1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/namei.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/namei.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 * Directory entry file type support and forward compatibility hooks
19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20 * Hash Tree Directory indexing (c)
21 * Daniel Phillips, 2001
22 * Hash Tree Directory indexing porting
23 * Christopher Li, 2002
24 * Hash Tree Directory indexing cleanup
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include <linux/iversion.h>
38 #include <linux/unicode.h>
40 #include "ext4_jbd2.h"
45 #include <trace/events/ext4.h>
47 * define how far ahead to read directories while searching them.
49 #define NAMEI_RA_CHUNKS 2
50 #define NAMEI_RA_BLOCKS 4
51 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
53 static struct buffer_head *ext4_append(handle_t *handle,
57 struct buffer_head *bh;
60 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
61 ((inode->i_size >> 10) >=
62 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
63 return ERR_PTR(-ENOSPC);
65 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
67 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
70 inode->i_size += inode->i_sb->s_blocksize;
71 EXT4_I(inode)->i_disksize = inode->i_size;
72 BUFFER_TRACE(bh, "get_write_access");
73 err = ext4_journal_get_write_access(handle, bh);
76 ext4_std_error(inode->i_sb, err);
82 static int ext4_dx_csum_verify(struct inode *inode,
83 struct ext4_dir_entry *dirent);
86 * Hints to ext4_read_dirblock regarding whether we expect a directory
87 * block being read to be an index block, or a block containing
88 * directory entries (and if the latter, whether it was found via a
89 * logical block in an htree index block). This is used to control
90 * what sort of sanity checkinig ext4_read_dirblock() will do on the
91 * directory block read from the storage device. EITHER will means
92 * the caller doesn't know what kind of directory block will be read,
93 * so no specific verification will be done.
96 EITHER, INDEX, DIRENT, DIRENT_HTREE
99 #define ext4_read_dirblock(inode, block, type) \
100 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
104 dirblock_type_t type,
108 struct buffer_head *bh;
109 struct ext4_dir_entry *dirent;
112 bh = ext4_bread(NULL, inode, block, 0);
114 __ext4_warning(inode->i_sb, func, line,
115 "inode #%lu: lblock %lu: comm %s: "
116 "error %ld reading directory block",
117 inode->i_ino, (unsigned long)block,
118 current->comm, PTR_ERR(bh));
122 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
123 ext4_error_inode(inode, func, line, block,
124 "Directory hole found for htree %s block",
125 (type == INDEX) ? "index" : "leaf");
126 return ERR_PTR(-EFSCORRUPTED);
130 dirent = (struct ext4_dir_entry *) bh->b_data;
131 /* Determine whether or not we have an index block */
135 else if (ext4_rec_len_from_disk(dirent->rec_len,
136 inode->i_sb->s_blocksize) ==
137 inode->i_sb->s_blocksize)
140 if (!is_dx_block && type == INDEX) {
141 ext4_error_inode(inode, func, line, block,
142 "directory leaf block found instead of index block");
144 return ERR_PTR(-EFSCORRUPTED);
146 if (!ext4_has_metadata_csum(inode->i_sb) ||
151 * An empty leaf block can get mistaken for a index block; for
152 * this reason, we can only check the index checksum when the
153 * caller is sure it should be an index block.
155 if (is_dx_block && type == INDEX) {
156 if (ext4_dx_csum_verify(inode, dirent))
157 set_buffer_verified(bh);
159 ext4_error_inode(inode, func, line, block,
160 "Directory index failed checksum");
162 return ERR_PTR(-EFSBADCRC);
166 if (ext4_dirblock_csum_verify(inode, bh))
167 set_buffer_verified(bh);
169 ext4_error_inode(inode, func, line, block,
170 "Directory block failed checksum");
172 return ERR_PTR(-EFSBADCRC);
179 #define assert(test) J_ASSERT(test)
183 #define dxtrace(command) command
185 #define dxtrace(command)
209 * dx_root_info is laid out so that if it should somehow get overlaid by a
210 * dirent the two low bits of the hash version will be zero. Therefore, the
211 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
216 struct fake_dirent dot;
218 struct fake_dirent dotdot;
222 __le32 reserved_zero;
224 u8 info_length; /* 8 */
229 struct dx_entry entries[0];
234 struct fake_dirent fake;
235 struct dx_entry entries[0];
241 struct buffer_head *bh;
242 struct dx_entry *entries;
254 * This goes at the end of each htree block.
258 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
261 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
262 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
263 static inline unsigned dx_get_hash(struct dx_entry *entry);
264 static void dx_set_hash(struct dx_entry *entry, unsigned value);
265 static unsigned dx_get_count(struct dx_entry *entries);
266 static unsigned dx_get_limit(struct dx_entry *entries);
267 static void dx_set_count(struct dx_entry *entries, unsigned value);
268 static void dx_set_limit(struct dx_entry *entries, unsigned value);
269 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
270 static unsigned dx_node_limit(struct inode *dir);
271 static struct dx_frame *dx_probe(struct ext4_filename *fname,
273 struct dx_hash_info *hinfo,
274 struct dx_frame *frame);
275 static void dx_release(struct dx_frame *frames);
276 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
277 unsigned blocksize, struct dx_hash_info *hinfo,
278 struct dx_map_entry map[]);
279 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
280 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
281 struct dx_map_entry *offsets, int count, unsigned blocksize);
282 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
283 static void dx_insert_block(struct dx_frame *frame,
284 u32 hash, ext4_lblk_t block);
285 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
286 struct dx_frame *frame,
287 struct dx_frame *frames,
289 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
290 struct ext4_filename *fname,
291 struct ext4_dir_entry_2 **res_dir);
292 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
293 struct inode *dir, struct inode *inode);
295 /* checksumming functions */
296 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
297 unsigned int blocksize)
299 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
300 t->det_rec_len = ext4_rec_len_to_disk(
301 sizeof(struct ext4_dir_entry_tail), blocksize);
302 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
305 /* Walk through a dirent block to find a checksum "dirent" at the tail */
306 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
307 struct buffer_head *bh)
309 struct ext4_dir_entry_tail *t;
312 struct ext4_dir_entry *d, *top;
314 d = (struct ext4_dir_entry *)bh->b_data;
315 top = (struct ext4_dir_entry *)(bh->b_data +
316 (EXT4_BLOCK_SIZE(inode->i_sb) -
317 sizeof(struct ext4_dir_entry_tail)));
318 while (d < top && d->rec_len)
319 d = (struct ext4_dir_entry *)(((void *)d) +
320 le16_to_cpu(d->rec_len));
325 t = (struct ext4_dir_entry_tail *)d;
327 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
330 if (t->det_reserved_zero1 ||
331 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
332 t->det_reserved_zero2 ||
333 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
339 static __le32 ext4_dirblock_csum(struct inode *inode, void *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_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
361 struct ext4_dir_entry_tail *t;
363 if (!ext4_has_metadata_csum(inode->i_sb))
366 t = get_dirent_tail(inode, bh);
368 warn_no_space_for_csum(inode);
372 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
373 (char *)t - bh->b_data))
379 static void ext4_dirblock_csum_set(struct inode *inode,
380 struct buffer_head *bh)
382 struct ext4_dir_entry_tail *t;
384 if (!ext4_has_metadata_csum(inode->i_sb))
387 t = get_dirent_tail(inode, bh);
389 warn_no_space_for_csum(inode);
393 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
394 (char *)t - bh->b_data);
397 int ext4_handle_dirty_dirblock(handle_t *handle,
399 struct buffer_head *bh)
401 ext4_dirblock_csum_set(inode, bh);
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_FS_ENCRYPTION
631 struct fscrypt_str fname_crypto_str =
637 if (IS_ENCRYPTED(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(dir, 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(dir, de->name,
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(dir, 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(dir, 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_FS_ENCRYPTION
1004 /* Check if the directory is encrypted */
1005 if (IS_ENCRYPTED(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(dir, 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 (!IS_ENCRYPTED(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_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(dir, 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);
1271 #ifdef CONFIG_UNICODE
1273 * Test whether a case-insensitive directory entry matches the filename
1274 * being searched for. If quick is set, assume the name being looked up
1275 * is already in the casefolded form.
1277 * Returns: 0 if the directory entry matches, more than 0 if it
1278 * doesn't match or less than zero on error.
1280 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1281 const struct qstr *entry, bool quick)
1283 const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1284 const struct unicode_map *um = sbi->s_encoding;
1288 ret = utf8_strncasecmp_folded(um, name, entry);
1290 ret = utf8_strncasecmp(um, name, entry);
1293 /* Handle invalid character sequence as either an error
1294 * or as an opaque byte sequence.
1296 if (ext4_has_strict_mode(sbi))
1299 if (name->len != entry->len)
1302 return !!memcmp(name->name, entry->name, name->len);
1308 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1309 struct fscrypt_str *cf_name)
1311 if (!IS_CASEFOLDED(dir)) {
1312 cf_name->name = NULL;
1316 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1320 cf_name->len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1321 iname, cf_name->name,
1323 if (cf_name->len <= 0) {
1324 kfree(cf_name->name);
1325 cf_name->name = NULL;
1331 * Test whether a directory entry matches the filename being searched for.
1333 * Return: %true if the directory entry matches, otherwise %false.
1335 static inline bool ext4_match(const struct inode *parent,
1336 const struct ext4_filename *fname,
1337 const struct ext4_dir_entry_2 *de)
1339 struct fscrypt_name f;
1340 #ifdef CONFIG_UNICODE
1341 const struct qstr entry = {.name = de->name, .len = de->name_len};
1347 f.usr_fname = fname->usr_fname;
1348 f.disk_name = fname->disk_name;
1349 #ifdef CONFIG_FS_ENCRYPTION
1350 f.crypto_buf = fname->crypto_buf;
1353 #ifdef CONFIG_UNICODE
1354 if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1355 if (fname->cf_name.name) {
1356 struct qstr cf = {.name = fname->cf_name.name,
1357 .len = fname->cf_name.len};
1358 return !ext4_ci_compare(parent, &cf, &entry, true);
1360 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1365 return fscrypt_match_name(&f, de->name, de->name_len);
1369 * Returns 0 if not found, -1 on failure, and 1 on success
1371 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1372 struct inode *dir, struct ext4_filename *fname,
1373 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1375 struct ext4_dir_entry_2 * de;
1379 de = (struct ext4_dir_entry_2 *)search_buf;
1380 dlimit = search_buf + buf_size;
1381 while ((char *) de < dlimit) {
1382 /* this code is executed quadratically often */
1383 /* do minimal checking `by hand' */
1384 if ((char *) de + de->name_len <= dlimit &&
1385 ext4_match(dir, fname, de)) {
1386 /* found a match - just to be sure, do
1388 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1389 bh->b_size, offset))
1394 /* prevent looping on a bad block */
1395 de_len = ext4_rec_len_from_disk(de->rec_len,
1396 dir->i_sb->s_blocksize);
1400 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1405 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1406 struct ext4_dir_entry *de)
1408 struct super_block *sb = dir->i_sb;
1414 if (de->inode == 0 &&
1415 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1422 * __ext4_find_entry()
1424 * finds an entry in the specified directory with the wanted name. It
1425 * returns the cache buffer in which the entry was found, and the entry
1426 * itself (as a parameter - res_dir). It does NOT read the inode of the
1427 * entry - you'll have to do that yourself if you want to.
1429 * The returned buffer_head has ->b_count elevated. The caller is expected
1430 * to brelse() it when appropriate.
1432 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1433 struct ext4_filename *fname,
1434 struct ext4_dir_entry_2 **res_dir,
1437 struct super_block *sb;
1438 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1439 struct buffer_head *bh, *ret = NULL;
1440 ext4_lblk_t start, block;
1441 const u8 *name = fname->usr_fname->name;
1442 size_t ra_max = 0; /* Number of bh's in the readahead
1444 size_t ra_ptr = 0; /* Current index into readahead
1446 ext4_lblk_t nblocks;
1447 int i, namelen, retval;
1451 namelen = fname->usr_fname->len;
1452 if (namelen > EXT4_NAME_LEN)
1455 if (ext4_has_inline_data(dir)) {
1456 int has_inline_data = 1;
1457 ret = ext4_find_inline_entry(dir, fname, res_dir,
1459 if (has_inline_data) {
1462 goto cleanup_and_exit;
1466 if ((namelen <= 2) && (name[0] == '.') &&
1467 (name[1] == '.' || name[1] == '\0')) {
1469 * "." or ".." will only be in the first block
1470 * NFS may look up ".."; "." should be handled by the VFS
1477 ret = ext4_dx_find_entry(dir, fname, res_dir);
1479 * On success, or if the error was file not found,
1480 * return. Otherwise, fall back to doing a search the
1481 * old fashioned way.
1483 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1484 goto cleanup_and_exit;
1485 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1489 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1492 goto cleanup_and_exit;
1494 start = EXT4_I(dir)->i_dir_start_lookup;
1495 if (start >= nblocks)
1501 * We deal with the read-ahead logic here.
1503 if (ra_ptr >= ra_max) {
1504 /* Refill the readahead buffer */
1507 ra_max = start - block;
1509 ra_max = nblocks - block;
1510 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1511 retval = ext4_bread_batch(dir, block, ra_max,
1512 false /* wait */, bh_use);
1514 ret = ERR_PTR(retval);
1516 goto cleanup_and_exit;
1519 if ((bh = bh_use[ra_ptr++]) == NULL)
1522 if (!buffer_uptodate(bh)) {
1523 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1524 (unsigned long) block);
1526 ret = ERR_PTR(-EIO);
1527 goto cleanup_and_exit;
1529 if (!buffer_verified(bh) &&
1530 !is_dx_internal_node(dir, block,
1531 (struct ext4_dir_entry *)bh->b_data) &&
1532 !ext4_dirblock_csum_verify(dir, bh)) {
1533 EXT4_ERROR_INODE(dir, "checksumming directory "
1534 "block %lu", (unsigned long)block);
1536 ret = ERR_PTR(-EFSBADCRC);
1537 goto cleanup_and_exit;
1539 set_buffer_verified(bh);
1540 i = search_dirblock(bh, dir, fname,
1541 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1543 EXT4_I(dir)->i_dir_start_lookup = block;
1545 goto cleanup_and_exit;
1549 goto cleanup_and_exit;
1552 if (++block >= nblocks)
1554 } while (block != start);
1557 * If the directory has grown while we were searching, then
1558 * search the last part of the directory before giving up.
1561 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1562 if (block < nblocks) {
1568 /* Clean up the read-ahead blocks */
1569 for (; ra_ptr < ra_max; ra_ptr++)
1570 brelse(bh_use[ra_ptr]);
1574 static struct buffer_head *ext4_find_entry(struct inode *dir,
1575 const struct qstr *d_name,
1576 struct ext4_dir_entry_2 **res_dir,
1580 struct ext4_filename fname;
1581 struct buffer_head *bh;
1583 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1587 return ERR_PTR(err);
1589 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1591 ext4_fname_free_filename(&fname);
1595 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1596 struct dentry *dentry,
1597 struct ext4_dir_entry_2 **res_dir)
1600 struct ext4_filename fname;
1601 struct buffer_head *bh;
1603 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1607 return ERR_PTR(err);
1609 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1611 ext4_fname_free_filename(&fname);
1615 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1616 struct ext4_filename *fname,
1617 struct ext4_dir_entry_2 **res_dir)
1619 struct super_block * sb = dir->i_sb;
1620 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1621 struct buffer_head *bh;
1625 #ifdef CONFIG_FS_ENCRYPTION
1628 frame = dx_probe(fname, dir, NULL, frames);
1630 return (struct buffer_head *) frame;
1632 block = dx_get_block(frame->at);
1633 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1637 retval = search_dirblock(bh, dir, fname,
1638 block << EXT4_BLOCK_SIZE_BITS(sb),
1644 bh = ERR_PTR(ERR_BAD_DX_DIR);
1648 /* Check to see if we should continue to search */
1649 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1652 ext4_warning_inode(dir,
1653 "error %d reading directory index block",
1655 bh = ERR_PTR(retval);
1658 } while (retval == 1);
1662 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1668 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1670 struct inode *inode;
1671 struct ext4_dir_entry_2 *de;
1672 struct buffer_head *bh;
1674 if (dentry->d_name.len > EXT4_NAME_LEN)
1675 return ERR_PTR(-ENAMETOOLONG);
1677 bh = ext4_lookup_entry(dir, dentry, &de);
1679 return ERR_CAST(bh);
1682 __u32 ino = le32_to_cpu(de->inode);
1684 if (!ext4_valid_inum(dir->i_sb, ino)) {
1685 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1686 return ERR_PTR(-EFSCORRUPTED);
1688 if (unlikely(ino == dir->i_ino)) {
1689 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1691 return ERR_PTR(-EFSCORRUPTED);
1693 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1694 if (inode == ERR_PTR(-ESTALE)) {
1695 EXT4_ERROR_INODE(dir,
1696 "deleted inode referenced: %u",
1698 return ERR_PTR(-EFSCORRUPTED);
1700 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1701 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1702 !fscrypt_has_permitted_context(dir, inode)) {
1703 ext4_warning(inode->i_sb,
1704 "Inconsistent encryption contexts: %lu/%lu",
1705 dir->i_ino, inode->i_ino);
1707 return ERR_PTR(-EPERM);
1711 #ifdef CONFIG_UNICODE
1712 if (!inode && IS_CASEFOLDED(dir)) {
1713 /* Eventually we want to call d_add_ci(dentry, NULL)
1714 * for negative dentries in the encoding case as
1715 * well. For now, prevent the negative dentry
1716 * from being cached.
1721 return d_splice_alias(inode, dentry);
1725 struct dentry *ext4_get_parent(struct dentry *child)
1728 static const struct qstr dotdot = QSTR_INIT("..", 2);
1729 struct ext4_dir_entry_2 * de;
1730 struct buffer_head *bh;
1732 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1734 return ERR_CAST(bh);
1736 return ERR_PTR(-ENOENT);
1737 ino = le32_to_cpu(de->inode);
1740 if (!ext4_valid_inum(child->d_sb, ino)) {
1741 EXT4_ERROR_INODE(d_inode(child),
1742 "bad parent inode number: %u", ino);
1743 return ERR_PTR(-EFSCORRUPTED);
1746 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1750 * Move count entries from end of map between two memory locations.
1751 * Returns pointer to last entry moved.
1753 static struct ext4_dir_entry_2 *
1754 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1757 unsigned rec_len = 0;
1760 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1761 (from + (map->offs<<2));
1762 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1763 memcpy (to, de, rec_len);
1764 ((struct ext4_dir_entry_2 *) to)->rec_len =
1765 ext4_rec_len_to_disk(rec_len, blocksize);
1770 return (struct ext4_dir_entry_2 *) (to - rec_len);
1774 * Compact each dir entry in the range to the minimal rec_len.
1775 * Returns pointer to last entry in range.
1777 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1779 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1780 unsigned rec_len = 0;
1783 while ((char*)de < base + blocksize) {
1784 next = ext4_next_entry(de, blocksize);
1785 if (de->inode && de->name_len) {
1786 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1788 memmove(to, de, rec_len);
1789 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1791 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1799 * Split a full leaf block to make room for a new dir entry.
1800 * Allocate a new block, and move entries so that they are approx. equally full.
1801 * Returns pointer to de in block into which the new entry will be inserted.
1803 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1804 struct buffer_head **bh,struct dx_frame *frame,
1805 struct dx_hash_info *hinfo)
1807 unsigned blocksize = dir->i_sb->s_blocksize;
1808 unsigned count, continued;
1809 struct buffer_head *bh2;
1810 ext4_lblk_t newblock;
1812 struct dx_map_entry *map;
1813 char *data1 = (*bh)->b_data, *data2;
1814 unsigned split, move, size;
1815 struct ext4_dir_entry_2 *de = NULL, *de2;
1816 struct ext4_dir_entry_tail *t;
1820 if (ext4_has_metadata_csum(dir->i_sb))
1821 csum_size = sizeof(struct ext4_dir_entry_tail);
1823 bh2 = ext4_append(handle, dir, &newblock);
1827 return (struct ext4_dir_entry_2 *) bh2;
1830 BUFFER_TRACE(*bh, "get_write_access");
1831 err = ext4_journal_get_write_access(handle, *bh);
1835 BUFFER_TRACE(frame->bh, "get_write_access");
1836 err = ext4_journal_get_write_access(handle, frame->bh);
1840 data2 = bh2->b_data;
1842 /* create map in the end of data2 block */
1843 map = (struct dx_map_entry *) (data2 + blocksize);
1844 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1845 blocksize, hinfo, map);
1847 dx_sort_map(map, count);
1848 /* Split the existing block in the middle, size-wise */
1851 for (i = count-1; i >= 0; i--) {
1852 /* is more than half of this entry in 2nd half of the block? */
1853 if (size + map[i].size/2 > blocksize/2)
1855 size += map[i].size;
1858 /* map index at which we will split */
1859 split = count - move;
1860 hash2 = map[split].hash;
1861 continued = hash2 == map[split - 1].hash;
1862 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1863 (unsigned long)dx_get_block(frame->at),
1864 hash2, split, count-split));
1866 /* Fancy dance to stay within two buffers */
1867 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1869 de = dx_pack_dirents(data1, blocksize);
1870 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1873 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1877 t = EXT4_DIRENT_TAIL(data2, blocksize);
1878 initialize_dirent_tail(t, blocksize);
1880 t = EXT4_DIRENT_TAIL(data1, blocksize);
1881 initialize_dirent_tail(t, blocksize);
1884 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1886 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1889 /* Which block gets the new entry? */
1890 if (hinfo->hash >= hash2) {
1894 dx_insert_block(frame, hash2 + continued, newblock);
1895 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1898 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1902 dxtrace(dx_show_index("frame", frame->entries));
1909 ext4_std_error(dir->i_sb, err);
1910 return ERR_PTR(err);
1913 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1914 struct buffer_head *bh,
1915 void *buf, int buf_size,
1916 struct ext4_filename *fname,
1917 struct ext4_dir_entry_2 **dest_de)
1919 struct ext4_dir_entry_2 *de;
1920 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1922 unsigned int offset = 0;
1925 de = (struct ext4_dir_entry_2 *)buf;
1926 top = buf + buf_size - reclen;
1927 while ((char *) de <= top) {
1928 if (ext4_check_dir_entry(dir, NULL, de, bh,
1929 buf, buf_size, offset))
1930 return -EFSCORRUPTED;
1931 if (ext4_match(dir, fname, de))
1933 nlen = EXT4_DIR_REC_LEN(de->name_len);
1934 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1935 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1937 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1940 if ((char *) de > top)
1947 void ext4_insert_dentry(struct inode *inode,
1948 struct ext4_dir_entry_2 *de,
1950 struct ext4_filename *fname)
1955 nlen = EXT4_DIR_REC_LEN(de->name_len);
1956 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1958 struct ext4_dir_entry_2 *de1 =
1959 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1960 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1961 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1964 de->file_type = EXT4_FT_UNKNOWN;
1965 de->inode = cpu_to_le32(inode->i_ino);
1966 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1967 de->name_len = fname_len(fname);
1968 memcpy(de->name, fname_name(fname), fname_len(fname));
1972 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1973 * it points to a directory entry which is guaranteed to be large
1974 * enough for new directory entry. If de is NULL, then
1975 * add_dirent_to_buf will attempt search the directory block for
1976 * space. It will return -ENOSPC if no space is available, and -EIO
1977 * and -EEXIST if directory entry already exists.
1979 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1981 struct inode *inode, struct ext4_dir_entry_2 *de,
1982 struct buffer_head *bh)
1984 unsigned int blocksize = dir->i_sb->s_blocksize;
1988 if (ext4_has_metadata_csum(inode->i_sb))
1989 csum_size = sizeof(struct ext4_dir_entry_tail);
1992 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1993 blocksize - csum_size, fname, &de);
1997 BUFFER_TRACE(bh, "get_write_access");
1998 err = ext4_journal_get_write_access(handle, bh);
2000 ext4_std_error(dir->i_sb, err);
2004 /* By now the buffer is marked for journaling */
2005 ext4_insert_dentry(inode, de, blocksize, fname);
2008 * XXX shouldn't update any times until successful
2009 * completion of syscall, but too many callers depend
2012 * XXX similarly, too many callers depend on
2013 * ext4_new_inode() setting the times, but error
2014 * recovery deletes the inode, so the worst that can
2015 * happen is that the times are slightly out of date
2016 * and/or different from the directory change time.
2018 dir->i_mtime = dir->i_ctime = current_time(dir);
2019 ext4_update_dx_flag(dir);
2020 inode_inc_iversion(dir);
2021 ext4_mark_inode_dirty(handle, dir);
2022 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2023 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2025 ext4_std_error(dir->i_sb, err);
2030 * This converts a one block unindexed directory to a 3 block indexed
2031 * directory, and adds the dentry to the indexed directory.
2033 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2035 struct inode *inode, struct buffer_head *bh)
2037 struct buffer_head *bh2;
2038 struct dx_root *root;
2039 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2040 struct dx_entry *entries;
2041 struct ext4_dir_entry_2 *de, *de2;
2042 struct ext4_dir_entry_tail *t;
2048 struct fake_dirent *fde;
2051 if (ext4_has_metadata_csum(inode->i_sb))
2052 csum_size = sizeof(struct ext4_dir_entry_tail);
2054 blocksize = dir->i_sb->s_blocksize;
2055 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2056 BUFFER_TRACE(bh, "get_write_access");
2057 retval = ext4_journal_get_write_access(handle, bh);
2059 ext4_std_error(dir->i_sb, retval);
2063 root = (struct dx_root *) bh->b_data;
2065 /* The 0th block becomes the root, move the dirents out */
2066 fde = &root->dotdot;
2067 de = (struct ext4_dir_entry_2 *)((char *)fde +
2068 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2069 if ((char *) de >= (((char *) root) + blocksize)) {
2070 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2072 return -EFSCORRUPTED;
2074 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2076 /* Allocate new block for the 0th block's dirents */
2077 bh2 = ext4_append(handle, dir, &block);
2080 return PTR_ERR(bh2);
2082 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2083 data1 = bh2->b_data;
2085 memcpy (data1, de, len);
2086 de = (struct ext4_dir_entry_2 *) data1;
2088 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2090 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2095 t = EXT4_DIRENT_TAIL(data1, blocksize);
2096 initialize_dirent_tail(t, blocksize);
2099 /* Initialize the root; the dot dirents already exist */
2100 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2101 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2103 memset (&root->info, 0, sizeof(root->info));
2104 root->info.info_length = sizeof(root->info);
2105 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2106 entries = root->entries;
2107 dx_set_block(entries, 1);
2108 dx_set_count(entries, 1);
2109 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2111 /* Initialize as for dx_probe */
2112 fname->hinfo.hash_version = root->info.hash_version;
2113 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2114 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2115 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2116 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2118 memset(frames, 0, sizeof(frames));
2120 frame->entries = entries;
2121 frame->at = entries;
2124 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2127 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2131 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2133 retval = PTR_ERR(de);
2137 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2140 * Even if the block split failed, we have to properly write
2141 * out all the changes we did so far. Otherwise we can end up
2142 * with corrupted filesystem.
2145 ext4_mark_inode_dirty(handle, dir);
2154 * adds a file entry to the specified directory, using the same
2155 * semantics as ext4_find_entry(). It returns NULL if it failed.
2157 * NOTE!! The inode part of 'de' is left at 0 - which means you
2158 * may not sleep between calling this and putting something into
2159 * the entry, as someone else might have used it while you slept.
2161 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2162 struct inode *inode)
2164 struct inode *dir = d_inode(dentry->d_parent);
2165 struct buffer_head *bh = NULL;
2166 struct ext4_dir_entry_2 *de;
2167 struct ext4_dir_entry_tail *t;
2168 struct super_block *sb;
2169 struct ext4_sb_info *sbi;
2170 struct ext4_filename fname;
2174 ext4_lblk_t block, blocks;
2177 if (ext4_has_metadata_csum(inode->i_sb))
2178 csum_size = sizeof(struct ext4_dir_entry_tail);
2182 blocksize = sb->s_blocksize;
2183 if (!dentry->d_name.len)
2186 #ifdef CONFIG_UNICODE
2187 if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2188 utf8_validate(sbi->s_encoding, &dentry->d_name))
2192 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2196 if (ext4_has_inline_data(dir)) {
2197 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2207 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2208 if (!retval || (retval != ERR_BAD_DX_DIR))
2210 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2212 ext4_mark_inode_dirty(handle, dir);
2214 blocks = dir->i_size >> sb->s_blocksize_bits;
2215 for (block = 0; block < blocks; block++) {
2216 bh = ext4_read_dirblock(dir, block, DIRENT);
2218 bh = ext4_bread(handle, dir, block,
2219 EXT4_GET_BLOCKS_CREATE);
2220 goto add_to_new_block;
2223 retval = PTR_ERR(bh);
2227 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2229 if (retval != -ENOSPC)
2232 if (blocks == 1 && !dx_fallback &&
2233 ext4_has_feature_dir_index(sb)) {
2234 retval = make_indexed_dir(handle, &fname, dir,
2236 bh = NULL; /* make_indexed_dir releases bh */
2241 bh = ext4_append(handle, dir, &block);
2244 retval = PTR_ERR(bh);
2248 de = (struct ext4_dir_entry_2 *) bh->b_data;
2250 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2253 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2254 initialize_dirent_tail(t, blocksize);
2257 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2259 ext4_fname_free_filename(&fname);
2262 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2267 * Returns 0 for success, or a negative error value
2269 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2270 struct inode *dir, struct inode *inode)
2272 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2273 struct dx_entry *entries, *at;
2274 struct buffer_head *bh;
2275 struct super_block *sb = dir->i_sb;
2276 struct ext4_dir_entry_2 *de;
2282 frame = dx_probe(fname, dir, NULL, frames);
2284 return PTR_ERR(frame);
2285 entries = frame->entries;
2287 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2294 BUFFER_TRACE(bh, "get_write_access");
2295 err = ext4_journal_get_write_access(handle, bh);
2299 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2304 /* Block full, should compress but for now just split */
2305 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2306 dx_get_count(entries), dx_get_limit(entries)));
2307 /* Need to split index? */
2308 if (dx_get_count(entries) == dx_get_limit(entries)) {
2309 ext4_lblk_t newblock;
2310 int levels = frame - frames + 1;
2311 unsigned int icount;
2313 struct dx_entry *entries2;
2314 struct dx_node *node2;
2315 struct buffer_head *bh2;
2317 while (frame > frames) {
2318 if (dx_get_count((frame - 1)->entries) <
2319 dx_get_limit((frame - 1)->entries)) {
2323 frame--; /* split higher index block */
2325 entries = frame->entries;
2328 if (add_level && levels == ext4_dir_htree_level(sb)) {
2329 ext4_warning(sb, "Directory (ino: %lu) index full, "
2330 "reach max htree level :%d",
2331 dir->i_ino, levels);
2332 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2333 ext4_warning(sb, "Large directory feature is "
2334 "not enabled on this "
2340 icount = dx_get_count(entries);
2341 bh2 = ext4_append(handle, dir, &newblock);
2346 node2 = (struct dx_node *)(bh2->b_data);
2347 entries2 = node2->entries;
2348 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2349 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2351 BUFFER_TRACE(frame->bh, "get_write_access");
2352 err = ext4_journal_get_write_access(handle, frame->bh);
2356 unsigned icount1 = icount/2, icount2 = icount - icount1;
2357 unsigned hash2 = dx_get_hash(entries + icount1);
2358 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2361 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2362 err = ext4_journal_get_write_access(handle,
2367 memcpy((char *) entries2, (char *) (entries + icount1),
2368 icount2 * sizeof(struct dx_entry));
2369 dx_set_count(entries, icount1);
2370 dx_set_count(entries2, icount2);
2371 dx_set_limit(entries2, dx_node_limit(dir));
2373 /* Which index block gets the new entry? */
2374 if (at - entries >= icount1) {
2375 frame->at = at = at - entries - icount1 + entries2;
2376 frame->entries = entries = entries2;
2377 swap(frame->bh, bh2);
2379 dx_insert_block((frame - 1), hash2, newblock);
2380 dxtrace(dx_show_index("node", frame->entries));
2381 dxtrace(dx_show_index("node",
2382 ((struct dx_node *) bh2->b_data)->entries));
2383 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2387 err = ext4_handle_dirty_dx_node(handle, dir,
2392 err = ext4_handle_dirty_dx_node(handle, dir,
2397 struct dx_root *dxroot;
2398 memcpy((char *) entries2, (char *) entries,
2399 icount * sizeof(struct dx_entry));
2400 dx_set_limit(entries2, dx_node_limit(dir));
2403 dx_set_count(entries, 1);
2404 dx_set_block(entries + 0, newblock);
2405 dxroot = (struct dx_root *)frames[0].bh->b_data;
2406 dxroot->info.indirect_levels += 1;
2407 dxtrace(printk(KERN_DEBUG
2408 "Creating %d level index...\n",
2409 dxroot->info.indirect_levels));
2410 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2413 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2419 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2424 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2428 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2432 /* @restart is true means htree-path has been changed, we need to
2433 * repeat dx_probe() to find out valid htree-path
2435 if (restart && err == 0)
2441 * ext4_generic_delete_entry deletes a directory entry by merging it
2442 * with the previous entry
2444 int ext4_generic_delete_entry(handle_t *handle,
2446 struct ext4_dir_entry_2 *de_del,
2447 struct buffer_head *bh,
2452 struct ext4_dir_entry_2 *de, *pde;
2453 unsigned int blocksize = dir->i_sb->s_blocksize;
2458 de = (struct ext4_dir_entry_2 *)entry_buf;
2459 while (i < buf_size - csum_size) {
2460 if (ext4_check_dir_entry(dir, NULL, de, bh,
2461 bh->b_data, bh->b_size, i))
2462 return -EFSCORRUPTED;
2465 pde->rec_len = ext4_rec_len_to_disk(
2466 ext4_rec_len_from_disk(pde->rec_len,
2468 ext4_rec_len_from_disk(de->rec_len,
2473 inode_inc_iversion(dir);
2476 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2478 de = ext4_next_entry(de, blocksize);
2483 static int ext4_delete_entry(handle_t *handle,
2485 struct ext4_dir_entry_2 *de_del,
2486 struct buffer_head *bh)
2488 int err, csum_size = 0;
2490 if (ext4_has_inline_data(dir)) {
2491 int has_inline_data = 1;
2492 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2494 if (has_inline_data)
2498 if (ext4_has_metadata_csum(dir->i_sb))
2499 csum_size = sizeof(struct ext4_dir_entry_tail);
2501 BUFFER_TRACE(bh, "get_write_access");
2502 err = ext4_journal_get_write_access(handle, bh);
2506 err = ext4_generic_delete_entry(handle, dir, de_del,
2508 dir->i_sb->s_blocksize, csum_size);
2512 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2513 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2520 ext4_std_error(dir->i_sb, err);
2525 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2526 * since this indicates that nlinks count was previously 1 to avoid overflowing
2527 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2528 * that subdirectory link counts are not being maintained accurately.
2530 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2531 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2532 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2533 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2535 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2539 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2540 set_nlink(inode, 1);
2544 * If a directory had nlink == 1, then we should let it be 1. This indicates
2545 * directory has >EXT4_LINK_MAX subdirs.
2547 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2549 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2554 static int ext4_add_nondir(handle_t *handle,
2555 struct dentry *dentry, struct inode *inode)
2557 int err = ext4_add_entry(handle, dentry, inode);
2559 ext4_mark_inode_dirty(handle, inode);
2560 d_instantiate_new(dentry, inode);
2564 unlock_new_inode(inode);
2570 * By the time this is called, we already have created
2571 * the directory cache entry for the new file, but it
2572 * is so far negative - it has no inode.
2574 * If the create succeeds, we fill in the inode information
2575 * with d_instantiate().
2577 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2581 struct inode *inode;
2582 int err, credits, retries = 0;
2584 err = dquot_initialize(dir);
2588 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2589 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2591 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2592 NULL, EXT4_HT_DIR, credits);
2593 handle = ext4_journal_current_handle();
2594 err = PTR_ERR(inode);
2595 if (!IS_ERR(inode)) {
2596 inode->i_op = &ext4_file_inode_operations;
2597 inode->i_fop = &ext4_file_operations;
2598 ext4_set_aops(inode);
2599 err = ext4_add_nondir(handle, dentry, inode);
2600 if (!err && IS_DIRSYNC(dir))
2601 ext4_handle_sync(handle);
2604 ext4_journal_stop(handle);
2605 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2610 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2611 umode_t mode, dev_t rdev)
2614 struct inode *inode;
2615 int err, credits, retries = 0;
2617 err = dquot_initialize(dir);
2621 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2622 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2624 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2625 NULL, EXT4_HT_DIR, credits);
2626 handle = ext4_journal_current_handle();
2627 err = PTR_ERR(inode);
2628 if (!IS_ERR(inode)) {
2629 init_special_inode(inode, inode->i_mode, rdev);
2630 inode->i_op = &ext4_special_inode_operations;
2631 err = ext4_add_nondir(handle, dentry, inode);
2632 if (!err && IS_DIRSYNC(dir))
2633 ext4_handle_sync(handle);
2636 ext4_journal_stop(handle);
2637 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2642 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2645 struct inode *inode;
2646 int err, retries = 0;
2648 err = dquot_initialize(dir);
2653 inode = ext4_new_inode_start_handle(dir, mode,
2656 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2657 4 + EXT4_XATTR_TRANS_BLOCKS);
2658 handle = ext4_journal_current_handle();
2659 err = PTR_ERR(inode);
2660 if (!IS_ERR(inode)) {
2661 inode->i_op = &ext4_file_inode_operations;
2662 inode->i_fop = &ext4_file_operations;
2663 ext4_set_aops(inode);
2664 d_tmpfile(dentry, inode);
2665 err = ext4_orphan_add(handle, inode);
2667 goto err_unlock_inode;
2668 mark_inode_dirty(inode);
2669 unlock_new_inode(inode);
2672 ext4_journal_stop(handle);
2673 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2677 ext4_journal_stop(handle);
2678 unlock_new_inode(inode);
2682 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2683 struct ext4_dir_entry_2 *de,
2684 int blocksize, int csum_size,
2685 unsigned int parent_ino, int dotdot_real_len)
2687 de->inode = cpu_to_le32(inode->i_ino);
2689 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2691 strcpy(de->name, ".");
2692 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2694 de = ext4_next_entry(de, blocksize);
2695 de->inode = cpu_to_le32(parent_ino);
2697 if (!dotdot_real_len)
2698 de->rec_len = ext4_rec_len_to_disk(blocksize -
2699 (csum_size + EXT4_DIR_REC_LEN(1)),
2702 de->rec_len = ext4_rec_len_to_disk(
2703 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2704 strcpy(de->name, "..");
2705 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2707 return ext4_next_entry(de, blocksize);
2710 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2711 struct inode *inode)
2713 struct buffer_head *dir_block = NULL;
2714 struct ext4_dir_entry_2 *de;
2715 struct ext4_dir_entry_tail *t;
2716 ext4_lblk_t block = 0;
2717 unsigned int blocksize = dir->i_sb->s_blocksize;
2721 if (ext4_has_metadata_csum(dir->i_sb))
2722 csum_size = sizeof(struct ext4_dir_entry_tail);
2724 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2725 err = ext4_try_create_inline_dir(handle, dir, inode);
2726 if (err < 0 && err != -ENOSPC)
2733 dir_block = ext4_append(handle, inode, &block);
2734 if (IS_ERR(dir_block))
2735 return PTR_ERR(dir_block);
2736 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2737 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2738 set_nlink(inode, 2);
2740 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2741 initialize_dirent_tail(t, blocksize);
2744 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2745 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2748 set_buffer_verified(dir_block);
2754 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2757 struct inode *inode;
2758 int err, credits, retries = 0;
2760 if (EXT4_DIR_LINK_MAX(dir))
2763 err = dquot_initialize(dir);
2767 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2768 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2770 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2772 0, NULL, EXT4_HT_DIR, credits);
2773 handle = ext4_journal_current_handle();
2774 err = PTR_ERR(inode);
2778 inode->i_op = &ext4_dir_inode_operations;
2779 inode->i_fop = &ext4_dir_operations;
2780 err = ext4_init_new_dir(handle, dir, inode);
2782 goto out_clear_inode;
2783 err = ext4_mark_inode_dirty(handle, inode);
2785 err = ext4_add_entry(handle, dentry, inode);
2789 unlock_new_inode(inode);
2790 ext4_mark_inode_dirty(handle, inode);
2794 ext4_inc_count(handle, dir);
2795 ext4_update_dx_flag(dir);
2796 err = ext4_mark_inode_dirty(handle, dir);
2798 goto out_clear_inode;
2799 d_instantiate_new(dentry, inode);
2800 if (IS_DIRSYNC(dir))
2801 ext4_handle_sync(handle);
2805 ext4_journal_stop(handle);
2806 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2812 * routine to check that the specified directory is empty (for rmdir)
2814 bool ext4_empty_dir(struct inode *inode)
2816 unsigned int offset;
2817 struct buffer_head *bh;
2818 struct ext4_dir_entry_2 *de, *de1;
2819 struct super_block *sb;
2821 if (ext4_has_inline_data(inode)) {
2822 int has_inline_data = 1;
2825 ret = empty_inline_dir(inode, &has_inline_data);
2826 if (has_inline_data)
2831 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2832 EXT4_ERROR_INODE(inode, "invalid size");
2835 /* The first directory block must not be a hole,
2836 * so treat it as DIRENT_HTREE
2838 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2842 de = (struct ext4_dir_entry_2 *) bh->b_data;
2843 de1 = ext4_next_entry(de, sb->s_blocksize);
2844 if (le32_to_cpu(de->inode) != inode->i_ino ||
2845 le32_to_cpu(de1->inode) == 0 ||
2846 strcmp(".", de->name) || strcmp("..", de1->name)) {
2847 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2851 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2852 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2853 de = ext4_next_entry(de1, sb->s_blocksize);
2854 while (offset < inode->i_size) {
2855 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2856 unsigned int lblock;
2858 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2859 bh = ext4_read_dirblock(inode, lblock, EITHER);
2861 offset += sb->s_blocksize;
2866 de = (struct ext4_dir_entry_2 *) bh->b_data;
2868 if (ext4_check_dir_entry(inode, NULL, de, bh,
2869 bh->b_data, bh->b_size, offset)) {
2870 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2872 offset = (offset | (sb->s_blocksize - 1)) + 1;
2875 if (le32_to_cpu(de->inode)) {
2879 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2880 de = ext4_next_entry(de, sb->s_blocksize);
2887 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2888 * such inodes, starting at the superblock, in case we crash before the
2889 * file is closed/deleted, or in case the inode truncate spans multiple
2890 * transactions and the last transaction is not recovered after a crash.
2892 * At filesystem recovery time, we walk this list deleting unlinked
2893 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2895 * Orphan list manipulation functions must be called under i_mutex unless
2896 * we are just creating the inode or deleting it.
2898 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2900 struct super_block *sb = inode->i_sb;
2901 struct ext4_sb_info *sbi = EXT4_SB(sb);
2902 struct ext4_iloc iloc;
2906 if (!sbi->s_journal || is_bad_inode(inode))
2909 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2910 !inode_is_locked(inode));
2912 * Exit early if inode already is on orphan list. This is a big speedup
2913 * since we don't have to contend on the global s_orphan_lock.
2915 if (!list_empty(&EXT4_I(inode)->i_orphan))
2919 * Orphan handling is only valid for files with data blocks
2920 * being truncated, or files being unlinked. Note that we either
2921 * hold i_mutex, or the inode can not be referenced from outside,
2922 * so i_nlink should not be bumped due to race
2924 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2925 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2927 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2928 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2932 err = ext4_reserve_inode_write(handle, inode, &iloc);
2936 mutex_lock(&sbi->s_orphan_lock);
2938 * Due to previous errors inode may be already a part of on-disk
2939 * orphan list. If so skip on-disk list modification.
2941 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2942 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2943 /* Insert this inode at the head of the on-disk orphan list */
2944 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2945 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2948 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2949 mutex_unlock(&sbi->s_orphan_lock);
2952 err = ext4_handle_dirty_super(handle, sb);
2953 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2958 * We have to remove inode from in-memory list if
2959 * addition to on disk orphan list failed. Stray orphan
2960 * list entries can cause panics at unmount time.
2962 mutex_lock(&sbi->s_orphan_lock);
2963 list_del_init(&EXT4_I(inode)->i_orphan);
2964 mutex_unlock(&sbi->s_orphan_lock);
2969 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2970 jbd_debug(4, "orphan inode %lu will point to %d\n",
2971 inode->i_ino, NEXT_ORPHAN(inode));
2973 ext4_std_error(sb, err);
2978 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2979 * of such inodes stored on disk, because it is finally being cleaned up.
2981 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2983 struct list_head *prev;
2984 struct ext4_inode_info *ei = EXT4_I(inode);
2985 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2987 struct ext4_iloc iloc;
2990 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2993 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2994 !inode_is_locked(inode));
2995 /* Do this quick check before taking global s_orphan_lock. */
2996 if (list_empty(&ei->i_orphan))
3000 /* Grab inode buffer early before taking global s_orphan_lock */
3001 err = ext4_reserve_inode_write(handle, inode, &iloc);
3004 mutex_lock(&sbi->s_orphan_lock);
3005 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3007 prev = ei->i_orphan.prev;
3008 list_del_init(&ei->i_orphan);
3010 /* If we're on an error path, we may not have a valid
3011 * transaction handle with which to update the orphan list on
3012 * disk, but we still need to remove the inode from the linked
3013 * list in memory. */
3014 if (!handle || err) {
3015 mutex_unlock(&sbi->s_orphan_lock);
3019 ino_next = NEXT_ORPHAN(inode);
3020 if (prev == &sbi->s_orphan) {
3021 jbd_debug(4, "superblock will point to %u\n", ino_next);
3022 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3023 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3025 mutex_unlock(&sbi->s_orphan_lock);
3028 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3029 mutex_unlock(&sbi->s_orphan_lock);
3030 err = ext4_handle_dirty_super(handle, inode->i_sb);
3032 struct ext4_iloc iloc2;
3033 struct inode *i_prev =
3034 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3036 jbd_debug(4, "orphan inode %lu will point to %u\n",
3037 i_prev->i_ino, ino_next);
3038 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3040 mutex_unlock(&sbi->s_orphan_lock);
3043 NEXT_ORPHAN(i_prev) = ino_next;
3044 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3045 mutex_unlock(&sbi->s_orphan_lock);
3049 NEXT_ORPHAN(inode) = 0;
3050 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3052 ext4_std_error(inode->i_sb, err);
3060 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3063 struct inode *inode;
3064 struct buffer_head *bh;
3065 struct ext4_dir_entry_2 *de;
3066 handle_t *handle = NULL;
3068 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3071 /* Initialize quotas before so that eventual writes go in
3072 * separate transaction */
3073 retval = dquot_initialize(dir);
3076 retval = dquot_initialize(d_inode(dentry));
3081 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3087 inode = d_inode(dentry);
3089 retval = -EFSCORRUPTED;
3090 if (le32_to_cpu(de->inode) != inode->i_ino)
3093 retval = -ENOTEMPTY;
3094 if (!ext4_empty_dir(inode))
3097 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3098 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3099 if (IS_ERR(handle)) {
3100 retval = PTR_ERR(handle);
3105 if (IS_DIRSYNC(dir))
3106 ext4_handle_sync(handle);
3108 retval = ext4_delete_entry(handle, dir, de, bh);
3111 if (!EXT4_DIR_LINK_EMPTY(inode))
3112 ext4_warning_inode(inode,
3113 "empty directory '%.*s' has too many links (%u)",
3114 dentry->d_name.len, dentry->d_name.name,
3116 inode_inc_iversion(inode);
3118 /* There's no need to set i_disksize: the fact that i_nlink is
3119 * zero will ensure that the right thing happens during any
3122 ext4_orphan_add(handle, inode);
3123 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3124 ext4_mark_inode_dirty(handle, inode);
3125 ext4_dec_count(handle, dir);
3126 ext4_update_dx_flag(dir);
3127 ext4_mark_inode_dirty(handle, dir);
3129 #ifdef CONFIG_UNICODE
3130 /* VFS negative dentries are incompatible with Encoding and
3131 * Case-insensitiveness. Eventually we'll want avoid
3132 * invalidating the dentries here, alongside with returning the
3133 * negative dentries at ext4_lookup(), when it is better
3134 * supported by the VFS for the CI case.
3136 if (IS_CASEFOLDED(dir))
3137 d_invalidate(dentry);
3143 ext4_journal_stop(handle);
3147 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3150 struct inode *inode;
3151 struct buffer_head *bh;
3152 struct ext4_dir_entry_2 *de;
3153 handle_t *handle = NULL;
3155 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3158 trace_ext4_unlink_enter(dir, dentry);
3159 /* Initialize quotas before so that eventual writes go
3160 * in separate transaction */
3161 retval = dquot_initialize(dir);
3164 retval = dquot_initialize(d_inode(dentry));
3169 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3175 inode = d_inode(dentry);
3177 retval = -EFSCORRUPTED;
3178 if (le32_to_cpu(de->inode) != inode->i_ino)
3181 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3182 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3183 if (IS_ERR(handle)) {
3184 retval = PTR_ERR(handle);
3189 if (IS_DIRSYNC(dir))
3190 ext4_handle_sync(handle);
3192 if (inode->i_nlink == 0) {
3193 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3194 dentry->d_name.len, dentry->d_name.name);
3195 set_nlink(inode, 1);
3197 retval = ext4_delete_entry(handle, dir, de, bh);
3200 dir->i_ctime = dir->i_mtime = current_time(dir);
3201 ext4_update_dx_flag(dir);
3202 ext4_mark_inode_dirty(handle, dir);
3204 if (!inode->i_nlink)
3205 ext4_orphan_add(handle, inode);
3206 inode->i_ctime = current_time(inode);
3207 ext4_mark_inode_dirty(handle, inode);
3209 #ifdef CONFIG_UNICODE
3210 /* VFS negative dentries are incompatible with Encoding and
3211 * Case-insensitiveness. Eventually we'll want avoid
3212 * invalidating the dentries here, alongside with returning the
3213 * negative dentries at ext4_lookup(), when it is better
3214 * supported by the VFS for the CI case.
3216 if (IS_CASEFOLDED(dir))
3217 d_invalidate(dentry);
3223 ext4_journal_stop(handle);
3224 trace_ext4_unlink_exit(dentry, retval);
3228 static int ext4_symlink(struct inode *dir,
3229 struct dentry *dentry, const char *symname)
3232 struct inode *inode;
3233 int err, len = strlen(symname);
3235 struct fscrypt_str disk_link;
3237 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3240 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3245 err = dquot_initialize(dir);
3249 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3251 * For non-fast symlinks, we just allocate inode and put it on
3252 * orphan list in the first transaction => we need bitmap,
3253 * group descriptor, sb, inode block, quota blocks, and
3254 * possibly selinux xattr blocks.
3256 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3257 EXT4_XATTR_TRANS_BLOCKS;
3260 * Fast symlink. We have to add entry to directory
3261 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3262 * allocate new inode (bitmap, group descriptor, inode block,
3263 * quota blocks, sb is already counted in previous macros).
3265 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3266 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3269 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3270 &dentry->d_name, 0, NULL,
3271 EXT4_HT_DIR, credits);
3272 handle = ext4_journal_current_handle();
3273 if (IS_ERR(inode)) {
3275 ext4_journal_stop(handle);
3276 return PTR_ERR(inode);
3279 if (IS_ENCRYPTED(inode)) {
3280 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3282 goto err_drop_inode;
3283 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3286 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3287 if (!IS_ENCRYPTED(inode))
3288 inode->i_op = &ext4_symlink_inode_operations;
3289 inode_nohighmem(inode);
3290 ext4_set_aops(inode);
3292 * We cannot call page_symlink() with transaction started
3293 * because it calls into ext4_write_begin() which can wait
3294 * for transaction commit if we are running out of space
3295 * and thus we deadlock. So we have to stop transaction now
3296 * and restart it when symlink contents is written.
3298 * To keep fs consistent in case of crash, we have to put inode
3299 * to orphan list in the mean time.
3302 err = ext4_orphan_add(handle, inode);
3303 ext4_journal_stop(handle);
3306 goto err_drop_inode;
3307 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3309 goto err_drop_inode;
3311 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3312 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3314 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3315 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3316 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3317 if (IS_ERR(handle)) {
3318 err = PTR_ERR(handle);
3320 goto err_drop_inode;
3322 set_nlink(inode, 1);
3323 err = ext4_orphan_del(handle, inode);
3325 goto err_drop_inode;
3327 /* clear the extent format for fast symlink */
3328 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3329 if (!IS_ENCRYPTED(inode)) {
3330 inode->i_op = &ext4_fast_symlink_inode_operations;
3331 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3333 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3335 inode->i_size = disk_link.len - 1;
3337 EXT4_I(inode)->i_disksize = inode->i_size;
3338 err = ext4_add_nondir(handle, dentry, inode);
3339 if (!err && IS_DIRSYNC(dir))
3340 ext4_handle_sync(handle);
3343 ext4_journal_stop(handle);
3344 goto out_free_encrypted_link;
3348 ext4_journal_stop(handle);
3350 unlock_new_inode(inode);
3352 out_free_encrypted_link:
3353 if (disk_link.name != (unsigned char *)symname)
3354 kfree(disk_link.name);
3358 static int ext4_link(struct dentry *old_dentry,
3359 struct inode *dir, struct dentry *dentry)
3362 struct inode *inode = d_inode(old_dentry);
3363 int err, retries = 0;
3365 if (inode->i_nlink >= EXT4_LINK_MAX)
3368 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3372 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3373 (!projid_eq(EXT4_I(dir)->i_projid,
3374 EXT4_I(old_dentry->d_inode)->i_projid)))
3377 err = dquot_initialize(dir);
3382 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3383 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3384 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3386 return PTR_ERR(handle);
3388 if (IS_DIRSYNC(dir))
3389 ext4_handle_sync(handle);
3391 inode->i_ctime = current_time(inode);
3392 ext4_inc_count(handle, inode);
3395 err = ext4_add_entry(handle, dentry, inode);
3397 ext4_mark_inode_dirty(handle, inode);
3398 /* this can happen only for tmpfile being
3399 * linked the first time
3401 if (inode->i_nlink == 1)
3402 ext4_orphan_del(handle, inode);
3403 d_instantiate(dentry, inode);
3408 ext4_journal_stop(handle);
3409 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3416 * Try to find buffer head where contains the parent block.
3417 * It should be the inode block if it is inlined or the 1st block
3418 * if it is a normal dir.
3420 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3421 struct inode *inode,
3423 struct ext4_dir_entry_2 **parent_de,
3426 struct buffer_head *bh;
3428 if (!ext4_has_inline_data(inode)) {
3429 /* The first directory block must not be a hole, so
3430 * treat it as DIRENT_HTREE
3432 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3434 *retval = PTR_ERR(bh);
3437 *parent_de = ext4_next_entry(
3438 (struct ext4_dir_entry_2 *)bh->b_data,
3439 inode->i_sb->s_blocksize);
3444 return ext4_get_first_inline_block(inode, parent_de, retval);
3447 struct ext4_renament {
3449 struct dentry *dentry;
3450 struct inode *inode;
3452 int dir_nlink_delta;
3454 /* entry for "dentry" */
3455 struct buffer_head *bh;
3456 struct ext4_dir_entry_2 *de;
3459 /* entry for ".." in inode if it's a directory */
3460 struct buffer_head *dir_bh;
3461 struct ext4_dir_entry_2 *parent_de;
3465 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3469 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3470 &retval, &ent->parent_de,
3474 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3475 return -EFSCORRUPTED;
3476 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3477 return ext4_journal_get_write_access(handle, ent->dir_bh);
3480 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3485 ent->parent_de->inode = cpu_to_le32(dir_ino);
3486 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3487 if (!ent->dir_inlined) {
3488 if (is_dx(ent->inode)) {
3489 retval = ext4_handle_dirty_dx_node(handle,
3493 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3497 retval = ext4_mark_inode_dirty(handle, ent->inode);
3500 ext4_std_error(ent->dir->i_sb, retval);
3506 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3507 unsigned ino, unsigned file_type)
3511 BUFFER_TRACE(ent->bh, "get write access");
3512 retval = ext4_journal_get_write_access(handle, ent->bh);
3515 ent->de->inode = cpu_to_le32(ino);
3516 if (ext4_has_feature_filetype(ent->dir->i_sb))
3517 ent->de->file_type = file_type;
3518 inode_inc_iversion(ent->dir);
3519 ent->dir->i_ctime = ent->dir->i_mtime =
3520 current_time(ent->dir);
3521 ext4_mark_inode_dirty(handle, ent->dir);
3522 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3523 if (!ent->inlined) {
3524 retval = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3525 if (unlikely(retval)) {
3526 ext4_std_error(ent->dir->i_sb, retval);
3536 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3537 const struct qstr *d_name)
3539 int retval = -ENOENT;
3540 struct buffer_head *bh;
3541 struct ext4_dir_entry_2 *de;
3543 bh = ext4_find_entry(dir, d_name, &de, NULL);
3547 retval = ext4_delete_entry(handle, dir, de, bh);
3553 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3558 * ent->de could have moved from under us during htree split, so make
3559 * sure that we are deleting the right entry. We might also be pointing
3560 * to a stale entry in the unused part of ent->bh so just checking inum
3561 * and the name isn't enough.
3563 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3564 ent->de->name_len != ent->dentry->d_name.len ||
3565 strncmp(ent->de->name, ent->dentry->d_name.name,
3566 ent->de->name_len) ||
3568 retval = ext4_find_delete_entry(handle, ent->dir,
3569 &ent->dentry->d_name);
3571 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3572 if (retval == -ENOENT) {
3573 retval = ext4_find_delete_entry(handle, ent->dir,
3574 &ent->dentry->d_name);
3579 ext4_warning_inode(ent->dir,
3580 "Deleting old file: nlink %d, error=%d",
3581 ent->dir->i_nlink, retval);
3585 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3587 if (ent->dir_nlink_delta) {
3588 if (ent->dir_nlink_delta == -1)
3589 ext4_dec_count(handle, ent->dir);
3591 ext4_inc_count(handle, ent->dir);
3592 ext4_mark_inode_dirty(handle, ent->dir);
3596 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3597 int credits, handle_t **h)
3604 * for inode block, sb block, group summaries,
3607 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3608 EXT4_XATTR_TRANS_BLOCKS + 4);
3610 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3611 &ent->dentry->d_name, 0, NULL,
3612 EXT4_HT_DIR, credits);
3614 handle = ext4_journal_current_handle();
3617 ext4_journal_stop(handle);
3618 if (PTR_ERR(wh) == -ENOSPC &&
3619 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3623 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3624 wh->i_op = &ext4_special_inode_operations;
3630 * Anybody can rename anything with this: the permission checks are left to the
3631 * higher-level routines.
3633 * n.b. old_{dentry,inode) refers to the source dentry/inode
3634 * while new_{dentry,inode) refers to the destination dentry/inode
3635 * This comes from rename(const char *oldpath, const char *newpath)
3637 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3638 struct inode *new_dir, struct dentry *new_dentry,
3641 handle_t *handle = NULL;
3642 struct ext4_renament old = {
3644 .dentry = old_dentry,
3645 .inode = d_inode(old_dentry),
3647 struct ext4_renament new = {
3649 .dentry = new_dentry,
3650 .inode = d_inode(new_dentry),
3654 struct inode *whiteout = NULL;
3658 if (new.inode && new.inode->i_nlink == 0) {
3659 EXT4_ERROR_INODE(new.inode,
3660 "target of rename is already freed");
3661 return -EFSCORRUPTED;
3664 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3665 (!projid_eq(EXT4_I(new_dir)->i_projid,
3666 EXT4_I(old_dentry->d_inode)->i_projid)))
3669 retval = dquot_initialize(old.dir);
3672 retval = dquot_initialize(new.dir);
3676 /* Initialize quotas before so that eventual writes go
3677 * in separate transaction */
3679 retval = dquot_initialize(new.inode);
3684 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3686 return PTR_ERR(old.bh);
3688 * Check for inode number is _not_ due to possible IO errors.
3689 * We might rmdir the source, keep it as pwd of some process
3690 * and merrily kill the link to whatever was created under the
3691 * same name. Goodbye sticky bit ;-<
3694 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3697 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3698 &new.de, &new.inlined);
3699 if (IS_ERR(new.bh)) {
3700 retval = PTR_ERR(new.bh);
3710 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3711 ext4_alloc_da_blocks(old.inode);
3713 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3714 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3715 if (!(flags & RENAME_WHITEOUT)) {
3716 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3717 if (IS_ERR(handle)) {
3718 retval = PTR_ERR(handle);
3723 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3724 if (IS_ERR(whiteout)) {
3725 retval = PTR_ERR(whiteout);
3731 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3732 ext4_handle_sync(handle);
3734 if (S_ISDIR(old.inode->i_mode)) {
3736 retval = -ENOTEMPTY;
3737 if (!ext4_empty_dir(new.inode))
3741 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3744 retval = ext4_rename_dir_prepare(handle, &old);
3749 * If we're renaming a file within an inline_data dir and adding or
3750 * setting the new dirent causes a conversion from inline_data to
3751 * extents/blockmap, we need to force the dirent delete code to
3752 * re-read the directory, or else we end up trying to delete a dirent
3753 * from what is now the extent tree root (or a block map).
3755 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3756 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3758 old_file_type = old.de->file_type;
3761 * Do this before adding a new entry, so the old entry is sure
3762 * to be still pointing to the valid old entry.
3764 retval = ext4_setent(handle, &old, whiteout->i_ino,
3768 ext4_mark_inode_dirty(handle, whiteout);
3771 retval = ext4_add_entry(handle, new.dentry, old.inode);
3775 retval = ext4_setent(handle, &new,
3776 old.inode->i_ino, old_file_type);
3781 force_reread = !ext4_test_inode_flag(new.dir,
3782 EXT4_INODE_INLINE_DATA);
3785 * Like most other Unix systems, set the ctime for inodes on a
3788 old.inode->i_ctime = current_time(old.inode);
3789 ext4_mark_inode_dirty(handle, old.inode);
3795 ext4_rename_delete(handle, &old, force_reread);
3799 ext4_dec_count(handle, new.inode);
3800 new.inode->i_ctime = current_time(new.inode);
3802 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3803 ext4_update_dx_flag(old.dir);
3805 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3809 ext4_dec_count(handle, old.dir);
3811 /* checked ext4_empty_dir above, can't have another
3812 * parent, ext4_dec_count() won't work for many-linked
3814 clear_nlink(new.inode);
3816 ext4_inc_count(handle, new.dir);
3817 ext4_update_dx_flag(new.dir);
3818 ext4_mark_inode_dirty(handle, new.dir);
3821 ext4_mark_inode_dirty(handle, old.dir);
3823 ext4_mark_inode_dirty(handle, new.inode);
3824 if (!new.inode->i_nlink)
3825 ext4_orphan_add(handle, new.inode);
3835 drop_nlink(whiteout);
3836 unlock_new_inode(whiteout);
3840 ext4_journal_stop(handle);
3844 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3845 struct inode *new_dir, struct dentry *new_dentry)
3847 handle_t *handle = NULL;
3848 struct ext4_renament old = {
3850 .dentry = old_dentry,
3851 .inode = d_inode(old_dentry),
3853 struct ext4_renament new = {
3855 .dentry = new_dentry,
3856 .inode = d_inode(new_dentry),
3860 struct timespec64 ctime;
3862 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3863 !projid_eq(EXT4_I(new_dir)->i_projid,
3864 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3865 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3866 !projid_eq(EXT4_I(old_dir)->i_projid,
3867 EXT4_I(new_dentry->d_inode)->i_projid)))
3870 retval = dquot_initialize(old.dir);
3873 retval = dquot_initialize(new.dir);
3877 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3878 &old.de, &old.inlined);
3880 return PTR_ERR(old.bh);
3882 * Check for inode number is _not_ due to possible IO errors.
3883 * We might rmdir the source, keep it as pwd of some process
3884 * and merrily kill the link to whatever was created under the
3885 * same name. Goodbye sticky bit ;-<
3888 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3891 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3892 &new.de, &new.inlined);
3893 if (IS_ERR(new.bh)) {
3894 retval = PTR_ERR(new.bh);
3899 /* RENAME_EXCHANGE case: old *and* new must both exist */
3900 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3903 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3904 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3905 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3906 if (IS_ERR(handle)) {
3907 retval = PTR_ERR(handle);
3912 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3913 ext4_handle_sync(handle);
3915 if (S_ISDIR(old.inode->i_mode)) {
3917 retval = ext4_rename_dir_prepare(handle, &old);
3921 if (S_ISDIR(new.inode->i_mode)) {
3923 retval = ext4_rename_dir_prepare(handle, &new);
3929 * Other than the special case of overwriting a directory, parents'
3930 * nlink only needs to be modified if this is a cross directory rename.
3932 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3933 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3934 new.dir_nlink_delta = -old.dir_nlink_delta;
3936 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3937 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3941 new_file_type = new.de->file_type;
3942 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3946 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3951 * Like most other Unix systems, set the ctime for inodes on a
3954 ctime = current_time(old.inode);
3955 old.inode->i_ctime = ctime;
3956 new.inode->i_ctime = ctime;
3957 ext4_mark_inode_dirty(handle, old.inode);
3958 ext4_mark_inode_dirty(handle, new.inode);
3961 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3966 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3970 ext4_update_dir_count(handle, &old);
3971 ext4_update_dir_count(handle, &new);
3980 ext4_journal_stop(handle);
3984 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3985 struct inode *new_dir, struct dentry *new_dentry,
3990 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3993 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3996 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4001 if (flags & RENAME_EXCHANGE) {
4002 return ext4_cross_rename(old_dir, old_dentry,
4003 new_dir, new_dentry);
4006 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4010 * directories can handle most operations...
4012 const struct inode_operations ext4_dir_inode_operations = {
4013 .create = ext4_create,
4014 .lookup = ext4_lookup,
4016 .unlink = ext4_unlink,
4017 .symlink = ext4_symlink,
4018 .mkdir = ext4_mkdir,
4019 .rmdir = ext4_rmdir,
4020 .mknod = ext4_mknod,
4021 .tmpfile = ext4_tmpfile,
4022 .rename = ext4_rename2,
4023 .setattr = ext4_setattr,
4024 .getattr = ext4_getattr,
4025 .listxattr = ext4_listxattr,
4026 .get_acl = ext4_get_acl,
4027 .set_acl = ext4_set_acl,
4028 .fiemap = ext4_fiemap,
4031 const struct inode_operations ext4_special_inode_operations = {
4032 .setattr = ext4_setattr,
4033 .getattr = ext4_getattr,
4034 .listxattr = ext4_listxattr,
4035 .get_acl = ext4_get_acl,
4036 .set_acl = ext4_set_acl,