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) & 0x0fffffff;
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 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
743 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
744 if (IS_ERR(frame->bh))
745 return (struct dx_frame *) frame->bh;
747 root = (struct dx_root *) frame->bh->b_data;
748 if (root->info.hash_version != DX_HASH_TEA &&
749 root->info.hash_version != DX_HASH_HALF_MD4 &&
750 root->info.hash_version != DX_HASH_LEGACY) {
751 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
752 root->info.hash_version);
756 hinfo = &fname->hinfo;
757 hinfo->hash_version = root->info.hash_version;
758 if (hinfo->hash_version <= DX_HASH_TEA)
759 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
760 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
761 if (fname && fname_name(fname))
762 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
765 if (root->info.unused_flags & 1) {
766 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
767 root->info.unused_flags);
771 indirect = root->info.indirect_levels;
772 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
773 ext4_warning(dir->i_sb,
774 "Directory (ino: %lu) htree depth %#06x exceed"
775 "supported value", dir->i_ino,
776 ext4_dir_htree_level(dir->i_sb));
777 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
778 ext4_warning(dir->i_sb, "Enable large directory "
779 "feature to access it");
784 entries = (struct dx_entry *)(((char *)&root->info) +
785 root->info.info_length);
787 if (dx_get_limit(entries) != dx_root_limit(dir,
788 root->info.info_length)) {
789 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
790 dx_get_limit(entries),
791 dx_root_limit(dir, root->info.info_length));
795 dxtrace(printk("Look up %x", hash));
797 count = dx_get_count(entries);
798 if (!count || count > dx_get_limit(entries)) {
799 ext4_warning_inode(dir,
800 "dx entry: count %u beyond limit %u",
801 count, dx_get_limit(entries));
806 q = entries + count - 1;
809 dxtrace(printk(KERN_CONT "."));
810 if (dx_get_hash(m) > hash)
816 if (0) { // linear search cross check
817 unsigned n = count - 1;
821 dxtrace(printk(KERN_CONT ","));
822 if (dx_get_hash(++at) > hash)
828 assert (at == p - 1);
832 dxtrace(printk(KERN_CONT " %x->%u\n",
833 at == entries ? 0 : dx_get_hash(at),
835 frame->entries = entries;
840 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
841 if (IS_ERR(frame->bh)) {
842 ret_err = (struct dx_frame *) frame->bh;
846 entries = ((struct dx_node *) frame->bh->b_data)->entries;
848 if (dx_get_limit(entries) != dx_node_limit(dir)) {
849 ext4_warning_inode(dir,
850 "dx entry: limit %u != node limit %u",
851 dx_get_limit(entries), dx_node_limit(dir));
856 while (frame >= frame_in) {
861 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
862 ext4_warning_inode(dir,
863 "Corrupt directory, running e2fsck is recommended");
867 static void dx_release(struct dx_frame *frames)
869 struct dx_root_info *info;
872 if (frames[0].bh == NULL)
875 info = &((struct dx_root *)frames[0].bh->b_data)->info;
876 for (i = 0; i <= info->indirect_levels; i++) {
877 if (frames[i].bh == NULL)
879 brelse(frames[i].bh);
885 * This function increments the frame pointer to search the next leaf
886 * block, and reads in the necessary intervening nodes if the search
887 * should be necessary. Whether or not the search is necessary is
888 * controlled by the hash parameter. If the hash value is even, then
889 * the search is only continued if the next block starts with that
890 * hash value. This is used if we are searching for a specific file.
892 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
894 * This function returns 1 if the caller should continue to search,
895 * or 0 if it should not. If there is an error reading one of the
896 * index blocks, it will a negative error code.
898 * If start_hash is non-null, it will be filled in with the starting
899 * hash of the next page.
901 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
902 struct dx_frame *frame,
903 struct dx_frame *frames,
907 struct buffer_head *bh;
913 * Find the next leaf page by incrementing the frame pointer.
914 * If we run out of entries in the interior node, loop around and
915 * increment pointer in the parent node. When we break out of
916 * this loop, num_frames indicates the number of interior
917 * nodes need to be read.
920 if (++(p->at) < p->entries + dx_get_count(p->entries))
929 * If the hash is 1, then continue only if the next page has a
930 * continuation hash of any value. This is used for readdir
931 * handling. Otherwise, check to see if the hash matches the
932 * desired contiuation hash. If it doesn't, return since
933 * there's no point to read in the successive index pages.
935 bhash = dx_get_hash(p->at);
938 if ((hash & 1) == 0) {
939 if ((bhash & ~1) != hash)
943 * If the hash is HASH_NB_ALWAYS, we always go to the next
944 * block so no check is necessary
946 while (num_frames--) {
947 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
953 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
960 * This function fills a red-black tree with information from a
961 * directory block. It returns the number directory entries loaded
962 * into the tree. If there is an error it is returned in err.
964 static int htree_dirblock_to_tree(struct file *dir_file,
965 struct inode *dir, ext4_lblk_t block,
966 struct dx_hash_info *hinfo,
967 __u32 start_hash, __u32 start_minor_hash)
969 struct buffer_head *bh;
970 struct ext4_dir_entry_2 *de, *top;
971 int err = 0, count = 0;
972 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
974 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
975 (unsigned long)block));
976 bh = ext4_read_dirblock(dir, block, DIRENT);
980 de = (struct ext4_dir_entry_2 *) bh->b_data;
981 top = (struct ext4_dir_entry_2 *) ((char *) de +
982 dir->i_sb->s_blocksize -
983 EXT4_DIR_REC_LEN(0));
984 #ifdef CONFIG_EXT4_FS_ENCRYPTION
985 /* Check if the directory is encrypted */
986 if (ext4_encrypted_inode(dir)) {
987 err = fscrypt_get_encryption_info(dir);
992 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1000 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1001 if (ext4_check_dir_entry(dir, NULL, de, bh,
1002 bh->b_data, bh->b_size,
1003 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1004 + ((char *)de - bh->b_data))) {
1005 /* silently ignore the rest of the block */
1008 ext4fs_dirhash(de->name, de->name_len, hinfo);
1009 if ((hinfo->hash < start_hash) ||
1010 ((hinfo->hash == start_hash) &&
1011 (hinfo->minor_hash < start_minor_hash)))
1015 if (!ext4_encrypted_inode(dir)) {
1016 tmp_str.name = de->name;
1017 tmp_str.len = de->name_len;
1018 err = ext4_htree_store_dirent(dir_file,
1019 hinfo->hash, hinfo->minor_hash, de,
1022 int save_len = fname_crypto_str.len;
1023 struct fscrypt_str de_name = FSTR_INIT(de->name,
1026 /* Directory is encrypted */
1027 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1028 hinfo->minor_hash, &de_name,
1034 err = ext4_htree_store_dirent(dir_file,
1035 hinfo->hash, hinfo->minor_hash, de,
1037 fname_crypto_str.len = save_len;
1047 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1048 fscrypt_fname_free_buffer(&fname_crypto_str);
1055 * This function fills a red-black tree with information from a
1056 * directory. We start scanning the directory in hash order, starting
1057 * at start_hash and start_minor_hash.
1059 * This function returns the number of entries inserted into the tree,
1060 * or a negative error code.
1062 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1063 __u32 start_minor_hash, __u32 *next_hash)
1065 struct dx_hash_info hinfo;
1066 struct ext4_dir_entry_2 *de;
1067 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1073 struct fscrypt_str tmp_str;
1075 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1076 start_hash, start_minor_hash));
1077 dir = file_inode(dir_file);
1078 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1079 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1080 if (hinfo.hash_version <= DX_HASH_TEA)
1081 hinfo.hash_version +=
1082 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1083 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1084 if (ext4_has_inline_data(dir)) {
1085 int has_inline_data = 1;
1086 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1090 if (has_inline_data) {
1095 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1096 start_hash, start_minor_hash);
1100 hinfo.hash = start_hash;
1101 hinfo.minor_hash = 0;
1102 frame = dx_probe(NULL, dir, &hinfo, frames);
1104 return PTR_ERR(frame);
1106 /* Add '.' and '..' from the htree header */
1107 if (!start_hash && !start_minor_hash) {
1108 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1109 tmp_str.name = de->name;
1110 tmp_str.len = de->name_len;
1111 err = ext4_htree_store_dirent(dir_file, 0, 0,
1117 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1118 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1119 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1120 tmp_str.name = de->name;
1121 tmp_str.len = de->name_len;
1122 err = ext4_htree_store_dirent(dir_file, 2, 0,
1130 if (fatal_signal_pending(current)) {
1135 block = dx_get_block(frame->at);
1136 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1137 start_hash, start_minor_hash);
1144 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1145 frame, frames, &hashval);
1146 *next_hash = hashval;
1152 * Stop if: (a) there are no more entries, or
1153 * (b) we have inserted at least one entry and the
1154 * next hash value is not a continuation
1157 (count && ((hashval & 1) == 0)))
1161 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1162 "next hash: %x\n", count, *next_hash));
1169 static inline int search_dirblock(struct buffer_head *bh,
1171 struct ext4_filename *fname,
1172 unsigned int offset,
1173 struct ext4_dir_entry_2 **res_dir)
1175 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1176 fname, offset, res_dir);
1180 * Directory block splitting, compacting
1184 * Create map of hash values, offsets, and sizes, stored at end of block.
1185 * Returns number of entries mapped.
1187 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1188 unsigned blocksize, struct dx_hash_info *hinfo,
1189 struct dx_map_entry *map_tail)
1192 char *base = (char *) de;
1193 struct dx_hash_info h = *hinfo;
1195 while ((char *) de < base + blocksize) {
1196 if (de->name_len && de->inode) {
1197 ext4fs_dirhash(de->name, de->name_len, &h);
1199 map_tail->hash = h.hash;
1200 map_tail->offs = ((char *) de - base)>>2;
1201 map_tail->size = le16_to_cpu(de->rec_len);
1205 /* XXX: do we need to check rec_len == 0 case? -Chris */
1206 de = ext4_next_entry(de, blocksize);
1211 /* Sort map by hash value */
1212 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1214 struct dx_map_entry *p, *q, *top = map + count - 1;
1216 /* Combsort until bubble sort doesn't suck */
1218 count = count*10/13;
1219 if (count - 9 < 2) /* 9, 10 -> 11 */
1221 for (p = top, q = p - count; q >= map; p--, q--)
1222 if (p->hash < q->hash)
1225 /* Garden variety bubble sort */
1230 if (q[1].hash >= q[0].hash)
1238 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1240 struct dx_entry *entries = frame->entries;
1241 struct dx_entry *old = frame->at, *new = old + 1;
1242 int count = dx_get_count(entries);
1244 assert(count < dx_get_limit(entries));
1245 assert(old < entries + count);
1246 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1247 dx_set_hash(new, hash);
1248 dx_set_block(new, block);
1249 dx_set_count(entries, count + 1);
1253 * Test whether a directory entry matches the filename being searched for.
1255 * Return: %true if the directory entry matches, otherwise %false.
1257 static inline bool ext4_match(const struct ext4_filename *fname,
1258 const struct ext4_dir_entry_2 *de)
1260 struct fscrypt_name f;
1265 f.usr_fname = fname->usr_fname;
1266 f.disk_name = fname->disk_name;
1267 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1268 f.crypto_buf = fname->crypto_buf;
1270 return fscrypt_match_name(&f, de->name, de->name_len);
1274 * Returns 0 if not found, -1 on failure, and 1 on success
1276 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1277 struct inode *dir, struct ext4_filename *fname,
1278 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1280 struct ext4_dir_entry_2 * de;
1284 de = (struct ext4_dir_entry_2 *)search_buf;
1285 dlimit = search_buf + buf_size;
1286 while ((char *) de < dlimit) {
1287 /* this code is executed quadratically often */
1288 /* do minimal checking `by hand' */
1289 if ((char *) de + de->name_len <= dlimit &&
1290 ext4_match(fname, de)) {
1291 /* found a match - just to be sure, do
1293 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1294 bh->b_size, offset))
1299 /* prevent looping on a bad block */
1300 de_len = ext4_rec_len_from_disk(de->rec_len,
1301 dir->i_sb->s_blocksize);
1305 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1310 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1311 struct ext4_dir_entry *de)
1313 struct super_block *sb = dir->i_sb;
1319 if (de->inode == 0 &&
1320 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1329 * finds an entry in the specified directory with the wanted name. It
1330 * returns the cache buffer in which the entry was found, and the entry
1331 * itself (as a parameter - res_dir). It does NOT read the inode of the
1332 * entry - you'll have to do that yourself if you want to.
1334 * The returned buffer_head has ->b_count elevated. The caller is expected
1335 * to brelse() it when appropriate.
1337 static struct buffer_head * ext4_find_entry (struct inode *dir,
1338 const struct qstr *d_name,
1339 struct ext4_dir_entry_2 **res_dir,
1342 struct super_block *sb;
1343 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1344 struct buffer_head *bh, *ret = NULL;
1345 ext4_lblk_t start, block, b;
1346 const u8 *name = d_name->name;
1347 int ra_max = 0; /* Number of bh's in the readahead
1349 int ra_ptr = 0; /* Current index into readahead
1352 ext4_lblk_t nblocks;
1353 int i, namelen, retval;
1354 struct ext4_filename fname;
1358 namelen = d_name->len;
1359 if (namelen > EXT4_NAME_LEN)
1362 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1363 if (retval == -ENOENT)
1366 return ERR_PTR(retval);
1368 if (ext4_has_inline_data(dir)) {
1369 int has_inline_data = 1;
1370 ret = ext4_find_inline_entry(dir, &fname, res_dir,
1372 if (has_inline_data) {
1375 goto cleanup_and_exit;
1379 if ((namelen <= 2) && (name[0] == '.') &&
1380 (name[1] == '.' || name[1] == '\0')) {
1382 * "." or ".." will only be in the first block
1383 * NFS may look up ".."; "." should be handled by the VFS
1390 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1392 * On success, or if the error was file not found,
1393 * return. Otherwise, fall back to doing a search the
1394 * old fashioned way.
1396 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1397 goto cleanup_and_exit;
1398 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1401 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1402 start = EXT4_I(dir)->i_dir_start_lookup;
1403 if (start >= nblocks)
1409 * We deal with the read-ahead logic here.
1411 if (ra_ptr >= ra_max) {
1412 /* Refill the readahead buffer */
1415 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1417 * Terminate if we reach the end of the
1418 * directory and must wrap, or if our
1419 * search has finished at this block.
1421 if (b >= nblocks || (num && block == start)) {
1422 bh_use[ra_max] = NULL;
1426 bh = ext4_getblk(NULL, dir, b++, 0);
1430 goto cleanup_and_exit;
1434 bh_use[ra_max] = bh;
1436 ll_rw_block(REQ_OP_READ,
1437 REQ_META | REQ_PRIO,
1441 if ((bh = bh_use[ra_ptr++]) == NULL)
1444 if (!buffer_uptodate(bh)) {
1445 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1446 (unsigned long) block);
1448 ret = ERR_PTR(-EIO);
1449 goto cleanup_and_exit;
1451 if (!buffer_verified(bh) &&
1452 !is_dx_internal_node(dir, block,
1453 (struct ext4_dir_entry *)bh->b_data) &&
1454 !ext4_dirent_csum_verify(dir,
1455 (struct ext4_dir_entry *)bh->b_data)) {
1456 EXT4_ERROR_INODE(dir, "checksumming directory "
1457 "block %lu", (unsigned long)block);
1459 ret = ERR_PTR(-EFSBADCRC);
1460 goto cleanup_and_exit;
1462 set_buffer_verified(bh);
1463 i = search_dirblock(bh, dir, &fname,
1464 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1466 EXT4_I(dir)->i_dir_start_lookup = block;
1468 goto cleanup_and_exit;
1472 goto cleanup_and_exit;
1475 if (++block >= nblocks)
1477 } while (block != start);
1480 * If the directory has grown while we were searching, then
1481 * search the last part of the directory before giving up.
1484 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1485 if (block < nblocks) {
1491 /* Clean up the read-ahead blocks */
1492 for (; ra_ptr < ra_max; ra_ptr++)
1493 brelse(bh_use[ra_ptr]);
1494 ext4_fname_free_filename(&fname);
1498 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1499 struct ext4_filename *fname,
1500 struct ext4_dir_entry_2 **res_dir)
1502 struct super_block * sb = dir->i_sb;
1503 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1504 struct buffer_head *bh;
1508 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1511 frame = dx_probe(fname, dir, NULL, frames);
1513 return (struct buffer_head *) frame;
1515 block = dx_get_block(frame->at);
1516 bh = ext4_read_dirblock(dir, block, DIRENT);
1520 retval = search_dirblock(bh, dir, fname,
1521 block << EXT4_BLOCK_SIZE_BITS(sb),
1527 bh = ERR_PTR(ERR_BAD_DX_DIR);
1531 /* Check to see if we should continue to search */
1532 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1535 ext4_warning_inode(dir,
1536 "error %d reading directory index block",
1538 bh = ERR_PTR(retval);
1541 } while (retval == 1);
1545 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1551 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1553 struct inode *inode;
1554 struct ext4_dir_entry_2 *de;
1555 struct buffer_head *bh;
1557 if (ext4_encrypted_inode(dir)) {
1558 int res = fscrypt_get_encryption_info(dir);
1561 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1562 * created while the directory was encrypted and we
1563 * have access to the key.
1565 if (fscrypt_has_encryption_key(dir))
1566 fscrypt_set_encrypted_dentry(dentry);
1567 fscrypt_set_d_op(dentry);
1568 if (res && res != -ENOKEY)
1569 return ERR_PTR(res);
1572 if (dentry->d_name.len > EXT4_NAME_LEN)
1573 return ERR_PTR(-ENAMETOOLONG);
1575 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1577 return (struct dentry *) bh;
1580 __u32 ino = le32_to_cpu(de->inode);
1582 if (!ext4_valid_inum(dir->i_sb, ino)) {
1583 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1584 return ERR_PTR(-EFSCORRUPTED);
1586 if (unlikely(ino == dir->i_ino)) {
1587 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1589 return ERR_PTR(-EFSCORRUPTED);
1591 inode = ext4_iget_normal(dir->i_sb, ino);
1592 if (inode == ERR_PTR(-ESTALE)) {
1593 EXT4_ERROR_INODE(dir,
1594 "deleted inode referenced: %u",
1596 return ERR_PTR(-EFSCORRUPTED);
1598 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1599 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1600 !fscrypt_has_permitted_context(dir, inode)) {
1601 ext4_warning(inode->i_sb,
1602 "Inconsistent encryption contexts: %lu/%lu",
1603 dir->i_ino, inode->i_ino);
1605 return ERR_PTR(-EPERM);
1608 return d_splice_alias(inode, dentry);
1612 struct dentry *ext4_get_parent(struct dentry *child)
1615 static const struct qstr dotdot = QSTR_INIT("..", 2);
1616 struct ext4_dir_entry_2 * de;
1617 struct buffer_head *bh;
1619 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1621 return (struct dentry *) bh;
1623 return ERR_PTR(-ENOENT);
1624 ino = le32_to_cpu(de->inode);
1627 if (!ext4_valid_inum(child->d_sb, ino)) {
1628 EXT4_ERROR_INODE(d_inode(child),
1629 "bad parent inode number: %u", ino);
1630 return ERR_PTR(-EFSCORRUPTED);
1633 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1637 * Move count entries from end of map between two memory locations.
1638 * Returns pointer to last entry moved.
1640 static struct ext4_dir_entry_2 *
1641 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1644 unsigned rec_len = 0;
1647 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1648 (from + (map->offs<<2));
1649 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1650 memcpy (to, de, rec_len);
1651 ((struct ext4_dir_entry_2 *) to)->rec_len =
1652 ext4_rec_len_to_disk(rec_len, blocksize);
1657 return (struct ext4_dir_entry_2 *) (to - rec_len);
1661 * Compact each dir entry in the range to the minimal rec_len.
1662 * Returns pointer to last entry in range.
1664 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1666 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1667 unsigned rec_len = 0;
1670 while ((char*)de < base + blocksize) {
1671 next = ext4_next_entry(de, blocksize);
1672 if (de->inode && de->name_len) {
1673 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1675 memmove(to, de, rec_len);
1676 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1678 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1686 * Split a full leaf block to make room for a new dir entry.
1687 * Allocate a new block, and move entries so that they are approx. equally full.
1688 * Returns pointer to de in block into which the new entry will be inserted.
1690 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1691 struct buffer_head **bh,struct dx_frame *frame,
1692 struct dx_hash_info *hinfo)
1694 unsigned blocksize = dir->i_sb->s_blocksize;
1695 unsigned count, continued;
1696 struct buffer_head *bh2;
1697 ext4_lblk_t newblock;
1699 struct dx_map_entry *map;
1700 char *data1 = (*bh)->b_data, *data2;
1701 unsigned split, move, size;
1702 struct ext4_dir_entry_2 *de = NULL, *de2;
1703 struct ext4_dir_entry_tail *t;
1707 if (ext4_has_metadata_csum(dir->i_sb))
1708 csum_size = sizeof(struct ext4_dir_entry_tail);
1710 bh2 = ext4_append(handle, dir, &newblock);
1714 return (struct ext4_dir_entry_2 *) bh2;
1717 BUFFER_TRACE(*bh, "get_write_access");
1718 err = ext4_journal_get_write_access(handle, *bh);
1722 BUFFER_TRACE(frame->bh, "get_write_access");
1723 err = ext4_journal_get_write_access(handle, frame->bh);
1727 data2 = bh2->b_data;
1729 /* create map in the end of data2 block */
1730 map = (struct dx_map_entry *) (data2 + blocksize);
1731 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1732 blocksize, hinfo, map);
1734 dx_sort_map(map, count);
1735 /* Split the existing block in the middle, size-wise */
1738 for (i = count-1; i >= 0; i--) {
1739 /* is more than half of this entry in 2nd half of the block? */
1740 if (size + map[i].size/2 > blocksize/2)
1742 size += map[i].size;
1745 /* map index at which we will split */
1746 split = count - move;
1747 hash2 = map[split].hash;
1748 continued = hash2 == map[split - 1].hash;
1749 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1750 (unsigned long)dx_get_block(frame->at),
1751 hash2, split, count-split));
1753 /* Fancy dance to stay within two buffers */
1754 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1756 de = dx_pack_dirents(data1, blocksize);
1757 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1760 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1764 t = EXT4_DIRENT_TAIL(data2, blocksize);
1765 initialize_dirent_tail(t, blocksize);
1767 t = EXT4_DIRENT_TAIL(data1, blocksize);
1768 initialize_dirent_tail(t, blocksize);
1771 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1773 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1776 /* Which block gets the new entry? */
1777 if (hinfo->hash >= hash2) {
1781 dx_insert_block(frame, hash2 + continued, newblock);
1782 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1785 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1789 dxtrace(dx_show_index("frame", frame->entries));
1796 ext4_std_error(dir->i_sb, err);
1797 return ERR_PTR(err);
1800 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1801 struct buffer_head *bh,
1802 void *buf, int buf_size,
1803 struct ext4_filename *fname,
1804 struct ext4_dir_entry_2 **dest_de)
1806 struct ext4_dir_entry_2 *de;
1807 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1809 unsigned int offset = 0;
1812 de = (struct ext4_dir_entry_2 *)buf;
1813 top = buf + buf_size - reclen;
1814 while ((char *) de <= top) {
1815 if (ext4_check_dir_entry(dir, NULL, de, bh,
1816 buf, buf_size, offset))
1817 return -EFSCORRUPTED;
1818 if (ext4_match(fname, de))
1820 nlen = EXT4_DIR_REC_LEN(de->name_len);
1821 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1822 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1824 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1827 if ((char *) de > top)
1834 void ext4_insert_dentry(struct inode *inode,
1835 struct ext4_dir_entry_2 *de,
1837 struct ext4_filename *fname)
1842 nlen = EXT4_DIR_REC_LEN(de->name_len);
1843 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1845 struct ext4_dir_entry_2 *de1 =
1846 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1847 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1848 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1851 de->file_type = EXT4_FT_UNKNOWN;
1852 de->inode = cpu_to_le32(inode->i_ino);
1853 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1854 de->name_len = fname_len(fname);
1855 memcpy(de->name, fname_name(fname), fname_len(fname));
1859 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1860 * it points to a directory entry which is guaranteed to be large
1861 * enough for new directory entry. If de is NULL, then
1862 * add_dirent_to_buf will attempt search the directory block for
1863 * space. It will return -ENOSPC if no space is available, and -EIO
1864 * and -EEXIST if directory entry already exists.
1866 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1868 struct inode *inode, struct ext4_dir_entry_2 *de,
1869 struct buffer_head *bh)
1871 unsigned int blocksize = dir->i_sb->s_blocksize;
1875 if (ext4_has_metadata_csum(inode->i_sb))
1876 csum_size = sizeof(struct ext4_dir_entry_tail);
1879 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1880 blocksize - csum_size, fname, &de);
1884 BUFFER_TRACE(bh, "get_write_access");
1885 err = ext4_journal_get_write_access(handle, bh);
1887 ext4_std_error(dir->i_sb, err);
1891 /* By now the buffer is marked for journaling */
1892 ext4_insert_dentry(inode, de, blocksize, fname);
1895 * XXX shouldn't update any times until successful
1896 * completion of syscall, but too many callers depend
1899 * XXX similarly, too many callers depend on
1900 * ext4_new_inode() setting the times, but error
1901 * recovery deletes the inode, so the worst that can
1902 * happen is that the times are slightly out of date
1903 * and/or different from the directory change time.
1905 dir->i_mtime = dir->i_ctime = current_time(dir);
1906 ext4_update_dx_flag(dir);
1907 inode_inc_iversion(dir);
1908 ext4_mark_inode_dirty(handle, dir);
1909 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1910 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1912 ext4_std_error(dir->i_sb, err);
1917 * This converts a one block unindexed directory to a 3 block indexed
1918 * directory, and adds the dentry to the indexed directory.
1920 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1922 struct inode *inode, struct buffer_head *bh)
1924 struct buffer_head *bh2;
1925 struct dx_root *root;
1926 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1927 struct dx_entry *entries;
1928 struct ext4_dir_entry_2 *de, *de2;
1929 struct ext4_dir_entry_tail *t;
1935 struct fake_dirent *fde;
1938 if (ext4_has_metadata_csum(inode->i_sb))
1939 csum_size = sizeof(struct ext4_dir_entry_tail);
1941 blocksize = dir->i_sb->s_blocksize;
1942 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1943 BUFFER_TRACE(bh, "get_write_access");
1944 retval = ext4_journal_get_write_access(handle, bh);
1946 ext4_std_error(dir->i_sb, retval);
1950 root = (struct dx_root *) bh->b_data;
1952 /* The 0th block becomes the root, move the dirents out */
1953 fde = &root->dotdot;
1954 de = (struct ext4_dir_entry_2 *)((char *)fde +
1955 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1956 if ((char *) de >= (((char *) root) + blocksize)) {
1957 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1959 return -EFSCORRUPTED;
1961 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1963 /* Allocate new block for the 0th block's dirents */
1964 bh2 = ext4_append(handle, dir, &block);
1967 return PTR_ERR(bh2);
1969 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1970 data1 = bh2->b_data;
1972 memcpy (data1, de, len);
1973 de = (struct ext4_dir_entry_2 *) data1;
1975 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1977 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1982 t = EXT4_DIRENT_TAIL(data1, blocksize);
1983 initialize_dirent_tail(t, blocksize);
1986 /* Initialize the root; the dot dirents already exist */
1987 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1988 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1990 memset (&root->info, 0, sizeof(root->info));
1991 root->info.info_length = sizeof(root->info);
1992 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1993 entries = root->entries;
1994 dx_set_block(entries, 1);
1995 dx_set_count(entries, 1);
1996 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1998 /* Initialize as for dx_probe */
1999 fname->hinfo.hash_version = root->info.hash_version;
2000 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2001 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2002 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2003 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2005 memset(frames, 0, sizeof(frames));
2007 frame->entries = entries;
2008 frame->at = entries;
2011 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2014 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2018 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2020 retval = PTR_ERR(de);
2024 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2027 * Even if the block split failed, we have to properly write
2028 * out all the changes we did so far. Otherwise we can end up
2029 * with corrupted filesystem.
2032 ext4_mark_inode_dirty(handle, dir);
2041 * adds a file entry to the specified directory, using the same
2042 * semantics as ext4_find_entry(). It returns NULL if it failed.
2044 * NOTE!! The inode part of 'de' is left at 0 - which means you
2045 * may not sleep between calling this and putting something into
2046 * the entry, as someone else might have used it while you slept.
2048 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2049 struct inode *inode)
2051 struct inode *dir = d_inode(dentry->d_parent);
2052 struct buffer_head *bh = NULL;
2053 struct ext4_dir_entry_2 *de;
2054 struct ext4_dir_entry_tail *t;
2055 struct super_block *sb;
2056 struct ext4_filename fname;
2060 ext4_lblk_t block, blocks;
2063 if (ext4_has_metadata_csum(inode->i_sb))
2064 csum_size = sizeof(struct ext4_dir_entry_tail);
2067 blocksize = sb->s_blocksize;
2068 if (!dentry->d_name.len)
2071 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2075 if (ext4_has_inline_data(dir)) {
2076 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2086 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2087 if (!retval || (retval != ERR_BAD_DX_DIR))
2089 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2091 ext4_mark_inode_dirty(handle, dir);
2093 blocks = dir->i_size >> sb->s_blocksize_bits;
2094 for (block = 0; block < blocks; block++) {
2095 bh = ext4_read_dirblock(dir, block, DIRENT);
2097 retval = PTR_ERR(bh);
2101 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2103 if (retval != -ENOSPC)
2106 if (blocks == 1 && !dx_fallback &&
2107 ext4_has_feature_dir_index(sb)) {
2108 retval = make_indexed_dir(handle, &fname, dir,
2110 bh = NULL; /* make_indexed_dir releases bh */
2115 bh = ext4_append(handle, dir, &block);
2117 retval = PTR_ERR(bh);
2121 de = (struct ext4_dir_entry_2 *) bh->b_data;
2123 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2126 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2127 initialize_dirent_tail(t, blocksize);
2130 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2132 ext4_fname_free_filename(&fname);
2135 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2140 * Returns 0 for success, or a negative error value
2142 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2143 struct inode *dir, struct inode *inode)
2145 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2146 struct dx_entry *entries, *at;
2147 struct buffer_head *bh;
2148 struct super_block *sb = dir->i_sb;
2149 struct ext4_dir_entry_2 *de;
2155 frame = dx_probe(fname, dir, NULL, frames);
2157 return PTR_ERR(frame);
2158 entries = frame->entries;
2160 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2167 BUFFER_TRACE(bh, "get_write_access");
2168 err = ext4_journal_get_write_access(handle, bh);
2172 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2177 /* Block full, should compress but for now just split */
2178 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2179 dx_get_count(entries), dx_get_limit(entries)));
2180 /* Need to split index? */
2181 if (dx_get_count(entries) == dx_get_limit(entries)) {
2182 ext4_lblk_t newblock;
2183 int levels = frame - frames + 1;
2184 unsigned int icount;
2186 struct dx_entry *entries2;
2187 struct dx_node *node2;
2188 struct buffer_head *bh2;
2190 while (frame > frames) {
2191 if (dx_get_count((frame - 1)->entries) <
2192 dx_get_limit((frame - 1)->entries)) {
2196 frame--; /* split higher index block */
2198 entries = frame->entries;
2201 if (add_level && levels == ext4_dir_htree_level(sb)) {
2202 ext4_warning(sb, "Directory (ino: %lu) index full, "
2203 "reach max htree level :%d",
2204 dir->i_ino, levels);
2205 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2206 ext4_warning(sb, "Large directory feature is "
2207 "not enabled on this "
2213 icount = dx_get_count(entries);
2214 bh2 = ext4_append(handle, dir, &newblock);
2219 node2 = (struct dx_node *)(bh2->b_data);
2220 entries2 = node2->entries;
2221 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2222 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2224 BUFFER_TRACE(frame->bh, "get_write_access");
2225 err = ext4_journal_get_write_access(handle, frame->bh);
2229 unsigned icount1 = icount/2, icount2 = icount - icount1;
2230 unsigned hash2 = dx_get_hash(entries + icount1);
2231 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2234 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2235 err = ext4_journal_get_write_access(handle,
2240 memcpy((char *) entries2, (char *) (entries + icount1),
2241 icount2 * sizeof(struct dx_entry));
2242 dx_set_count(entries, icount1);
2243 dx_set_count(entries2, icount2);
2244 dx_set_limit(entries2, dx_node_limit(dir));
2246 /* Which index block gets the new entry? */
2247 if (at - entries >= icount1) {
2248 frame->at = at = at - entries - icount1 + entries2;
2249 frame->entries = entries = entries2;
2250 swap(frame->bh, bh2);
2252 dx_insert_block((frame - 1), hash2, newblock);
2253 dxtrace(dx_show_index("node", frame->entries));
2254 dxtrace(dx_show_index("node",
2255 ((struct dx_node *) bh2->b_data)->entries));
2256 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2260 err = ext4_handle_dirty_dx_node(handle, dir,
2265 err = ext4_handle_dirty_dx_node(handle, dir,
2270 struct dx_root *dxroot;
2271 memcpy((char *) entries2, (char *) entries,
2272 icount * sizeof(struct dx_entry));
2273 dx_set_limit(entries2, dx_node_limit(dir));
2276 dx_set_count(entries, 1);
2277 dx_set_block(entries + 0, newblock);
2278 dxroot = (struct dx_root *)frames[0].bh->b_data;
2279 dxroot->info.indirect_levels += 1;
2280 dxtrace(printk(KERN_DEBUG
2281 "Creating %d level index...\n",
2282 info->indirect_levels));
2283 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2286 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2292 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2297 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2301 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2305 /* @restart is true means htree-path has been changed, we need to
2306 * repeat dx_probe() to find out valid htree-path
2308 if (restart && err == 0)
2314 * ext4_generic_delete_entry deletes a directory entry by merging it
2315 * with the previous entry
2317 int ext4_generic_delete_entry(handle_t *handle,
2319 struct ext4_dir_entry_2 *de_del,
2320 struct buffer_head *bh,
2325 struct ext4_dir_entry_2 *de, *pde;
2326 unsigned int blocksize = dir->i_sb->s_blocksize;
2331 de = (struct ext4_dir_entry_2 *)entry_buf;
2332 while (i < buf_size - csum_size) {
2333 if (ext4_check_dir_entry(dir, NULL, de, bh,
2334 bh->b_data, bh->b_size, i))
2335 return -EFSCORRUPTED;
2338 pde->rec_len = ext4_rec_len_to_disk(
2339 ext4_rec_len_from_disk(pde->rec_len,
2341 ext4_rec_len_from_disk(de->rec_len,
2346 inode_inc_iversion(dir);
2349 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2351 de = ext4_next_entry(de, blocksize);
2356 static int ext4_delete_entry(handle_t *handle,
2358 struct ext4_dir_entry_2 *de_del,
2359 struct buffer_head *bh)
2361 int err, csum_size = 0;
2363 if (ext4_has_inline_data(dir)) {
2364 int has_inline_data = 1;
2365 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2367 if (has_inline_data)
2371 if (ext4_has_metadata_csum(dir->i_sb))
2372 csum_size = sizeof(struct ext4_dir_entry_tail);
2374 BUFFER_TRACE(bh, "get_write_access");
2375 err = ext4_journal_get_write_access(handle, bh);
2379 err = ext4_generic_delete_entry(handle, dir, de_del,
2381 dir->i_sb->s_blocksize, csum_size);
2385 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2386 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2393 ext4_std_error(dir->i_sb, err);
2398 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2399 * since this indicates that nlinks count was previously 1.
2401 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2404 if (is_dx(inode) && inode->i_nlink > 1) {
2405 /* limit is 16-bit i_links_count */
2406 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2407 set_nlink(inode, 1);
2408 ext4_set_feature_dir_nlink(inode->i_sb);
2414 * If a directory had nlink == 1, then we should let it be 1. This indicates
2415 * directory has >EXT4_LINK_MAX subdirs.
2417 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2419 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2424 static int ext4_add_nondir(handle_t *handle,
2425 struct dentry *dentry, struct inode *inode)
2427 int err = ext4_add_entry(handle, dentry, inode);
2429 ext4_mark_inode_dirty(handle, inode);
2430 unlock_new_inode(inode);
2431 d_instantiate(dentry, inode);
2435 unlock_new_inode(inode);
2441 * By the time this is called, we already have created
2442 * the directory cache entry for the new file, but it
2443 * is so far negative - it has no inode.
2445 * If the create succeeds, we fill in the inode information
2446 * with d_instantiate().
2448 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2452 struct inode *inode;
2453 int err, credits, retries = 0;
2455 err = dquot_initialize(dir);
2459 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2460 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2462 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2463 NULL, EXT4_HT_DIR, credits);
2464 handle = ext4_journal_current_handle();
2465 err = PTR_ERR(inode);
2466 if (!IS_ERR(inode)) {
2467 inode->i_op = &ext4_file_inode_operations;
2468 inode->i_fop = &ext4_file_operations;
2469 ext4_set_aops(inode);
2470 err = ext4_add_nondir(handle, dentry, inode);
2471 if (!err && IS_DIRSYNC(dir))
2472 ext4_handle_sync(handle);
2475 ext4_journal_stop(handle);
2476 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2481 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2482 umode_t mode, dev_t rdev)
2485 struct inode *inode;
2486 int err, credits, retries = 0;
2488 err = dquot_initialize(dir);
2492 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2493 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2495 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2496 NULL, EXT4_HT_DIR, credits);
2497 handle = ext4_journal_current_handle();
2498 err = PTR_ERR(inode);
2499 if (!IS_ERR(inode)) {
2500 init_special_inode(inode, inode->i_mode, rdev);
2501 inode->i_op = &ext4_special_inode_operations;
2502 err = ext4_add_nondir(handle, dentry, inode);
2503 if (!err && IS_DIRSYNC(dir))
2504 ext4_handle_sync(handle);
2507 ext4_journal_stop(handle);
2508 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2513 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2516 struct inode *inode;
2517 int err, retries = 0;
2519 err = dquot_initialize(dir);
2524 inode = ext4_new_inode_start_handle(dir, mode,
2527 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2528 4 + EXT4_XATTR_TRANS_BLOCKS);
2529 handle = ext4_journal_current_handle();
2530 err = PTR_ERR(inode);
2531 if (!IS_ERR(inode)) {
2532 inode->i_op = &ext4_file_inode_operations;
2533 inode->i_fop = &ext4_file_operations;
2534 ext4_set_aops(inode);
2535 d_tmpfile(dentry, inode);
2536 err = ext4_orphan_add(handle, inode);
2538 goto err_unlock_inode;
2539 mark_inode_dirty(inode);
2540 unlock_new_inode(inode);
2543 ext4_journal_stop(handle);
2544 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2548 ext4_journal_stop(handle);
2549 unlock_new_inode(inode);
2553 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2554 struct ext4_dir_entry_2 *de,
2555 int blocksize, int csum_size,
2556 unsigned int parent_ino, int dotdot_real_len)
2558 de->inode = cpu_to_le32(inode->i_ino);
2560 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2562 strcpy(de->name, ".");
2563 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2565 de = ext4_next_entry(de, blocksize);
2566 de->inode = cpu_to_le32(parent_ino);
2568 if (!dotdot_real_len)
2569 de->rec_len = ext4_rec_len_to_disk(blocksize -
2570 (csum_size + EXT4_DIR_REC_LEN(1)),
2573 de->rec_len = ext4_rec_len_to_disk(
2574 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2575 strcpy(de->name, "..");
2576 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2578 return ext4_next_entry(de, blocksize);
2581 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2582 struct inode *inode)
2584 struct buffer_head *dir_block = NULL;
2585 struct ext4_dir_entry_2 *de;
2586 struct ext4_dir_entry_tail *t;
2587 ext4_lblk_t block = 0;
2588 unsigned int blocksize = dir->i_sb->s_blocksize;
2592 if (ext4_has_metadata_csum(dir->i_sb))
2593 csum_size = sizeof(struct ext4_dir_entry_tail);
2595 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2596 err = ext4_try_create_inline_dir(handle, dir, inode);
2597 if (err < 0 && err != -ENOSPC)
2604 dir_block = ext4_append(handle, inode, &block);
2605 if (IS_ERR(dir_block))
2606 return PTR_ERR(dir_block);
2607 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2608 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2609 set_nlink(inode, 2);
2611 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2612 initialize_dirent_tail(t, blocksize);
2615 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2616 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2619 set_buffer_verified(dir_block);
2625 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2628 struct inode *inode;
2629 int err, credits, retries = 0;
2631 if (EXT4_DIR_LINK_MAX(dir))
2634 err = dquot_initialize(dir);
2638 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2639 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2641 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2643 0, NULL, EXT4_HT_DIR, credits);
2644 handle = ext4_journal_current_handle();
2645 err = PTR_ERR(inode);
2649 inode->i_op = &ext4_dir_inode_operations;
2650 inode->i_fop = &ext4_dir_operations;
2651 err = ext4_init_new_dir(handle, dir, inode);
2653 goto out_clear_inode;
2654 err = ext4_mark_inode_dirty(handle, inode);
2656 err = ext4_add_entry(handle, dentry, inode);
2660 unlock_new_inode(inode);
2661 ext4_mark_inode_dirty(handle, inode);
2665 ext4_inc_count(handle, dir);
2666 ext4_update_dx_flag(dir);
2667 err = ext4_mark_inode_dirty(handle, dir);
2669 goto out_clear_inode;
2670 unlock_new_inode(inode);
2671 d_instantiate(dentry, inode);
2672 if (IS_DIRSYNC(dir))
2673 ext4_handle_sync(handle);
2677 ext4_journal_stop(handle);
2678 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2684 * routine to check that the specified directory is empty (for rmdir)
2686 bool ext4_empty_dir(struct inode *inode)
2688 unsigned int offset;
2689 struct buffer_head *bh;
2690 struct ext4_dir_entry_2 *de, *de1;
2691 struct super_block *sb;
2693 if (ext4_has_inline_data(inode)) {
2694 int has_inline_data = 1;
2697 ret = empty_inline_dir(inode, &has_inline_data);
2698 if (has_inline_data)
2703 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2704 EXT4_ERROR_INODE(inode, "invalid size");
2707 bh = ext4_read_dirblock(inode, 0, EITHER);
2711 de = (struct ext4_dir_entry_2 *) bh->b_data;
2712 de1 = ext4_next_entry(de, sb->s_blocksize);
2713 if (le32_to_cpu(de->inode) != inode->i_ino ||
2714 le32_to_cpu(de1->inode) == 0 ||
2715 strcmp(".", de->name) || strcmp("..", de1->name)) {
2716 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2720 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2721 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2722 de = ext4_next_entry(de1, sb->s_blocksize);
2723 while (offset < inode->i_size) {
2724 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2725 unsigned int lblock;
2727 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2728 bh = ext4_read_dirblock(inode, lblock, EITHER);
2731 de = (struct ext4_dir_entry_2 *) bh->b_data;
2733 if (ext4_check_dir_entry(inode, NULL, de, bh,
2734 bh->b_data, bh->b_size, offset)) {
2735 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2737 offset = (offset | (sb->s_blocksize - 1)) + 1;
2740 if (le32_to_cpu(de->inode)) {
2744 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2745 de = ext4_next_entry(de, sb->s_blocksize);
2752 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2753 * such inodes, starting at the superblock, in case we crash before the
2754 * file is closed/deleted, or in case the inode truncate spans multiple
2755 * transactions and the last transaction is not recovered after a crash.
2757 * At filesystem recovery time, we walk this list deleting unlinked
2758 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2760 * Orphan list manipulation functions must be called under i_mutex unless
2761 * we are just creating the inode or deleting it.
2763 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2765 struct super_block *sb = inode->i_sb;
2766 struct ext4_sb_info *sbi = EXT4_SB(sb);
2767 struct ext4_iloc iloc;
2771 if (!sbi->s_journal || is_bad_inode(inode))
2774 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2775 !inode_is_locked(inode));
2777 * Exit early if inode already is on orphan list. This is a big speedup
2778 * since we don't have to contend on the global s_orphan_lock.
2780 if (!list_empty(&EXT4_I(inode)->i_orphan))
2784 * Orphan handling is only valid for files with data blocks
2785 * being truncated, or files being unlinked. Note that we either
2786 * hold i_mutex, or the inode can not be referenced from outside,
2787 * so i_nlink should not be bumped due to race
2789 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2790 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2792 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2793 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2797 err = ext4_reserve_inode_write(handle, inode, &iloc);
2801 mutex_lock(&sbi->s_orphan_lock);
2803 * Due to previous errors inode may be already a part of on-disk
2804 * orphan list. If so skip on-disk list modification.
2806 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2807 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2808 /* Insert this inode at the head of the on-disk orphan list */
2809 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2810 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2813 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2814 mutex_unlock(&sbi->s_orphan_lock);
2817 err = ext4_handle_dirty_super(handle, sb);
2818 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2823 * We have to remove inode from in-memory list if
2824 * addition to on disk orphan list failed. Stray orphan
2825 * list entries can cause panics at unmount time.
2827 mutex_lock(&sbi->s_orphan_lock);
2828 list_del_init(&EXT4_I(inode)->i_orphan);
2829 mutex_unlock(&sbi->s_orphan_lock);
2832 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2833 jbd_debug(4, "orphan inode %lu will point to %d\n",
2834 inode->i_ino, NEXT_ORPHAN(inode));
2836 ext4_std_error(sb, err);
2841 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2842 * of such inodes stored on disk, because it is finally being cleaned up.
2844 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2846 struct list_head *prev;
2847 struct ext4_inode_info *ei = EXT4_I(inode);
2848 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2850 struct ext4_iloc iloc;
2853 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2856 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2857 !inode_is_locked(inode));
2858 /* Do this quick check before taking global s_orphan_lock. */
2859 if (list_empty(&ei->i_orphan))
2863 /* Grab inode buffer early before taking global s_orphan_lock */
2864 err = ext4_reserve_inode_write(handle, inode, &iloc);
2867 mutex_lock(&sbi->s_orphan_lock);
2868 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2870 prev = ei->i_orphan.prev;
2871 list_del_init(&ei->i_orphan);
2873 /* If we're on an error path, we may not have a valid
2874 * transaction handle with which to update the orphan list on
2875 * disk, but we still need to remove the inode from the linked
2876 * list in memory. */
2877 if (!handle || err) {
2878 mutex_unlock(&sbi->s_orphan_lock);
2882 ino_next = NEXT_ORPHAN(inode);
2883 if (prev == &sbi->s_orphan) {
2884 jbd_debug(4, "superblock will point to %u\n", ino_next);
2885 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2886 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2888 mutex_unlock(&sbi->s_orphan_lock);
2891 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2892 mutex_unlock(&sbi->s_orphan_lock);
2893 err = ext4_handle_dirty_super(handle, inode->i_sb);
2895 struct ext4_iloc iloc2;
2896 struct inode *i_prev =
2897 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2899 jbd_debug(4, "orphan inode %lu will point to %u\n",
2900 i_prev->i_ino, ino_next);
2901 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2903 mutex_unlock(&sbi->s_orphan_lock);
2906 NEXT_ORPHAN(i_prev) = ino_next;
2907 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2908 mutex_unlock(&sbi->s_orphan_lock);
2912 NEXT_ORPHAN(inode) = 0;
2913 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2915 ext4_std_error(inode->i_sb, err);
2923 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2926 struct inode *inode;
2927 struct buffer_head *bh;
2928 struct ext4_dir_entry_2 *de;
2929 handle_t *handle = NULL;
2931 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2934 /* Initialize quotas before so that eventual writes go in
2935 * separate transaction */
2936 retval = dquot_initialize(dir);
2939 retval = dquot_initialize(d_inode(dentry));
2944 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2950 inode = d_inode(dentry);
2952 retval = -EFSCORRUPTED;
2953 if (le32_to_cpu(de->inode) != inode->i_ino)
2956 retval = -ENOTEMPTY;
2957 if (!ext4_empty_dir(inode))
2960 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2961 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2962 if (IS_ERR(handle)) {
2963 retval = PTR_ERR(handle);
2968 if (IS_DIRSYNC(dir))
2969 ext4_handle_sync(handle);
2971 retval = ext4_delete_entry(handle, dir, de, bh);
2974 if (!EXT4_DIR_LINK_EMPTY(inode))
2975 ext4_warning_inode(inode,
2976 "empty directory '%.*s' has too many links (%u)",
2977 dentry->d_name.len, dentry->d_name.name,
2981 /* There's no need to set i_disksize: the fact that i_nlink is
2982 * zero will ensure that the right thing happens during any
2985 ext4_orphan_add(handle, inode);
2986 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2987 ext4_mark_inode_dirty(handle, inode);
2988 ext4_dec_count(handle, dir);
2989 ext4_update_dx_flag(dir);
2990 ext4_mark_inode_dirty(handle, dir);
2995 ext4_journal_stop(handle);
2999 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3002 struct inode *inode;
3003 struct buffer_head *bh;
3004 struct ext4_dir_entry_2 *de;
3005 handle_t *handle = NULL;
3007 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3010 trace_ext4_unlink_enter(dir, dentry);
3011 /* Initialize quotas before so that eventual writes go
3012 * in separate transaction */
3013 retval = dquot_initialize(dir);
3016 retval = dquot_initialize(d_inode(dentry));
3021 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3027 inode = d_inode(dentry);
3029 retval = -EFSCORRUPTED;
3030 if (le32_to_cpu(de->inode) != inode->i_ino)
3033 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3034 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3035 if (IS_ERR(handle)) {
3036 retval = PTR_ERR(handle);
3041 if (IS_DIRSYNC(dir))
3042 ext4_handle_sync(handle);
3044 if (inode->i_nlink == 0) {
3045 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3046 dentry->d_name.len, dentry->d_name.name);
3047 set_nlink(inode, 1);
3049 retval = ext4_delete_entry(handle, dir, de, bh);
3052 dir->i_ctime = dir->i_mtime = current_time(dir);
3053 ext4_update_dx_flag(dir);
3054 ext4_mark_inode_dirty(handle, dir);
3056 if (!inode->i_nlink)
3057 ext4_orphan_add(handle, inode);
3058 inode->i_ctime = current_time(inode);
3059 ext4_mark_inode_dirty(handle, inode);
3064 ext4_journal_stop(handle);
3065 trace_ext4_unlink_exit(dentry, retval);
3069 static int ext4_symlink(struct inode *dir,
3070 struct dentry *dentry, const char *symname)
3073 struct inode *inode;
3074 int err, len = strlen(symname);
3076 bool encryption_required;
3077 struct fscrypt_str disk_link;
3078 struct fscrypt_symlink_data *sd = NULL;
3080 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3083 disk_link.len = len + 1;
3084 disk_link.name = (char *) symname;
3086 encryption_required = (ext4_encrypted_inode(dir) ||
3087 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3088 if (encryption_required) {
3089 err = fscrypt_get_encryption_info(dir);
3092 if (!fscrypt_has_encryption_key(dir))
3094 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3095 sizeof(struct fscrypt_symlink_data));
3096 sd = kzalloc(disk_link.len, GFP_KERNEL);
3101 if (disk_link.len > dir->i_sb->s_blocksize) {
3102 err = -ENAMETOOLONG;
3106 err = dquot_initialize(dir);
3110 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3112 * For non-fast symlinks, we just allocate inode and put it on
3113 * orphan list in the first transaction => we need bitmap,
3114 * group descriptor, sb, inode block, quota blocks, and
3115 * possibly selinux xattr blocks.
3117 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3118 EXT4_XATTR_TRANS_BLOCKS;
3121 * Fast symlink. We have to add entry to directory
3122 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3123 * allocate new inode (bitmap, group descriptor, inode block,
3124 * quota blocks, sb is already counted in previous macros).
3126 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3127 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3130 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3131 &dentry->d_name, 0, NULL,
3132 EXT4_HT_DIR, credits);
3133 handle = ext4_journal_current_handle();
3134 if (IS_ERR(inode)) {
3136 ext4_journal_stop(handle);
3137 err = PTR_ERR(inode);
3141 if (encryption_required) {
3143 struct fscrypt_str ostr =
3144 FSTR_INIT(sd->encrypted_path, disk_link.len);
3146 istr.name = (const unsigned char *) symname;
3148 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3150 goto err_drop_inode;
3151 sd->len = cpu_to_le16(ostr.len);
3152 disk_link.name = (char *) sd;
3153 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3156 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3157 if (!encryption_required)
3158 inode->i_op = &ext4_symlink_inode_operations;
3159 inode_nohighmem(inode);
3160 ext4_set_aops(inode);
3162 * We cannot call page_symlink() with transaction started
3163 * because it calls into ext4_write_begin() which can wait
3164 * for transaction commit if we are running out of space
3165 * and thus we deadlock. So we have to stop transaction now
3166 * and restart it when symlink contents is written.
3168 * To keep fs consistent in case of crash, we have to put inode
3169 * to orphan list in the mean time.
3172 err = ext4_orphan_add(handle, inode);
3173 ext4_journal_stop(handle);
3176 goto err_drop_inode;
3177 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3179 goto err_drop_inode;
3181 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3182 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3184 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3185 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3186 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3187 if (IS_ERR(handle)) {
3188 err = PTR_ERR(handle);
3190 goto err_drop_inode;
3192 set_nlink(inode, 1);
3193 err = ext4_orphan_del(handle, inode);
3195 goto err_drop_inode;
3197 /* clear the extent format for fast symlink */
3198 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3199 if (!encryption_required) {
3200 inode->i_op = &ext4_fast_symlink_inode_operations;
3201 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3203 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3205 inode->i_size = disk_link.len - 1;
3207 EXT4_I(inode)->i_disksize = inode->i_size;
3208 err = ext4_add_nondir(handle, dentry, inode);
3209 if (!err && IS_DIRSYNC(dir))
3210 ext4_handle_sync(handle);
3213 ext4_journal_stop(handle);
3218 ext4_journal_stop(handle);
3220 unlock_new_inode(inode);
3227 static int ext4_link(struct dentry *old_dentry,
3228 struct inode *dir, struct dentry *dentry)
3231 struct inode *inode = d_inode(old_dentry);
3232 int err, retries = 0;
3234 if (inode->i_nlink >= EXT4_LINK_MAX)
3236 if (ext4_encrypted_inode(dir) &&
3237 !fscrypt_has_permitted_context(dir, inode))
3240 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3241 (!projid_eq(EXT4_I(dir)->i_projid,
3242 EXT4_I(old_dentry->d_inode)->i_projid)))
3245 err = dquot_initialize(dir);
3250 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3251 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3252 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3254 return PTR_ERR(handle);
3256 if (IS_DIRSYNC(dir))
3257 ext4_handle_sync(handle);
3259 inode->i_ctime = current_time(inode);
3260 ext4_inc_count(handle, inode);
3263 err = ext4_add_entry(handle, dentry, inode);
3265 ext4_mark_inode_dirty(handle, inode);
3266 /* this can happen only for tmpfile being
3267 * linked the first time
3269 if (inode->i_nlink == 1)
3270 ext4_orphan_del(handle, inode);
3271 d_instantiate(dentry, inode);
3276 ext4_journal_stop(handle);
3277 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3284 * Try to find buffer head where contains the parent block.
3285 * It should be the inode block if it is inlined or the 1st block
3286 * if it is a normal dir.
3288 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3289 struct inode *inode,
3291 struct ext4_dir_entry_2 **parent_de,
3294 struct buffer_head *bh;
3296 if (!ext4_has_inline_data(inode)) {
3297 bh = ext4_read_dirblock(inode, 0, EITHER);
3299 *retval = PTR_ERR(bh);
3302 *parent_de = ext4_next_entry(
3303 (struct ext4_dir_entry_2 *)bh->b_data,
3304 inode->i_sb->s_blocksize);
3309 return ext4_get_first_inline_block(inode, parent_de, retval);
3312 struct ext4_renament {
3314 struct dentry *dentry;
3315 struct inode *inode;
3317 int dir_nlink_delta;
3319 /* entry for "dentry" */
3320 struct buffer_head *bh;
3321 struct ext4_dir_entry_2 *de;
3324 /* entry for ".." in inode if it's a directory */
3325 struct buffer_head *dir_bh;
3326 struct ext4_dir_entry_2 *parent_de;
3330 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3334 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3335 &retval, &ent->parent_de,
3339 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3340 return -EFSCORRUPTED;
3341 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3342 return ext4_journal_get_write_access(handle, ent->dir_bh);
3345 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3350 ent->parent_de->inode = cpu_to_le32(dir_ino);
3351 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3352 if (!ent->dir_inlined) {
3353 if (is_dx(ent->inode)) {
3354 retval = ext4_handle_dirty_dx_node(handle,
3358 retval = ext4_handle_dirty_dirent_node(handle,
3363 retval = ext4_mark_inode_dirty(handle, ent->inode);
3366 ext4_std_error(ent->dir->i_sb, retval);
3372 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3373 unsigned ino, unsigned file_type)
3377 BUFFER_TRACE(ent->bh, "get write access");
3378 retval = ext4_journal_get_write_access(handle, ent->bh);
3381 ent->de->inode = cpu_to_le32(ino);
3382 if (ext4_has_feature_filetype(ent->dir->i_sb))
3383 ent->de->file_type = file_type;
3384 ent->dir->i_version++;
3385 ent->dir->i_ctime = ent->dir->i_mtime =
3386 current_time(ent->dir);
3387 ext4_mark_inode_dirty(handle, ent->dir);
3388 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3389 if (!ent->inlined) {
3390 retval = ext4_handle_dirty_dirent_node(handle,
3392 if (unlikely(retval)) {
3393 ext4_std_error(ent->dir->i_sb, retval);
3403 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3404 const struct qstr *d_name)
3406 int retval = -ENOENT;
3407 struct buffer_head *bh;
3408 struct ext4_dir_entry_2 *de;
3410 bh = ext4_find_entry(dir, d_name, &de, NULL);
3414 retval = ext4_delete_entry(handle, dir, de, bh);
3420 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3425 * ent->de could have moved from under us during htree split, so make
3426 * sure that we are deleting the right entry. We might also be pointing
3427 * to a stale entry in the unused part of ent->bh so just checking inum
3428 * and the name isn't enough.
3430 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3431 ent->de->name_len != ent->dentry->d_name.len ||
3432 strncmp(ent->de->name, ent->dentry->d_name.name,
3433 ent->de->name_len) ||
3435 retval = ext4_find_delete_entry(handle, ent->dir,
3436 &ent->dentry->d_name);
3438 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3439 if (retval == -ENOENT) {
3440 retval = ext4_find_delete_entry(handle, ent->dir,
3441 &ent->dentry->d_name);
3446 ext4_warning_inode(ent->dir,
3447 "Deleting old file: nlink %d, error=%d",
3448 ent->dir->i_nlink, retval);
3452 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3454 if (ent->dir_nlink_delta) {
3455 if (ent->dir_nlink_delta == -1)
3456 ext4_dec_count(handle, ent->dir);
3458 ext4_inc_count(handle, ent->dir);
3459 ext4_mark_inode_dirty(handle, ent->dir);
3463 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3464 int credits, handle_t **h)
3471 * for inode block, sb block, group summaries,
3474 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3475 EXT4_XATTR_TRANS_BLOCKS + 4);
3477 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3478 &ent->dentry->d_name, 0, NULL,
3479 EXT4_HT_DIR, credits);
3481 handle = ext4_journal_current_handle();
3484 ext4_journal_stop(handle);
3485 if (PTR_ERR(wh) == -ENOSPC &&
3486 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3490 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3491 wh->i_op = &ext4_special_inode_operations;
3497 * Anybody can rename anything with this: the permission checks are left to the
3498 * higher-level routines.
3500 * n.b. old_{dentry,inode) refers to the source dentry/inode
3501 * while new_{dentry,inode) refers to the destination dentry/inode
3502 * This comes from rename(const char *oldpath, const char *newpath)
3504 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3505 struct inode *new_dir, struct dentry *new_dentry,
3508 handle_t *handle = NULL;
3509 struct ext4_renament old = {
3511 .dentry = old_dentry,
3512 .inode = d_inode(old_dentry),
3514 struct ext4_renament new = {
3516 .dentry = new_dentry,
3517 .inode = d_inode(new_dentry),
3521 struct inode *whiteout = NULL;
3525 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3526 (!projid_eq(EXT4_I(new_dir)->i_projid,
3527 EXT4_I(old_dentry->d_inode)->i_projid)))
3530 if ((ext4_encrypted_inode(old_dir) &&
3531 !fscrypt_has_encryption_key(old_dir)) ||
3532 (ext4_encrypted_inode(new_dir) &&
3533 !fscrypt_has_encryption_key(new_dir)))
3536 retval = dquot_initialize(old.dir);
3539 retval = dquot_initialize(new.dir);
3543 /* Initialize quotas before so that eventual writes go
3544 * in separate transaction */
3546 retval = dquot_initialize(new.inode);
3551 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3553 return PTR_ERR(old.bh);
3555 * Check for inode number is _not_ due to possible IO errors.
3556 * We might rmdir the source, keep it as pwd of some process
3557 * and merrily kill the link to whatever was created under the
3558 * same name. Goodbye sticky bit ;-<
3561 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3564 if ((old.dir != new.dir) &&
3565 ext4_encrypted_inode(new.dir) &&
3566 !fscrypt_has_permitted_context(new.dir, old.inode)) {
3571 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3572 &new.de, &new.inlined);
3573 if (IS_ERR(new.bh)) {
3574 retval = PTR_ERR(new.bh);
3584 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3585 ext4_alloc_da_blocks(old.inode);
3587 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3588 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3589 if (!(flags & RENAME_WHITEOUT)) {
3590 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3591 if (IS_ERR(handle)) {
3592 retval = PTR_ERR(handle);
3597 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3598 if (IS_ERR(whiteout)) {
3599 retval = PTR_ERR(whiteout);
3605 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3606 ext4_handle_sync(handle);
3608 if (S_ISDIR(old.inode->i_mode)) {
3610 retval = -ENOTEMPTY;
3611 if (!ext4_empty_dir(new.inode))
3615 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3618 retval = ext4_rename_dir_prepare(handle, &old);
3623 * If we're renaming a file within an inline_data dir and adding or
3624 * setting the new dirent causes a conversion from inline_data to
3625 * extents/blockmap, we need to force the dirent delete code to
3626 * re-read the directory, or else we end up trying to delete a dirent
3627 * from what is now the extent tree root (or a block map).
3629 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3630 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3632 old_file_type = old.de->file_type;
3635 * Do this before adding a new entry, so the old entry is sure
3636 * to be still pointing to the valid old entry.
3638 retval = ext4_setent(handle, &old, whiteout->i_ino,
3642 ext4_mark_inode_dirty(handle, whiteout);
3645 retval = ext4_add_entry(handle, new.dentry, old.inode);
3649 retval = ext4_setent(handle, &new,
3650 old.inode->i_ino, old_file_type);
3655 force_reread = !ext4_test_inode_flag(new.dir,
3656 EXT4_INODE_INLINE_DATA);
3659 * Like most other Unix systems, set the ctime for inodes on a
3662 old.inode->i_ctime = current_time(old.inode);
3663 ext4_mark_inode_dirty(handle, old.inode);
3669 ext4_rename_delete(handle, &old, force_reread);
3673 ext4_dec_count(handle, new.inode);
3674 new.inode->i_ctime = current_time(new.inode);
3676 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3677 ext4_update_dx_flag(old.dir);
3679 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3683 ext4_dec_count(handle, old.dir);
3685 /* checked ext4_empty_dir above, can't have another
3686 * parent, ext4_dec_count() won't work for many-linked
3688 clear_nlink(new.inode);
3690 ext4_inc_count(handle, new.dir);
3691 ext4_update_dx_flag(new.dir);
3692 ext4_mark_inode_dirty(handle, new.dir);
3695 ext4_mark_inode_dirty(handle, old.dir);
3697 ext4_mark_inode_dirty(handle, new.inode);
3698 if (!new.inode->i_nlink)
3699 ext4_orphan_add(handle, new.inode);
3709 drop_nlink(whiteout);
3710 unlock_new_inode(whiteout);
3714 ext4_journal_stop(handle);
3718 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3719 struct inode *new_dir, struct dentry *new_dentry)
3721 handle_t *handle = NULL;
3722 struct ext4_renament old = {
3724 .dentry = old_dentry,
3725 .inode = d_inode(old_dentry),
3727 struct ext4_renament new = {
3729 .dentry = new_dentry,
3730 .inode = d_inode(new_dentry),
3734 struct timespec ctime;
3736 if ((ext4_encrypted_inode(old_dir) &&
3737 !fscrypt_has_encryption_key(old_dir)) ||
3738 (ext4_encrypted_inode(new_dir) &&
3739 !fscrypt_has_encryption_key(new_dir)))
3742 if ((ext4_encrypted_inode(old_dir) ||
3743 ext4_encrypted_inode(new_dir)) &&
3744 (old_dir != new_dir) &&
3745 (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3746 !fscrypt_has_permitted_context(old_dir, new.inode)))
3749 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3750 !projid_eq(EXT4_I(new_dir)->i_projid,
3751 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3752 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3753 !projid_eq(EXT4_I(old_dir)->i_projid,
3754 EXT4_I(new_dentry->d_inode)->i_projid)))
3757 retval = dquot_initialize(old.dir);
3760 retval = dquot_initialize(new.dir);
3764 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3765 &old.de, &old.inlined);
3767 return PTR_ERR(old.bh);
3769 * Check for inode number is _not_ due to possible IO errors.
3770 * We might rmdir the source, keep it as pwd of some process
3771 * and merrily kill the link to whatever was created under the
3772 * same name. Goodbye sticky bit ;-<
3775 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3778 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3779 &new.de, &new.inlined);
3780 if (IS_ERR(new.bh)) {
3781 retval = PTR_ERR(new.bh);
3786 /* RENAME_EXCHANGE case: old *and* new must both exist */
3787 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3790 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3791 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3792 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3793 if (IS_ERR(handle)) {
3794 retval = PTR_ERR(handle);
3799 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3800 ext4_handle_sync(handle);
3802 if (S_ISDIR(old.inode->i_mode)) {
3804 retval = ext4_rename_dir_prepare(handle, &old);
3808 if (S_ISDIR(new.inode->i_mode)) {
3810 retval = ext4_rename_dir_prepare(handle, &new);
3816 * Other than the special case of overwriting a directory, parents'
3817 * nlink only needs to be modified if this is a cross directory rename.
3819 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3820 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3821 new.dir_nlink_delta = -old.dir_nlink_delta;
3823 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3824 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3828 new_file_type = new.de->file_type;
3829 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3833 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3838 * Like most other Unix systems, set the ctime for inodes on a
3841 ctime = current_time(old.inode);
3842 old.inode->i_ctime = ctime;
3843 new.inode->i_ctime = ctime;
3844 ext4_mark_inode_dirty(handle, old.inode);
3845 ext4_mark_inode_dirty(handle, new.inode);
3848 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3853 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3857 ext4_update_dir_count(handle, &old);
3858 ext4_update_dir_count(handle, &new);
3867 ext4_journal_stop(handle);
3871 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3872 struct inode *new_dir, struct dentry *new_dentry,
3875 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3878 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3881 if (flags & RENAME_EXCHANGE) {
3882 return ext4_cross_rename(old_dir, old_dentry,
3883 new_dir, new_dentry);
3886 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3890 * directories can handle most operations...
3892 const struct inode_operations ext4_dir_inode_operations = {
3893 .create = ext4_create,
3894 .lookup = ext4_lookup,
3896 .unlink = ext4_unlink,
3897 .symlink = ext4_symlink,
3898 .mkdir = ext4_mkdir,
3899 .rmdir = ext4_rmdir,
3900 .mknod = ext4_mknod,
3901 .tmpfile = ext4_tmpfile,
3902 .rename = ext4_rename2,
3903 .setattr = ext4_setattr,
3904 .getattr = ext4_getattr,
3905 .listxattr = ext4_listxattr,
3906 .get_acl = ext4_get_acl,
3907 .set_acl = ext4_set_acl,
3908 .fiemap = ext4_fiemap,
3911 const struct inode_operations ext4_special_inode_operations = {
3912 .setattr = ext4_setattr,
3913 .getattr = ext4_getattr,
3914 .listxattr = ext4_listxattr,
3915 .get_acl = ext4_get_acl,
3916 .set_acl = ext4_set_acl,