1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/namei.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/namei.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 * Directory entry file type support and forward compatibility hooks
19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20 * Hash Tree Directory indexing (c)
21 * Daniel Phillips, 2001
22 * Hash Tree Directory indexing porting
23 * Christopher Li, 2002
24 * Hash Tree Directory indexing cleanup
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include <linux/iversion.h>
39 #include "ext4_jbd2.h"
44 #include <trace/events/ext4.h>
46 * define how far ahead to read directories while searching them.
48 #define NAMEI_RA_CHUNKS 2
49 #define NAMEI_RA_BLOCKS 4
50 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52 static struct buffer_head *ext4_append(handle_t *handle,
56 struct buffer_head *bh;
59 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
60 ((inode->i_size >> 10) >=
61 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
62 return ERR_PTR(-ENOSPC);
64 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
66 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
69 inode->i_size += inode->i_sb->s_blocksize;
70 EXT4_I(inode)->i_disksize = inode->i_size;
71 BUFFER_TRACE(bh, "get_write_access");
72 err = ext4_journal_get_write_access(handle, bh);
75 ext4_std_error(inode->i_sb, err);
81 static int ext4_dx_csum_verify(struct inode *inode,
82 struct ext4_dir_entry *dirent);
85 * Hints to ext4_read_dirblock regarding whether we expect a directory
86 * block being read to be an index block, or a block containing
87 * directory entries (and if the latter, whether it was found via a
88 * logical block in an htree index block). This is used to control
89 * what sort of sanity checkinig ext4_read_dirblock() will do on the
90 * directory block read from the storage device. EITHER will means
91 * the caller doesn't know what kind of directory block will be read,
92 * so no specific verification will be done.
95 EITHER, INDEX, DIRENT, DIRENT_HTREE
98 #define ext4_read_dirblock(inode, block, type) \
99 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
101 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
103 dirblock_type_t type,
107 struct buffer_head *bh;
108 struct ext4_dir_entry *dirent;
111 bh = ext4_bread(NULL, inode, block, 0);
113 __ext4_warning(inode->i_sb, func, line,
114 "inode #%lu: lblock %lu: comm %s: "
115 "error %ld reading directory block",
116 inode->i_ino, (unsigned long)block,
117 current->comm, PTR_ERR(bh));
121 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
122 ext4_error_inode(inode, func, line, block,
123 "Directory hole found for htree %s block",
124 (type == INDEX) ? "index" : "leaf");
125 return ERR_PTR(-EFSCORRUPTED);
129 dirent = (struct ext4_dir_entry *) bh->b_data;
130 /* Determine whether or not we have an index block */
134 else if (ext4_rec_len_from_disk(dirent->rec_len,
135 inode->i_sb->s_blocksize) ==
136 inode->i_sb->s_blocksize)
139 if (!is_dx_block && type == INDEX) {
140 ext4_error_inode(inode, func, line, block,
141 "directory leaf block found instead of index block");
143 return ERR_PTR(-EFSCORRUPTED);
145 if (!ext4_has_metadata_csum(inode->i_sb) ||
150 * An empty leaf block can get mistaken for a index block; for
151 * this reason, we can only check the index checksum when the
152 * caller is sure it should be an index block.
154 if (is_dx_block && type == INDEX) {
155 if (ext4_dx_csum_verify(inode, dirent))
156 set_buffer_verified(bh);
158 ext4_error_inode(inode, func, line, block,
159 "Directory index failed checksum");
161 return ERR_PTR(-EFSBADCRC);
165 if (ext4_dirent_csum_verify(inode, dirent))
166 set_buffer_verified(bh);
168 ext4_error_inode(inode, func, line, block,
169 "Directory block failed checksum");
171 return ERR_PTR(-EFSBADCRC);
178 #define assert(test) J_ASSERT(test)
182 #define dxtrace(command) command
184 #define dxtrace(command)
208 * dx_root_info is laid out so that if it should somehow get overlaid by a
209 * dirent the two low bits of the hash version will be zero. Therefore, the
210 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
215 struct fake_dirent dot;
217 struct fake_dirent dotdot;
221 __le32 reserved_zero;
223 u8 info_length; /* 8 */
228 struct dx_entry entries[0];
233 struct fake_dirent fake;
234 struct dx_entry entries[0];
240 struct buffer_head *bh;
241 struct dx_entry *entries;
253 * This goes at the end of each htree block.
257 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
260 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
261 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
262 static inline unsigned dx_get_hash(struct dx_entry *entry);
263 static void dx_set_hash(struct dx_entry *entry, unsigned value);
264 static unsigned dx_get_count(struct dx_entry *entries);
265 static unsigned dx_get_limit(struct dx_entry *entries);
266 static void dx_set_count(struct dx_entry *entries, unsigned value);
267 static void dx_set_limit(struct dx_entry *entries, unsigned value);
268 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
269 static unsigned dx_node_limit(struct inode *dir);
270 static struct dx_frame *dx_probe(struct ext4_filename *fname,
272 struct dx_hash_info *hinfo,
273 struct dx_frame *frame);
274 static void dx_release(struct dx_frame *frames);
275 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
276 unsigned blocksize, struct dx_hash_info *hinfo,
277 struct dx_map_entry map[]);
278 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
279 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
280 struct dx_map_entry *offsets, int count, unsigned blocksize);
281 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
282 static void dx_insert_block(struct dx_frame *frame,
283 u32 hash, ext4_lblk_t block);
284 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
285 struct dx_frame *frame,
286 struct dx_frame *frames,
288 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
289 struct ext4_filename *fname,
290 struct ext4_dir_entry_2 **res_dir);
291 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
292 struct inode *dir, struct inode *inode);
294 /* checksumming functions */
295 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
296 unsigned int blocksize)
298 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
299 t->det_rec_len = ext4_rec_len_to_disk(
300 sizeof(struct ext4_dir_entry_tail), blocksize);
301 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
304 /* Walk through a dirent block to find a checksum "dirent" at the tail */
305 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
306 struct ext4_dir_entry *de)
308 struct ext4_dir_entry_tail *t;
311 struct ext4_dir_entry *d, *top;
314 top = (struct ext4_dir_entry *)(((void *)de) +
315 (EXT4_BLOCK_SIZE(inode->i_sb) -
316 sizeof(struct ext4_dir_entry_tail)));
317 while (d < top && d->rec_len)
318 d = (struct ext4_dir_entry *)(((void *)d) +
319 le16_to_cpu(d->rec_len));
324 t = (struct ext4_dir_entry_tail *)d;
326 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
329 if (t->det_reserved_zero1 ||
330 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
331 t->det_reserved_zero2 ||
332 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
338 static __le32 ext4_dirent_csum(struct inode *inode,
339 struct ext4_dir_entry *dirent, int size)
341 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
342 struct ext4_inode_info *ei = EXT4_I(inode);
345 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
346 return cpu_to_le32(csum);
349 #define warn_no_space_for_csum(inode) \
350 __warn_no_space_for_csum((inode), __func__, __LINE__)
352 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
355 __ext4_warning_inode(inode, func, line,
356 "No space for directory leaf checksum. Please run e2fsck -D.");
359 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
361 struct ext4_dir_entry_tail *t;
363 if (!ext4_has_metadata_csum(inode->i_sb))
366 t = get_dirent_tail(inode, dirent);
368 warn_no_space_for_csum(inode);
372 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
373 (void *)t - (void *)dirent))
379 static void ext4_dirent_csum_set(struct inode *inode,
380 struct ext4_dir_entry *dirent)
382 struct ext4_dir_entry_tail *t;
384 if (!ext4_has_metadata_csum(inode->i_sb))
387 t = get_dirent_tail(inode, dirent);
389 warn_no_space_for_csum(inode);
393 t->det_checksum = ext4_dirent_csum(inode, dirent,
394 (void *)t - (void *)dirent);
397 int ext4_handle_dirty_dirent_node(handle_t *handle,
399 struct buffer_head *bh)
401 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
402 return ext4_handle_dirty_metadata(handle, inode, bh);
405 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
406 struct ext4_dir_entry *dirent,
409 struct ext4_dir_entry *dp;
410 struct dx_root_info *root;
413 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
415 else if (le16_to_cpu(dirent->rec_len) == 12) {
416 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
417 if (le16_to_cpu(dp->rec_len) !=
418 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
420 root = (struct dx_root_info *)(((void *)dp + 12));
421 if (root->reserved_zero ||
422 root->info_length != sizeof(struct dx_root_info))
429 *offset = count_offset;
430 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
433 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
434 int count_offset, int count, struct dx_tail *t)
436 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
437 struct ext4_inode_info *ei = EXT4_I(inode);
440 __u32 dummy_csum = 0;
441 int offset = offsetof(struct dx_tail, dt_checksum);
443 size = count_offset + (count * sizeof(struct dx_entry));
444 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
445 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
446 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
448 return cpu_to_le32(csum);
451 static int ext4_dx_csum_verify(struct inode *inode,
452 struct ext4_dir_entry *dirent)
454 struct dx_countlimit *c;
456 int count_offset, limit, count;
458 if (!ext4_has_metadata_csum(inode->i_sb))
461 c = get_dx_countlimit(inode, dirent, &count_offset);
463 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
466 limit = le16_to_cpu(c->limit);
467 count = le16_to_cpu(c->count);
468 if (count_offset + (limit * sizeof(struct dx_entry)) >
469 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
470 warn_no_space_for_csum(inode);
473 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
475 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
481 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
483 struct dx_countlimit *c;
485 int count_offset, limit, count;
487 if (!ext4_has_metadata_csum(inode->i_sb))
490 c = get_dx_countlimit(inode, dirent, &count_offset);
492 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
495 limit = le16_to_cpu(c->limit);
496 count = le16_to_cpu(c->count);
497 if (count_offset + (limit * sizeof(struct dx_entry)) >
498 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
499 warn_no_space_for_csum(inode);
502 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
504 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
507 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
509 struct buffer_head *bh)
511 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
512 return ext4_handle_dirty_metadata(handle, inode, bh);
516 * p is at least 6 bytes before the end of page
518 static inline struct ext4_dir_entry_2 *
519 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
521 return (struct ext4_dir_entry_2 *)((char *)p +
522 ext4_rec_len_from_disk(p->rec_len, blocksize));
526 * Future: use high four bits of block for coalesce-on-delete flags
527 * Mask them off for now.
530 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
532 return le32_to_cpu(entry->block) & 0x0fffffff;
535 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
537 entry->block = cpu_to_le32(value);
540 static inline unsigned dx_get_hash(struct dx_entry *entry)
542 return le32_to_cpu(entry->hash);
545 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
547 entry->hash = cpu_to_le32(value);
550 static inline unsigned dx_get_count(struct dx_entry *entries)
552 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
555 static inline unsigned dx_get_limit(struct dx_entry *entries)
557 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
560 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
562 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
565 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
567 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
570 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
572 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
573 EXT4_DIR_REC_LEN(2) - infosize;
575 if (ext4_has_metadata_csum(dir->i_sb))
576 entry_space -= sizeof(struct dx_tail);
577 return entry_space / sizeof(struct dx_entry);
580 static inline unsigned dx_node_limit(struct inode *dir)
582 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
584 if (ext4_has_metadata_csum(dir->i_sb))
585 entry_space -= sizeof(struct dx_tail);
586 return entry_space / sizeof(struct dx_entry);
593 static void dx_show_index(char * label, struct dx_entry *entries)
595 int i, n = dx_get_count (entries);
596 printk(KERN_DEBUG "%s index", label);
597 for (i = 0; i < n; i++) {
598 printk(KERN_CONT " %x->%lu",
599 i ? dx_get_hash(entries + i) : 0,
600 (unsigned long)dx_get_block(entries + i));
602 printk(KERN_CONT "\n");
612 static struct stats dx_show_leaf(struct inode *dir,
613 struct dx_hash_info *hinfo,
614 struct ext4_dir_entry_2 *de,
615 int size, int show_names)
617 unsigned names = 0, space = 0;
618 char *base = (char *) de;
619 struct dx_hash_info h = *hinfo;
622 while ((char *) de < base + size)
628 #ifdef CONFIG_EXT4_FS_ENCRYPTION
631 struct fscrypt_str fname_crypto_str =
637 if (ext4_encrypted_inode(dir))
638 res = fscrypt_get_encryption_info(dir);
640 printk(KERN_WARNING "Error setting up"
641 " fname crypto: %d\n", res);
643 if (!fscrypt_has_encryption_key(dir)) {
644 /* Directory is not encrypted */
645 ext4fs_dirhash(de->name,
647 printk("%*.s:(U)%x.%u ", len,
649 (unsigned) ((char *) de
652 struct fscrypt_str de_name =
653 FSTR_INIT(name, len);
655 /* Directory is encrypted */
656 res = fscrypt_fname_alloc_buffer(
660 printk(KERN_WARNING "Error "
664 res = fscrypt_fname_disk_to_usr(dir,
668 printk(KERN_WARNING "Error "
669 "converting filename "
675 name = fname_crypto_str.name;
676 len = fname_crypto_str.len;
678 ext4fs_dirhash(de->name, de->name_len,
680 printk("%*.s:(E)%x.%u ", len, name,
681 h.hash, (unsigned) ((char *) de
683 fscrypt_fname_free_buffer(
687 int len = de->name_len;
688 char *name = de->name;
689 ext4fs_dirhash(de->name, de->name_len, &h);
690 printk("%*.s:%x.%u ", len, name, h.hash,
691 (unsigned) ((char *) de - base));
694 space += EXT4_DIR_REC_LEN(de->name_len);
697 de = ext4_next_entry(de, size);
699 printk(KERN_CONT "(%i)\n", names);
700 return (struct stats) { names, space, 1 };
703 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
704 struct dx_entry *entries, int levels)
706 unsigned blocksize = dir->i_sb->s_blocksize;
707 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
709 struct buffer_head *bh;
710 printk("%i indexed blocks...\n", count);
711 for (i = 0; i < count; i++, entries++)
713 ext4_lblk_t block = dx_get_block(entries);
714 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
715 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
717 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
718 bh = ext4_bread(NULL,dir, block, 0);
719 if (!bh || IS_ERR(bh))
722 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
723 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
724 bh->b_data, blocksize, 0);
725 names += stats.names;
726 space += stats.space;
727 bcount += stats.bcount;
731 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
732 levels ? "" : " ", names, space/bcount,
733 (space/bcount)*100/blocksize);
734 return (struct stats) { names, space, bcount};
736 #endif /* DX_DEBUG */
739 * Probe for a directory leaf block to search.
741 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
742 * error in the directory index, and the caller should fall back to
743 * searching the directory normally. The callers of dx_probe **MUST**
744 * check for this error code, and make sure it never gets reflected
747 static struct dx_frame *
748 dx_probe(struct ext4_filename *fname, struct inode *dir,
749 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
751 unsigned count, indirect;
752 struct dx_entry *at, *entries, *p, *q, *m;
753 struct dx_root *root;
754 struct dx_frame *frame = frame_in;
755 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
758 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
759 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
760 if (IS_ERR(frame->bh))
761 return (struct dx_frame *) frame->bh;
763 root = (struct dx_root *) frame->bh->b_data;
764 if (root->info.hash_version != DX_HASH_TEA &&
765 root->info.hash_version != DX_HASH_HALF_MD4 &&
766 root->info.hash_version != DX_HASH_LEGACY) {
767 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
768 root->info.hash_version);
772 hinfo = &fname->hinfo;
773 hinfo->hash_version = root->info.hash_version;
774 if (hinfo->hash_version <= DX_HASH_TEA)
775 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
776 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
777 if (fname && fname_name(fname))
778 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
781 if (root->info.unused_flags & 1) {
782 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
783 root->info.unused_flags);
787 indirect = root->info.indirect_levels;
788 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
789 ext4_warning(dir->i_sb,
790 "Directory (ino: %lu) htree depth %#06x exceed"
791 "supported value", dir->i_ino,
792 ext4_dir_htree_level(dir->i_sb));
793 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
794 ext4_warning(dir->i_sb, "Enable large directory "
795 "feature to access it");
800 entries = (struct dx_entry *)(((char *)&root->info) +
801 root->info.info_length);
803 if (dx_get_limit(entries) != dx_root_limit(dir,
804 root->info.info_length)) {
805 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
806 dx_get_limit(entries),
807 dx_root_limit(dir, root->info.info_length));
811 dxtrace(printk("Look up %x", hash));
813 count = dx_get_count(entries);
814 if (!count || count > dx_get_limit(entries)) {
815 ext4_warning_inode(dir,
816 "dx entry: count %u beyond limit %u",
817 count, dx_get_limit(entries));
822 q = entries + count - 1;
825 dxtrace(printk(KERN_CONT "."));
826 if (dx_get_hash(m) > hash)
832 if (0) { // linear search cross check
833 unsigned n = count - 1;
837 dxtrace(printk(KERN_CONT ","));
838 if (dx_get_hash(++at) > hash)
844 assert (at == p - 1);
848 dxtrace(printk(KERN_CONT " %x->%u\n",
849 at == entries ? 0 : dx_get_hash(at),
851 frame->entries = entries;
856 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
857 if (IS_ERR(frame->bh)) {
858 ret_err = (struct dx_frame *) frame->bh;
862 entries = ((struct dx_node *) frame->bh->b_data)->entries;
864 if (dx_get_limit(entries) != dx_node_limit(dir)) {
865 ext4_warning_inode(dir,
866 "dx entry: limit %u != node limit %u",
867 dx_get_limit(entries), dx_node_limit(dir));
872 while (frame >= frame_in) {
877 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
878 ext4_warning_inode(dir,
879 "Corrupt directory, running e2fsck is recommended");
883 static void dx_release(struct dx_frame *frames)
885 struct dx_root_info *info;
887 unsigned int indirect_levels;
889 if (frames[0].bh == NULL)
892 info = &((struct dx_root *)frames[0].bh->b_data)->info;
893 /* save local copy, "info" may be freed after brelse() */
894 indirect_levels = info->indirect_levels;
895 for (i = 0; i <= indirect_levels; i++) {
896 if (frames[i].bh == NULL)
898 brelse(frames[i].bh);
904 * This function increments the frame pointer to search the next leaf
905 * block, and reads in the necessary intervening nodes if the search
906 * should be necessary. Whether or not the search is necessary is
907 * controlled by the hash parameter. If the hash value is even, then
908 * the search is only continued if the next block starts with that
909 * hash value. This is used if we are searching for a specific file.
911 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
913 * This function returns 1 if the caller should continue to search,
914 * or 0 if it should not. If there is an error reading one of the
915 * index blocks, it will a negative error code.
917 * If start_hash is non-null, it will be filled in with the starting
918 * hash of the next page.
920 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
921 struct dx_frame *frame,
922 struct dx_frame *frames,
926 struct buffer_head *bh;
932 * Find the next leaf page by incrementing the frame pointer.
933 * If we run out of entries in the interior node, loop around and
934 * increment pointer in the parent node. When we break out of
935 * this loop, num_frames indicates the number of interior
936 * nodes need to be read.
939 if (++(p->at) < p->entries + dx_get_count(p->entries))
948 * If the hash is 1, then continue only if the next page has a
949 * continuation hash of any value. This is used for readdir
950 * handling. Otherwise, check to see if the hash matches the
951 * desired contiuation hash. If it doesn't, return since
952 * there's no point to read in the successive index pages.
954 bhash = dx_get_hash(p->at);
957 if ((hash & 1) == 0) {
958 if ((bhash & ~1) != hash)
962 * If the hash is HASH_NB_ALWAYS, we always go to the next
963 * block so no check is necessary
965 while (num_frames--) {
966 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
972 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
979 * This function fills a red-black tree with information from a
980 * directory block. It returns the number directory entries loaded
981 * into the tree. If there is an error it is returned in err.
983 static int htree_dirblock_to_tree(struct file *dir_file,
984 struct inode *dir, ext4_lblk_t block,
985 struct dx_hash_info *hinfo,
986 __u32 start_hash, __u32 start_minor_hash)
988 struct buffer_head *bh;
989 struct ext4_dir_entry_2 *de, *top;
990 int err = 0, count = 0;
991 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
993 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
994 (unsigned long)block));
995 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
999 de = (struct ext4_dir_entry_2 *) bh->b_data;
1000 top = (struct ext4_dir_entry_2 *) ((char *) de +
1001 dir->i_sb->s_blocksize -
1002 EXT4_DIR_REC_LEN(0));
1003 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1004 /* Check if the directory is encrypted */
1005 if (ext4_encrypted_inode(dir)) {
1006 err = fscrypt_get_encryption_info(dir);
1011 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1019 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1020 if (ext4_check_dir_entry(dir, NULL, de, bh,
1021 bh->b_data, bh->b_size,
1022 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1023 + ((char *)de - bh->b_data))) {
1024 /* silently ignore the rest of the block */
1027 ext4fs_dirhash(de->name, de->name_len, hinfo);
1028 if ((hinfo->hash < start_hash) ||
1029 ((hinfo->hash == start_hash) &&
1030 (hinfo->minor_hash < start_minor_hash)))
1034 if (!ext4_encrypted_inode(dir)) {
1035 tmp_str.name = de->name;
1036 tmp_str.len = de->name_len;
1037 err = ext4_htree_store_dirent(dir_file,
1038 hinfo->hash, hinfo->minor_hash, de,
1041 int save_len = fname_crypto_str.len;
1042 struct fscrypt_str de_name = FSTR_INIT(de->name,
1045 /* Directory is encrypted */
1046 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1047 hinfo->minor_hash, &de_name,
1053 err = ext4_htree_store_dirent(dir_file,
1054 hinfo->hash, hinfo->minor_hash, de,
1056 fname_crypto_str.len = save_len;
1066 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1067 fscrypt_fname_free_buffer(&fname_crypto_str);
1074 * This function fills a red-black tree with information from a
1075 * directory. We start scanning the directory in hash order, starting
1076 * at start_hash and start_minor_hash.
1078 * This function returns the number of entries inserted into the tree,
1079 * or a negative error code.
1081 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1082 __u32 start_minor_hash, __u32 *next_hash)
1084 struct dx_hash_info hinfo;
1085 struct ext4_dir_entry_2 *de;
1086 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1092 struct fscrypt_str tmp_str;
1094 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1095 start_hash, start_minor_hash));
1096 dir = file_inode(dir_file);
1097 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1098 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1099 if (hinfo.hash_version <= DX_HASH_TEA)
1100 hinfo.hash_version +=
1101 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1102 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1103 if (ext4_has_inline_data(dir)) {
1104 int has_inline_data = 1;
1105 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1109 if (has_inline_data) {
1114 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1115 start_hash, start_minor_hash);
1119 hinfo.hash = start_hash;
1120 hinfo.minor_hash = 0;
1121 frame = dx_probe(NULL, dir, &hinfo, frames);
1123 return PTR_ERR(frame);
1125 /* Add '.' and '..' from the htree header */
1126 if (!start_hash && !start_minor_hash) {
1127 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1128 tmp_str.name = de->name;
1129 tmp_str.len = de->name_len;
1130 err = ext4_htree_store_dirent(dir_file, 0, 0,
1136 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1137 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1138 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1139 tmp_str.name = de->name;
1140 tmp_str.len = de->name_len;
1141 err = ext4_htree_store_dirent(dir_file, 2, 0,
1149 if (fatal_signal_pending(current)) {
1154 block = dx_get_block(frame->at);
1155 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1156 start_hash, start_minor_hash);
1163 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1164 frame, frames, &hashval);
1165 *next_hash = hashval;
1171 * Stop if: (a) there are no more entries, or
1172 * (b) we have inserted at least one entry and the
1173 * next hash value is not a continuation
1176 (count && ((hashval & 1) == 0)))
1180 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1181 "next hash: %x\n", count, *next_hash));
1188 static inline int search_dirblock(struct buffer_head *bh,
1190 struct ext4_filename *fname,
1191 unsigned int offset,
1192 struct ext4_dir_entry_2 **res_dir)
1194 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1195 fname, offset, res_dir);
1199 * Directory block splitting, compacting
1203 * Create map of hash values, offsets, and sizes, stored at end of block.
1204 * Returns number of entries mapped.
1206 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1207 unsigned blocksize, struct dx_hash_info *hinfo,
1208 struct dx_map_entry *map_tail)
1211 char *base = (char *) de;
1212 struct dx_hash_info h = *hinfo;
1214 while ((char *) de < base + blocksize) {
1215 if (de->name_len && de->inode) {
1216 ext4fs_dirhash(de->name, de->name_len, &h);
1218 map_tail->hash = h.hash;
1219 map_tail->offs = ((char *) de - base)>>2;
1220 map_tail->size = le16_to_cpu(de->rec_len);
1224 /* XXX: do we need to check rec_len == 0 case? -Chris */
1225 de = ext4_next_entry(de, blocksize);
1230 /* Sort map by hash value */
1231 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1233 struct dx_map_entry *p, *q, *top = map + count - 1;
1235 /* Combsort until bubble sort doesn't suck */
1237 count = count*10/13;
1238 if (count - 9 < 2) /* 9, 10 -> 11 */
1240 for (p = top, q = p - count; q >= map; p--, q--)
1241 if (p->hash < q->hash)
1244 /* Garden variety bubble sort */
1249 if (q[1].hash >= q[0].hash)
1257 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1259 struct dx_entry *entries = frame->entries;
1260 struct dx_entry *old = frame->at, *new = old + 1;
1261 int count = dx_get_count(entries);
1263 assert(count < dx_get_limit(entries));
1264 assert(old < entries + count);
1265 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1266 dx_set_hash(new, hash);
1267 dx_set_block(new, block);
1268 dx_set_count(entries, count + 1);
1272 * Test whether a directory entry matches the filename being searched for.
1274 * Return: %true if the directory entry matches, otherwise %false.
1276 static inline bool ext4_match(const struct ext4_filename *fname,
1277 const struct ext4_dir_entry_2 *de)
1279 struct fscrypt_name f;
1284 f.usr_fname = fname->usr_fname;
1285 f.disk_name = fname->disk_name;
1286 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1287 f.crypto_buf = fname->crypto_buf;
1289 return fscrypt_match_name(&f, de->name, de->name_len);
1293 * Returns 0 if not found, -1 on failure, and 1 on success
1295 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1296 struct inode *dir, struct ext4_filename *fname,
1297 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1299 struct ext4_dir_entry_2 * de;
1303 de = (struct ext4_dir_entry_2 *)search_buf;
1304 dlimit = search_buf + buf_size;
1305 while ((char *) de < dlimit) {
1306 /* this code is executed quadratically often */
1307 /* do minimal checking `by hand' */
1308 if ((char *) de + de->name_len <= dlimit &&
1309 ext4_match(fname, de)) {
1310 /* found a match - just to be sure, do
1312 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1318 /* prevent looping on a bad block */
1319 de_len = ext4_rec_len_from_disk(de->rec_len,
1320 dir->i_sb->s_blocksize);
1324 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1329 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1330 struct ext4_dir_entry *de)
1332 struct super_block *sb = dir->i_sb;
1338 if (de->inode == 0 &&
1339 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1346 * __ext4_find_entry()
1348 * finds an entry in the specified directory with the wanted name. It
1349 * returns the cache buffer in which the entry was found, and the entry
1350 * itself (as a parameter - res_dir). It does NOT read the inode of the
1351 * entry - you'll have to do that yourself if you want to.
1353 * The returned buffer_head has ->b_count elevated. The caller is expected
1354 * to brelse() it when appropriate.
1356 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1357 struct ext4_filename *fname,
1358 struct ext4_dir_entry_2 **res_dir,
1361 struct super_block *sb;
1362 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1363 struct buffer_head *bh, *ret = NULL;
1364 ext4_lblk_t start, block;
1365 const u8 *name = fname->usr_fname->name;
1366 size_t ra_max = 0; /* Number of bh's in the readahead
1368 size_t ra_ptr = 0; /* Current index into readahead
1370 ext4_lblk_t nblocks;
1371 int i, namelen, retval;
1375 namelen = fname->usr_fname->len;
1376 if (namelen > EXT4_NAME_LEN)
1379 if (ext4_has_inline_data(dir)) {
1380 int has_inline_data = 1;
1381 ret = ext4_find_inline_entry(dir, fname, res_dir,
1383 if (has_inline_data) {
1386 goto cleanup_and_exit;
1390 if ((namelen <= 2) && (name[0] == '.') &&
1391 (name[1] == '.' || name[1] == '\0')) {
1393 * "." or ".." will only be in the first block
1394 * NFS may look up ".."; "." should be handled by the VFS
1401 ret = ext4_dx_find_entry(dir, fname, res_dir);
1403 * On success, or if the error was file not found,
1404 * return. Otherwise, fall back to doing a search the
1405 * old fashioned way.
1407 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1408 goto cleanup_and_exit;
1409 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1413 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1416 goto cleanup_and_exit;
1418 start = EXT4_I(dir)->i_dir_start_lookup;
1419 if (start >= nblocks)
1425 * We deal with the read-ahead logic here.
1428 if (ra_ptr >= ra_max) {
1429 /* Refill the readahead buffer */
1432 ra_max = start - block;
1434 ra_max = nblocks - block;
1435 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1436 retval = ext4_bread_batch(dir, block, ra_max,
1437 false /* wait */, bh_use);
1439 ret = ERR_PTR(retval);
1441 goto cleanup_and_exit;
1444 if ((bh = bh_use[ra_ptr++]) == NULL)
1447 if (!buffer_uptodate(bh)) {
1448 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1449 (unsigned long) block);
1451 ret = ERR_PTR(-EIO);
1452 goto cleanup_and_exit;
1454 if (!buffer_verified(bh) &&
1455 !is_dx_internal_node(dir, block,
1456 (struct ext4_dir_entry *)bh->b_data) &&
1457 !ext4_dirent_csum_verify(dir,
1458 (struct ext4_dir_entry *)bh->b_data)) {
1459 EXT4_ERROR_INODE(dir, "checksumming directory "
1460 "block %lu", (unsigned long)block);
1462 ret = ERR_PTR(-EFSBADCRC);
1463 goto cleanup_and_exit;
1465 set_buffer_verified(bh);
1466 i = search_dirblock(bh, dir, fname,
1467 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1469 EXT4_I(dir)->i_dir_start_lookup = block;
1471 goto cleanup_and_exit;
1475 goto cleanup_and_exit;
1478 if (++block >= nblocks)
1480 } while (block != start);
1483 * If the directory has grown while we were searching, then
1484 * search the last part of the directory before giving up.
1487 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1488 if (block < nblocks) {
1494 /* Clean up the read-ahead blocks */
1495 for (; ra_ptr < ra_max; ra_ptr++)
1496 brelse(bh_use[ra_ptr]);
1500 static struct buffer_head *ext4_find_entry(struct inode *dir,
1501 const struct qstr *d_name,
1502 struct ext4_dir_entry_2 **res_dir,
1506 struct ext4_filename fname;
1507 struct buffer_head *bh;
1509 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1513 return ERR_PTR(err);
1515 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1517 ext4_fname_free_filename(&fname);
1521 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1522 struct dentry *dentry,
1523 struct ext4_dir_entry_2 **res_dir)
1526 struct ext4_filename fname;
1527 struct buffer_head *bh;
1529 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1533 return ERR_PTR(err);
1535 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1537 ext4_fname_free_filename(&fname);
1541 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1542 struct ext4_filename *fname,
1543 struct ext4_dir_entry_2 **res_dir)
1545 struct super_block * sb = dir->i_sb;
1546 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1547 struct buffer_head *bh;
1551 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1554 frame = dx_probe(fname, dir, NULL, frames);
1556 return (struct buffer_head *) frame;
1558 block = dx_get_block(frame->at);
1559 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1563 retval = search_dirblock(bh, dir, fname,
1564 block << EXT4_BLOCK_SIZE_BITS(sb),
1570 bh = ERR_PTR(ERR_BAD_DX_DIR);
1574 /* Check to see if we should continue to search */
1575 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1578 ext4_warning_inode(dir,
1579 "error %d reading directory index block",
1581 bh = ERR_PTR(retval);
1584 } while (retval == 1);
1588 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1594 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1596 struct inode *inode;
1597 struct ext4_dir_entry_2 *de;
1598 struct buffer_head *bh;
1600 if (dentry->d_name.len > EXT4_NAME_LEN)
1601 return ERR_PTR(-ENAMETOOLONG);
1603 bh = ext4_lookup_entry(dir, dentry, &de);
1605 return (struct dentry *) bh;
1608 __u32 ino = le32_to_cpu(de->inode);
1610 if (!ext4_valid_inum(dir->i_sb, ino)) {
1611 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1612 return ERR_PTR(-EFSCORRUPTED);
1614 if (unlikely(ino == dir->i_ino)) {
1615 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1617 return ERR_PTR(-EFSCORRUPTED);
1619 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1620 if (inode == ERR_PTR(-ESTALE)) {
1621 EXT4_ERROR_INODE(dir,
1622 "deleted inode referenced: %u",
1624 return ERR_PTR(-EFSCORRUPTED);
1626 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1627 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1628 !fscrypt_has_permitted_context(dir, inode)) {
1629 ext4_warning(inode->i_sb,
1630 "Inconsistent encryption contexts: %lu/%lu",
1631 dir->i_ino, inode->i_ino);
1633 return ERR_PTR(-EPERM);
1636 return d_splice_alias(inode, dentry);
1640 struct dentry *ext4_get_parent(struct dentry *child)
1643 static const struct qstr dotdot = QSTR_INIT("..", 2);
1644 struct ext4_dir_entry_2 * de;
1645 struct buffer_head *bh;
1647 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1649 return (struct dentry *) bh;
1651 return ERR_PTR(-ENOENT);
1652 ino = le32_to_cpu(de->inode);
1655 if (!ext4_valid_inum(child->d_sb, ino)) {
1656 EXT4_ERROR_INODE(d_inode(child),
1657 "bad parent inode number: %u", ino);
1658 return ERR_PTR(-EFSCORRUPTED);
1661 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1665 * Move count entries from end of map between two memory locations.
1666 * Returns pointer to last entry moved.
1668 static struct ext4_dir_entry_2 *
1669 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1672 unsigned rec_len = 0;
1675 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1676 (from + (map->offs<<2));
1677 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1678 memcpy (to, de, rec_len);
1679 ((struct ext4_dir_entry_2 *) to)->rec_len =
1680 ext4_rec_len_to_disk(rec_len, blocksize);
1685 return (struct ext4_dir_entry_2 *) (to - rec_len);
1689 * Compact each dir entry in the range to the minimal rec_len.
1690 * Returns pointer to last entry in range.
1692 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1694 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1695 unsigned rec_len = 0;
1698 while ((char*)de < base + blocksize) {
1699 next = ext4_next_entry(de, blocksize);
1700 if (de->inode && de->name_len) {
1701 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1703 memmove(to, de, rec_len);
1704 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1706 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1714 * Split a full leaf block to make room for a new dir entry.
1715 * Allocate a new block, and move entries so that they are approx. equally full.
1716 * Returns pointer to de in block into which the new entry will be inserted.
1718 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1719 struct buffer_head **bh,struct dx_frame *frame,
1720 struct dx_hash_info *hinfo)
1722 unsigned blocksize = dir->i_sb->s_blocksize;
1723 unsigned count, continued;
1724 struct buffer_head *bh2;
1725 ext4_lblk_t newblock;
1727 struct dx_map_entry *map;
1728 char *data1 = (*bh)->b_data, *data2;
1729 unsigned split, move, size;
1730 struct ext4_dir_entry_2 *de = NULL, *de2;
1731 struct ext4_dir_entry_tail *t;
1735 if (ext4_has_metadata_csum(dir->i_sb))
1736 csum_size = sizeof(struct ext4_dir_entry_tail);
1738 bh2 = ext4_append(handle, dir, &newblock);
1742 return (struct ext4_dir_entry_2 *) bh2;
1745 BUFFER_TRACE(*bh, "get_write_access");
1746 err = ext4_journal_get_write_access(handle, *bh);
1750 BUFFER_TRACE(frame->bh, "get_write_access");
1751 err = ext4_journal_get_write_access(handle, frame->bh);
1755 data2 = bh2->b_data;
1757 /* create map in the end of data2 block */
1758 map = (struct dx_map_entry *) (data2 + blocksize);
1759 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1760 blocksize, hinfo, map);
1762 dx_sort_map(map, count);
1763 /* Ensure that neither split block is over half full */
1766 for (i = count-1; i >= 0; i--) {
1767 /* is more than half of this entry in 2nd half of the block? */
1768 if (size + map[i].size/2 > blocksize/2)
1770 size += map[i].size;
1774 * map index at which we will split
1776 * If the sum of active entries didn't exceed half the block size, just
1777 * split it in half by count; each resulting block will have at least
1778 * half the space free.
1781 split = count - move;
1785 hash2 = map[split].hash;
1786 continued = hash2 == map[split - 1].hash;
1787 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1788 (unsigned long)dx_get_block(frame->at),
1789 hash2, split, count-split));
1791 /* Fancy dance to stay within two buffers */
1792 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1794 de = dx_pack_dirents(data1, blocksize);
1795 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1798 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1802 t = EXT4_DIRENT_TAIL(data2, blocksize);
1803 initialize_dirent_tail(t, blocksize);
1805 t = EXT4_DIRENT_TAIL(data1, blocksize);
1806 initialize_dirent_tail(t, blocksize);
1809 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1811 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1814 /* Which block gets the new entry? */
1815 if (hinfo->hash >= hash2) {
1819 dx_insert_block(frame, hash2 + continued, newblock);
1820 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1823 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1827 dxtrace(dx_show_index("frame", frame->entries));
1834 ext4_std_error(dir->i_sb, err);
1835 return ERR_PTR(err);
1838 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1839 struct buffer_head *bh,
1840 void *buf, int buf_size,
1841 struct ext4_filename *fname,
1842 struct ext4_dir_entry_2 **dest_de)
1844 struct ext4_dir_entry_2 *de;
1845 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1847 unsigned int offset = 0;
1850 de = (struct ext4_dir_entry_2 *)buf;
1851 top = buf + buf_size - reclen;
1852 while ((char *) de <= top) {
1853 if (ext4_check_dir_entry(dir, NULL, de, bh,
1854 buf, buf_size, offset))
1855 return -EFSCORRUPTED;
1856 if (ext4_match(fname, de))
1858 nlen = EXT4_DIR_REC_LEN(de->name_len);
1859 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1860 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1862 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1865 if ((char *) de > top)
1872 void ext4_insert_dentry(struct inode *inode,
1873 struct ext4_dir_entry_2 *de,
1875 struct ext4_filename *fname)
1880 nlen = EXT4_DIR_REC_LEN(de->name_len);
1881 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1883 struct ext4_dir_entry_2 *de1 =
1884 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1885 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1886 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1889 de->file_type = EXT4_FT_UNKNOWN;
1890 de->inode = cpu_to_le32(inode->i_ino);
1891 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1892 de->name_len = fname_len(fname);
1893 memcpy(de->name, fname_name(fname), fname_len(fname));
1897 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1898 * it points to a directory entry which is guaranteed to be large
1899 * enough for new directory entry. If de is NULL, then
1900 * add_dirent_to_buf will attempt search the directory block for
1901 * space. It will return -ENOSPC if no space is available, and -EIO
1902 * and -EEXIST if directory entry already exists.
1904 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1906 struct inode *inode, struct ext4_dir_entry_2 *de,
1907 struct buffer_head *bh)
1909 unsigned int blocksize = dir->i_sb->s_blocksize;
1913 if (ext4_has_metadata_csum(inode->i_sb))
1914 csum_size = sizeof(struct ext4_dir_entry_tail);
1917 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1918 blocksize - csum_size, fname, &de);
1922 BUFFER_TRACE(bh, "get_write_access");
1923 err = ext4_journal_get_write_access(handle, bh);
1925 ext4_std_error(dir->i_sb, err);
1929 /* By now the buffer is marked for journaling */
1930 ext4_insert_dentry(inode, de, blocksize, fname);
1933 * XXX shouldn't update any times until successful
1934 * completion of syscall, but too many callers depend
1937 * XXX similarly, too many callers depend on
1938 * ext4_new_inode() setting the times, but error
1939 * recovery deletes the inode, so the worst that can
1940 * happen is that the times are slightly out of date
1941 * and/or different from the directory change time.
1943 dir->i_mtime = dir->i_ctime = current_time(dir);
1944 ext4_update_dx_flag(dir);
1945 inode_inc_iversion(dir);
1946 ext4_mark_inode_dirty(handle, dir);
1947 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1948 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1950 ext4_std_error(dir->i_sb, err);
1955 * This converts a one block unindexed directory to a 3 block indexed
1956 * directory, and adds the dentry to the indexed directory.
1958 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1960 struct inode *inode, struct buffer_head *bh)
1962 struct buffer_head *bh2;
1963 struct dx_root *root;
1964 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1965 struct dx_entry *entries;
1966 struct ext4_dir_entry_2 *de, *de2;
1967 struct ext4_dir_entry_tail *t;
1973 struct fake_dirent *fde;
1976 if (ext4_has_metadata_csum(inode->i_sb))
1977 csum_size = sizeof(struct ext4_dir_entry_tail);
1979 blocksize = dir->i_sb->s_blocksize;
1980 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1981 BUFFER_TRACE(bh, "get_write_access");
1982 retval = ext4_journal_get_write_access(handle, bh);
1984 ext4_std_error(dir->i_sb, retval);
1988 root = (struct dx_root *) bh->b_data;
1990 /* The 0th block becomes the root, move the dirents out */
1991 fde = &root->dotdot;
1992 de = (struct ext4_dir_entry_2 *)((char *)fde +
1993 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1994 if ((char *) de >= (((char *) root) + blocksize)) {
1995 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1997 return -EFSCORRUPTED;
1999 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2001 /* Allocate new block for the 0th block's dirents */
2002 bh2 = ext4_append(handle, dir, &block);
2005 return PTR_ERR(bh2);
2007 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2008 data1 = bh2->b_data;
2010 memcpy (data1, de, len);
2011 de = (struct ext4_dir_entry_2 *) data1;
2013 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2015 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2020 t = EXT4_DIRENT_TAIL(data1, blocksize);
2021 initialize_dirent_tail(t, blocksize);
2024 /* Initialize the root; the dot dirents already exist */
2025 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2026 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2028 memset (&root->info, 0, sizeof(root->info));
2029 root->info.info_length = sizeof(root->info);
2030 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2031 entries = root->entries;
2032 dx_set_block(entries, 1);
2033 dx_set_count(entries, 1);
2034 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2036 /* Initialize as for dx_probe */
2037 fname->hinfo.hash_version = root->info.hash_version;
2038 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2039 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2040 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2041 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2043 memset(frames, 0, sizeof(frames));
2045 frame->entries = entries;
2046 frame->at = entries;
2049 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2052 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2056 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2058 retval = PTR_ERR(de);
2062 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2065 * Even if the block split failed, we have to properly write
2066 * out all the changes we did so far. Otherwise we can end up
2067 * with corrupted filesystem.
2070 ext4_mark_inode_dirty(handle, dir);
2079 * adds a file entry to the specified directory, using the same
2080 * semantics as ext4_find_entry(). It returns NULL if it failed.
2082 * NOTE!! The inode part of 'de' is left at 0 - which means you
2083 * may not sleep between calling this and putting something into
2084 * the entry, as someone else might have used it while you slept.
2086 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2087 struct inode *inode)
2089 struct inode *dir = d_inode(dentry->d_parent);
2090 struct buffer_head *bh = NULL;
2091 struct ext4_dir_entry_2 *de;
2092 struct ext4_dir_entry_tail *t;
2093 struct super_block *sb;
2094 struct ext4_filename fname;
2098 ext4_lblk_t block, blocks;
2101 if (ext4_has_metadata_csum(inode->i_sb))
2102 csum_size = sizeof(struct ext4_dir_entry_tail);
2105 blocksize = sb->s_blocksize;
2106 if (!dentry->d_name.len)
2109 if (fscrypt_is_nokey_name(dentry))
2112 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2116 if (ext4_has_inline_data(dir)) {
2117 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2127 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2128 if (!retval || (retval != ERR_BAD_DX_DIR))
2130 /* Can we just ignore htree data? */
2131 if (ext4_has_metadata_csum(sb)) {
2132 EXT4_ERROR_INODE(dir,
2133 "Directory has corrupted htree index.");
2134 retval = -EFSCORRUPTED;
2137 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2139 ext4_mark_inode_dirty(handle, dir);
2141 blocks = dir->i_size >> sb->s_blocksize_bits;
2142 for (block = 0; block < blocks; block++) {
2143 bh = ext4_read_dirblock(dir, block, DIRENT);
2145 bh = ext4_bread(handle, dir, block,
2146 EXT4_GET_BLOCKS_CREATE);
2147 goto add_to_new_block;
2150 retval = PTR_ERR(bh);
2154 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2156 if (retval != -ENOSPC)
2159 if (blocks == 1 && !dx_fallback &&
2160 ext4_has_feature_dir_index(sb)) {
2161 retval = make_indexed_dir(handle, &fname, dir,
2163 bh = NULL; /* make_indexed_dir releases bh */
2168 bh = ext4_append(handle, dir, &block);
2171 retval = PTR_ERR(bh);
2175 de = (struct ext4_dir_entry_2 *) bh->b_data;
2177 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2180 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2181 initialize_dirent_tail(t, blocksize);
2184 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2186 ext4_fname_free_filename(&fname);
2189 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2194 * Returns 0 for success, or a negative error value
2196 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2197 struct inode *dir, struct inode *inode)
2199 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2200 struct dx_entry *entries, *at;
2201 struct buffer_head *bh;
2202 struct super_block *sb = dir->i_sb;
2203 struct ext4_dir_entry_2 *de;
2209 frame = dx_probe(fname, dir, NULL, frames);
2211 return PTR_ERR(frame);
2212 entries = frame->entries;
2214 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2221 BUFFER_TRACE(bh, "get_write_access");
2222 err = ext4_journal_get_write_access(handle, bh);
2226 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2231 /* Block full, should compress but for now just split */
2232 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2233 dx_get_count(entries), dx_get_limit(entries)));
2234 /* Need to split index? */
2235 if (dx_get_count(entries) == dx_get_limit(entries)) {
2236 ext4_lblk_t newblock;
2237 int levels = frame - frames + 1;
2238 unsigned int icount;
2240 struct dx_entry *entries2;
2241 struct dx_node *node2;
2242 struct buffer_head *bh2;
2244 while (frame > frames) {
2245 if (dx_get_count((frame - 1)->entries) <
2246 dx_get_limit((frame - 1)->entries)) {
2250 frame--; /* split higher index block */
2252 entries = frame->entries;
2255 if (add_level && levels == ext4_dir_htree_level(sb)) {
2256 ext4_warning(sb, "Directory (ino: %lu) index full, "
2257 "reach max htree level :%d",
2258 dir->i_ino, levels);
2259 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2260 ext4_warning(sb, "Large directory feature is "
2261 "not enabled on this "
2267 icount = dx_get_count(entries);
2268 bh2 = ext4_append(handle, dir, &newblock);
2273 node2 = (struct dx_node *)(bh2->b_data);
2274 entries2 = node2->entries;
2275 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2276 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2278 BUFFER_TRACE(frame->bh, "get_write_access");
2279 err = ext4_journal_get_write_access(handle, frame->bh);
2283 unsigned icount1 = icount/2, icount2 = icount - icount1;
2284 unsigned hash2 = dx_get_hash(entries + icount1);
2285 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2288 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2289 err = ext4_journal_get_write_access(handle,
2294 memcpy((char *) entries2, (char *) (entries + icount1),
2295 icount2 * sizeof(struct dx_entry));
2296 dx_set_count(entries, icount1);
2297 dx_set_count(entries2, icount2);
2298 dx_set_limit(entries2, dx_node_limit(dir));
2300 /* Which index block gets the new entry? */
2301 if (at - entries >= icount1) {
2302 frame->at = at = at - entries - icount1 + entries2;
2303 frame->entries = entries = entries2;
2304 swap(frame->bh, bh2);
2306 dx_insert_block((frame - 1), hash2, newblock);
2307 dxtrace(dx_show_index("node", frame->entries));
2308 dxtrace(dx_show_index("node",
2309 ((struct dx_node *) bh2->b_data)->entries));
2310 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2314 err = ext4_handle_dirty_dx_node(handle, dir,
2318 err = ext4_handle_dirty_dx_node(handle, dir,
2323 struct dx_root *dxroot;
2324 memcpy((char *) entries2, (char *) entries,
2325 icount * sizeof(struct dx_entry));
2326 dx_set_limit(entries2, dx_node_limit(dir));
2329 dx_set_count(entries, 1);
2330 dx_set_block(entries + 0, newblock);
2331 dxroot = (struct dx_root *)frames[0].bh->b_data;
2332 dxroot->info.indirect_levels += 1;
2333 dxtrace(printk(KERN_DEBUG
2334 "Creating %d level index...\n",
2335 dxroot->info.indirect_levels));
2336 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2339 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2345 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2350 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2354 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2358 /* @restart is true means htree-path has been changed, we need to
2359 * repeat dx_probe() to find out valid htree-path
2361 if (restart && err == 0)
2367 * ext4_generic_delete_entry deletes a directory entry by merging it
2368 * with the previous entry
2370 int ext4_generic_delete_entry(handle_t *handle,
2372 struct ext4_dir_entry_2 *de_del,
2373 struct buffer_head *bh,
2378 struct ext4_dir_entry_2 *de, *pde;
2379 unsigned int blocksize = dir->i_sb->s_blocksize;
2384 de = (struct ext4_dir_entry_2 *)entry_buf;
2385 while (i < buf_size - csum_size) {
2386 if (ext4_check_dir_entry(dir, NULL, de, bh,
2387 entry_buf, buf_size, i))
2388 return -EFSCORRUPTED;
2391 pde->rec_len = ext4_rec_len_to_disk(
2392 ext4_rec_len_from_disk(pde->rec_len,
2394 ext4_rec_len_from_disk(de->rec_len,
2399 inode_inc_iversion(dir);
2402 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2404 de = ext4_next_entry(de, blocksize);
2409 static int ext4_delete_entry(handle_t *handle,
2411 struct ext4_dir_entry_2 *de_del,
2412 struct buffer_head *bh)
2414 int err, csum_size = 0;
2416 if (ext4_has_inline_data(dir)) {
2417 int has_inline_data = 1;
2418 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2420 if (has_inline_data)
2424 if (ext4_has_metadata_csum(dir->i_sb))
2425 csum_size = sizeof(struct ext4_dir_entry_tail);
2427 BUFFER_TRACE(bh, "get_write_access");
2428 err = ext4_journal_get_write_access(handle, bh);
2432 err = ext4_generic_delete_entry(handle, dir, de_del,
2434 dir->i_sb->s_blocksize, csum_size);
2438 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2439 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2446 ext4_std_error(dir->i_sb, err);
2451 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2452 * since this indicates that nlinks count was previously 1 to avoid overflowing
2453 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2454 * that subdirectory link counts are not being maintained accurately.
2456 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2457 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2458 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2459 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2461 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2465 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2466 set_nlink(inode, 1);
2470 * If a directory had nlink == 1, then we should let it be 1. This indicates
2471 * directory has >EXT4_LINK_MAX subdirs.
2473 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2475 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2480 static int ext4_add_nondir(handle_t *handle,
2481 struct dentry *dentry, struct inode *inode)
2483 int err = ext4_add_entry(handle, dentry, inode);
2485 ext4_mark_inode_dirty(handle, inode);
2486 d_instantiate_new(dentry, inode);
2490 unlock_new_inode(inode);
2496 * By the time this is called, we already have created
2497 * the directory cache entry for the new file, but it
2498 * is so far negative - it has no inode.
2500 * If the create succeeds, we fill in the inode information
2501 * with d_instantiate().
2503 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2507 struct inode *inode;
2508 int err, credits, retries = 0;
2510 err = dquot_initialize(dir);
2514 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2515 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2517 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2518 NULL, EXT4_HT_DIR, credits);
2519 handle = ext4_journal_current_handle();
2520 err = PTR_ERR(inode);
2521 if (!IS_ERR(inode)) {
2522 inode->i_op = &ext4_file_inode_operations;
2523 inode->i_fop = &ext4_file_operations;
2524 ext4_set_aops(inode);
2525 err = ext4_add_nondir(handle, dentry, inode);
2526 if (!err && IS_DIRSYNC(dir))
2527 ext4_handle_sync(handle);
2530 ext4_journal_stop(handle);
2531 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2536 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2537 umode_t mode, dev_t rdev)
2540 struct inode *inode;
2541 int err, credits, retries = 0;
2543 err = dquot_initialize(dir);
2547 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2548 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2550 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2551 NULL, EXT4_HT_DIR, credits);
2552 handle = ext4_journal_current_handle();
2553 err = PTR_ERR(inode);
2554 if (!IS_ERR(inode)) {
2555 init_special_inode(inode, inode->i_mode, rdev);
2556 inode->i_op = &ext4_special_inode_operations;
2557 err = ext4_add_nondir(handle, dentry, inode);
2558 if (!err && IS_DIRSYNC(dir))
2559 ext4_handle_sync(handle);
2562 ext4_journal_stop(handle);
2563 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2568 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2571 struct inode *inode;
2572 int err, retries = 0;
2574 err = dquot_initialize(dir);
2579 inode = ext4_new_inode_start_handle(dir, mode,
2582 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2583 4 + EXT4_XATTR_TRANS_BLOCKS);
2584 handle = ext4_journal_current_handle();
2585 err = PTR_ERR(inode);
2586 if (!IS_ERR(inode)) {
2587 inode->i_op = &ext4_file_inode_operations;
2588 inode->i_fop = &ext4_file_operations;
2589 ext4_set_aops(inode);
2590 d_tmpfile(dentry, inode);
2591 err = ext4_orphan_add(handle, inode);
2593 goto err_unlock_inode;
2594 mark_inode_dirty(inode);
2595 unlock_new_inode(inode);
2598 ext4_journal_stop(handle);
2599 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2603 ext4_journal_stop(handle);
2604 unlock_new_inode(inode);
2608 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2609 struct ext4_dir_entry_2 *de,
2610 int blocksize, int csum_size,
2611 unsigned int parent_ino, int dotdot_real_len)
2613 de->inode = cpu_to_le32(inode->i_ino);
2615 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2617 strcpy(de->name, ".");
2618 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2620 de = ext4_next_entry(de, blocksize);
2621 de->inode = cpu_to_le32(parent_ino);
2623 if (!dotdot_real_len)
2624 de->rec_len = ext4_rec_len_to_disk(blocksize -
2625 (csum_size + EXT4_DIR_REC_LEN(1)),
2628 de->rec_len = ext4_rec_len_to_disk(
2629 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2630 strcpy(de->name, "..");
2631 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2633 return ext4_next_entry(de, blocksize);
2636 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2637 struct inode *inode)
2639 struct buffer_head *dir_block = NULL;
2640 struct ext4_dir_entry_2 *de;
2641 struct ext4_dir_entry_tail *t;
2642 ext4_lblk_t block = 0;
2643 unsigned int blocksize = dir->i_sb->s_blocksize;
2647 if (ext4_has_metadata_csum(dir->i_sb))
2648 csum_size = sizeof(struct ext4_dir_entry_tail);
2650 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2651 err = ext4_try_create_inline_dir(handle, dir, inode);
2652 if (err < 0 && err != -ENOSPC)
2659 dir_block = ext4_append(handle, inode, &block);
2660 if (IS_ERR(dir_block))
2661 return PTR_ERR(dir_block);
2662 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2663 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2664 set_nlink(inode, 2);
2666 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2667 initialize_dirent_tail(t, blocksize);
2670 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2671 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2674 set_buffer_verified(dir_block);
2680 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2683 struct inode *inode;
2684 int err, credits, retries = 0;
2686 if (EXT4_DIR_LINK_MAX(dir))
2689 err = dquot_initialize(dir);
2693 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2694 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2696 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2698 0, NULL, EXT4_HT_DIR, credits);
2699 handle = ext4_journal_current_handle();
2700 err = PTR_ERR(inode);
2704 inode->i_op = &ext4_dir_inode_operations;
2705 inode->i_fop = &ext4_dir_operations;
2706 err = ext4_init_new_dir(handle, dir, inode);
2708 goto out_clear_inode;
2709 err = ext4_mark_inode_dirty(handle, inode);
2711 err = ext4_add_entry(handle, dentry, inode);
2715 unlock_new_inode(inode);
2716 ext4_mark_inode_dirty(handle, inode);
2720 ext4_inc_count(handle, dir);
2721 ext4_update_dx_flag(dir);
2722 err = ext4_mark_inode_dirty(handle, dir);
2724 goto out_clear_inode;
2725 d_instantiate_new(dentry, inode);
2726 if (IS_DIRSYNC(dir))
2727 ext4_handle_sync(handle);
2731 ext4_journal_stop(handle);
2732 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2738 * routine to check that the specified directory is empty (for rmdir)
2740 bool ext4_empty_dir(struct inode *inode)
2742 unsigned int offset;
2743 struct buffer_head *bh;
2744 struct ext4_dir_entry_2 *de;
2745 struct super_block *sb;
2747 if (ext4_has_inline_data(inode)) {
2748 int has_inline_data = 1;
2751 ret = empty_inline_dir(inode, &has_inline_data);
2752 if (has_inline_data)
2757 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2758 EXT4_ERROR_INODE(inode, "invalid size");
2761 /* The first directory block must not be a hole,
2762 * so treat it as DIRENT_HTREE
2764 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2768 de = (struct ext4_dir_entry_2 *) bh->b_data;
2769 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2771 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2772 ext4_warning_inode(inode, "directory missing '.'");
2776 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2777 de = ext4_next_entry(de, sb->s_blocksize);
2778 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2780 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2781 ext4_warning_inode(inode, "directory missing '..'");
2785 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2786 while (offset < inode->i_size) {
2787 if (!(offset & (sb->s_blocksize - 1))) {
2788 unsigned int lblock;
2790 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2791 bh = ext4_read_dirblock(inode, lblock, EITHER);
2793 offset += sb->s_blocksize;
2799 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2800 (offset & (sb->s_blocksize - 1)));
2801 if (ext4_check_dir_entry(inode, NULL, de, bh,
2802 bh->b_data, bh->b_size, offset)) {
2803 offset = (offset | (sb->s_blocksize - 1)) + 1;
2806 if (le32_to_cpu(de->inode)) {
2810 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2817 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2818 * such inodes, starting at the superblock, in case we crash before the
2819 * file is closed/deleted, or in case the inode truncate spans multiple
2820 * transactions and the last transaction is not recovered after a crash.
2822 * At filesystem recovery time, we walk this list deleting unlinked
2823 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2825 * Orphan list manipulation functions must be called under i_mutex unless
2826 * we are just creating the inode or deleting it.
2828 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2830 struct super_block *sb = inode->i_sb;
2831 struct ext4_sb_info *sbi = EXT4_SB(sb);
2832 struct ext4_iloc iloc;
2836 if (!sbi->s_journal || is_bad_inode(inode))
2839 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2840 !inode_is_locked(inode));
2842 * Exit early if inode already is on orphan list. This is a big speedup
2843 * since we don't have to contend on the global s_orphan_lock.
2845 if (!list_empty(&EXT4_I(inode)->i_orphan))
2849 * Orphan handling is only valid for files with data blocks
2850 * being truncated, or files being unlinked. Note that we either
2851 * hold i_mutex, or the inode can not be referenced from outside,
2852 * so i_nlink should not be bumped due to race
2854 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2855 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2857 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2858 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2862 err = ext4_reserve_inode_write(handle, inode, &iloc);
2866 mutex_lock(&sbi->s_orphan_lock);
2868 * Due to previous errors inode may be already a part of on-disk
2869 * orphan list. If so skip on-disk list modification.
2871 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2872 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2873 /* Insert this inode at the head of the on-disk orphan list */
2874 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2875 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2878 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2879 mutex_unlock(&sbi->s_orphan_lock);
2882 err = ext4_handle_dirty_super(handle, sb);
2883 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2888 * We have to remove inode from in-memory list if
2889 * addition to on disk orphan list failed. Stray orphan
2890 * list entries can cause panics at unmount time.
2892 mutex_lock(&sbi->s_orphan_lock);
2893 list_del_init(&EXT4_I(inode)->i_orphan);
2894 mutex_unlock(&sbi->s_orphan_lock);
2899 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2900 jbd_debug(4, "orphan inode %lu will point to %d\n",
2901 inode->i_ino, NEXT_ORPHAN(inode));
2903 ext4_std_error(sb, err);
2908 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2909 * of such inodes stored on disk, because it is finally being cleaned up.
2911 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2913 struct list_head *prev;
2914 struct ext4_inode_info *ei = EXT4_I(inode);
2915 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2917 struct ext4_iloc iloc;
2920 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2923 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2924 !inode_is_locked(inode));
2925 /* Do this quick check before taking global s_orphan_lock. */
2926 if (list_empty(&ei->i_orphan))
2930 /* Grab inode buffer early before taking global s_orphan_lock */
2931 err = ext4_reserve_inode_write(handle, inode, &iloc);
2934 mutex_lock(&sbi->s_orphan_lock);
2935 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2937 prev = ei->i_orphan.prev;
2938 list_del_init(&ei->i_orphan);
2940 /* If we're on an error path, we may not have a valid
2941 * transaction handle with which to update the orphan list on
2942 * disk, but we still need to remove the inode from the linked
2943 * list in memory. */
2944 if (!handle || err) {
2945 mutex_unlock(&sbi->s_orphan_lock);
2949 ino_next = NEXT_ORPHAN(inode);
2950 if (prev == &sbi->s_orphan) {
2951 jbd_debug(4, "superblock will point to %u\n", ino_next);
2952 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2953 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2955 mutex_unlock(&sbi->s_orphan_lock);
2958 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2959 mutex_unlock(&sbi->s_orphan_lock);
2960 err = ext4_handle_dirty_super(handle, inode->i_sb);
2962 struct ext4_iloc iloc2;
2963 struct inode *i_prev =
2964 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2966 jbd_debug(4, "orphan inode %lu will point to %u\n",
2967 i_prev->i_ino, ino_next);
2968 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2970 mutex_unlock(&sbi->s_orphan_lock);
2973 NEXT_ORPHAN(i_prev) = ino_next;
2974 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2975 mutex_unlock(&sbi->s_orphan_lock);
2979 NEXT_ORPHAN(inode) = 0;
2980 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2982 ext4_std_error(inode->i_sb, err);
2990 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2993 struct inode *inode;
2994 struct buffer_head *bh;
2995 struct ext4_dir_entry_2 *de;
2996 handle_t *handle = NULL;
2998 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3001 /* Initialize quotas before so that eventual writes go in
3002 * separate transaction */
3003 retval = dquot_initialize(dir);
3006 retval = dquot_initialize(d_inode(dentry));
3011 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3017 inode = d_inode(dentry);
3019 retval = -EFSCORRUPTED;
3020 if (le32_to_cpu(de->inode) != inode->i_ino)
3023 retval = -ENOTEMPTY;
3024 if (!ext4_empty_dir(inode))
3027 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3028 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3029 if (IS_ERR(handle)) {
3030 retval = PTR_ERR(handle);
3035 if (IS_DIRSYNC(dir))
3036 ext4_handle_sync(handle);
3038 retval = ext4_delete_entry(handle, dir, de, bh);
3041 if (!EXT4_DIR_LINK_EMPTY(inode))
3042 ext4_warning_inode(inode,
3043 "empty directory '%.*s' has too many links (%u)",
3044 dentry->d_name.len, dentry->d_name.name,
3046 inode_inc_iversion(inode);
3048 /* There's no need to set i_disksize: the fact that i_nlink is
3049 * zero will ensure that the right thing happens during any
3052 ext4_orphan_add(handle, inode);
3053 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3054 ext4_mark_inode_dirty(handle, inode);
3055 ext4_dec_count(handle, dir);
3056 ext4_update_dx_flag(dir);
3057 ext4_mark_inode_dirty(handle, dir);
3062 ext4_journal_stop(handle);
3066 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3069 struct inode *inode;
3070 struct buffer_head *bh;
3071 struct ext4_dir_entry_2 *de;
3072 handle_t *handle = NULL;
3074 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3077 trace_ext4_unlink_enter(dir, dentry);
3078 /* Initialize quotas before so that eventual writes go
3079 * in separate transaction */
3080 retval = dquot_initialize(dir);
3083 retval = dquot_initialize(d_inode(dentry));
3088 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3094 inode = d_inode(dentry);
3096 retval = -EFSCORRUPTED;
3097 if (le32_to_cpu(de->inode) != inode->i_ino)
3100 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3101 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3102 if (IS_ERR(handle)) {
3103 retval = PTR_ERR(handle);
3108 if (IS_DIRSYNC(dir))
3109 ext4_handle_sync(handle);
3111 retval = ext4_delete_entry(handle, dir, de, bh);
3114 dir->i_ctime = dir->i_mtime = current_time(dir);
3115 ext4_update_dx_flag(dir);
3116 ext4_mark_inode_dirty(handle, dir);
3117 if (inode->i_nlink == 0)
3118 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3119 dentry->d_name.len, dentry->d_name.name);
3122 if (!inode->i_nlink)
3123 ext4_orphan_add(handle, inode);
3124 inode->i_ctime = current_time(inode);
3125 ext4_mark_inode_dirty(handle, inode);
3130 ext4_journal_stop(handle);
3131 trace_ext4_unlink_exit(dentry, retval);
3135 static int ext4_symlink(struct inode *dir,
3136 struct dentry *dentry, const char *symname)
3139 struct inode *inode;
3140 int err, len = strlen(symname);
3142 struct fscrypt_str disk_link;
3144 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3147 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3152 err = dquot_initialize(dir);
3156 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3158 * For non-fast symlinks, we just allocate inode and put it on
3159 * orphan list in the first transaction => we need bitmap,
3160 * group descriptor, sb, inode block, quota blocks, and
3161 * possibly selinux xattr blocks.
3163 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3164 EXT4_XATTR_TRANS_BLOCKS;
3167 * Fast symlink. We have to add entry to directory
3168 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3169 * allocate new inode (bitmap, group descriptor, inode block,
3170 * quota blocks, sb is already counted in previous macros).
3172 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3173 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3176 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3177 &dentry->d_name, 0, NULL,
3178 EXT4_HT_DIR, credits);
3179 handle = ext4_journal_current_handle();
3180 if (IS_ERR(inode)) {
3182 ext4_journal_stop(handle);
3183 return PTR_ERR(inode);
3186 if (IS_ENCRYPTED(inode)) {
3187 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3189 goto err_drop_inode;
3190 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3193 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3194 if (!IS_ENCRYPTED(inode))
3195 inode->i_op = &ext4_symlink_inode_operations;
3196 inode_nohighmem(inode);
3197 ext4_set_aops(inode);
3199 * We cannot call page_symlink() with transaction started
3200 * because it calls into ext4_write_begin() which can wait
3201 * for transaction commit if we are running out of space
3202 * and thus we deadlock. So we have to stop transaction now
3203 * and restart it when symlink contents is written.
3205 * To keep fs consistent in case of crash, we have to put inode
3206 * to orphan list in the mean time.
3209 err = ext4_orphan_add(handle, inode);
3210 ext4_journal_stop(handle);
3213 goto err_drop_inode;
3214 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3216 goto err_drop_inode;
3218 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3219 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3221 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3222 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3223 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3224 if (IS_ERR(handle)) {
3225 err = PTR_ERR(handle);
3227 goto err_drop_inode;
3229 set_nlink(inode, 1);
3230 err = ext4_orphan_del(handle, inode);
3232 goto err_drop_inode;
3234 /* clear the extent format for fast symlink */
3235 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3236 if (!IS_ENCRYPTED(inode)) {
3237 inode->i_op = &ext4_fast_symlink_inode_operations;
3238 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3240 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3242 inode->i_size = disk_link.len - 1;
3244 EXT4_I(inode)->i_disksize = inode->i_size;
3245 err = ext4_add_nondir(handle, dentry, inode);
3246 if (!err && IS_DIRSYNC(dir))
3247 ext4_handle_sync(handle);
3250 ext4_journal_stop(handle);
3251 goto out_free_encrypted_link;
3255 ext4_journal_stop(handle);
3257 unlock_new_inode(inode);
3259 out_free_encrypted_link:
3260 if (disk_link.name != (unsigned char *)symname)
3261 kfree(disk_link.name);
3265 static int ext4_link(struct dentry *old_dentry,
3266 struct inode *dir, struct dentry *dentry)
3269 struct inode *inode = d_inode(old_dentry);
3270 int err, retries = 0;
3272 if (inode->i_nlink >= EXT4_LINK_MAX)
3275 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3279 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3280 (!projid_eq(EXT4_I(dir)->i_projid,
3281 EXT4_I(old_dentry->d_inode)->i_projid)))
3284 err = dquot_initialize(dir);
3289 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3290 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3291 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3293 return PTR_ERR(handle);
3295 if (IS_DIRSYNC(dir))
3296 ext4_handle_sync(handle);
3298 inode->i_ctime = current_time(inode);
3299 ext4_inc_count(handle, inode);
3302 err = ext4_add_entry(handle, dentry, inode);
3304 ext4_mark_inode_dirty(handle, inode);
3305 /* this can happen only for tmpfile being
3306 * linked the first time
3308 if (inode->i_nlink == 1)
3309 ext4_orphan_del(handle, inode);
3310 d_instantiate(dentry, inode);
3315 ext4_journal_stop(handle);
3316 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3323 * Try to find buffer head where contains the parent block.
3324 * It should be the inode block if it is inlined or the 1st block
3325 * if it is a normal dir.
3327 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3328 struct inode *inode,
3330 struct ext4_dir_entry_2 **parent_de,
3333 struct buffer_head *bh;
3335 if (!ext4_has_inline_data(inode)) {
3336 /* The first directory block must not be a hole, so
3337 * treat it as DIRENT_HTREE
3339 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3341 *retval = PTR_ERR(bh);
3344 *parent_de = ext4_next_entry(
3345 (struct ext4_dir_entry_2 *)bh->b_data,
3346 inode->i_sb->s_blocksize);
3351 return ext4_get_first_inline_block(inode, parent_de, retval);
3354 struct ext4_renament {
3356 struct dentry *dentry;
3357 struct inode *inode;
3359 int dir_nlink_delta;
3361 /* entry for "dentry" */
3362 struct buffer_head *bh;
3363 struct ext4_dir_entry_2 *de;
3366 /* entry for ".." in inode if it's a directory */
3367 struct buffer_head *dir_bh;
3368 struct ext4_dir_entry_2 *parent_de;
3372 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3376 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3377 &retval, &ent->parent_de,
3381 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3382 return -EFSCORRUPTED;
3383 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3384 return ext4_journal_get_write_access(handle, ent->dir_bh);
3387 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3392 ent->parent_de->inode = cpu_to_le32(dir_ino);
3393 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3394 if (!ent->dir_inlined) {
3395 if (is_dx(ent->inode)) {
3396 retval = ext4_handle_dirty_dx_node(handle,
3400 retval = ext4_handle_dirty_dirent_node(handle,
3405 retval = ext4_mark_inode_dirty(handle, ent->inode);
3408 ext4_std_error(ent->dir->i_sb, retval);
3414 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3415 unsigned ino, unsigned file_type)
3419 BUFFER_TRACE(ent->bh, "get write access");
3420 retval = ext4_journal_get_write_access(handle, ent->bh);
3423 ent->de->inode = cpu_to_le32(ino);
3424 if (ext4_has_feature_filetype(ent->dir->i_sb))
3425 ent->de->file_type = file_type;
3426 inode_inc_iversion(ent->dir);
3427 ent->dir->i_ctime = ent->dir->i_mtime =
3428 current_time(ent->dir);
3429 ext4_mark_inode_dirty(handle, ent->dir);
3430 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3431 if (!ent->inlined) {
3432 retval = ext4_handle_dirty_dirent_node(handle,
3434 if (unlikely(retval)) {
3435 ext4_std_error(ent->dir->i_sb, retval);
3443 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3444 unsigned ino, unsigned file_type)
3446 struct ext4_renament old = *ent;
3450 * old->de could have moved from under us during make indexed dir,
3451 * so the old->de may no longer valid and need to find it again
3452 * before reset old inode info.
3454 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3456 retval = PTR_ERR(old.bh);
3460 ext4_std_error(old.dir->i_sb, retval);
3464 ext4_setent(handle, &old, ino, file_type);
3468 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3469 const struct qstr *d_name)
3471 int retval = -ENOENT;
3472 struct buffer_head *bh;
3473 struct ext4_dir_entry_2 *de;
3475 bh = ext4_find_entry(dir, d_name, &de, NULL);
3479 retval = ext4_delete_entry(handle, dir, de, bh);
3485 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3490 * ent->de could have moved from under us during htree split, so make
3491 * sure that we are deleting the right entry. We might also be pointing
3492 * to a stale entry in the unused part of ent->bh so just checking inum
3493 * and the name isn't enough.
3495 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3496 ent->de->name_len != ent->dentry->d_name.len ||
3497 strncmp(ent->de->name, ent->dentry->d_name.name,
3498 ent->de->name_len) ||
3500 retval = ext4_find_delete_entry(handle, ent->dir,
3501 &ent->dentry->d_name);
3503 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3504 if (retval == -ENOENT) {
3505 retval = ext4_find_delete_entry(handle, ent->dir,
3506 &ent->dentry->d_name);
3511 ext4_warning_inode(ent->dir,
3512 "Deleting old file: nlink %d, error=%d",
3513 ent->dir->i_nlink, retval);
3517 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3519 if (ent->dir_nlink_delta) {
3520 if (ent->dir_nlink_delta == -1)
3521 ext4_dec_count(handle, ent->dir);
3523 ext4_inc_count(handle, ent->dir);
3524 ext4_mark_inode_dirty(handle, ent->dir);
3528 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3529 int credits, handle_t **h)
3536 * for inode block, sb block, group summaries,
3539 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3540 EXT4_XATTR_TRANS_BLOCKS + 4);
3542 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3543 &ent->dentry->d_name, 0, NULL,
3544 EXT4_HT_DIR, credits);
3546 handle = ext4_journal_current_handle();
3549 ext4_journal_stop(handle);
3550 if (PTR_ERR(wh) == -ENOSPC &&
3551 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3555 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3556 wh->i_op = &ext4_special_inode_operations;
3562 * Anybody can rename anything with this: the permission checks are left to the
3563 * higher-level routines.
3565 * n.b. old_{dentry,inode) refers to the source dentry/inode
3566 * while new_{dentry,inode) refers to the destination dentry/inode
3567 * This comes from rename(const char *oldpath, const char *newpath)
3569 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3570 struct inode *new_dir, struct dentry *new_dentry,
3573 handle_t *handle = NULL;
3574 struct ext4_renament old = {
3576 .dentry = old_dentry,
3577 .inode = d_inode(old_dentry),
3579 struct ext4_renament new = {
3581 .dentry = new_dentry,
3582 .inode = d_inode(new_dentry),
3586 struct inode *whiteout = NULL;
3590 if (new.inode && new.inode->i_nlink == 0) {
3591 EXT4_ERROR_INODE(new.inode,
3592 "target of rename is already freed");
3593 return -EFSCORRUPTED;
3596 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3597 (!projid_eq(EXT4_I(new_dir)->i_projid,
3598 EXT4_I(old_dentry->d_inode)->i_projid)))
3601 retval = dquot_initialize(old.dir);
3604 retval = dquot_initialize(new.dir);
3608 /* Initialize quotas before so that eventual writes go
3609 * in separate transaction */
3611 retval = dquot_initialize(new.inode);
3616 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3618 return PTR_ERR(old.bh);
3620 * Check for inode number is _not_ due to possible IO errors.
3621 * We might rmdir the source, keep it as pwd of some process
3622 * and merrily kill the link to whatever was created under the
3623 * same name. Goodbye sticky bit ;-<
3626 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3629 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3630 &new.de, &new.inlined);
3631 if (IS_ERR(new.bh)) {
3632 retval = PTR_ERR(new.bh);
3642 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3643 ext4_alloc_da_blocks(old.inode);
3645 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3646 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3647 if (!(flags & RENAME_WHITEOUT)) {
3648 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3649 if (IS_ERR(handle)) {
3650 retval = PTR_ERR(handle);
3654 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3655 if (IS_ERR(whiteout)) {
3656 retval = PTR_ERR(whiteout);
3661 old_file_type = old.de->file_type;
3662 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3663 ext4_handle_sync(handle);
3665 if (S_ISDIR(old.inode->i_mode)) {
3667 retval = -ENOTEMPTY;
3668 if (!ext4_empty_dir(new.inode))
3672 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3675 retval = ext4_rename_dir_prepare(handle, &old);
3680 * If we're renaming a file within an inline_data dir and adding or
3681 * setting the new dirent causes a conversion from inline_data to
3682 * extents/blockmap, we need to force the dirent delete code to
3683 * re-read the directory, or else we end up trying to delete a dirent
3684 * from what is now the extent tree root (or a block map).
3686 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3687 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3691 * Do this before adding a new entry, so the old entry is sure
3692 * to be still pointing to the valid old entry.
3694 retval = ext4_setent(handle, &old, whiteout->i_ino,
3698 ext4_mark_inode_dirty(handle, whiteout);
3701 retval = ext4_add_entry(handle, new.dentry, old.inode);
3705 retval = ext4_setent(handle, &new,
3706 old.inode->i_ino, old_file_type);
3711 force_reread = !ext4_test_inode_flag(new.dir,
3712 EXT4_INODE_INLINE_DATA);
3715 * Like most other Unix systems, set the ctime for inodes on a
3718 old.inode->i_ctime = current_time(old.inode);
3719 ext4_mark_inode_dirty(handle, old.inode);
3725 ext4_rename_delete(handle, &old, force_reread);
3729 ext4_dec_count(handle, new.inode);
3730 new.inode->i_ctime = current_time(new.inode);
3732 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3733 ext4_update_dx_flag(old.dir);
3735 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3739 ext4_dec_count(handle, old.dir);
3741 /* checked ext4_empty_dir above, can't have another
3742 * parent, ext4_dec_count() won't work for many-linked
3744 clear_nlink(new.inode);
3746 ext4_inc_count(handle, new.dir);
3747 ext4_update_dx_flag(new.dir);
3748 ext4_mark_inode_dirty(handle, new.dir);
3751 ext4_mark_inode_dirty(handle, old.dir);
3753 ext4_mark_inode_dirty(handle, new.inode);
3754 if (!new.inode->i_nlink)
3755 ext4_orphan_add(handle, new.inode);
3762 ext4_resetent(handle, &old,
3763 old.inode->i_ino, old_file_type);
3764 drop_nlink(whiteout);
3765 ext4_orphan_add(handle, whiteout);
3767 unlock_new_inode(whiteout);
3768 ext4_journal_stop(handle);
3771 ext4_journal_stop(handle);
3780 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3781 struct inode *new_dir, struct dentry *new_dentry)
3783 handle_t *handle = NULL;
3784 struct ext4_renament old = {
3786 .dentry = old_dentry,
3787 .inode = d_inode(old_dentry),
3789 struct ext4_renament new = {
3791 .dentry = new_dentry,
3792 .inode = d_inode(new_dentry),
3796 struct timespec64 ctime;
3798 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3799 !projid_eq(EXT4_I(new_dir)->i_projid,
3800 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3801 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3802 !projid_eq(EXT4_I(old_dir)->i_projid,
3803 EXT4_I(new_dentry->d_inode)->i_projid)))
3806 retval = dquot_initialize(old.dir);
3809 retval = dquot_initialize(new.dir);
3813 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3814 &old.de, &old.inlined);
3816 return PTR_ERR(old.bh);
3818 * Check for inode number is _not_ due to possible IO errors.
3819 * We might rmdir the source, keep it as pwd of some process
3820 * and merrily kill the link to whatever was created under the
3821 * same name. Goodbye sticky bit ;-<
3824 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3827 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3828 &new.de, &new.inlined);
3829 if (IS_ERR(new.bh)) {
3830 retval = PTR_ERR(new.bh);
3835 /* RENAME_EXCHANGE case: old *and* new must both exist */
3836 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3839 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3840 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3841 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3842 if (IS_ERR(handle)) {
3843 retval = PTR_ERR(handle);
3848 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3849 ext4_handle_sync(handle);
3851 if (S_ISDIR(old.inode->i_mode)) {
3853 retval = ext4_rename_dir_prepare(handle, &old);
3857 if (S_ISDIR(new.inode->i_mode)) {
3859 retval = ext4_rename_dir_prepare(handle, &new);
3865 * Other than the special case of overwriting a directory, parents'
3866 * nlink only needs to be modified if this is a cross directory rename.
3868 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3869 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3870 new.dir_nlink_delta = -old.dir_nlink_delta;
3872 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3873 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3877 new_file_type = new.de->file_type;
3878 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3882 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3887 * Like most other Unix systems, set the ctime for inodes on a
3890 ctime = current_time(old.inode);
3891 old.inode->i_ctime = ctime;
3892 new.inode->i_ctime = ctime;
3893 ext4_mark_inode_dirty(handle, old.inode);
3894 ext4_mark_inode_dirty(handle, new.inode);
3897 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3902 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3906 ext4_update_dir_count(handle, &old);
3907 ext4_update_dir_count(handle, &new);
3916 ext4_journal_stop(handle);
3920 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3921 struct inode *new_dir, struct dentry *new_dentry,
3926 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3929 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3932 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
3937 if (flags & RENAME_EXCHANGE) {
3938 return ext4_cross_rename(old_dir, old_dentry,
3939 new_dir, new_dentry);
3942 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3946 * directories can handle most operations...
3948 const struct inode_operations ext4_dir_inode_operations = {
3949 .create = ext4_create,
3950 .lookup = ext4_lookup,
3952 .unlink = ext4_unlink,
3953 .symlink = ext4_symlink,
3954 .mkdir = ext4_mkdir,
3955 .rmdir = ext4_rmdir,
3956 .mknod = ext4_mknod,
3957 .tmpfile = ext4_tmpfile,
3958 .rename = ext4_rename2,
3959 .setattr = ext4_setattr,
3960 .getattr = ext4_getattr,
3961 .listxattr = ext4_listxattr,
3962 .get_acl = ext4_get_acl,
3963 .set_acl = ext4_set_acl,
3964 .fiemap = ext4_fiemap,
3967 const struct inode_operations ext4_special_inode_operations = {
3968 .setattr = ext4_setattr,
3969 .getattr = ext4_getattr,
3970 .listxattr = ext4_listxattr,
3971 .get_acl = ext4_get_acl,
3972 .set_acl = ext4_set_acl,