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 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
115 bh = ext4_bread(NULL, inode, block, 0);
117 __ext4_warning(inode->i_sb, func, line,
118 "inode #%lu: lblock %lu: comm %s: "
119 "error %ld reading directory block",
120 inode->i_ino, (unsigned long)block,
121 current->comm, PTR_ERR(bh));
125 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
126 ext4_error_inode(inode, func, line, block,
127 "Directory hole found for htree %s block",
128 (type == INDEX) ? "index" : "leaf");
129 return ERR_PTR(-EFSCORRUPTED);
133 dirent = (struct ext4_dir_entry *) bh->b_data;
134 /* Determine whether or not we have an index block */
138 else if (ext4_rec_len_from_disk(dirent->rec_len,
139 inode->i_sb->s_blocksize) ==
140 inode->i_sb->s_blocksize)
143 if (!is_dx_block && type == INDEX) {
144 ext4_error_inode(inode, func, line, block,
145 "directory leaf block found instead of index block");
147 return ERR_PTR(-EFSCORRUPTED);
149 if (!ext4_has_metadata_csum(inode->i_sb) ||
154 * An empty leaf block can get mistaken for a index block; for
155 * this reason, we can only check the index checksum when the
156 * caller is sure it should be an index block.
158 if (is_dx_block && type == INDEX) {
159 if (ext4_dx_csum_verify(inode, dirent) &&
160 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
161 set_buffer_verified(bh);
163 ext4_error_inode_err(inode, func, line, block,
165 "Directory index failed checksum");
167 return ERR_PTR(-EFSBADCRC);
171 if (ext4_dirblock_csum_verify(inode, bh) &&
172 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
173 set_buffer_verified(bh);
175 ext4_error_inode_err(inode, func, line, block,
177 "Directory block failed checksum");
179 return ERR_PTR(-EFSBADCRC);
186 #define assert(test) J_ASSERT(test)
190 #define dxtrace(command) command
192 #define dxtrace(command)
216 * dx_root_info is laid out so that if it should somehow get overlaid by a
217 * dirent the two low bits of the hash version will be zero. Therefore, the
218 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
223 struct fake_dirent dot;
225 struct fake_dirent dotdot;
229 __le32 reserved_zero;
231 u8 info_length; /* 8 */
236 struct dx_entry entries[];
241 struct fake_dirent fake;
242 struct dx_entry entries[];
248 struct buffer_head *bh;
249 struct dx_entry *entries;
261 * This goes at the end of each htree block.
265 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
268 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
269 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
270 static inline unsigned dx_get_hash(struct dx_entry *entry);
271 static void dx_set_hash(struct dx_entry *entry, unsigned value);
272 static unsigned dx_get_count(struct dx_entry *entries);
273 static unsigned dx_get_limit(struct dx_entry *entries);
274 static void dx_set_count(struct dx_entry *entries, unsigned value);
275 static void dx_set_limit(struct dx_entry *entries, unsigned value);
276 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
277 static unsigned dx_node_limit(struct inode *dir);
278 static struct dx_frame *dx_probe(struct ext4_filename *fname,
280 struct dx_hash_info *hinfo,
281 struct dx_frame *frame);
282 static void dx_release(struct dx_frame *frames);
283 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
284 unsigned blocksize, struct dx_hash_info *hinfo,
285 struct dx_map_entry map[]);
286 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
287 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
288 struct dx_map_entry *offsets, int count, unsigned blocksize);
289 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
290 static void dx_insert_block(struct dx_frame *frame,
291 u32 hash, ext4_lblk_t block);
292 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
293 struct dx_frame *frame,
294 struct dx_frame *frames,
296 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
297 struct ext4_filename *fname,
298 struct ext4_dir_entry_2 **res_dir);
299 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
300 struct inode *dir, struct inode *inode);
302 /* checksumming functions */
303 void ext4_initialize_dirent_tail(struct buffer_head *bh,
304 unsigned int blocksize)
306 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
308 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
309 t->det_rec_len = ext4_rec_len_to_disk(
310 sizeof(struct ext4_dir_entry_tail), blocksize);
311 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
314 /* Walk through a dirent block to find a checksum "dirent" at the tail */
315 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
316 struct buffer_head *bh)
318 struct ext4_dir_entry_tail *t;
321 struct ext4_dir_entry *d, *top;
323 d = (struct ext4_dir_entry *)bh->b_data;
324 top = (struct ext4_dir_entry *)(bh->b_data +
325 (EXT4_BLOCK_SIZE(inode->i_sb) -
326 sizeof(struct ext4_dir_entry_tail)));
327 while (d < top && d->rec_len)
328 d = (struct ext4_dir_entry *)(((void *)d) +
329 le16_to_cpu(d->rec_len));
334 t = (struct ext4_dir_entry_tail *)d;
336 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
339 if (t->det_reserved_zero1 ||
340 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
341 t->det_reserved_zero2 ||
342 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
348 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
350 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
351 struct ext4_inode_info *ei = EXT4_I(inode);
354 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
355 return cpu_to_le32(csum);
358 #define warn_no_space_for_csum(inode) \
359 __warn_no_space_for_csum((inode), __func__, __LINE__)
361 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
364 __ext4_warning_inode(inode, func, line,
365 "No space for directory leaf checksum. Please run e2fsck -D.");
368 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
370 struct ext4_dir_entry_tail *t;
372 if (!ext4_has_metadata_csum(inode->i_sb))
375 t = get_dirent_tail(inode, bh);
377 warn_no_space_for_csum(inode);
381 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
382 (char *)t - bh->b_data))
388 static void ext4_dirblock_csum_set(struct inode *inode,
389 struct buffer_head *bh)
391 struct ext4_dir_entry_tail *t;
393 if (!ext4_has_metadata_csum(inode->i_sb))
396 t = get_dirent_tail(inode, bh);
398 warn_no_space_for_csum(inode);
402 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
403 (char *)t - bh->b_data);
406 int ext4_handle_dirty_dirblock(handle_t *handle,
408 struct buffer_head *bh)
410 ext4_dirblock_csum_set(inode, bh);
411 return ext4_handle_dirty_metadata(handle, inode, bh);
414 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
415 struct ext4_dir_entry *dirent,
418 struct ext4_dir_entry *dp;
419 struct dx_root_info *root;
422 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
424 else if (le16_to_cpu(dirent->rec_len) == 12) {
425 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
426 if (le16_to_cpu(dp->rec_len) !=
427 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
429 root = (struct dx_root_info *)(((void *)dp + 12));
430 if (root->reserved_zero ||
431 root->info_length != sizeof(struct dx_root_info))
438 *offset = count_offset;
439 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
442 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
443 int count_offset, int count, struct dx_tail *t)
445 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
446 struct ext4_inode_info *ei = EXT4_I(inode);
449 __u32 dummy_csum = 0;
450 int offset = offsetof(struct dx_tail, dt_checksum);
452 size = count_offset + (count * sizeof(struct dx_entry));
453 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
454 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
455 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
457 return cpu_to_le32(csum);
460 static int ext4_dx_csum_verify(struct inode *inode,
461 struct ext4_dir_entry *dirent)
463 struct dx_countlimit *c;
465 int count_offset, limit, count;
467 if (!ext4_has_metadata_csum(inode->i_sb))
470 c = get_dx_countlimit(inode, dirent, &count_offset);
472 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
475 limit = le16_to_cpu(c->limit);
476 count = le16_to_cpu(c->count);
477 if (count_offset + (limit * sizeof(struct dx_entry)) >
478 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
479 warn_no_space_for_csum(inode);
482 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
484 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
490 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
492 struct dx_countlimit *c;
494 int count_offset, limit, count;
496 if (!ext4_has_metadata_csum(inode->i_sb))
499 c = get_dx_countlimit(inode, dirent, &count_offset);
501 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
504 limit = le16_to_cpu(c->limit);
505 count = le16_to_cpu(c->count);
506 if (count_offset + (limit * sizeof(struct dx_entry)) >
507 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
508 warn_no_space_for_csum(inode);
511 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
513 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
516 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
518 struct buffer_head *bh)
520 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
521 return ext4_handle_dirty_metadata(handle, inode, bh);
525 * p is at least 6 bytes before the end of page
527 static inline struct ext4_dir_entry_2 *
528 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
530 return (struct ext4_dir_entry_2 *)((char *)p +
531 ext4_rec_len_from_disk(p->rec_len, blocksize));
535 * Future: use high four bits of block for coalesce-on-delete flags
536 * Mask them off for now.
539 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
541 return le32_to_cpu(entry->block) & 0x0fffffff;
544 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
546 entry->block = cpu_to_le32(value);
549 static inline unsigned dx_get_hash(struct dx_entry *entry)
551 return le32_to_cpu(entry->hash);
554 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
556 entry->hash = cpu_to_le32(value);
559 static inline unsigned dx_get_count(struct dx_entry *entries)
561 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
564 static inline unsigned dx_get_limit(struct dx_entry *entries)
566 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
569 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
571 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
574 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
576 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
579 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
581 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
582 EXT4_DIR_REC_LEN(2) - infosize;
584 if (ext4_has_metadata_csum(dir->i_sb))
585 entry_space -= sizeof(struct dx_tail);
586 return entry_space / sizeof(struct dx_entry);
589 static inline unsigned dx_node_limit(struct inode *dir)
591 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
593 if (ext4_has_metadata_csum(dir->i_sb))
594 entry_space -= sizeof(struct dx_tail);
595 return entry_space / sizeof(struct dx_entry);
602 static void dx_show_index(char * label, struct dx_entry *entries)
604 int i, n = dx_get_count (entries);
605 printk(KERN_DEBUG "%s index", label);
606 for (i = 0; i < n; i++) {
607 printk(KERN_CONT " %x->%lu",
608 i ? dx_get_hash(entries + i) : 0,
609 (unsigned long)dx_get_block(entries + i));
611 printk(KERN_CONT "\n");
621 static struct stats dx_show_leaf(struct inode *dir,
622 struct dx_hash_info *hinfo,
623 struct ext4_dir_entry_2 *de,
624 int size, int show_names)
626 unsigned names = 0, space = 0;
627 char *base = (char *) de;
628 struct dx_hash_info h = *hinfo;
631 while ((char *) de < base + size)
637 #ifdef CONFIG_FS_ENCRYPTION
640 struct fscrypt_str fname_crypto_str =
646 if (!IS_ENCRYPTED(dir)) {
647 /* Directory is not encrypted */
648 ext4fs_dirhash(dir, de->name,
650 printk("%*.s:(U)%x.%u ", len,
652 (unsigned) ((char *) de
655 struct fscrypt_str de_name =
656 FSTR_INIT(name, len);
658 /* Directory is encrypted */
659 res = fscrypt_fname_alloc_buffer(
660 len, &fname_crypto_str);
662 printk(KERN_WARNING "Error "
666 res = fscrypt_fname_disk_to_usr(dir,
670 printk(KERN_WARNING "Error "
671 "converting filename "
677 name = fname_crypto_str.name;
678 len = fname_crypto_str.len;
680 ext4fs_dirhash(dir, de->name,
682 printk("%*.s:(E)%x.%u ", len, name,
683 h.hash, (unsigned) ((char *) de
685 fscrypt_fname_free_buffer(
689 int len = de->name_len;
690 char *name = de->name;
691 ext4fs_dirhash(dir, de->name, de->name_len, &h);
692 printk("%*.s:%x.%u ", len, name, h.hash,
693 (unsigned) ((char *) de - base));
696 space += EXT4_DIR_REC_LEN(de->name_len);
699 de = ext4_next_entry(de, size);
701 printk(KERN_CONT "(%i)\n", names);
702 return (struct stats) { names, space, 1 };
705 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
706 struct dx_entry *entries, int levels)
708 unsigned blocksize = dir->i_sb->s_blocksize;
709 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
711 struct buffer_head *bh;
712 printk("%i indexed blocks...\n", count);
713 for (i = 0; i < count; i++, entries++)
715 ext4_lblk_t block = dx_get_block(entries);
716 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
717 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
719 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
720 bh = ext4_bread(NULL,dir, block, 0);
721 if (!bh || IS_ERR(bh))
724 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
725 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
726 bh->b_data, blocksize, 0);
727 names += stats.names;
728 space += stats.space;
729 bcount += stats.bcount;
733 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
734 levels ? "" : " ", names, space/bcount,
735 (space/bcount)*100/blocksize);
736 return (struct stats) { names, space, bcount};
738 #endif /* DX_DEBUG */
741 * Probe for a directory leaf block to search.
743 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
744 * error in the directory index, and the caller should fall back to
745 * searching the directory normally. The callers of dx_probe **MUST**
746 * check for this error code, and make sure it never gets reflected
749 static struct dx_frame *
750 dx_probe(struct ext4_filename *fname, struct inode *dir,
751 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
753 unsigned count, indirect;
754 struct dx_entry *at, *entries, *p, *q, *m;
755 struct dx_root *root;
756 struct dx_frame *frame = frame_in;
757 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
760 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
761 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
762 if (IS_ERR(frame->bh))
763 return (struct dx_frame *) frame->bh;
765 root = (struct dx_root *) frame->bh->b_data;
766 if (root->info.hash_version != DX_HASH_TEA &&
767 root->info.hash_version != DX_HASH_HALF_MD4 &&
768 root->info.hash_version != DX_HASH_LEGACY) {
769 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
770 root->info.hash_version);
774 hinfo = &fname->hinfo;
775 hinfo->hash_version = root->info.hash_version;
776 if (hinfo->hash_version <= DX_HASH_TEA)
777 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
778 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
779 if (fname && fname_name(fname))
780 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
783 if (root->info.unused_flags & 1) {
784 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
785 root->info.unused_flags);
789 indirect = root->info.indirect_levels;
790 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
791 ext4_warning(dir->i_sb,
792 "Directory (ino: %lu) htree depth %#06x exceed"
793 "supported value", dir->i_ino,
794 ext4_dir_htree_level(dir->i_sb));
795 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
796 ext4_warning(dir->i_sb, "Enable large directory "
797 "feature to access it");
802 entries = (struct dx_entry *)(((char *)&root->info) +
803 root->info.info_length);
805 if (dx_get_limit(entries) != dx_root_limit(dir,
806 root->info.info_length)) {
807 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
808 dx_get_limit(entries),
809 dx_root_limit(dir, root->info.info_length));
813 dxtrace(printk("Look up %x", hash));
815 count = dx_get_count(entries);
816 if (!count || count > dx_get_limit(entries)) {
817 ext4_warning_inode(dir,
818 "dx entry: count %u beyond limit %u",
819 count, dx_get_limit(entries));
824 q = entries + count - 1;
827 dxtrace(printk(KERN_CONT "."));
828 if (dx_get_hash(m) > hash)
834 if (0) { // linear search cross check
835 unsigned n = count - 1;
839 dxtrace(printk(KERN_CONT ","));
840 if (dx_get_hash(++at) > hash)
846 assert (at == p - 1);
850 dxtrace(printk(KERN_CONT " %x->%u\n",
851 at == entries ? 0 : dx_get_hash(at),
853 frame->entries = entries;
858 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
859 if (IS_ERR(frame->bh)) {
860 ret_err = (struct dx_frame *) frame->bh;
864 entries = ((struct dx_node *) frame->bh->b_data)->entries;
866 if (dx_get_limit(entries) != dx_node_limit(dir)) {
867 ext4_warning_inode(dir,
868 "dx entry: limit %u != node limit %u",
869 dx_get_limit(entries), dx_node_limit(dir));
874 while (frame >= frame_in) {
879 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
880 ext4_warning_inode(dir,
881 "Corrupt directory, running e2fsck is recommended");
885 static void dx_release(struct dx_frame *frames)
887 struct dx_root_info *info;
889 unsigned int indirect_levels;
891 if (frames[0].bh == NULL)
894 info = &((struct dx_root *)frames[0].bh->b_data)->info;
895 /* save local copy, "info" may be freed after brelse() */
896 indirect_levels = info->indirect_levels;
897 for (i = 0; i <= indirect_levels; i++) {
898 if (frames[i].bh == NULL)
900 brelse(frames[i].bh);
906 * This function increments the frame pointer to search the next leaf
907 * block, and reads in the necessary intervening nodes if the search
908 * should be necessary. Whether or not the search is necessary is
909 * controlled by the hash parameter. If the hash value is even, then
910 * the search is only continued if the next block starts with that
911 * hash value. This is used if we are searching for a specific file.
913 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
915 * This function returns 1 if the caller should continue to search,
916 * or 0 if it should not. If there is an error reading one of the
917 * index blocks, it will a negative error code.
919 * If start_hash is non-null, it will be filled in with the starting
920 * hash of the next page.
922 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
923 struct dx_frame *frame,
924 struct dx_frame *frames,
928 struct buffer_head *bh;
934 * Find the next leaf page by incrementing the frame pointer.
935 * If we run out of entries in the interior node, loop around and
936 * increment pointer in the parent node. When we break out of
937 * this loop, num_frames indicates the number of interior
938 * nodes need to be read.
941 if (++(p->at) < p->entries + dx_get_count(p->entries))
950 * If the hash is 1, then continue only if the next page has a
951 * continuation hash of any value. This is used for readdir
952 * handling. Otherwise, check to see if the hash matches the
953 * desired contiuation hash. If it doesn't, return since
954 * there's no point to read in the successive index pages.
956 bhash = dx_get_hash(p->at);
959 if ((hash & 1) == 0) {
960 if ((bhash & ~1) != hash)
964 * If the hash is HASH_NB_ALWAYS, we always go to the next
965 * block so no check is necessary
967 while (num_frames--) {
968 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
974 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
981 * This function fills a red-black tree with information from a
982 * directory block. It returns the number directory entries loaded
983 * into the tree. If there is an error it is returned in err.
985 static int htree_dirblock_to_tree(struct file *dir_file,
986 struct inode *dir, ext4_lblk_t block,
987 struct dx_hash_info *hinfo,
988 __u32 start_hash, __u32 start_minor_hash)
990 struct buffer_head *bh;
991 struct ext4_dir_entry_2 *de, *top;
992 int err = 0, count = 0;
993 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
995 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
996 (unsigned long)block));
997 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1001 de = (struct ext4_dir_entry_2 *) bh->b_data;
1002 top = (struct ext4_dir_entry_2 *) ((char *) de +
1003 dir->i_sb->s_blocksize -
1004 EXT4_DIR_REC_LEN(0));
1005 /* Check if the directory is encrypted */
1006 if (IS_ENCRYPTED(dir)) {
1007 err = fscrypt_prepare_readdir(dir);
1012 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1020 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1021 if (ext4_check_dir_entry(dir, NULL, de, bh,
1022 bh->b_data, bh->b_size,
1023 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1024 + ((char *)de - bh->b_data))) {
1025 /* silently ignore the rest of the block */
1028 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1029 if ((hinfo->hash < start_hash) ||
1030 ((hinfo->hash == start_hash) &&
1031 (hinfo->minor_hash < start_minor_hash)))
1035 if (!IS_ENCRYPTED(dir)) {
1036 tmp_str.name = de->name;
1037 tmp_str.len = de->name_len;
1038 err = ext4_htree_store_dirent(dir_file,
1039 hinfo->hash, hinfo->minor_hash, de,
1042 int save_len = fname_crypto_str.len;
1043 struct fscrypt_str de_name = FSTR_INIT(de->name,
1046 /* Directory is encrypted */
1047 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1048 hinfo->minor_hash, &de_name,
1054 err = ext4_htree_store_dirent(dir_file,
1055 hinfo->hash, hinfo->minor_hash, de,
1057 fname_crypto_str.len = save_len;
1067 fscrypt_fname_free_buffer(&fname_crypto_str);
1073 * This function fills a red-black tree with information from a
1074 * directory. We start scanning the directory in hash order, starting
1075 * at start_hash and start_minor_hash.
1077 * This function returns the number of entries inserted into the tree,
1078 * or a negative error code.
1080 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1081 __u32 start_minor_hash, __u32 *next_hash)
1083 struct dx_hash_info hinfo;
1084 struct ext4_dir_entry_2 *de;
1085 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1091 struct fscrypt_str tmp_str;
1093 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1094 start_hash, start_minor_hash));
1095 dir = file_inode(dir_file);
1096 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1097 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1098 if (hinfo.hash_version <= DX_HASH_TEA)
1099 hinfo.hash_version +=
1100 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1101 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1102 if (ext4_has_inline_data(dir)) {
1103 int has_inline_data = 1;
1104 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1108 if (has_inline_data) {
1113 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1114 start_hash, start_minor_hash);
1118 hinfo.hash = start_hash;
1119 hinfo.minor_hash = 0;
1120 frame = dx_probe(NULL, dir, &hinfo, frames);
1122 return PTR_ERR(frame);
1124 /* Add '.' and '..' from the htree header */
1125 if (!start_hash && !start_minor_hash) {
1126 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1127 tmp_str.name = de->name;
1128 tmp_str.len = de->name_len;
1129 err = ext4_htree_store_dirent(dir_file, 0, 0,
1135 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1136 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1137 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1138 tmp_str.name = de->name;
1139 tmp_str.len = de->name_len;
1140 err = ext4_htree_store_dirent(dir_file, 2, 0,
1148 if (fatal_signal_pending(current)) {
1153 block = dx_get_block(frame->at);
1154 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1155 start_hash, start_minor_hash);
1162 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1163 frame, frames, &hashval);
1164 *next_hash = hashval;
1170 * Stop if: (a) there are no more entries, or
1171 * (b) we have inserted at least one entry and the
1172 * next hash value is not a continuation
1175 (count && ((hashval & 1) == 0)))
1179 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1180 "next hash: %x\n", count, *next_hash));
1187 static inline int search_dirblock(struct buffer_head *bh,
1189 struct ext4_filename *fname,
1190 unsigned int offset,
1191 struct ext4_dir_entry_2 **res_dir)
1193 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1194 fname, offset, res_dir);
1198 * Directory block splitting, compacting
1202 * Create map of hash values, offsets, and sizes, stored at end of block.
1203 * Returns number of entries mapped.
1205 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1206 unsigned blocksize, struct dx_hash_info *hinfo,
1207 struct dx_map_entry *map_tail)
1210 char *base = (char *) de;
1211 struct dx_hash_info h = *hinfo;
1213 while ((char *) de < base + blocksize) {
1214 if (de->name_len && de->inode) {
1215 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1217 map_tail->hash = h.hash;
1218 map_tail->offs = ((char *) de - base)>>2;
1219 map_tail->size = le16_to_cpu(de->rec_len);
1223 /* XXX: do we need to check rec_len == 0 case? -Chris */
1224 de = ext4_next_entry(de, blocksize);
1229 /* Sort map by hash value */
1230 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1232 struct dx_map_entry *p, *q, *top = map + count - 1;
1234 /* Combsort until bubble sort doesn't suck */
1236 count = count*10/13;
1237 if (count - 9 < 2) /* 9, 10 -> 11 */
1239 for (p = top, q = p - count; q >= map; p--, q--)
1240 if (p->hash < q->hash)
1243 /* Garden variety bubble sort */
1248 if (q[1].hash >= q[0].hash)
1256 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1258 struct dx_entry *entries = frame->entries;
1259 struct dx_entry *old = frame->at, *new = old + 1;
1260 int count = dx_get_count(entries);
1262 assert(count < dx_get_limit(entries));
1263 assert(old < entries + count);
1264 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1265 dx_set_hash(new, hash);
1266 dx_set_block(new, block);
1267 dx_set_count(entries, count + 1);
1270 #ifdef CONFIG_UNICODE
1272 * Test whether a case-insensitive directory entry matches the filename
1273 * being searched for. If quick is set, assume the name being looked up
1274 * is already in the casefolded form.
1276 * Returns: 0 if the directory entry matches, more than 0 if it
1277 * doesn't match or less than zero on error.
1279 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1280 const struct qstr *entry, bool quick)
1282 const struct super_block *sb = parent->i_sb;
1283 const struct unicode_map *um = sb->s_encoding;
1287 ret = utf8_strncasecmp_folded(um, name, entry);
1289 ret = utf8_strncasecmp(um, name, entry);
1292 /* Handle invalid character sequence as either an error
1293 * or as an opaque byte sequence.
1295 if (sb_has_strict_encoding(sb))
1298 if (name->len != entry->len)
1301 return !!memcmp(name->name, entry->name, name->len);
1307 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1308 struct fscrypt_str *cf_name)
1312 if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) {
1313 cf_name->name = NULL;
1317 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1321 len = utf8_casefold(dir->i_sb->s_encoding,
1322 iname, cf_name->name,
1325 kfree(cf_name->name);
1326 cf_name->name = NULL;
1329 cf_name->len = (unsigned) len;
1335 * Test whether a directory entry matches the filename being searched for.
1337 * Return: %true if the directory entry matches, otherwise %false.
1339 static inline bool ext4_match(const struct inode *parent,
1340 const struct ext4_filename *fname,
1341 const struct ext4_dir_entry_2 *de)
1343 struct fscrypt_name f;
1344 #ifdef CONFIG_UNICODE
1345 const struct qstr entry = {.name = de->name, .len = de->name_len};
1351 f.usr_fname = fname->usr_fname;
1352 f.disk_name = fname->disk_name;
1353 #ifdef CONFIG_FS_ENCRYPTION
1354 f.crypto_buf = fname->crypto_buf;
1357 #ifdef CONFIG_UNICODE
1358 if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) {
1359 if (fname->cf_name.name) {
1360 struct qstr cf = {.name = fname->cf_name.name,
1361 .len = fname->cf_name.len};
1362 return !ext4_ci_compare(parent, &cf, &entry, true);
1364 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1369 return fscrypt_match_name(&f, de->name, de->name_len);
1373 * Returns 0 if not found, -1 on failure, and 1 on success
1375 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1376 struct inode *dir, struct ext4_filename *fname,
1377 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1379 struct ext4_dir_entry_2 * de;
1383 de = (struct ext4_dir_entry_2 *)search_buf;
1384 dlimit = search_buf + buf_size;
1385 while ((char *) de < dlimit) {
1386 /* this code is executed quadratically often */
1387 /* do minimal checking `by hand' */
1388 if ((char *) de + de->name_len <= dlimit &&
1389 ext4_match(dir, fname, de)) {
1390 /* found a match - just to be sure, do
1392 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1398 /* prevent looping on a bad block */
1399 de_len = ext4_rec_len_from_disk(de->rec_len,
1400 dir->i_sb->s_blocksize);
1404 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1409 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1410 struct ext4_dir_entry *de)
1412 struct super_block *sb = dir->i_sb;
1418 if (de->inode == 0 &&
1419 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1426 * __ext4_find_entry()
1428 * finds an entry in the specified directory with the wanted name. It
1429 * returns the cache buffer in which the entry was found, and the entry
1430 * itself (as a parameter - res_dir). It does NOT read the inode of the
1431 * entry - you'll have to do that yourself if you want to.
1433 * The returned buffer_head has ->b_count elevated. The caller is expected
1434 * to brelse() it when appropriate.
1436 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1437 struct ext4_filename *fname,
1438 struct ext4_dir_entry_2 **res_dir,
1441 struct super_block *sb;
1442 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1443 struct buffer_head *bh, *ret = NULL;
1444 ext4_lblk_t start, block;
1445 const u8 *name = fname->usr_fname->name;
1446 size_t ra_max = 0; /* Number of bh's in the readahead
1448 size_t ra_ptr = 0; /* Current index into readahead
1450 ext4_lblk_t nblocks;
1451 int i, namelen, retval;
1455 namelen = fname->usr_fname->len;
1456 if (namelen > EXT4_NAME_LEN)
1459 if (ext4_has_inline_data(dir)) {
1460 int has_inline_data = 1;
1461 ret = ext4_find_inline_entry(dir, fname, res_dir,
1463 if (has_inline_data) {
1466 goto cleanup_and_exit;
1470 if ((namelen <= 2) && (name[0] == '.') &&
1471 (name[1] == '.' || name[1] == '\0')) {
1473 * "." or ".." will only be in the first block
1474 * NFS may look up ".."; "." should be handled by the VFS
1481 ret = ext4_dx_find_entry(dir, fname, res_dir);
1483 * On success, or if the error was file not found,
1484 * return. Otherwise, fall back to doing a search the
1485 * old fashioned way.
1487 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1488 goto cleanup_and_exit;
1489 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1493 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1496 goto cleanup_and_exit;
1498 start = EXT4_I(dir)->i_dir_start_lookup;
1499 if (start >= nblocks)
1505 * We deal with the read-ahead logic here.
1508 if (ra_ptr >= ra_max) {
1509 /* Refill the readahead buffer */
1512 ra_max = start - block;
1514 ra_max = nblocks - block;
1515 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1516 retval = ext4_bread_batch(dir, block, ra_max,
1517 false /* wait */, bh_use);
1519 ret = ERR_PTR(retval);
1521 goto cleanup_and_exit;
1524 if ((bh = bh_use[ra_ptr++]) == NULL)
1527 if (!buffer_uptodate(bh)) {
1528 EXT4_ERROR_INODE_ERR(dir, EIO,
1529 "reading directory lblock %lu",
1530 (unsigned long) block);
1532 ret = ERR_PTR(-EIO);
1533 goto cleanup_and_exit;
1535 if (!buffer_verified(bh) &&
1536 !is_dx_internal_node(dir, block,
1537 (struct ext4_dir_entry *)bh->b_data) &&
1538 !ext4_dirblock_csum_verify(dir, bh)) {
1539 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1540 "checksumming directory "
1541 "block %lu", (unsigned long)block);
1543 ret = ERR_PTR(-EFSBADCRC);
1544 goto cleanup_and_exit;
1546 set_buffer_verified(bh);
1547 i = search_dirblock(bh, dir, fname,
1548 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1550 EXT4_I(dir)->i_dir_start_lookup = block;
1552 goto cleanup_and_exit;
1556 goto cleanup_and_exit;
1559 if (++block >= nblocks)
1561 } while (block != start);
1564 * If the directory has grown while we were searching, then
1565 * search the last part of the directory before giving up.
1568 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1569 if (block < nblocks) {
1575 /* Clean up the read-ahead blocks */
1576 for (; ra_ptr < ra_max; ra_ptr++)
1577 brelse(bh_use[ra_ptr]);
1581 static struct buffer_head *ext4_find_entry(struct inode *dir,
1582 const struct qstr *d_name,
1583 struct ext4_dir_entry_2 **res_dir,
1587 struct ext4_filename fname;
1588 struct buffer_head *bh;
1590 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1594 return ERR_PTR(err);
1596 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1598 ext4_fname_free_filename(&fname);
1602 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1603 struct dentry *dentry,
1604 struct ext4_dir_entry_2 **res_dir)
1607 struct ext4_filename fname;
1608 struct buffer_head *bh;
1610 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1614 return ERR_PTR(err);
1616 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1618 ext4_fname_free_filename(&fname);
1622 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1623 struct ext4_filename *fname,
1624 struct ext4_dir_entry_2 **res_dir)
1626 struct super_block * sb = dir->i_sb;
1627 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1628 struct buffer_head *bh;
1632 #ifdef CONFIG_FS_ENCRYPTION
1635 frame = dx_probe(fname, dir, NULL, frames);
1637 return (struct buffer_head *) frame;
1639 block = dx_get_block(frame->at);
1640 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1644 retval = search_dirblock(bh, dir, fname,
1645 block << EXT4_BLOCK_SIZE_BITS(sb),
1651 bh = ERR_PTR(ERR_BAD_DX_DIR);
1655 /* Check to see if we should continue to search */
1656 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1659 ext4_warning_inode(dir,
1660 "error %d reading directory index block",
1662 bh = ERR_PTR(retval);
1665 } while (retval == 1);
1669 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1675 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1677 struct inode *inode;
1678 struct ext4_dir_entry_2 *de;
1679 struct buffer_head *bh;
1681 if (dentry->d_name.len > EXT4_NAME_LEN)
1682 return ERR_PTR(-ENAMETOOLONG);
1684 bh = ext4_lookup_entry(dir, dentry, &de);
1686 return ERR_CAST(bh);
1689 __u32 ino = le32_to_cpu(de->inode);
1691 if (!ext4_valid_inum(dir->i_sb, ino)) {
1692 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1693 return ERR_PTR(-EFSCORRUPTED);
1695 if (unlikely(ino == dir->i_ino)) {
1696 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1698 return ERR_PTR(-EFSCORRUPTED);
1700 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1701 if (inode == ERR_PTR(-ESTALE)) {
1702 EXT4_ERROR_INODE(dir,
1703 "deleted inode referenced: %u",
1705 return ERR_PTR(-EFSCORRUPTED);
1707 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1708 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1709 !fscrypt_has_permitted_context(dir, inode)) {
1710 ext4_warning(inode->i_sb,
1711 "Inconsistent encryption contexts: %lu/%lu",
1712 dir->i_ino, inode->i_ino);
1714 return ERR_PTR(-EPERM);
1718 #ifdef CONFIG_UNICODE
1719 if (!inode && IS_CASEFOLDED(dir)) {
1720 /* Eventually we want to call d_add_ci(dentry, NULL)
1721 * for negative dentries in the encoding case as
1722 * well. For now, prevent the negative dentry
1723 * from being cached.
1728 return d_splice_alias(inode, dentry);
1732 struct dentry *ext4_get_parent(struct dentry *child)
1735 static const struct qstr dotdot = QSTR_INIT("..", 2);
1736 struct ext4_dir_entry_2 * de;
1737 struct buffer_head *bh;
1739 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1741 return ERR_CAST(bh);
1743 return ERR_PTR(-ENOENT);
1744 ino = le32_to_cpu(de->inode);
1747 if (!ext4_valid_inum(child->d_sb, ino)) {
1748 EXT4_ERROR_INODE(d_inode(child),
1749 "bad parent inode number: %u", ino);
1750 return ERR_PTR(-EFSCORRUPTED);
1753 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1757 * Move count entries from end of map between two memory locations.
1758 * Returns pointer to last entry moved.
1760 static struct ext4_dir_entry_2 *
1761 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1764 unsigned rec_len = 0;
1767 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1768 (from + (map->offs<<2));
1769 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1770 memcpy (to, de, rec_len);
1771 ((struct ext4_dir_entry_2 *) to)->rec_len =
1772 ext4_rec_len_to_disk(rec_len, blocksize);
1777 return (struct ext4_dir_entry_2 *) (to - rec_len);
1781 * Compact each dir entry in the range to the minimal rec_len.
1782 * Returns pointer to last entry in range.
1784 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1786 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1787 unsigned rec_len = 0;
1790 while ((char*)de < base + blocksize) {
1791 next = ext4_next_entry(de, blocksize);
1792 if (de->inode && de->name_len) {
1793 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1795 memmove(to, de, rec_len);
1796 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1798 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1806 * Split a full leaf block to make room for a new dir entry.
1807 * Allocate a new block, and move entries so that they are approx. equally full.
1808 * Returns pointer to de in block into which the new entry will be inserted.
1810 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1811 struct buffer_head **bh,struct dx_frame *frame,
1812 struct dx_hash_info *hinfo)
1814 unsigned blocksize = dir->i_sb->s_blocksize;
1815 unsigned count, continued;
1816 struct buffer_head *bh2;
1817 ext4_lblk_t newblock;
1819 struct dx_map_entry *map;
1820 char *data1 = (*bh)->b_data, *data2;
1821 unsigned split, move, size;
1822 struct ext4_dir_entry_2 *de = NULL, *de2;
1826 if (ext4_has_metadata_csum(dir->i_sb))
1827 csum_size = sizeof(struct ext4_dir_entry_tail);
1829 bh2 = ext4_append(handle, dir, &newblock);
1833 return (struct ext4_dir_entry_2 *) bh2;
1836 BUFFER_TRACE(*bh, "get_write_access");
1837 err = ext4_journal_get_write_access(handle, *bh);
1841 BUFFER_TRACE(frame->bh, "get_write_access");
1842 err = ext4_journal_get_write_access(handle, frame->bh);
1846 data2 = bh2->b_data;
1848 /* create map in the end of data2 block */
1849 map = (struct dx_map_entry *) (data2 + blocksize);
1850 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1851 blocksize, hinfo, map);
1853 dx_sort_map(map, count);
1854 /* Ensure that neither split block is over half full */
1857 for (i = count-1; i >= 0; i--) {
1858 /* is more than half of this entry in 2nd half of the block? */
1859 if (size + map[i].size/2 > blocksize/2)
1861 size += map[i].size;
1865 * map index at which we will split
1867 * If the sum of active entries didn't exceed half the block size, just
1868 * split it in half by count; each resulting block will have at least
1869 * half the space free.
1872 split = count - move;
1876 hash2 = map[split].hash;
1877 continued = hash2 == map[split - 1].hash;
1878 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1879 (unsigned long)dx_get_block(frame->at),
1880 hash2, split, count-split));
1882 /* Fancy dance to stay within two buffers */
1883 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1885 de = dx_pack_dirents(data1, blocksize);
1886 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1889 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1893 ext4_initialize_dirent_tail(*bh, blocksize);
1894 ext4_initialize_dirent_tail(bh2, blocksize);
1897 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1899 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1902 /* Which block gets the new entry? */
1903 if (hinfo->hash >= hash2) {
1907 dx_insert_block(frame, hash2 + continued, newblock);
1908 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1911 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1915 dxtrace(dx_show_index("frame", frame->entries));
1922 ext4_std_error(dir->i_sb, err);
1923 return ERR_PTR(err);
1926 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1927 struct buffer_head *bh,
1928 void *buf, int buf_size,
1929 struct ext4_filename *fname,
1930 struct ext4_dir_entry_2 **dest_de)
1932 struct ext4_dir_entry_2 *de;
1933 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1935 unsigned int offset = 0;
1938 de = (struct ext4_dir_entry_2 *)buf;
1939 top = buf + buf_size - reclen;
1940 while ((char *) de <= top) {
1941 if (ext4_check_dir_entry(dir, NULL, de, bh,
1942 buf, buf_size, offset))
1943 return -EFSCORRUPTED;
1944 if (ext4_match(dir, fname, de))
1946 nlen = EXT4_DIR_REC_LEN(de->name_len);
1947 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1948 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1950 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1953 if ((char *) de > top)
1960 void ext4_insert_dentry(struct inode *inode,
1961 struct ext4_dir_entry_2 *de,
1963 struct ext4_filename *fname)
1968 nlen = EXT4_DIR_REC_LEN(de->name_len);
1969 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1971 struct ext4_dir_entry_2 *de1 =
1972 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1973 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1974 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1977 de->file_type = EXT4_FT_UNKNOWN;
1978 de->inode = cpu_to_le32(inode->i_ino);
1979 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1980 de->name_len = fname_len(fname);
1981 memcpy(de->name, fname_name(fname), fname_len(fname));
1985 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1986 * it points to a directory entry which is guaranteed to be large
1987 * enough for new directory entry. If de is NULL, then
1988 * add_dirent_to_buf will attempt search the directory block for
1989 * space. It will return -ENOSPC if no space is available, and -EIO
1990 * and -EEXIST if directory entry already exists.
1992 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1994 struct inode *inode, struct ext4_dir_entry_2 *de,
1995 struct buffer_head *bh)
1997 unsigned int blocksize = dir->i_sb->s_blocksize;
2001 if (ext4_has_metadata_csum(inode->i_sb))
2002 csum_size = sizeof(struct ext4_dir_entry_tail);
2005 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2006 blocksize - csum_size, fname, &de);
2010 BUFFER_TRACE(bh, "get_write_access");
2011 err = ext4_journal_get_write_access(handle, bh);
2013 ext4_std_error(dir->i_sb, err);
2017 /* By now the buffer is marked for journaling */
2018 ext4_insert_dentry(inode, de, blocksize, fname);
2021 * XXX shouldn't update any times until successful
2022 * completion of syscall, but too many callers depend
2025 * XXX similarly, too many callers depend on
2026 * ext4_new_inode() setting the times, but error
2027 * recovery deletes the inode, so the worst that can
2028 * happen is that the times are slightly out of date
2029 * and/or different from the directory change time.
2031 dir->i_mtime = dir->i_ctime = current_time(dir);
2032 ext4_update_dx_flag(dir);
2033 inode_inc_iversion(dir);
2034 err2 = ext4_mark_inode_dirty(handle, dir);
2035 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2036 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2038 ext4_std_error(dir->i_sb, err);
2039 return err ? err : err2;
2043 * This converts a one block unindexed directory to a 3 block indexed
2044 * directory, and adds the dentry to the indexed directory.
2046 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2048 struct inode *inode, struct buffer_head *bh)
2050 struct buffer_head *bh2;
2051 struct dx_root *root;
2052 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2053 struct dx_entry *entries;
2054 struct ext4_dir_entry_2 *de, *de2;
2060 struct fake_dirent *fde;
2063 if (ext4_has_metadata_csum(inode->i_sb))
2064 csum_size = sizeof(struct ext4_dir_entry_tail);
2066 blocksize = dir->i_sb->s_blocksize;
2067 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2068 BUFFER_TRACE(bh, "get_write_access");
2069 retval = ext4_journal_get_write_access(handle, bh);
2071 ext4_std_error(dir->i_sb, retval);
2075 root = (struct dx_root *) bh->b_data;
2077 /* The 0th block becomes the root, move the dirents out */
2078 fde = &root->dotdot;
2079 de = (struct ext4_dir_entry_2 *)((char *)fde +
2080 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2081 if ((char *) de >= (((char *) root) + blocksize)) {
2082 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2084 return -EFSCORRUPTED;
2086 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2088 /* Allocate new block for the 0th block's dirents */
2089 bh2 = ext4_append(handle, dir, &block);
2092 return PTR_ERR(bh2);
2094 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2095 data2 = bh2->b_data;
2097 memcpy(data2, de, len);
2098 de = (struct ext4_dir_entry_2 *) data2;
2100 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2102 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2103 (char *) de, blocksize);
2106 ext4_initialize_dirent_tail(bh2, blocksize);
2108 /* Initialize the root; the dot dirents already exist */
2109 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2110 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2112 memset (&root->info, 0, sizeof(root->info));
2113 root->info.info_length = sizeof(root->info);
2114 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2115 entries = root->entries;
2116 dx_set_block(entries, 1);
2117 dx_set_count(entries, 1);
2118 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2120 /* Initialize as for dx_probe */
2121 fname->hinfo.hash_version = root->info.hash_version;
2122 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2123 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2124 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2125 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2127 memset(frames, 0, sizeof(frames));
2129 frame->entries = entries;
2130 frame->at = entries;
2133 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2136 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2140 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2142 retval = PTR_ERR(de);
2146 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2149 * Even if the block split failed, we have to properly write
2150 * out all the changes we did so far. Otherwise we can end up
2151 * with corrupted filesystem.
2154 ext4_mark_inode_dirty(handle, dir);
2163 * adds a file entry to the specified directory, using the same
2164 * semantics as ext4_find_entry(). It returns NULL if it failed.
2166 * NOTE!! The inode part of 'de' is left at 0 - which means you
2167 * may not sleep between calling this and putting something into
2168 * the entry, as someone else might have used it while you slept.
2170 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2171 struct inode *inode)
2173 struct inode *dir = d_inode(dentry->d_parent);
2174 struct buffer_head *bh = NULL;
2175 struct ext4_dir_entry_2 *de;
2176 struct super_block *sb;
2177 struct ext4_filename fname;
2181 ext4_lblk_t block, blocks;
2184 if (ext4_has_metadata_csum(inode->i_sb))
2185 csum_size = sizeof(struct ext4_dir_entry_tail);
2188 blocksize = sb->s_blocksize;
2189 if (!dentry->d_name.len)
2192 if (fscrypt_is_nokey_name(dentry))
2195 #ifdef CONFIG_UNICODE
2196 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2197 sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2201 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2205 if (ext4_has_inline_data(dir)) {
2206 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2216 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2217 if (!retval || (retval != ERR_BAD_DX_DIR))
2219 /* Can we just ignore htree data? */
2220 if (ext4_has_metadata_csum(sb)) {
2221 EXT4_ERROR_INODE(dir,
2222 "Directory has corrupted htree index.");
2223 retval = -EFSCORRUPTED;
2226 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2228 retval = ext4_mark_inode_dirty(handle, dir);
2229 if (unlikely(retval))
2232 blocks = dir->i_size >> sb->s_blocksize_bits;
2233 for (block = 0; block < blocks; block++) {
2234 bh = ext4_read_dirblock(dir, block, DIRENT);
2236 bh = ext4_bread(handle, dir, block,
2237 EXT4_GET_BLOCKS_CREATE);
2238 goto add_to_new_block;
2241 retval = PTR_ERR(bh);
2245 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2247 if (retval != -ENOSPC)
2250 if (blocks == 1 && !dx_fallback &&
2251 ext4_has_feature_dir_index(sb)) {
2252 retval = make_indexed_dir(handle, &fname, dir,
2254 bh = NULL; /* make_indexed_dir releases bh */
2259 bh = ext4_append(handle, dir, &block);
2262 retval = PTR_ERR(bh);
2266 de = (struct ext4_dir_entry_2 *) bh->b_data;
2268 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2271 ext4_initialize_dirent_tail(bh, blocksize);
2273 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2275 ext4_fname_free_filename(&fname);
2278 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2283 * Returns 0 for success, or a negative error value
2285 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2286 struct inode *dir, struct inode *inode)
2288 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2289 struct dx_entry *entries, *at;
2290 struct buffer_head *bh;
2291 struct super_block *sb = dir->i_sb;
2292 struct ext4_dir_entry_2 *de;
2298 frame = dx_probe(fname, dir, NULL, frames);
2300 return PTR_ERR(frame);
2301 entries = frame->entries;
2303 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2310 BUFFER_TRACE(bh, "get_write_access");
2311 err = ext4_journal_get_write_access(handle, bh);
2315 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2320 /* Block full, should compress but for now just split */
2321 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2322 dx_get_count(entries), dx_get_limit(entries)));
2323 /* Need to split index? */
2324 if (dx_get_count(entries) == dx_get_limit(entries)) {
2325 ext4_lblk_t newblock;
2326 int levels = frame - frames + 1;
2327 unsigned int icount;
2329 struct dx_entry *entries2;
2330 struct dx_node *node2;
2331 struct buffer_head *bh2;
2333 while (frame > frames) {
2334 if (dx_get_count((frame - 1)->entries) <
2335 dx_get_limit((frame - 1)->entries)) {
2339 frame--; /* split higher index block */
2341 entries = frame->entries;
2344 if (add_level && levels == ext4_dir_htree_level(sb)) {
2345 ext4_warning(sb, "Directory (ino: %lu) index full, "
2346 "reach max htree level :%d",
2347 dir->i_ino, levels);
2348 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2349 ext4_warning(sb, "Large directory feature is "
2350 "not enabled on this "
2356 icount = dx_get_count(entries);
2357 bh2 = ext4_append(handle, dir, &newblock);
2362 node2 = (struct dx_node *)(bh2->b_data);
2363 entries2 = node2->entries;
2364 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2365 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2367 BUFFER_TRACE(frame->bh, "get_write_access");
2368 err = ext4_journal_get_write_access(handle, frame->bh);
2372 unsigned icount1 = icount/2, icount2 = icount - icount1;
2373 unsigned hash2 = dx_get_hash(entries + icount1);
2374 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2377 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2378 err = ext4_journal_get_write_access(handle,
2383 memcpy((char *) entries2, (char *) (entries + icount1),
2384 icount2 * sizeof(struct dx_entry));
2385 dx_set_count(entries, icount1);
2386 dx_set_count(entries2, icount2);
2387 dx_set_limit(entries2, dx_node_limit(dir));
2389 /* Which index block gets the new entry? */
2390 if (at - entries >= icount1) {
2391 frame->at = at = at - entries - icount1 + entries2;
2392 frame->entries = entries = entries2;
2393 swap(frame->bh, bh2);
2395 dx_insert_block((frame - 1), hash2, newblock);
2396 dxtrace(dx_show_index("node", frame->entries));
2397 dxtrace(dx_show_index("node",
2398 ((struct dx_node *) bh2->b_data)->entries));
2399 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2403 err = ext4_handle_dirty_dx_node(handle, dir,
2408 err = ext4_handle_dirty_dx_node(handle, dir,
2413 struct dx_root *dxroot;
2414 memcpy((char *) entries2, (char *) entries,
2415 icount * sizeof(struct dx_entry));
2416 dx_set_limit(entries2, dx_node_limit(dir));
2419 dx_set_count(entries, 1);
2420 dx_set_block(entries + 0, newblock);
2421 dxroot = (struct dx_root *)frames[0].bh->b_data;
2422 dxroot->info.indirect_levels += 1;
2423 dxtrace(printk(KERN_DEBUG
2424 "Creating %d level index...\n",
2425 dxroot->info.indirect_levels));
2426 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2429 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2435 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2440 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2444 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2448 /* @restart is true means htree-path has been changed, we need to
2449 * repeat dx_probe() to find out valid htree-path
2451 if (restart && err == 0)
2457 * ext4_generic_delete_entry deletes a directory entry by merging it
2458 * with the previous entry
2460 int ext4_generic_delete_entry(struct inode *dir,
2461 struct ext4_dir_entry_2 *de_del,
2462 struct buffer_head *bh,
2467 struct ext4_dir_entry_2 *de, *pde;
2468 unsigned int blocksize = dir->i_sb->s_blocksize;
2473 de = (struct ext4_dir_entry_2 *)entry_buf;
2474 while (i < buf_size - csum_size) {
2475 if (ext4_check_dir_entry(dir, NULL, de, bh,
2476 entry_buf, buf_size, i))
2477 return -EFSCORRUPTED;
2480 pde->rec_len = ext4_rec_len_to_disk(
2481 ext4_rec_len_from_disk(pde->rec_len,
2483 ext4_rec_len_from_disk(de->rec_len,
2488 inode_inc_iversion(dir);
2491 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2493 de = ext4_next_entry(de, blocksize);
2498 static int ext4_delete_entry(handle_t *handle,
2500 struct ext4_dir_entry_2 *de_del,
2501 struct buffer_head *bh)
2503 int err, csum_size = 0;
2505 if (ext4_has_inline_data(dir)) {
2506 int has_inline_data = 1;
2507 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2509 if (has_inline_data)
2513 if (ext4_has_metadata_csum(dir->i_sb))
2514 csum_size = sizeof(struct ext4_dir_entry_tail);
2516 BUFFER_TRACE(bh, "get_write_access");
2517 err = ext4_journal_get_write_access(handle, bh);
2521 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2522 dir->i_sb->s_blocksize, csum_size);
2526 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2527 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2534 ext4_std_error(dir->i_sb, err);
2539 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2540 * since this indicates that nlinks count was previously 1 to avoid overflowing
2541 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2542 * that subdirectory link counts are not being maintained accurately.
2544 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2545 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2546 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2547 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2549 static void ext4_inc_count(struct inode *inode)
2553 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2554 set_nlink(inode, 1);
2558 * If a directory had nlink == 1, then we should let it be 1. This indicates
2559 * directory has >EXT4_LINK_MAX subdirs.
2561 static void ext4_dec_count(struct inode *inode)
2563 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2569 * Add non-directory inode to a directory. On success, the inode reference is
2570 * consumed by dentry is instantiation. This is also indicated by clearing of
2571 * *inodep pointer. On failure, the caller is responsible for dropping the
2572 * inode reference in the safe context.
2574 static int ext4_add_nondir(handle_t *handle,
2575 struct dentry *dentry, struct inode **inodep)
2577 struct inode *dir = d_inode(dentry->d_parent);
2578 struct inode *inode = *inodep;
2579 int err = ext4_add_entry(handle, dentry, inode);
2581 err = ext4_mark_inode_dirty(handle, inode);
2582 if (IS_DIRSYNC(dir))
2583 ext4_handle_sync(handle);
2584 d_instantiate_new(dentry, inode);
2589 ext4_orphan_add(handle, inode);
2590 unlock_new_inode(inode);
2595 * By the time this is called, we already have created
2596 * the directory cache entry for the new file, but it
2597 * is so far negative - it has no inode.
2599 * If the create succeeds, we fill in the inode information
2600 * with d_instantiate().
2602 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2606 struct inode *inode;
2607 int err, credits, retries = 0;
2609 err = dquot_initialize(dir);
2613 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2614 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2616 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2617 NULL, EXT4_HT_DIR, credits);
2618 handle = ext4_journal_current_handle();
2619 err = PTR_ERR(inode);
2620 if (!IS_ERR(inode)) {
2621 inode->i_op = &ext4_file_inode_operations;
2622 inode->i_fop = &ext4_file_operations;
2623 ext4_set_aops(inode);
2624 err = ext4_add_nondir(handle, dentry, &inode);
2626 ext4_fc_track_create(handle, dentry);
2629 ext4_journal_stop(handle);
2630 if (!IS_ERR_OR_NULL(inode))
2632 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2637 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2638 umode_t mode, dev_t rdev)
2641 struct inode *inode;
2642 int err, credits, retries = 0;
2644 err = dquot_initialize(dir);
2648 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2649 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2651 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2652 NULL, EXT4_HT_DIR, credits);
2653 handle = ext4_journal_current_handle();
2654 err = PTR_ERR(inode);
2655 if (!IS_ERR(inode)) {
2656 init_special_inode(inode, inode->i_mode, rdev);
2657 inode->i_op = &ext4_special_inode_operations;
2658 err = ext4_add_nondir(handle, dentry, &inode);
2660 ext4_fc_track_create(handle, dentry);
2663 ext4_journal_stop(handle);
2664 if (!IS_ERR_OR_NULL(inode))
2666 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2671 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2674 struct inode *inode;
2675 int err, retries = 0;
2677 err = dquot_initialize(dir);
2682 inode = ext4_new_inode_start_handle(dir, mode,
2685 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2686 4 + EXT4_XATTR_TRANS_BLOCKS);
2687 handle = ext4_journal_current_handle();
2688 err = PTR_ERR(inode);
2689 if (!IS_ERR(inode)) {
2690 inode->i_op = &ext4_file_inode_operations;
2691 inode->i_fop = &ext4_file_operations;
2692 ext4_set_aops(inode);
2693 d_tmpfile(dentry, inode);
2694 err = ext4_orphan_add(handle, inode);
2696 goto err_unlock_inode;
2697 mark_inode_dirty(inode);
2698 unlock_new_inode(inode);
2701 ext4_journal_stop(handle);
2702 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2706 ext4_journal_stop(handle);
2707 unlock_new_inode(inode);
2711 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2712 struct ext4_dir_entry_2 *de,
2713 int blocksize, int csum_size,
2714 unsigned int parent_ino, int dotdot_real_len)
2716 de->inode = cpu_to_le32(inode->i_ino);
2718 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2720 strcpy(de->name, ".");
2721 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2723 de = ext4_next_entry(de, blocksize);
2724 de->inode = cpu_to_le32(parent_ino);
2726 if (!dotdot_real_len)
2727 de->rec_len = ext4_rec_len_to_disk(blocksize -
2728 (csum_size + EXT4_DIR_REC_LEN(1)),
2731 de->rec_len = ext4_rec_len_to_disk(
2732 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2733 strcpy(de->name, "..");
2734 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2736 return ext4_next_entry(de, blocksize);
2739 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2740 struct inode *inode)
2742 struct buffer_head *dir_block = NULL;
2743 struct ext4_dir_entry_2 *de;
2744 ext4_lblk_t block = 0;
2745 unsigned int blocksize = dir->i_sb->s_blocksize;
2749 if (ext4_has_metadata_csum(dir->i_sb))
2750 csum_size = sizeof(struct ext4_dir_entry_tail);
2752 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2753 err = ext4_try_create_inline_dir(handle, dir, inode);
2754 if (err < 0 && err != -ENOSPC)
2761 dir_block = ext4_append(handle, inode, &block);
2762 if (IS_ERR(dir_block))
2763 return PTR_ERR(dir_block);
2764 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2765 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2766 set_nlink(inode, 2);
2768 ext4_initialize_dirent_tail(dir_block, blocksize);
2770 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2771 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2774 set_buffer_verified(dir_block);
2780 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2783 struct inode *inode;
2784 int err, err2 = 0, credits, retries = 0;
2786 if (EXT4_DIR_LINK_MAX(dir))
2789 err = dquot_initialize(dir);
2793 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2794 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2796 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2798 0, NULL, EXT4_HT_DIR, credits);
2799 handle = ext4_journal_current_handle();
2800 err = PTR_ERR(inode);
2804 inode->i_op = &ext4_dir_inode_operations;
2805 inode->i_fop = &ext4_dir_operations;
2806 err = ext4_init_new_dir(handle, dir, inode);
2808 goto out_clear_inode;
2809 err = ext4_mark_inode_dirty(handle, inode);
2811 err = ext4_add_entry(handle, dentry, inode);
2815 ext4_orphan_add(handle, inode);
2816 unlock_new_inode(inode);
2817 err2 = ext4_mark_inode_dirty(handle, inode);
2820 ext4_journal_stop(handle);
2824 ext4_inc_count(dir);
2826 ext4_update_dx_flag(dir);
2827 err = ext4_mark_inode_dirty(handle, dir);
2829 goto out_clear_inode;
2830 d_instantiate_new(dentry, inode);
2831 ext4_fc_track_create(handle, dentry);
2832 if (IS_DIRSYNC(dir))
2833 ext4_handle_sync(handle);
2837 ext4_journal_stop(handle);
2839 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2845 * routine to check that the specified directory is empty (for rmdir)
2847 bool ext4_empty_dir(struct inode *inode)
2849 unsigned int offset;
2850 struct buffer_head *bh;
2851 struct ext4_dir_entry_2 *de;
2852 struct super_block *sb;
2854 if (ext4_has_inline_data(inode)) {
2855 int has_inline_data = 1;
2858 ret = empty_inline_dir(inode, &has_inline_data);
2859 if (has_inline_data)
2864 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2865 EXT4_ERROR_INODE(inode, "invalid size");
2868 /* The first directory block must not be a hole,
2869 * so treat it as DIRENT_HTREE
2871 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2875 de = (struct ext4_dir_entry_2 *) bh->b_data;
2876 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2878 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2879 ext4_warning_inode(inode, "directory missing '.'");
2883 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2884 de = ext4_next_entry(de, sb->s_blocksize);
2885 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2887 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2888 ext4_warning_inode(inode, "directory missing '..'");
2892 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2893 while (offset < inode->i_size) {
2894 if (!(offset & (sb->s_blocksize - 1))) {
2895 unsigned int lblock;
2897 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2898 bh = ext4_read_dirblock(inode, lblock, EITHER);
2900 offset += sb->s_blocksize;
2906 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2907 (offset & (sb->s_blocksize - 1)));
2908 if (ext4_check_dir_entry(inode, NULL, de, bh,
2909 bh->b_data, bh->b_size, offset)) {
2910 offset = (offset | (sb->s_blocksize - 1)) + 1;
2913 if (le32_to_cpu(de->inode)) {
2917 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2924 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2925 * such inodes, starting at the superblock, in case we crash before the
2926 * file is closed/deleted, or in case the inode truncate spans multiple
2927 * transactions and the last transaction is not recovered after a crash.
2929 * At filesystem recovery time, we walk this list deleting unlinked
2930 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2932 * Orphan list manipulation functions must be called under i_mutex unless
2933 * we are just creating the inode or deleting it.
2935 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2937 struct super_block *sb = inode->i_sb;
2938 struct ext4_sb_info *sbi = EXT4_SB(sb);
2939 struct ext4_iloc iloc;
2943 if (!sbi->s_journal || is_bad_inode(inode))
2946 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2947 !inode_is_locked(inode));
2949 * Exit early if inode already is on orphan list. This is a big speedup
2950 * since we don't have to contend on the global s_orphan_lock.
2952 if (!list_empty(&EXT4_I(inode)->i_orphan))
2956 * Orphan handling is only valid for files with data blocks
2957 * being truncated, or files being unlinked. Note that we either
2958 * hold i_mutex, or the inode can not be referenced from outside,
2959 * so i_nlink should not be bumped due to race
2961 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2962 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2964 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2965 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2969 err = ext4_reserve_inode_write(handle, inode, &iloc);
2973 mutex_lock(&sbi->s_orphan_lock);
2975 * Due to previous errors inode may be already a part of on-disk
2976 * orphan list. If so skip on-disk list modification.
2978 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2979 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2980 /* Insert this inode at the head of the on-disk orphan list */
2981 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2982 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2985 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2986 mutex_unlock(&sbi->s_orphan_lock);
2989 err = ext4_handle_dirty_super(handle, sb);
2990 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2995 * We have to remove inode from in-memory list if
2996 * addition to on disk orphan list failed. Stray orphan
2997 * list entries can cause panics at unmount time.
2999 mutex_lock(&sbi->s_orphan_lock);
3000 list_del_init(&EXT4_I(inode)->i_orphan);
3001 mutex_unlock(&sbi->s_orphan_lock);
3006 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3007 jbd_debug(4, "orphan inode %lu will point to %d\n",
3008 inode->i_ino, NEXT_ORPHAN(inode));
3010 ext4_std_error(sb, err);
3015 * ext4_orphan_del() removes an unlinked or truncated inode from the list
3016 * of such inodes stored on disk, because it is finally being cleaned up.
3018 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3020 struct list_head *prev;
3021 struct ext4_inode_info *ei = EXT4_I(inode);
3022 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3024 struct ext4_iloc iloc;
3027 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3030 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3031 !inode_is_locked(inode));
3032 /* Do this quick check before taking global s_orphan_lock. */
3033 if (list_empty(&ei->i_orphan))
3037 /* Grab inode buffer early before taking global s_orphan_lock */
3038 err = ext4_reserve_inode_write(handle, inode, &iloc);
3041 mutex_lock(&sbi->s_orphan_lock);
3042 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3044 prev = ei->i_orphan.prev;
3045 list_del_init(&ei->i_orphan);
3047 /* If we're on an error path, we may not have a valid
3048 * transaction handle with which to update the orphan list on
3049 * disk, but we still need to remove the inode from the linked
3050 * list in memory. */
3051 if (!handle || err) {
3052 mutex_unlock(&sbi->s_orphan_lock);
3056 ino_next = NEXT_ORPHAN(inode);
3057 if (prev == &sbi->s_orphan) {
3058 jbd_debug(4, "superblock will point to %u\n", ino_next);
3059 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3060 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3062 mutex_unlock(&sbi->s_orphan_lock);
3065 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3066 mutex_unlock(&sbi->s_orphan_lock);
3067 err = ext4_handle_dirty_super(handle, inode->i_sb);
3069 struct ext4_iloc iloc2;
3070 struct inode *i_prev =
3071 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3073 jbd_debug(4, "orphan inode %lu will point to %u\n",
3074 i_prev->i_ino, ino_next);
3075 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3077 mutex_unlock(&sbi->s_orphan_lock);
3080 NEXT_ORPHAN(i_prev) = ino_next;
3081 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3082 mutex_unlock(&sbi->s_orphan_lock);
3086 NEXT_ORPHAN(inode) = 0;
3087 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3089 ext4_std_error(inode->i_sb, err);
3097 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3100 struct inode *inode;
3101 struct buffer_head *bh;
3102 struct ext4_dir_entry_2 *de;
3103 handle_t *handle = NULL;
3105 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3108 /* Initialize quotas before so that eventual writes go in
3109 * separate transaction */
3110 retval = dquot_initialize(dir);
3113 retval = dquot_initialize(d_inode(dentry));
3118 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3124 inode = d_inode(dentry);
3126 retval = -EFSCORRUPTED;
3127 if (le32_to_cpu(de->inode) != inode->i_ino)
3130 retval = -ENOTEMPTY;
3131 if (!ext4_empty_dir(inode))
3134 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3135 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3136 if (IS_ERR(handle)) {
3137 retval = PTR_ERR(handle);
3142 if (IS_DIRSYNC(dir))
3143 ext4_handle_sync(handle);
3145 retval = ext4_delete_entry(handle, dir, de, bh);
3148 if (!EXT4_DIR_LINK_EMPTY(inode))
3149 ext4_warning_inode(inode,
3150 "empty directory '%.*s' has too many links (%u)",
3151 dentry->d_name.len, dentry->d_name.name,
3153 inode_inc_iversion(inode);
3155 /* There's no need to set i_disksize: the fact that i_nlink is
3156 * zero will ensure that the right thing happens during any
3159 ext4_orphan_add(handle, inode);
3160 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3161 retval = ext4_mark_inode_dirty(handle, inode);
3164 ext4_dec_count(dir);
3165 ext4_update_dx_flag(dir);
3166 ext4_fc_track_unlink(handle, dentry);
3167 retval = ext4_mark_inode_dirty(handle, dir);
3169 #ifdef CONFIG_UNICODE
3170 /* VFS negative dentries are incompatible with Encoding and
3171 * Case-insensitiveness. Eventually we'll want avoid
3172 * invalidating the dentries here, alongside with returning the
3173 * negative dentries at ext4_lookup(), when it is better
3174 * supported by the VFS for the CI case.
3176 if (IS_CASEFOLDED(dir))
3177 d_invalidate(dentry);
3183 ext4_journal_stop(handle);
3187 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3188 struct inode *inode)
3190 int retval = -ENOENT;
3191 struct buffer_head *bh;
3192 struct ext4_dir_entry_2 *de;
3193 int skip_remove_dentry = 0;
3195 bh = ext4_find_entry(dir, d_name, &de, NULL);
3202 if (le32_to_cpu(de->inode) != inode->i_ino) {
3204 * It's okay if we find dont find dentry which matches
3205 * the inode. That's because it might have gotten
3206 * renamed to a different inode number
3208 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3209 skip_remove_dentry = 1;
3214 if (IS_DIRSYNC(dir))
3215 ext4_handle_sync(handle);
3217 if (!skip_remove_dentry) {
3218 retval = ext4_delete_entry(handle, dir, de, bh);
3221 dir->i_ctime = dir->i_mtime = current_time(dir);
3222 ext4_update_dx_flag(dir);
3223 retval = ext4_mark_inode_dirty(handle, dir);
3229 if (inode->i_nlink == 0)
3230 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3231 d_name->len, d_name->name);
3234 if (!inode->i_nlink)
3235 ext4_orphan_add(handle, inode);
3236 inode->i_ctime = current_time(inode);
3237 retval = ext4_mark_inode_dirty(handle, inode);
3244 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3249 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3252 trace_ext4_unlink_enter(dir, dentry);
3254 * Initialize quotas before so that eventual writes go
3255 * in separate transaction
3257 retval = dquot_initialize(dir);
3260 retval = dquot_initialize(d_inode(dentry));
3264 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3265 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3266 if (IS_ERR(handle)) {
3267 retval = PTR_ERR(handle);
3271 retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3273 ext4_fc_track_unlink(handle, dentry);
3274 #ifdef CONFIG_UNICODE
3275 /* VFS negative dentries are incompatible with Encoding and
3276 * Case-insensitiveness. Eventually we'll want avoid
3277 * invalidating the dentries here, alongside with returning the
3278 * negative dentries at ext4_lookup(), when it is better
3279 * supported by the VFS for the CI case.
3281 if (IS_CASEFOLDED(dir))
3282 d_invalidate(dentry);
3285 ext4_journal_stop(handle);
3288 trace_ext4_unlink_exit(dentry, retval);
3292 static int ext4_symlink(struct inode *dir,
3293 struct dentry *dentry, const char *symname)
3296 struct inode *inode;
3297 int err, len = strlen(symname);
3299 struct fscrypt_str disk_link;
3301 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3304 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3309 err = dquot_initialize(dir);
3313 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3315 * For non-fast symlinks, we just allocate inode and put it on
3316 * orphan list in the first transaction => we need bitmap,
3317 * group descriptor, sb, inode block, quota blocks, and
3318 * possibly selinux xattr blocks.
3320 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3321 EXT4_XATTR_TRANS_BLOCKS;
3324 * Fast symlink. We have to add entry to directory
3325 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3326 * allocate new inode (bitmap, group descriptor, inode block,
3327 * quota blocks, sb is already counted in previous macros).
3329 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3330 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3333 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3334 &dentry->d_name, 0, NULL,
3335 EXT4_HT_DIR, credits);
3336 handle = ext4_journal_current_handle();
3337 if (IS_ERR(inode)) {
3339 ext4_journal_stop(handle);
3340 return PTR_ERR(inode);
3343 if (IS_ENCRYPTED(inode)) {
3344 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3346 goto err_drop_inode;
3347 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3350 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3351 if (!IS_ENCRYPTED(inode))
3352 inode->i_op = &ext4_symlink_inode_operations;
3353 inode_nohighmem(inode);
3354 ext4_set_aops(inode);
3356 * We cannot call page_symlink() with transaction started
3357 * because it calls into ext4_write_begin() which can wait
3358 * for transaction commit if we are running out of space
3359 * and thus we deadlock. So we have to stop transaction now
3360 * and restart it when symlink contents is written.
3362 * To keep fs consistent in case of crash, we have to put inode
3363 * to orphan list in the mean time.
3366 err = ext4_orphan_add(handle, inode);
3368 ext4_journal_stop(handle);
3371 goto err_drop_inode;
3372 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3374 goto err_drop_inode;
3376 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3377 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3379 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3380 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3381 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3382 if (IS_ERR(handle)) {
3383 err = PTR_ERR(handle);
3385 goto err_drop_inode;
3387 set_nlink(inode, 1);
3388 err = ext4_orphan_del(handle, inode);
3390 goto err_drop_inode;
3392 /* clear the extent format for fast symlink */
3393 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3394 if (!IS_ENCRYPTED(inode)) {
3395 inode->i_op = &ext4_fast_symlink_inode_operations;
3396 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3398 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3400 inode->i_size = disk_link.len - 1;
3402 EXT4_I(inode)->i_disksize = inode->i_size;
3403 err = ext4_add_nondir(handle, dentry, &inode);
3405 ext4_journal_stop(handle);
3408 goto out_free_encrypted_link;
3412 ext4_journal_stop(handle);
3414 unlock_new_inode(inode);
3416 out_free_encrypted_link:
3417 if (disk_link.name != (unsigned char *)symname)
3418 kfree(disk_link.name);
3422 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3425 int err, retries = 0;
3427 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3428 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3429 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3431 return PTR_ERR(handle);
3433 if (IS_DIRSYNC(dir))
3434 ext4_handle_sync(handle);
3436 inode->i_ctime = current_time(inode);
3437 ext4_inc_count(inode);
3440 err = ext4_add_entry(handle, dentry, inode);
3442 err = ext4_mark_inode_dirty(handle, inode);
3443 /* this can happen only for tmpfile being
3444 * linked the first time
3446 if (inode->i_nlink == 1)
3447 ext4_orphan_del(handle, inode);
3448 d_instantiate(dentry, inode);
3449 ext4_fc_track_link(handle, dentry);
3454 ext4_journal_stop(handle);
3455 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3460 static int ext4_link(struct dentry *old_dentry,
3461 struct inode *dir, struct dentry *dentry)
3463 struct inode *inode = d_inode(old_dentry);
3466 if (inode->i_nlink >= EXT4_LINK_MAX)
3469 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3473 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3474 (!projid_eq(EXT4_I(dir)->i_projid,
3475 EXT4_I(old_dentry->d_inode)->i_projid)))
3478 err = dquot_initialize(dir);
3481 return __ext4_link(dir, inode, dentry);
3485 * Try to find buffer head where contains the parent block.
3486 * It should be the inode block if it is inlined or the 1st block
3487 * if it is a normal dir.
3489 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3490 struct inode *inode,
3492 struct ext4_dir_entry_2 **parent_de,
3495 struct buffer_head *bh;
3497 if (!ext4_has_inline_data(inode)) {
3498 /* The first directory block must not be a hole, so
3499 * treat it as DIRENT_HTREE
3501 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3503 *retval = PTR_ERR(bh);
3506 *parent_de = ext4_next_entry(
3507 (struct ext4_dir_entry_2 *)bh->b_data,
3508 inode->i_sb->s_blocksize);
3513 return ext4_get_first_inline_block(inode, parent_de, retval);
3516 struct ext4_renament {
3518 struct dentry *dentry;
3519 struct inode *inode;
3521 int dir_nlink_delta;
3523 /* entry for "dentry" */
3524 struct buffer_head *bh;
3525 struct ext4_dir_entry_2 *de;
3528 /* entry for ".." in inode if it's a directory */
3529 struct buffer_head *dir_bh;
3530 struct ext4_dir_entry_2 *parent_de;
3534 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3538 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3539 &retval, &ent->parent_de,
3543 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3544 return -EFSCORRUPTED;
3545 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3546 return ext4_journal_get_write_access(handle, ent->dir_bh);
3549 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3554 ent->parent_de->inode = cpu_to_le32(dir_ino);
3555 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3556 if (!ent->dir_inlined) {
3557 if (is_dx(ent->inode)) {
3558 retval = ext4_handle_dirty_dx_node(handle,
3562 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3566 retval = ext4_mark_inode_dirty(handle, ent->inode);
3569 ext4_std_error(ent->dir->i_sb, retval);
3575 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3576 unsigned ino, unsigned file_type)
3578 int retval, retval2;
3580 BUFFER_TRACE(ent->bh, "get write access");
3581 retval = ext4_journal_get_write_access(handle, ent->bh);
3584 ent->de->inode = cpu_to_le32(ino);
3585 if (ext4_has_feature_filetype(ent->dir->i_sb))
3586 ent->de->file_type = file_type;
3587 inode_inc_iversion(ent->dir);
3588 ent->dir->i_ctime = ent->dir->i_mtime =
3589 current_time(ent->dir);
3590 retval = ext4_mark_inode_dirty(handle, ent->dir);
3591 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3592 if (!ent->inlined) {
3593 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3594 if (unlikely(retval2)) {
3595 ext4_std_error(ent->dir->i_sb, retval2);
3605 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3606 const struct qstr *d_name)
3608 int retval = -ENOENT;
3609 struct buffer_head *bh;
3610 struct ext4_dir_entry_2 *de;
3612 bh = ext4_find_entry(dir, d_name, &de, NULL);
3616 retval = ext4_delete_entry(handle, dir, de, bh);
3622 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3627 * ent->de could have moved from under us during htree split, so make
3628 * sure that we are deleting the right entry. We might also be pointing
3629 * to a stale entry in the unused part of ent->bh so just checking inum
3630 * and the name isn't enough.
3632 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3633 ent->de->name_len != ent->dentry->d_name.len ||
3634 strncmp(ent->de->name, ent->dentry->d_name.name,
3635 ent->de->name_len) ||
3637 retval = ext4_find_delete_entry(handle, ent->dir,
3638 &ent->dentry->d_name);
3640 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3641 if (retval == -ENOENT) {
3642 retval = ext4_find_delete_entry(handle, ent->dir,
3643 &ent->dentry->d_name);
3648 ext4_warning_inode(ent->dir,
3649 "Deleting old file: nlink %d, error=%d",
3650 ent->dir->i_nlink, retval);
3654 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3656 if (ent->dir_nlink_delta) {
3657 if (ent->dir_nlink_delta == -1)
3658 ext4_dec_count(ent->dir);
3660 ext4_inc_count(ent->dir);
3661 ext4_mark_inode_dirty(handle, ent->dir);
3665 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3666 int credits, handle_t **h)
3673 * for inode block, sb block, group summaries,
3676 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3677 EXT4_XATTR_TRANS_BLOCKS + 4);
3679 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3680 &ent->dentry->d_name, 0, NULL,
3681 EXT4_HT_DIR, credits);
3683 handle = ext4_journal_current_handle();
3686 ext4_journal_stop(handle);
3687 if (PTR_ERR(wh) == -ENOSPC &&
3688 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3692 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3693 wh->i_op = &ext4_special_inode_operations;
3699 * Anybody can rename anything with this: the permission checks are left to the
3700 * higher-level routines.
3702 * n.b. old_{dentry,inode) refers to the source dentry/inode
3703 * while new_{dentry,inode) refers to the destination dentry/inode
3704 * This comes from rename(const char *oldpath, const char *newpath)
3706 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3707 struct inode *new_dir, struct dentry *new_dentry,
3710 handle_t *handle = NULL;
3711 struct ext4_renament old = {
3713 .dentry = old_dentry,
3714 .inode = d_inode(old_dentry),
3716 struct ext4_renament new = {
3718 .dentry = new_dentry,
3719 .inode = d_inode(new_dentry),
3723 struct inode *whiteout = NULL;
3727 if (new.inode && new.inode->i_nlink == 0) {
3728 EXT4_ERROR_INODE(new.inode,
3729 "target of rename is already freed");
3730 return -EFSCORRUPTED;
3733 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3734 (!projid_eq(EXT4_I(new_dir)->i_projid,
3735 EXT4_I(old_dentry->d_inode)->i_projid)))
3738 retval = dquot_initialize(old.dir);
3741 retval = dquot_initialize(new.dir);
3745 /* Initialize quotas before so that eventual writes go
3746 * in separate transaction */
3748 retval = dquot_initialize(new.inode);
3753 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3755 return PTR_ERR(old.bh);
3757 * Check for inode number is _not_ due to possible IO errors.
3758 * We might rmdir the source, keep it as pwd of some process
3759 * and merrily kill the link to whatever was created under the
3760 * same name. Goodbye sticky bit ;-<
3763 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3766 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3767 &new.de, &new.inlined);
3768 if (IS_ERR(new.bh)) {
3769 retval = PTR_ERR(new.bh);
3779 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3780 ext4_alloc_da_blocks(old.inode);
3782 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3783 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3784 if (!(flags & RENAME_WHITEOUT)) {
3785 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3786 if (IS_ERR(handle)) {
3787 retval = PTR_ERR(handle);
3792 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3793 if (IS_ERR(whiteout)) {
3794 retval = PTR_ERR(whiteout);
3800 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3801 ext4_handle_sync(handle);
3803 if (S_ISDIR(old.inode->i_mode)) {
3805 retval = -ENOTEMPTY;
3806 if (!ext4_empty_dir(new.inode))
3810 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3813 retval = ext4_rename_dir_prepare(handle, &old);
3818 * If we're renaming a file within an inline_data dir and adding or
3819 * setting the new dirent causes a conversion from inline_data to
3820 * extents/blockmap, we need to force the dirent delete code to
3821 * re-read the directory, or else we end up trying to delete a dirent
3822 * from what is now the extent tree root (or a block map).
3824 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3825 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3827 old_file_type = old.de->file_type;
3830 * Do this before adding a new entry, so the old entry is sure
3831 * to be still pointing to the valid old entry.
3833 retval = ext4_setent(handle, &old, whiteout->i_ino,
3837 retval = ext4_mark_inode_dirty(handle, whiteout);
3838 if (unlikely(retval))
3842 retval = ext4_add_entry(handle, new.dentry, old.inode);
3846 retval = ext4_setent(handle, &new,
3847 old.inode->i_ino, old_file_type);
3852 force_reread = !ext4_test_inode_flag(new.dir,
3853 EXT4_INODE_INLINE_DATA);
3856 * Like most other Unix systems, set the ctime for inodes on a
3859 old.inode->i_ctime = current_time(old.inode);
3860 retval = ext4_mark_inode_dirty(handle, old.inode);
3861 if (unlikely(retval))
3868 ext4_rename_delete(handle, &old, force_reread);
3872 ext4_dec_count(new.inode);
3873 new.inode->i_ctime = current_time(new.inode);
3875 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3876 ext4_update_dx_flag(old.dir);
3878 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3882 ext4_dec_count(old.dir);
3884 /* checked ext4_empty_dir above, can't have another
3885 * parent, ext4_dec_count() won't work for many-linked
3887 clear_nlink(new.inode);
3889 ext4_inc_count(new.dir);
3890 ext4_update_dx_flag(new.dir);
3891 retval = ext4_mark_inode_dirty(handle, new.dir);
3892 if (unlikely(retval))
3896 retval = ext4_mark_inode_dirty(handle, old.dir);
3897 if (unlikely(retval))
3900 if (S_ISDIR(old.inode->i_mode)) {
3902 * We disable fast commits here that's because the
3903 * replay code is not yet capable of changing dot dot
3904 * dirents in directories.
3906 ext4_fc_mark_ineligible(old.inode->i_sb,
3907 EXT4_FC_REASON_RENAME_DIR);
3910 ext4_fc_track_unlink(handle, new.dentry);
3911 __ext4_fc_track_link(handle, old.inode, new.dentry);
3912 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3916 retval = ext4_mark_inode_dirty(handle, new.inode);
3917 if (unlikely(retval))
3919 if (!new.inode->i_nlink)
3920 ext4_orphan_add(handle, new.inode);
3930 drop_nlink(whiteout);
3931 unlock_new_inode(whiteout);
3935 ext4_journal_stop(handle);
3939 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3940 struct inode *new_dir, struct dentry *new_dentry)
3942 handle_t *handle = NULL;
3943 struct ext4_renament old = {
3945 .dentry = old_dentry,
3946 .inode = d_inode(old_dentry),
3948 struct ext4_renament new = {
3950 .dentry = new_dentry,
3951 .inode = d_inode(new_dentry),
3955 struct timespec64 ctime;
3957 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3958 !projid_eq(EXT4_I(new_dir)->i_projid,
3959 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3960 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3961 !projid_eq(EXT4_I(old_dir)->i_projid,
3962 EXT4_I(new_dentry->d_inode)->i_projid)))
3965 retval = dquot_initialize(old.dir);
3968 retval = dquot_initialize(new.dir);
3972 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3973 &old.de, &old.inlined);
3975 return PTR_ERR(old.bh);
3977 * Check for inode number is _not_ due to possible IO errors.
3978 * We might rmdir the source, keep it as pwd of some process
3979 * and merrily kill the link to whatever was created under the
3980 * same name. Goodbye sticky bit ;-<
3983 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3986 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3987 &new.de, &new.inlined);
3988 if (IS_ERR(new.bh)) {
3989 retval = PTR_ERR(new.bh);
3994 /* RENAME_EXCHANGE case: old *and* new must both exist */
3995 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3998 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3999 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4000 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4001 if (IS_ERR(handle)) {
4002 retval = PTR_ERR(handle);
4007 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4008 ext4_handle_sync(handle);
4010 if (S_ISDIR(old.inode->i_mode)) {
4012 retval = ext4_rename_dir_prepare(handle, &old);
4016 if (S_ISDIR(new.inode->i_mode)) {
4018 retval = ext4_rename_dir_prepare(handle, &new);
4024 * Other than the special case of overwriting a directory, parents'
4025 * nlink only needs to be modified if this is a cross directory rename.
4027 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4028 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4029 new.dir_nlink_delta = -old.dir_nlink_delta;
4031 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4032 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4036 new_file_type = new.de->file_type;
4037 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4041 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4046 * Like most other Unix systems, set the ctime for inodes on a
4049 ctime = current_time(old.inode);
4050 old.inode->i_ctime = ctime;
4051 new.inode->i_ctime = ctime;
4052 retval = ext4_mark_inode_dirty(handle, old.inode);
4053 if (unlikely(retval))
4055 retval = ext4_mark_inode_dirty(handle, new.inode);
4056 if (unlikely(retval))
4058 ext4_fc_mark_ineligible(new.inode->i_sb,
4059 EXT4_FC_REASON_CROSS_RENAME);
4061 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4066 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4070 ext4_update_dir_count(handle, &old);
4071 ext4_update_dir_count(handle, &new);
4080 ext4_journal_stop(handle);
4084 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4085 struct inode *new_dir, struct dentry *new_dentry,
4090 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4093 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4096 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4101 if (flags & RENAME_EXCHANGE) {
4102 return ext4_cross_rename(old_dir, old_dentry,
4103 new_dir, new_dentry);
4106 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4110 * directories can handle most operations...
4112 const struct inode_operations ext4_dir_inode_operations = {
4113 .create = ext4_create,
4114 .lookup = ext4_lookup,
4116 .unlink = ext4_unlink,
4117 .symlink = ext4_symlink,
4118 .mkdir = ext4_mkdir,
4119 .rmdir = ext4_rmdir,
4120 .mknod = ext4_mknod,
4121 .tmpfile = ext4_tmpfile,
4122 .rename = ext4_rename2,
4123 .setattr = ext4_setattr,
4124 .getattr = ext4_getattr,
4125 .listxattr = ext4_listxattr,
4126 .get_acl = ext4_get_acl,
4127 .set_acl = ext4_set_acl,
4128 .fiemap = ext4_fiemap,
4131 const struct inode_operations ext4_special_inode_operations = {
4132 .setattr = ext4_setattr,
4133 .getattr = ext4_getattr,
4134 .listxattr = ext4_listxattr,
4135 .get_acl = ext4_get_acl,
4136 .set_acl = ext4_set_acl,