1323 static struct buffer_head * ext4_find_entry (struct inode *dir,
1324 const struct qstr *d_name,
1325 struct ext4_dir_entry_2 **res_dir,
1328 struct super_block *sb;
1329 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1330 struct buffer_head *bh, *ret = NULL;
1331 ext4_lblk_t start, block, b;
1332 const u8 *name = d_name->name;
1333 int ra_max = 0; /* Number of bh's in the readahead
1335 int ra_ptr = 0; /* Current index into readahead
1338 ext4_lblk_t nblocks;
1339 int i, namelen, retval;
1340 struct ext4_filename fname;
1344 namelen = d_name->len;
1345 if (namelen > EXT4_NAME_LEN)
1348 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1349 if (retval == -ENOENT)
1352 return ERR_PTR(retval);
1354 if (ext4_has_inline_data(dir)) {
1355 int has_inline_data = 1;
1356 ret = ext4_find_inline_entry(dir, &fname, res_dir,
1358 if (has_inline_data) {
1361 goto cleanup_and_exit;
1365 if ((namelen <= 2) && (name[0] == '.') &&
1366 (name[1] == '.' || name[1] == '\0')) {
1368 * "." or ".." will only be in the first block
1369 * NFS may look up ".."; "." should be handled by the VFS
1376 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1378 * On success, or if the error was file not found,
1379 * return. Otherwise, fall back to doing a search the
1380 * old fashioned way.
1382 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1383 goto cleanup_and_exit;
1384 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1387 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1388 start = EXT4_I(dir)->i_dir_start_lookup;
1389 if (start >= nblocks)
1395 * We deal with the read-ahead logic here.
1397 if (ra_ptr >= ra_max) {
1398 /* Refill the readahead buffer */
1401 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1403 * Terminate if we reach the end of the
1404 * directory and must wrap, or if our
1405 * search has finished at this block.
1407 if (b >= nblocks || (num && block == start)) {
1408 bh_use[ra_max] = NULL;
1412 bh = ext4_getblk(NULL, dir, b++, 0);
1416 goto cleanup_and_exit;
1420 bh_use[ra_max] = bh;
1422 ll_rw_block(REQ_OP_READ,
1423 REQ_META | REQ_PRIO,
1427 if ((bh = bh_use[ra_ptr++]) == NULL)
1430 if (!buffer_uptodate(bh)) {
1431 /* read error, skip block & hope for the best */
1432 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1433 (unsigned long) block);
1437 if (!buffer_verified(bh) &&
1438 !is_dx_internal_node(dir, block,
1439 (struct ext4_dir_entry *)bh->b_data) &&
1440 !ext4_dirent_csum_verify(dir,
1441 (struct ext4_dir_entry *)bh->b_data)) {
1442 EXT4_ERROR_INODE(dir, "checksumming directory "
1443 "block %lu", (unsigned long)block);
1447 set_buffer_verified(bh);
1448 i = search_dirblock(bh, dir, &fname,
1449 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1451 EXT4_I(dir)->i_dir_start_lookup = block;
1453 goto cleanup_and_exit;
1457 goto cleanup_and_exit;
1460 if (++block >= nblocks)
1462 } while (block != start);
1465 * If the directory has grown while we were searching, then
1466 * search the last part of the directory before giving up.
1469 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1470 if (block < nblocks) {
1476 /* Clean up the read-ahead blocks */
1477 for (; ra_ptr < ra_max; ra_ptr++)
1478 brelse(bh_use[ra_ptr]);
1479 ext4_fname_free_filename(&fname);
1483 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1484 struct ext4_filename *fname,
1485 struct ext4_dir_entry_2 **res_dir)
1487 struct super_block * sb = dir->i_sb;
1488 struct dx_frame frames[2], *frame;
1489 struct buffer_head *bh;
1493 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1496 frame = dx_probe(fname, dir, NULL, frames);
1498 return (struct buffer_head *) frame;
1500 block = dx_get_block(frame->at);
1501 bh = ext4_read_dirblock(dir, block, DIRENT);
1505 retval = search_dirblock(bh, dir, fname,
1506 block << EXT4_BLOCK_SIZE_BITS(sb),
1512 bh = ERR_PTR(ERR_BAD_DX_DIR);
1516 /* Check to see if we should continue to search */
1517 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1520 ext4_warning_inode(dir,
1521 "error %d reading directory index block",
1523 bh = ERR_PTR(retval);
1526 } while (retval == 1);
1530 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1536 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1538 struct inode *inode;
1539 struct ext4_dir_entry_2 *de;
1540 struct buffer_head *bh;
1542 if (ext4_encrypted_inode(dir)) {
1543 int res = fscrypt_get_encryption_info(dir);
1546 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1547 * created while the directory was encrypted and we
1548 * have access to the key.
1550 if (fscrypt_has_encryption_key(dir))
1551 fscrypt_set_encrypted_dentry(dentry);
1552 fscrypt_set_d_op(dentry);
1553 if (res && res != -ENOKEY)
1554 return ERR_PTR(res);
1557 if (dentry->d_name.len > EXT4_NAME_LEN)
1558 return ERR_PTR(-ENAMETOOLONG);
1560 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1562 return (struct dentry *) bh;
1565 __u32 ino = le32_to_cpu(de->inode);
1567 if (!ext4_valid_inum(dir->i_sb, ino)) {
1568 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1569 return ERR_PTR(-EFSCORRUPTED);
1571 if (unlikely(ino == dir->i_ino)) {
1572 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1574 return ERR_PTR(-EFSCORRUPTED);
1576 inode = ext4_iget_normal(dir->i_sb, ino);
1577 if (inode == ERR_PTR(-ESTALE)) {
1578 EXT4_ERROR_INODE(dir,
1579 "deleted inode referenced: %u",
1581 return ERR_PTR(-EFSCORRUPTED);
1583 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1584 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1585 !fscrypt_has_permitted_context(dir, inode)) {
1586 ext4_warning(inode->i_sb,
1587 "Inconsistent encryption contexts: %lu/%lu",
1588 dir->i_ino, inode->i_ino);
1590 return ERR_PTR(-EPERM);
1593 return d_splice_alias(inode, dentry);
1597 struct dentry *ext4_get_parent(struct dentry *child)
1600 static const struct qstr dotdot = QSTR_INIT("..", 2);
1601 struct ext4_dir_entry_2 * de;
1602 struct buffer_head *bh;
1604 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1606 return (struct dentry *) bh;
1608 return ERR_PTR(-ENOENT);
1609 ino = le32_to_cpu(de->inode);
1612 if (!ext4_valid_inum(child->d_sb, ino)) {
1613 EXT4_ERROR_INODE(d_inode(child),
1614 "bad parent inode number: %u", ino);
1615 return ERR_PTR(-EFSCORRUPTED);
1618 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1622 * Move count entries from end of map between two memory locations.
1623 * Returns pointer to last entry moved.
1625 static struct ext4_dir_entry_2 *
1626 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1629 unsigned rec_len = 0;
1632 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1633 (from + (map->offs<<2));
1634 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1635 memcpy (to, de, rec_len);
1636 ((struct ext4_dir_entry_2 *) to)->rec_len =
1637 ext4_rec_len_to_disk(rec_len, blocksize);
1642 return (struct ext4_dir_entry_2 *) (to - rec_len);
1646 * Compact each dir entry in the range to the minimal rec_len.
1647 * Returns pointer to last entry in range.
1649 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1651 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1652 unsigned rec_len = 0;
1655 while ((char*)de < base + blocksize) {
1656 next = ext4_next_entry(de, blocksize);
1657 if (de->inode && de->name_len) {
1658 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1660 memmove(to, de, rec_len);
1661 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1663 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1671 * Split a full leaf block to make room for a new dir entry.
1672 * Allocate a new block, and move entries so that they are approx. equally full.
1673 * Returns pointer to de in block into which the new entry will be inserted.
1675 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1676 struct buffer_head **bh,struct dx_frame *frame,
1677 struct dx_hash_info *hinfo)
1679 unsigned blocksize = dir->i_sb->s_blocksize;
1680 unsigned count, continued;
1681 struct buffer_head *bh2;
1682 ext4_lblk_t newblock;
1684 struct dx_map_entry *map;
1685 char *data1 = (*bh)->b_data, *data2;
1686 unsigned split, move, size;
1687 struct ext4_dir_entry_2 *de = NULL, *de2;
1688 struct ext4_dir_entry_tail *t;
1692 if (ext4_has_metadata_csum(dir->i_sb))
1693 csum_size = sizeof(struct ext4_dir_entry_tail);
1695 bh2 = ext4_append(handle, dir, &newblock);
1699 return (struct ext4_dir_entry_2 *) bh2;
1702 BUFFER_TRACE(*bh, "get_write_access");
1703 err = ext4_journal_get_write_access(handle, *bh);
1707 BUFFER_TRACE(frame->bh, "get_write_access");
1708 err = ext4_journal_get_write_access(handle, frame->bh);
1712 data2 = bh2->b_data;
1714 /* create map in the end of data2 block */
1715 map = (struct dx_map_entry *) (data2 + blocksize);
1716 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1717 blocksize, hinfo, map);
1719 dx_sort_map(map, count);
1720 /* Split the existing block in the middle, size-wise */
1723 for (i = count-1; i >= 0; i--) {
1724 /* is more than half of this entry in 2nd half of the block? */
1725 if (size + map[i].size/2 > blocksize/2)
1727 size += map[i].size;
1730 /* map index at which we will split */
1731 split = count - move;
1732 hash2 = map[split].hash;
1733 continued = hash2 == map[split - 1].hash;
1734 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1735 (unsigned long)dx_get_block(frame->at),
1736 hash2, split, count-split));
1738 /* Fancy dance to stay within two buffers */
1739 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1741 de = dx_pack_dirents(data1, blocksize);
1742 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1745 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1749 t = EXT4_DIRENT_TAIL(data2, blocksize);
1750 initialize_dirent_tail(t, blocksize);
1752 t = EXT4_DIRENT_TAIL(data1, blocksize);
1753 initialize_dirent_tail(t, blocksize);
1756 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1758 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1761 /* Which block gets the new entry? */
1762 if (hinfo->hash >= hash2) {
1766 dx_insert_block(frame, hash2 + continued, newblock);
1767 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1770 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1774 dxtrace(dx_show_index("frame", frame->entries));
1781 ext4_std_error(dir->i_sb, err);
1782 return ERR_PTR(err);
1785 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1786 struct buffer_head *bh,
1787 void *buf, int buf_size,
1788 struct ext4_filename *fname,
1789 struct ext4_dir_entry_2 **dest_de)
1791 struct ext4_dir_entry_2 *de;
1792 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1794 unsigned int offset = 0;
1797 de = (struct ext4_dir_entry_2 *)buf;
1798 top = buf + buf_size - reclen;
1799 while ((char *) de <= top) {
1800 if (ext4_check_dir_entry(dir, NULL, de, bh,
1801 buf, buf_size, offset))
1802 return -EFSCORRUPTED;
1803 if (ext4_match(fname, de))
1805 nlen = EXT4_DIR_REC_LEN(de->name_len);
1806 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1807 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1809 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1812 if ((char *) de > top)
1819 void ext4_insert_dentry(struct inode *inode,
1820 struct ext4_dir_entry_2 *de,
1822 struct ext4_filename *fname)
1827 nlen = EXT4_DIR_REC_LEN(de->name_len);
1828 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1830 struct ext4_dir_entry_2 *de1 =
1831 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1832 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1833 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1836 de->file_type = EXT4_FT_UNKNOWN;
1837 de->inode = cpu_to_le32(inode->i_ino);
1838 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1839 de->name_len = fname_len(fname);
1840 memcpy(de->name, fname_name(fname), fname_len(fname));
1844 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1845 * it points to a directory entry which is guaranteed to be large
1846 * enough for new directory entry. If de is NULL, then
1847 * add_dirent_to_buf will attempt search the directory block for
1848 * space. It will return -ENOSPC if no space is available, and -EIO
1849 * and -EEXIST if directory entry already exists.
1851 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1853 struct inode *inode, struct ext4_dir_entry_2 *de,
1854 struct buffer_head *bh)
1856 unsigned int blocksize = dir->i_sb->s_blocksize;
1860 if (ext4_has_metadata_csum(inode->i_sb))
1861 csum_size = sizeof(struct ext4_dir_entry_tail);
1864 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1865 blocksize - csum_size, fname, &de);
1869 BUFFER_TRACE(bh, "get_write_access");
1870 err = ext4_journal_get_write_access(handle, bh);
1872 ext4_std_error(dir->i_sb, err);
1876 /* By now the buffer is marked for journaling */
1877 ext4_insert_dentry(inode, de, blocksize, fname);
1880 * XXX shouldn't update any times until successful
1881 * completion of syscall, but too many callers depend
1884 * XXX similarly, too many callers depend on
1885 * ext4_new_inode() setting the times, but error
1886 * recovery deletes the inode, so the worst that can
1887 * happen is that the times are slightly out of date
1888 * and/or different from the directory change time.
1890 dir->i_mtime = dir->i_ctime = current_time(dir);
1891 ext4_update_dx_flag(dir);
1893 ext4_mark_inode_dirty(handle, dir);
1894 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1895 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1897 ext4_std_error(dir->i_sb, err);
1902 * This converts a one block unindexed directory to a 3 block indexed
1903 * directory, and adds the dentry to the indexed directory.
1905 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1907 struct inode *inode, struct buffer_head *bh)
1909 struct buffer_head *bh2;
1910 struct dx_root *root;
1911 struct dx_frame frames[2], *frame;
1912 struct dx_entry *entries;
1913 struct ext4_dir_entry_2 *de, *de2;
1914 struct ext4_dir_entry_tail *t;
1920 struct fake_dirent *fde;
1923 if (ext4_has_metadata_csum(inode->i_sb))
1924 csum_size = sizeof(struct ext4_dir_entry_tail);
1926 blocksize = dir->i_sb->s_blocksize;
1927 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1928 BUFFER_TRACE(bh, "get_write_access");
1929 retval = ext4_journal_get_write_access(handle, bh);
1931 ext4_std_error(dir->i_sb, retval);
1935 root = (struct dx_root *) bh->b_data;
1937 /* The 0th block becomes the root, move the dirents out */
1938 fde = &root->dotdot;
1939 de = (struct ext4_dir_entry_2 *)((char *)fde +
1940 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1941 if ((char *) de >= (((char *) root) + blocksize)) {
1942 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1944 return -EFSCORRUPTED;
1946 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1948 /* Allocate new block for the 0th block's dirents */
1949 bh2 = ext4_append(handle, dir, &block);
1952 return PTR_ERR(bh2);
1954 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1955 data1 = bh2->b_data;
1957 memcpy (data1, de, len);
1958 de = (struct ext4_dir_entry_2 *) data1;
1960 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1962 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1967 t = EXT4_DIRENT_TAIL(data1, blocksize);
1968 initialize_dirent_tail(t, blocksize);
1971 /* Initialize the root; the dot dirents already exist */
1972 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1973 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1975 memset (&root->info, 0, sizeof(root->info));
1976 root->info.info_length = sizeof(root->info);
1977 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1978 entries = root->entries;
1979 dx_set_block(entries, 1);
1980 dx_set_count(entries, 1);
1981 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1983 /* Initialize as for dx_probe */
1984 fname->hinfo.hash_version = root->info.hash_version;
1985 if (fname->hinfo.hash_version <= DX_HASH_TEA)
1986 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1987 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1988 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1990 memset(frames, 0, sizeof(frames));
1992 frame->entries = entries;
1993 frame->at = entries;
1996 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1999 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2003 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2005 retval = PTR_ERR(de);
2009 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2012 * Even if the block split failed, we have to properly write
2013 * out all the changes we did so far. Otherwise we can end up
2014 * with corrupted filesystem.
2017 ext4_mark_inode_dirty(handle, dir);
2026 * adds a file entry to the specified directory, using the same
2027 * semantics as ext4_find_entry(). It returns NULL if it failed.
2029 * NOTE!! The inode part of 'de' is left at 0 - which means you
2030 * may not sleep between calling this and putting something into
2031 * the entry, as someone else might have used it while you slept.
2033 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2034 struct inode *inode)
2036 struct inode *dir = d_inode(dentry->d_parent);
2037 struct buffer_head *bh = NULL;
2038 struct ext4_dir_entry_2 *de;
2039 struct ext4_dir_entry_tail *t;
2040 struct super_block *sb;
2041 struct ext4_filename fname;
2045 ext4_lblk_t block, blocks;
2048 if (ext4_has_metadata_csum(inode->i_sb))
2049 csum_size = sizeof(struct ext4_dir_entry_tail);
2052 blocksize = sb->s_blocksize;
2053 if (!dentry->d_name.len)
2056 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2060 if (ext4_has_inline_data(dir)) {
2061 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2071 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2072 if (!retval || (retval != ERR_BAD_DX_DIR))
2074 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2076 ext4_mark_inode_dirty(handle, dir);
2078 blocks = dir->i_size >> sb->s_blocksize_bits;
2079 for (block = 0; block < blocks; block++) {
2080 bh = ext4_read_dirblock(dir, block, DIRENT);
2082 retval = PTR_ERR(bh);
2086 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2088 if (retval != -ENOSPC)
2091 if (blocks == 1 && !dx_fallback &&
2092 ext4_has_feature_dir_index(sb)) {
2093 retval = make_indexed_dir(handle, &fname, dir,
2095 bh = NULL; /* make_indexed_dir releases bh */
2100 bh = ext4_append(handle, dir, &block);
2102 retval = PTR_ERR(bh);
2106 de = (struct ext4_dir_entry_2 *) bh->b_data;
2108 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2111 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2112 initialize_dirent_tail(t, blocksize);
2115 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2117 ext4_fname_free_filename(&fname);
2120 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2125 * Returns 0 for success, or a negative error value
2127 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2128 struct inode *dir, struct inode *inode)
2130 struct dx_frame frames[2], *frame;
2131 struct dx_entry *entries, *at;
2132 struct buffer_head *bh;
2133 struct super_block *sb = dir->i_sb;
2134 struct ext4_dir_entry_2 *de;
2137 frame = dx_probe(fname, dir, NULL, frames);
2139 return PTR_ERR(frame);
2140 entries = frame->entries;
2142 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2149 BUFFER_TRACE(bh, "get_write_access");
2150 err = ext4_journal_get_write_access(handle, bh);
2154 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2158 /* Block full, should compress but for now just split */
2159 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2160 dx_get_count(entries), dx_get_limit(entries)));
2161 /* Need to split index? */
2162 if (dx_get_count(entries) == dx_get_limit(entries)) {
2163 ext4_lblk_t newblock;
2164 unsigned icount = dx_get_count(entries);
2165 int levels = frame - frames;
2166 struct dx_entry *entries2;
2167 struct dx_node *node2;
2168 struct buffer_head *bh2;
2170 if (levels && (dx_get_count(frames->entries) ==
2171 dx_get_limit(frames->entries))) {
2172 ext4_warning_inode(dir, "Directory index full!");
2176 bh2 = ext4_append(handle, dir, &newblock);
2181 node2 = (struct dx_node *)(bh2->b_data);
2182 entries2 = node2->entries;
2183 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2184 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2186 BUFFER_TRACE(frame->bh, "get_write_access");
2187 err = ext4_journal_get_write_access(handle, frame->bh);
2191 unsigned icount1 = icount/2, icount2 = icount - icount1;
2192 unsigned hash2 = dx_get_hash(entries + icount1);
2193 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2196 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2197 err = ext4_journal_get_write_access(handle,
2202 memcpy((char *) entries2, (char *) (entries + icount1),
2203 icount2 * sizeof(struct dx_entry));
2204 dx_set_count(entries, icount1);
2205 dx_set_count(entries2, icount2);
2206 dx_set_limit(entries2, dx_node_limit(dir));
2208 /* Which index block gets the new entry? */
2209 if (at - entries >= icount1) {
2210 frame->at = at = at - entries - icount1 + entries2;
2211 frame->entries = entries = entries2;
2212 swap(frame->bh, bh2);
2214 dx_insert_block(frames + 0, hash2, newblock);
2215 dxtrace(dx_show_index("node", frames[1].entries));
2216 dxtrace(dx_show_index("node",
2217 ((struct dx_node *) bh2->b_data)->entries));
2218 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2223 dxtrace(printk(KERN_DEBUG
2224 "Creating second level index...\n"));
2225 memcpy((char *) entries2, (char *) entries,
2226 icount * sizeof(struct dx_entry));
2227 dx_set_limit(entries2, dx_node_limit(dir));
2230 dx_set_count(entries, 1);
2231 dx_set_block(entries + 0, newblock);
2232 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2234 /* Add new access path frame */
2236 frame->at = at = at - entries + entries2;
2237 frame->entries = entries = entries2;
2239 err = ext4_journal_get_write_access(handle,
2244 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2246 ext4_std_error(inode->i_sb, err);
2250 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2255 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2259 ext4_std_error(dir->i_sb, err);
2267 * ext4_generic_delete_entry deletes a directory entry by merging it
2268 * with the previous entry
2270 int ext4_generic_delete_entry(handle_t *handle,
2272 struct ext4_dir_entry_2 *de_del,
2273 struct buffer_head *bh,
2278 struct ext4_dir_entry_2 *de, *pde;
2279 unsigned int blocksize = dir->i_sb->s_blocksize;
2284 de = (struct ext4_dir_entry_2 *)entry_buf;
2285 while (i < buf_size - csum_size) {
2286 if (ext4_check_dir_entry(dir, NULL, de, bh,
2287 bh->b_data, bh->b_size, i))
2288 return -EFSCORRUPTED;
2291 pde->rec_len = ext4_rec_len_to_disk(
2292 ext4_rec_len_from_disk(pde->rec_len,
2294 ext4_rec_len_from_disk(de->rec_len,
2302 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2304 de = ext4_next_entry(de, blocksize);
2309 static int ext4_delete_entry(handle_t *handle,
2311 struct ext4_dir_entry_2 *de_del,
2312 struct buffer_head *bh)
2314 int err, csum_size = 0;
2316 if (ext4_has_inline_data(dir)) {
2317 int has_inline_data = 1;
2318 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2320 if (has_inline_data)
2324 if (ext4_has_metadata_csum(dir->i_sb))
2325 csum_size = sizeof(struct ext4_dir_entry_tail);
2327 BUFFER_TRACE(bh, "get_write_access");
2328 err = ext4_journal_get_write_access(handle, bh);
2332 err = ext4_generic_delete_entry(handle, dir, de_del,
2334 dir->i_sb->s_blocksize, csum_size);
2338 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2339 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2346 ext4_std_error(dir->i_sb, err);
2351 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2352 * since this indicates that nlinks count was previously 1.
2354 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2357 if (is_dx(inode) && inode->i_nlink > 1) {
2358 /* limit is 16-bit i_links_count */
2359 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2360 set_nlink(inode, 1);
2361 ext4_set_feature_dir_nlink(inode->i_sb);
2367 * If a directory had nlink == 1, then we should let it be 1. This indicates
2368 * directory has >EXT4_LINK_MAX subdirs.
2370 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2372 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2377 static int ext4_add_nondir(handle_t *handle,
2378 struct dentry *dentry, struct inode *inode)
2380 int err = ext4_add_entry(handle, dentry, inode);
2382 ext4_mark_inode_dirty(handle, inode);
2383 unlock_new_inode(inode);
2384 d_instantiate(dentry, inode);
2388 unlock_new_inode(inode);
2394 * By the time this is called, we already have created
2395 * the directory cache entry for the new file, but it
2396 * is so far negative - it has no inode.
2398 * If the create succeeds, we fill in the inode information
2399 * with d_instantiate().
2401 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2405 struct inode *inode;
2406 int err, credits, retries = 0;
2408 err = dquot_initialize(dir);
2412 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2413 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2415 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2416 NULL, EXT4_HT_DIR, credits);
2417 handle = ext4_journal_current_handle();
2418 err = PTR_ERR(inode);
2419 if (!IS_ERR(inode)) {
2420 inode->i_op = &ext4_file_inode_operations;
2421 inode->i_fop = &ext4_file_operations;
2422 ext4_set_aops(inode);
2423 err = ext4_add_nondir(handle, dentry, inode);
2424 if (!err && IS_DIRSYNC(dir))
2425 ext4_handle_sync(handle);
2428 ext4_journal_stop(handle);
2429 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2434 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2435 umode_t mode, dev_t rdev)
2438 struct inode *inode;
2439 int err, credits, retries = 0;
2441 err = dquot_initialize(dir);
2445 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2446 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2448 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2449 NULL, EXT4_HT_DIR, credits);
2450 handle = ext4_journal_current_handle();
2451 err = PTR_ERR(inode);
2452 if (!IS_ERR(inode)) {
2453 init_special_inode(inode, inode->i_mode, rdev);
2454 inode->i_op = &ext4_special_inode_operations;
2455 err = ext4_add_nondir(handle, dentry, inode);
2456 if (!err && IS_DIRSYNC(dir))
2457 ext4_handle_sync(handle);
2460 ext4_journal_stop(handle);
2461 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2466 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2469 struct inode *inode;
2470 int err, retries = 0;
2472 err = dquot_initialize(dir);
2477 inode = ext4_new_inode_start_handle(dir, mode,
2480 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2481 4 + EXT4_XATTR_TRANS_BLOCKS);
2482 handle = ext4_journal_current_handle();
2483 err = PTR_ERR(inode);
2484 if (!IS_ERR(inode)) {
2485 inode->i_op = &ext4_file_inode_operations;
2486 inode->i_fop = &ext4_file_operations;
2487 ext4_set_aops(inode);
2488 d_tmpfile(dentry, inode);
2489 err = ext4_orphan_add(handle, inode);
2491 goto err_unlock_inode;
2492 mark_inode_dirty(inode);
2493 unlock_new_inode(inode);
2496 ext4_journal_stop(handle);
2497 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2501 ext4_journal_stop(handle);
2502 unlock_new_inode(inode);
2506 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2507 struct ext4_dir_entry_2 *de,
2508 int blocksize, int csum_size,
2509 unsigned int parent_ino, int dotdot_real_len)
2511 de->inode = cpu_to_le32(inode->i_ino);
2513 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2515 strcpy(de->name, ".");
2516 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2518 de = ext4_next_entry(de, blocksize);
2519 de->inode = cpu_to_le32(parent_ino);
2521 if (!dotdot_real_len)
2522 de->rec_len = ext4_rec_len_to_disk(blocksize -
2523 (csum_size + EXT4_DIR_REC_LEN(1)),
2526 de->rec_len = ext4_rec_len_to_disk(
2527 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2528 strcpy(de->name, "..");
2529 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2531 return ext4_next_entry(de, blocksize);
2534 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2535 struct inode *inode)
2537 struct buffer_head *dir_block = NULL;
2538 struct ext4_dir_entry_2 *de;
2539 struct ext4_dir_entry_tail *t;
2540 ext4_lblk_t block = 0;
2541 unsigned int blocksize = dir->i_sb->s_blocksize;
2545 if (ext4_has_metadata_csum(dir->i_sb))
2546 csum_size = sizeof(struct ext4_dir_entry_tail);
2548 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2549 err = ext4_try_create_inline_dir(handle, dir, inode);
2550 if (err < 0 && err != -ENOSPC)
2557 dir_block = ext4_append(handle, inode, &block);
2558 if (IS_ERR(dir_block))
2559 return PTR_ERR(dir_block);
2560 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2561 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2562 set_nlink(inode, 2);
2564 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2565 initialize_dirent_tail(t, blocksize);
2568 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2569 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2572 set_buffer_verified(dir_block);
2578 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2581 struct inode *inode;
2582 int err, credits, retries = 0;
2584 if (EXT4_DIR_LINK_MAX(dir))
2587 err = dquot_initialize(dir);
2591 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2592 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2594 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2596 0, NULL, EXT4_HT_DIR, credits);
2597 handle = ext4_journal_current_handle();
2598 err = PTR_ERR(inode);
2602 inode->i_op = &ext4_dir_inode_operations;
2603 inode->i_fop = &ext4_dir_operations;
2604 err = ext4_init_new_dir(handle, dir, inode);
2606 goto out_clear_inode;
2607 err = ext4_mark_inode_dirty(handle, inode);
2609 err = ext4_add_entry(handle, dentry, inode);
2613 unlock_new_inode(inode);
2614 ext4_mark_inode_dirty(handle, inode);
2618 ext4_inc_count(handle, dir);
2619 ext4_update_dx_flag(dir);
2620 err = ext4_mark_inode_dirty(handle, dir);
2622 goto out_clear_inode;
2623 unlock_new_inode(inode);
2624 d_instantiate(dentry, inode);
2625 if (IS_DIRSYNC(dir))
2626 ext4_handle_sync(handle);
2630 ext4_journal_stop(handle);
2631 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2637 * routine to check that the specified directory is empty (for rmdir)
2639 bool ext4_empty_dir(struct inode *inode)
2641 unsigned int offset;
2642 struct buffer_head *bh;
2643 struct ext4_dir_entry_2 *de, *de1;
2644 struct super_block *sb;
2646 if (ext4_has_inline_data(inode)) {
2647 int has_inline_data = 1;
2650 ret = empty_inline_dir(inode, &has_inline_data);
2651 if (has_inline_data)
2656 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2657 EXT4_ERROR_INODE(inode, "invalid size");
2660 bh = ext4_read_dirblock(inode, 0, EITHER);
2664 de = (struct ext4_dir_entry_2 *) bh->b_data;
2665 de1 = ext4_next_entry(de, sb->s_blocksize);
2666 if (le32_to_cpu(de->inode) != inode->i_ino ||
2667 le32_to_cpu(de1->inode) == 0 ||
2668 strcmp(".", de->name) || strcmp("..", de1->name)) {
2669 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2673 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2674 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2675 de = ext4_next_entry(de1, sb->s_blocksize);
2676 while (offset < inode->i_size) {
2677 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2678 unsigned int lblock;
2680 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2681 bh = ext4_read_dirblock(inode, lblock, EITHER);
2684 de = (struct ext4_dir_entry_2 *) bh->b_data;
2686 if (ext4_check_dir_entry(inode, NULL, de, bh,
2687 bh->b_data, bh->b_size, offset)) {
2688 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2690 offset = (offset | (sb->s_blocksize - 1)) + 1;
2693 if (le32_to_cpu(de->inode)) {
2697 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2698 de = ext4_next_entry(de, sb->s_blocksize);
2705 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2706 * such inodes, starting at the superblock, in case we crash before the
2707 * file is closed/deleted, or in case the inode truncate spans multiple
2708 * transactions and the last transaction is not recovered after a crash.
2710 * At filesystem recovery time, we walk this list deleting unlinked
2711 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2713 * Orphan list manipulation functions must be called under i_mutex unless
2714 * we are just creating the inode or deleting it.
2716 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2718 struct super_block *sb = inode->i_sb;
2719 struct ext4_sb_info *sbi = EXT4_SB(sb);
2720 struct ext4_iloc iloc;
2724 if (!sbi->s_journal || is_bad_inode(inode))
2727 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2728 !inode_is_locked(inode));
2730 * Exit early if inode already is on orphan list. This is a big speedup
2731 * since we don't have to contend on the global s_orphan_lock.
2733 if (!list_empty(&EXT4_I(inode)->i_orphan))
2737 * Orphan handling is only valid for files with data blocks
2738 * being truncated, or files being unlinked. Note that we either
2739 * hold i_mutex, or the inode can not be referenced from outside,
2740 * so i_nlink should not be bumped due to race
2742 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2743 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2745 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2746 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2750 err = ext4_reserve_inode_write(handle, inode, &iloc);
2754 mutex_lock(&sbi->s_orphan_lock);
2756 * Due to previous errors inode may be already a part of on-disk
2757 * orphan list. If so skip on-disk list modification.
2759 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2760 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2761 /* Insert this inode at the head of the on-disk orphan list */
2762 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2763 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2766 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2767 mutex_unlock(&sbi->s_orphan_lock);
2770 err = ext4_handle_dirty_super(handle, sb);
2771 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2776 * We have to remove inode from in-memory list if
2777 * addition to on disk orphan list failed. Stray orphan
2778 * list entries can cause panics at unmount time.
2780 mutex_lock(&sbi->s_orphan_lock);
2781 list_del_init(&EXT4_I(inode)->i_orphan);
2782 mutex_unlock(&sbi->s_orphan_lock);
2785 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2786 jbd_debug(4, "orphan inode %lu will point to %d\n",
2787 inode->i_ino, NEXT_ORPHAN(inode));
2789 ext4_std_error(sb, err);
2794 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2795 * of such inodes stored on disk, because it is finally being cleaned up.
2797 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2799 struct list_head *prev;
2800 struct ext4_inode_info *ei = EXT4_I(inode);
2801 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2803 struct ext4_iloc iloc;
2806 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2809 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2810 !inode_is_locked(inode));
2811 /* Do this quick check before taking global s_orphan_lock. */
2812 if (list_empty(&ei->i_orphan))
2816 /* Grab inode buffer early before taking global s_orphan_lock */
2817 err = ext4_reserve_inode_write(handle, inode, &iloc);
2820 mutex_lock(&sbi->s_orphan_lock);
2821 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2823 prev = ei->i_orphan.prev;
2824 list_del_init(&ei->i_orphan);
2826 /* If we're on an error path, we may not have a valid
2827 * transaction handle with which to update the orphan list on
2828 * disk, but we still need to remove the inode from the linked
2829 * list in memory. */
2830 if (!handle || err) {
2831 mutex_unlock(&sbi->s_orphan_lock);
2835 ino_next = NEXT_ORPHAN(inode);
2836 if (prev == &sbi->s_orphan) {
2837 jbd_debug(4, "superblock will point to %u\n", ino_next);
2838 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2839 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2841 mutex_unlock(&sbi->s_orphan_lock);
2844 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2845 mutex_unlock(&sbi->s_orphan_lock);
2846 err = ext4_handle_dirty_super(handle, inode->i_sb);
2848 struct ext4_iloc iloc2;
2849 struct inode *i_prev =
2850 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2852 jbd_debug(4, "orphan inode %lu will point to %u\n",
2853 i_prev->i_ino, ino_next);
2854 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2856 mutex_unlock(&sbi->s_orphan_lock);
2859 NEXT_ORPHAN(i_prev) = ino_next;
2860 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2861 mutex_unlock(&sbi->s_orphan_lock);
2865 NEXT_ORPHAN(inode) = 0;
2866 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2868 ext4_std_error(inode->i_sb, err);
2876 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2879 struct inode *inode;
2880 struct buffer_head *bh;
2881 struct ext4_dir_entry_2 *de;
2882 handle_t *handle = NULL;
2884 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2887 /* Initialize quotas before so that eventual writes go in
2888 * separate transaction */
2889 retval = dquot_initialize(dir);
2892 retval = dquot_initialize(d_inode(dentry));
2897 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2903 inode = d_inode(dentry);
2905 retval = -EFSCORRUPTED;
2906 if (le32_to_cpu(de->inode) != inode->i_ino)
2909 retval = -ENOTEMPTY;
2910 if (!ext4_empty_dir(inode))
2913 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2914 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2915 if (IS_ERR(handle)) {
2916 retval = PTR_ERR(handle);
2921 if (IS_DIRSYNC(dir))
2922 ext4_handle_sync(handle);
2924 retval = ext4_delete_entry(handle, dir, de, bh);
2927 if (!EXT4_DIR_LINK_EMPTY(inode))
2928 ext4_warning_inode(inode,
2929 "empty directory '%.*s' has too many links (%u)",
2930 dentry->d_name.len, dentry->d_name.name,
2934 /* There's no need to set i_disksize: the fact that i_nlink is
2935 * zero will ensure that the right thing happens during any
2938 ext4_orphan_add(handle, inode);
2939 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2940 ext4_mark_inode_dirty(handle, inode);
2941 ext4_dec_count(handle, dir);
2942 ext4_update_dx_flag(dir);
2943 ext4_mark_inode_dirty(handle, dir);
2948 ext4_journal_stop(handle);
2952 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2955 struct inode *inode;
2956 struct buffer_head *bh;
2957 struct ext4_dir_entry_2 *de;
2958 handle_t *handle = NULL;
2960 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2963 trace_ext4_unlink_enter(dir, dentry);
2964 /* Initialize quotas before so that eventual writes go
2965 * in separate transaction */
2966 retval = dquot_initialize(dir);
2969 retval = dquot_initialize(d_inode(dentry));
2974 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2980 inode = d_inode(dentry);
2982 retval = -EFSCORRUPTED;
2983 if (le32_to_cpu(de->inode) != inode->i_ino)
2986 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2987 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2988 if (IS_ERR(handle)) {
2989 retval = PTR_ERR(handle);
2994 if (IS_DIRSYNC(dir))
2995 ext4_handle_sync(handle);
2997 if (inode->i_nlink == 0) {
2998 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
2999 dentry->d_name.len, dentry->d_name.name);
3000 set_nlink(inode, 1);
3002 retval = ext4_delete_entry(handle, dir, de, bh);
3005 dir->i_ctime = dir->i_mtime = current_time(dir);
3006 ext4_update_dx_flag(dir);
3007 ext4_mark_inode_dirty(handle, dir);
3009 if (!inode->i_nlink)
3010 ext4_orphan_add(handle, inode);
3011 inode->i_ctime = current_time(inode);
3012 ext4_mark_inode_dirty(handle, inode);
3017 ext4_journal_stop(handle);
3018 trace_ext4_unlink_exit(dentry, retval);
3022 static int ext4_symlink(struct inode *dir,
3023 struct dentry *dentry, const char *symname)
3026 struct inode *inode;
3027 int err, len = strlen(symname);
3029 bool encryption_required;
3030 struct fscrypt_str disk_link;
3031 struct fscrypt_symlink_data *sd = NULL;
3033 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3036 disk_link.len = len + 1;
3037 disk_link.name = (char *) symname;
3039 encryption_required = (ext4_encrypted_inode(dir) ||
3040 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3041 if (encryption_required) {
3042 err = fscrypt_get_encryption_info(dir);
3045 if (!fscrypt_has_encryption_key(dir))
3047 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3048 sizeof(struct fscrypt_symlink_data));
3049 sd = kzalloc(disk_link.len, GFP_KERNEL);
3054 if (disk_link.len > dir->i_sb->s_blocksize) {
3055 err = -ENAMETOOLONG;
3059 err = dquot_initialize(dir);
3063 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3065 * For non-fast symlinks, we just allocate inode and put it on
3066 * orphan list in the first transaction => we need bitmap,
3067 * group descriptor, sb, inode block, quota blocks, and
3068 * possibly selinux xattr blocks.
3070 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3071 EXT4_XATTR_TRANS_BLOCKS;
3074 * Fast symlink. We have to add entry to directory
3075 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3076 * allocate new inode (bitmap, group descriptor, inode block,
3077 * quota blocks, sb is already counted in previous macros).
3079 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3080 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3083 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3084 &dentry->d_name, 0, NULL,
3085 EXT4_HT_DIR, credits);
3086 handle = ext4_journal_current_handle();
3087 if (IS_ERR(inode)) {
3089 ext4_journal_stop(handle);
3090 err = PTR_ERR(inode);
3094 if (encryption_required) {
3096 struct fscrypt_str ostr =
3097 FSTR_INIT(sd->encrypted_path, disk_link.len);
3099 istr.name = (const unsigned char *) symname;
3101 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3103 goto err_drop_inode;
3104 sd->len = cpu_to_le16(ostr.len);
3105 disk_link.name = (char *) sd;
3106 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3109 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3110 if (!encryption_required)
3111 inode->i_op = &ext4_symlink_inode_operations;
3112 inode_nohighmem(inode);
3113 ext4_set_aops(inode);
3115 * We cannot call page_symlink() with transaction started
3116 * because it calls into ext4_write_begin() which can wait
3117 * for transaction commit if we are running out of space
3118 * and thus we deadlock. So we have to stop transaction now
3119 * and restart it when symlink contents is written.
3121 * To keep fs consistent in case of crash, we have to put inode
3122 * to orphan list in the mean time.
3125 err = ext4_orphan_add(handle, inode);
3126 ext4_journal_stop(handle);
3129 goto err_drop_inode;
3130 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3132 goto err_drop_inode;
3134 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3135 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3137 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3138 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3139 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3140 if (IS_ERR(handle)) {
3141 err = PTR_ERR(handle);
3143 goto err_drop_inode;
3145 set_nlink(inode, 1);
3146 err = ext4_orphan_del(handle, inode);
3148 goto err_drop_inode;
3150 /* clear the extent format for fast symlink */
3151 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3152 if (!encryption_required) {
3153 inode->i_op = &ext4_fast_symlink_inode_operations;
3154 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3156 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3158 inode->i_size = disk_link.len - 1;
3160 EXT4_I(inode)->i_disksize = inode->i_size;
3161 err = ext4_add_nondir(handle, dentry, inode);
3162 if (!err && IS_DIRSYNC(dir))
3163 ext4_handle_sync(handle);
3166 ext4_journal_stop(handle);
3171 ext4_journal_stop(handle);
3173 unlock_new_inode(inode);
3180 static int ext4_link(struct dentry *old_dentry,
3181 struct inode *dir, struct dentry *dentry)
3184 struct inode *inode = d_inode(old_dentry);
3185 int err, retries = 0;
3187 if (inode->i_nlink >= EXT4_LINK_MAX)
3189 if (ext4_encrypted_inode(dir) &&
3190 !fscrypt_has_permitted_context(dir, inode))
3193 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3194 (!projid_eq(EXT4_I(dir)->i_projid,
3195 EXT4_I(old_dentry->d_inode)->i_projid)))
3198 err = dquot_initialize(dir);
3203 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3204 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3205 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3207 return PTR_ERR(handle);
3209 if (IS_DIRSYNC(dir))
3210 ext4_handle_sync(handle);
3212 inode->i_ctime = current_time(inode);
3213 ext4_inc_count(handle, inode);
3216 err = ext4_add_entry(handle, dentry, inode);
3218 ext4_mark_inode_dirty(handle, inode);
3219 /* this can happen only for tmpfile being
3220 * linked the first time
3222 if (inode->i_nlink == 1)
3223 ext4_orphan_del(handle, inode);
3224 d_instantiate(dentry, inode);
3229 ext4_journal_stop(handle);
3230 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3237 * Try to find buffer head where contains the parent block.
3238 * It should be the inode block if it is inlined or the 1st block
3239 * if it is a normal dir.
3241 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3242 struct inode *inode,
3244 struct ext4_dir_entry_2 **parent_de,
3247 struct buffer_head *bh;
3249 if (!ext4_has_inline_data(inode)) {
3250 bh = ext4_read_dirblock(inode, 0, EITHER);
3252 *retval = PTR_ERR(bh);
3255 *parent_de = ext4_next_entry(
3256 (struct ext4_dir_entry_2 *)bh->b_data,
3257 inode->i_sb->s_blocksize);
3262 return ext4_get_first_inline_block(inode, parent_de, retval);
3265 struct ext4_renament {
3267 struct dentry *dentry;
3268 struct inode *inode;
3270 int dir_nlink_delta;
3272 /* entry for "dentry" */
3273 struct buffer_head *bh;
3274 struct ext4_dir_entry_2 *de;
3277 /* entry for ".." in inode if it's a directory */
3278 struct buffer_head *dir_bh;
3279 struct ext4_dir_entry_2 *parent_de;
3283 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3287 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3288 &retval, &ent->parent_de,
3292 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3293 return -EFSCORRUPTED;
3294 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3295 return ext4_journal_get_write_access(handle, ent->dir_bh);
3298 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3303 ent->parent_de->inode = cpu_to_le32(dir_ino);
3304 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3305 if (!ent->dir_inlined) {
3306 if (is_dx(ent->inode)) {
3307 retval = ext4_handle_dirty_dx_node(handle,
3311 retval = ext4_handle_dirty_dirent_node(handle,
3316 retval = ext4_mark_inode_dirty(handle, ent->inode);
3319 ext4_std_error(ent->dir->i_sb, retval);
3325 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3326 unsigned ino, unsigned file_type)
3330 BUFFER_TRACE(ent->bh, "get write access");
3331 retval = ext4_journal_get_write_access(handle, ent->bh);
3334 ent->de->inode = cpu_to_le32(ino);
3335 if (ext4_has_feature_filetype(ent->dir->i_sb))
3336 ent->de->file_type = file_type;
3337 ent->dir->i_version++;
3338 ent->dir->i_ctime = ent->dir->i_mtime =
3339 current_time(ent->dir);
3340 ext4_mark_inode_dirty(handle, ent->dir);
3341 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3342 if (!ent->inlined) {
3343 retval = ext4_handle_dirty_dirent_node(handle,
3345 if (unlikely(retval)) {
3346 ext4_std_error(ent->dir->i_sb, retval);
3356 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3357 const struct qstr *d_name)
3359 int retval = -ENOENT;
3360 struct buffer_head *bh;
3361 struct ext4_dir_entry_2 *de;
3363 bh = ext4_find_entry(dir, d_name, &de, NULL);
3367 retval = ext4_delete_entry(handle, dir, de, bh);
3373 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3378 * ent->de could have moved from under us during htree split, so make
3379 * sure that we are deleting the right entry. We might also be pointing
3380 * to a stale entry in the unused part of ent->bh so just checking inum
3381 * and the name isn't enough.
3383 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3384 ent->de->name_len != ent->dentry->d_name.len ||
3385 strncmp(ent->de->name, ent->dentry->d_name.name,
3386 ent->de->name_len) ||
3388 retval = ext4_find_delete_entry(handle, ent->dir,
3389 &ent->dentry->d_name);
3391 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3392 if (retval == -ENOENT) {
3393 retval = ext4_find_delete_entry(handle, ent->dir,
3394 &ent->dentry->d_name);
3399 ext4_warning_inode(ent->dir,
3400 "Deleting old file: nlink %d, error=%d",
3401 ent->dir->i_nlink, retval);
3405 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3407 if (ent->dir_nlink_delta) {
3408 if (ent->dir_nlink_delta == -1)
3409 ext4_dec_count(handle, ent->dir);
3411 ext4_inc_count(handle, ent->dir);
3412 ext4_mark_inode_dirty(handle, ent->dir);
3416 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3417 int credits, handle_t **h)
3424 * for inode block, sb block, group summaries,
3427 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3428 EXT4_XATTR_TRANS_BLOCKS + 4);
3430 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3431 &ent->dentry->d_name, 0, NULL,
3432 EXT4_HT_DIR, credits);
3434 handle = ext4_journal_current_handle();
3437 ext4_journal_stop(handle);
3438 if (PTR_ERR(wh) == -ENOSPC &&
3439 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3443 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3444 wh->i_op = &ext4_special_inode_operations;
3450 * Anybody can rename anything with this: the permission checks are left to the
3451 * higher-level routines.
3453 * n.b. old_{dentry,inode) refers to the source dentry/inode
3454 * while new_{dentry,inode) refers to the destination dentry/inode
3455 * This comes from rename(const char *oldpath, const char *newpath)
3457 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3458 struct inode *new_dir, struct dentry *new_dentry,
3461 handle_t *handle = NULL;
3462 struct ext4_renament old = {
3464 .dentry = old_dentry,
3465 .inode = d_inode(old_dentry),
3467 struct ext4_renament new = {
3469 .dentry = new_dentry,
3470 .inode = d_inode(new_dentry),
3474 struct inode *whiteout = NULL;
3478 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3479 (!projid_eq(EXT4_I(new_dir)->i_projid,
3480 EXT4_I(old_dentry->d_inode)->i_projid)))
3483 if ((ext4_encrypted_inode(old_dir) &&
3484 !fscrypt_has_encryption_key(old_dir)) ||
3485 (ext4_encrypted_inode(new_dir) &&
3486 !fscrypt_has_encryption_key(new_dir)))
3489 retval = dquot_initialize(old.dir);
3492 retval = dquot_initialize(new.dir);
3496 /* Initialize quotas before so that eventual writes go
3497 * in separate transaction */
3499 retval = dquot_initialize(new.inode);
3504 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3506 return PTR_ERR(old.bh);
3508 * Check for inode number is _not_ due to possible IO errors.
3509 * We might rmdir the source, keep it as pwd of some process
3510 * and merrily kill the link to whatever was created under the
3511 * same name. Goodbye sticky bit ;-<
3514 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3517 if ((old.dir != new.dir) &&
3518 ext4_encrypted_inode(new.dir) &&
3519 !fscrypt_has_permitted_context(new.dir, old.inode)) {
3524 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3525 &new.de, &new.inlined);
3526 if (IS_ERR(new.bh)) {
3527 retval = PTR_ERR(new.bh);
3537 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3538 ext4_alloc_da_blocks(old.inode);
3540 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3541 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3542 if (!(flags & RENAME_WHITEOUT)) {
3543 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3544 if (IS_ERR(handle)) {
3545 retval = PTR_ERR(handle);
3550 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3551 if (IS_ERR(whiteout)) {
3552 retval = PTR_ERR(whiteout);
3558 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3559 ext4_handle_sync(handle);
3561 if (S_ISDIR(old.inode->i_mode)) {
3563 retval = -ENOTEMPTY;
3564 if (!ext4_empty_dir(new.inode))
3568 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3571 retval = ext4_rename_dir_prepare(handle, &old);
3576 * If we're renaming a file within an inline_data dir and adding or
3577 * setting the new dirent causes a conversion from inline_data to
3578 * extents/blockmap, we need to force the dirent delete code to
3579 * re-read the directory, or else we end up trying to delete a dirent
3580 * from what is now the extent tree root (or a block map).
3582 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3583 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3585 old_file_type = old.de->file_type;
3588 * Do this before adding a new entry, so the old entry is sure
3589 * to be still pointing to the valid old entry.
3591 retval = ext4_setent(handle, &old, whiteout->i_ino,
3595 ext4_mark_inode_dirty(handle, whiteout);
3598 retval = ext4_add_entry(handle, new.dentry, old.inode);
3602 retval = ext4_setent(handle, &new,
3603 old.inode->i_ino, old_file_type);
3608 force_reread = !ext4_test_inode_flag(new.dir,
3609 EXT4_INODE_INLINE_DATA);
3612 * Like most other Unix systems, set the ctime for inodes on a
3615 old.inode->i_ctime = current_time(old.inode);
3616 ext4_mark_inode_dirty(handle, old.inode);
3622 ext4_rename_delete(handle, &old, force_reread);
3626 ext4_dec_count(handle, new.inode);
3627 new.inode->i_ctime = current_time(new.inode);
3629 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3630 ext4_update_dx_flag(old.dir);
3632 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3636 ext4_dec_count(handle, old.dir);
3638 /* checked ext4_empty_dir above, can't have another
3639 * parent, ext4_dec_count() won't work for many-linked
3641 clear_nlink(new.inode);
3643 ext4_inc_count(handle, new.dir);
3644 ext4_update_dx_flag(new.dir);
3645 ext4_mark_inode_dirty(handle, new.dir);
3648 ext4_mark_inode_dirty(handle, old.dir);
3650 ext4_mark_inode_dirty(handle, new.inode);
3651 if (!new.inode->i_nlink)
3652 ext4_orphan_add(handle, new.inode);
3662 drop_nlink(whiteout);
3663 unlock_new_inode(whiteout);
3667 ext4_journal_stop(handle);
3671 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3672 struct inode *new_dir, struct dentry *new_dentry)
3674 handle_t *handle = NULL;
3675 struct ext4_renament old = {
3677 .dentry = old_dentry,
3678 .inode = d_inode(old_dentry),
3680 struct ext4_renament new = {
3682 .dentry = new_dentry,
3683 .inode = d_inode(new_dentry),
3687 struct timespec ctime;
3689 if ((ext4_encrypted_inode(old_dir) &&
3690 !fscrypt_has_encryption_key(old_dir)) ||
3691 (ext4_encrypted_inode(new_dir) &&
3692 !fscrypt_has_encryption_key(new_dir)))
3695 if ((ext4_encrypted_inode(old_dir) ||
3696 ext4_encrypted_inode(new_dir)) &&
3697 (old_dir != new_dir) &&
3698 (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3699 !fscrypt_has_permitted_context(old_dir, new.inode)))
3702 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3703 !projid_eq(EXT4_I(new_dir)->i_projid,
3704 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3705 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3706 !projid_eq(EXT4_I(old_dir)->i_projid,
3707 EXT4_I(new_dentry->d_inode)->i_projid)))
3710 retval = dquot_initialize(old.dir);
3713 retval = dquot_initialize(new.dir);
3717 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3718 &old.de, &old.inlined);
3720 return PTR_ERR(old.bh);
3722 * Check for inode number is _not_ due to possible IO errors.
3723 * We might rmdir the source, keep it as pwd of some process
3724 * and merrily kill the link to whatever was created under the
3725 * same name. Goodbye sticky bit ;-<
3728 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3731 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3732 &new.de, &new.inlined);
3733 if (IS_ERR(new.bh)) {
3734 retval = PTR_ERR(new.bh);
3739 /* RENAME_EXCHANGE case: old *and* new must both exist */
3740 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3743 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3744 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3745 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3746 if (IS_ERR(handle)) {
3747 retval = PTR_ERR(handle);
3752 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3753 ext4_handle_sync(handle);
3755 if (S_ISDIR(old.inode->i_mode)) {
3757 retval = ext4_rename_dir_prepare(handle, &old);
3761 if (S_ISDIR(new.inode->i_mode)) {
3763 retval = ext4_rename_dir_prepare(handle, &new);
3769 * Other than the special case of overwriting a directory, parents'
3770 * nlink only needs to be modified if this is a cross directory rename.
3772 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3773 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3774 new.dir_nlink_delta = -old.dir_nlink_delta;
3776 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3777 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3781 new_file_type = new.de->file_type;
3782 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3786 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3791 * Like most other Unix systems, set the ctime for inodes on a
3794 ctime = current_time(old.inode);
3795 old.inode->i_ctime = ctime;
3796 new.inode->i_ctime = ctime;
3797 ext4_mark_inode_dirty(handle, old.inode);
3798 ext4_mark_inode_dirty(handle, new.inode);
3801 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3806 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3810 ext4_update_dir_count(handle, &old);
3811 ext4_update_dir_count(handle, &new);
3820 ext4_journal_stop(handle);
3824 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3825 struct inode *new_dir, struct dentry *new_dentry,
3828 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3831 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3834 if (flags & RENAME_EXCHANGE) {
3835 return ext4_cross_rename(old_dir, old_dentry,
3836 new_dir, new_dentry);
3839 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3843 * directories can handle most operations...
3845 const struct inode_operations ext4_dir_inode_operations = {
3846 .create = ext4_create,
3847 .lookup = ext4_lookup,
3849 .unlink = ext4_unlink,
3850 .symlink = ext4_symlink,
3851 .mkdir = ext4_mkdir,
3852 .rmdir = ext4_rmdir,
3853 .mknod = ext4_mknod,
3854 .tmpfile = ext4_tmpfile,
3855 .rename = ext4_rename2,
3856 .setattr = ext4_setattr,
3857 .getattr = ext4_getattr,
3858 .listxattr = ext4_listxattr,
3859 .get_acl = ext4_get_acl,
3860 .set_acl = ext4_set_acl,
3861 .fiemap = ext4_fiemap,
3864 const struct inode_operations ext4_special_inode_operations = {
3865 .setattr = ext4_setattr,
3866 .getattr = ext4_getattr,
3867 .listxattr = ext4_listxattr,
3868 .get_acl = ext4_get_acl,
3869 .set_acl = ext4_set_acl,