Merge tag 'regmap-fix-v5.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux/fpc-iii.git] / fs / ext4 / namei.c
blobb17a082b7db15298e99813cb22d07fd4d0c2e458
1 // SPDX-License-Identifier: GPL-2.0
2 /*
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)
10 * from
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
25 * Theodore Ts'o, 2002
28 #include <linux/fs.h>
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>
39 #include "ext4.h"
40 #include "ext4_jbd2.h"
42 #include "xattr.h"
43 #include "acl.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,
54 struct inode *inode,
55 ext4_lblk_t *block)
57 struct buffer_head *bh;
58 int err;
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);
68 if (IS_ERR(bh))
69 return bh;
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);
74 if (err) {
75 brelse(bh);
76 ext4_std_error(inode->i_sb, err);
77 return ERR_PTR(err);
79 return bh;
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.
95 typedef enum {
96 EITHER, INDEX, DIRENT, DIRENT_HTREE
97 } dirblock_type_t;
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,
103 ext4_lblk_t block,
104 dirblock_type_t type,
105 const char *func,
106 unsigned int line)
108 struct buffer_head *bh;
109 struct ext4_dir_entry *dirent;
110 int is_dx_block = 0;
112 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
113 bh = ERR_PTR(-EIO);
114 else
115 bh = ext4_bread(NULL, inode, block, 0);
116 if (IS_ERR(bh)) {
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));
123 return 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);
131 if (!bh)
132 return NULL;
133 dirent = (struct ext4_dir_entry *) bh->b_data;
134 /* Determine whether or not we have an index block */
135 if (is_dx(inode)) {
136 if (block == 0)
137 is_dx_block = 1;
138 else if (ext4_rec_len_from_disk(dirent->rec_len,
139 inode->i_sb->s_blocksize) ==
140 inode->i_sb->s_blocksize)
141 is_dx_block = 1;
143 if (!is_dx_block && type == INDEX) {
144 ext4_error_inode(inode, func, line, block,
145 "directory leaf block found instead of index block");
146 brelse(bh);
147 return ERR_PTR(-EFSCORRUPTED);
149 if (!ext4_has_metadata_csum(inode->i_sb) ||
150 buffer_verified(bh))
151 return bh;
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);
162 else {
163 ext4_error_inode_err(inode, func, line, block,
164 EFSBADCRC,
165 "Directory index failed checksum");
166 brelse(bh);
167 return ERR_PTR(-EFSBADCRC);
170 if (!is_dx_block) {
171 if (ext4_dirblock_csum_verify(inode, bh) &&
172 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
173 set_buffer_verified(bh);
174 else {
175 ext4_error_inode_err(inode, func, line, block,
176 EFSBADCRC,
177 "Directory block failed checksum");
178 brelse(bh);
179 return ERR_PTR(-EFSBADCRC);
182 return bh;
185 #ifdef DX_DEBUG
186 #define dxtrace(command) command
187 #else
188 #define dxtrace(command)
189 #endif
191 struct fake_dirent
193 __le32 inode;
194 __le16 rec_len;
195 u8 name_len;
196 u8 file_type;
199 struct dx_countlimit
201 __le16 limit;
202 __le16 count;
205 struct dx_entry
207 __le32 hash;
208 __le32 block;
212 * dx_root_info is laid out so that if it should somehow get overlaid by a
213 * dirent the two low bits of the hash version will be zero. Therefore, the
214 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
217 struct dx_root
219 struct fake_dirent dot;
220 char dot_name[4];
221 struct fake_dirent dotdot;
222 char dotdot_name[4];
223 struct dx_root_info
225 __le32 reserved_zero;
226 u8 hash_version;
227 u8 info_length; /* 8 */
228 u8 indirect_levels;
229 u8 unused_flags;
231 info;
232 struct dx_entry entries[];
235 struct dx_node
237 struct fake_dirent fake;
238 struct dx_entry entries[];
242 struct dx_frame
244 struct buffer_head *bh;
245 struct dx_entry *entries;
246 struct dx_entry *at;
249 struct dx_map_entry
251 u32 hash;
252 u16 offs;
253 u16 size;
257 * This goes at the end of each htree block.
259 struct dx_tail {
260 u32 dt_reserved;
261 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
264 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
265 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
266 static inline unsigned dx_get_hash(struct dx_entry *entry);
267 static void dx_set_hash(struct dx_entry *entry, unsigned value);
268 static unsigned dx_get_count(struct dx_entry *entries);
269 static unsigned dx_get_limit(struct dx_entry *entries);
270 static void dx_set_count(struct dx_entry *entries, unsigned value);
271 static void dx_set_limit(struct dx_entry *entries, unsigned value);
272 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
273 static unsigned dx_node_limit(struct inode *dir);
274 static struct dx_frame *dx_probe(struct ext4_filename *fname,
275 struct inode *dir,
276 struct dx_hash_info *hinfo,
277 struct dx_frame *frame);
278 static void dx_release(struct dx_frame *frames);
279 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
280 unsigned blocksize, struct dx_hash_info *hinfo,
281 struct dx_map_entry map[]);
282 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
283 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
284 struct dx_map_entry *offsets, int count, unsigned blocksize);
285 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
286 static void dx_insert_block(struct dx_frame *frame,
287 u32 hash, ext4_lblk_t block);
288 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
289 struct dx_frame *frame,
290 struct dx_frame *frames,
291 __u32 *start_hash);
292 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
293 struct ext4_filename *fname,
294 struct ext4_dir_entry_2 **res_dir);
295 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
296 struct inode *dir, struct inode *inode);
298 /* checksumming functions */
299 void ext4_initialize_dirent_tail(struct buffer_head *bh,
300 unsigned int blocksize)
302 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
304 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
305 t->det_rec_len = ext4_rec_len_to_disk(
306 sizeof(struct ext4_dir_entry_tail), blocksize);
307 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
310 /* Walk through a dirent block to find a checksum "dirent" at the tail */
311 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
312 struct buffer_head *bh)
314 struct ext4_dir_entry_tail *t;
316 #ifdef PARANOID
317 struct ext4_dir_entry *d, *top;
319 d = (struct ext4_dir_entry *)bh->b_data;
320 top = (struct ext4_dir_entry *)(bh->b_data +
321 (EXT4_BLOCK_SIZE(inode->i_sb) -
322 sizeof(struct ext4_dir_entry_tail)));
323 while (d < top && d->rec_len)
324 d = (struct ext4_dir_entry *)(((void *)d) +
325 le16_to_cpu(d->rec_len));
327 if (d != top)
328 return NULL;
330 t = (struct ext4_dir_entry_tail *)d;
331 #else
332 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
333 #endif
335 if (t->det_reserved_zero1 ||
336 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
337 t->det_reserved_zero2 ||
338 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
339 return NULL;
341 return t;
344 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
346 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
347 struct ext4_inode_info *ei = EXT4_I(inode);
348 __u32 csum;
350 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
351 return cpu_to_le32(csum);
354 #define warn_no_space_for_csum(inode) \
355 __warn_no_space_for_csum((inode), __func__, __LINE__)
357 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
358 unsigned int line)
360 __ext4_warning_inode(inode, func, line,
361 "No space for directory leaf checksum. Please run e2fsck -D.");
364 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
366 struct ext4_dir_entry_tail *t;
368 if (!ext4_has_metadata_csum(inode->i_sb))
369 return 1;
371 t = get_dirent_tail(inode, bh);
372 if (!t) {
373 warn_no_space_for_csum(inode);
374 return 0;
377 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
378 (char *)t - bh->b_data))
379 return 0;
381 return 1;
384 static void ext4_dirblock_csum_set(struct inode *inode,
385 struct buffer_head *bh)
387 struct ext4_dir_entry_tail *t;
389 if (!ext4_has_metadata_csum(inode->i_sb))
390 return;
392 t = get_dirent_tail(inode, bh);
393 if (!t) {
394 warn_no_space_for_csum(inode);
395 return;
398 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
399 (char *)t - bh->b_data);
402 int ext4_handle_dirty_dirblock(handle_t *handle,
403 struct inode *inode,
404 struct buffer_head *bh)
406 ext4_dirblock_csum_set(inode, bh);
407 return ext4_handle_dirty_metadata(handle, inode, bh);
410 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
411 struct ext4_dir_entry *dirent,
412 int *offset)
414 struct ext4_dir_entry *dp;
415 struct dx_root_info *root;
416 int count_offset;
418 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
419 count_offset = 8;
420 else if (le16_to_cpu(dirent->rec_len) == 12) {
421 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
422 if (le16_to_cpu(dp->rec_len) !=
423 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
424 return NULL;
425 root = (struct dx_root_info *)(((void *)dp + 12));
426 if (root->reserved_zero ||
427 root->info_length != sizeof(struct dx_root_info))
428 return NULL;
429 count_offset = 32;
430 } else
431 return NULL;
433 if (offset)
434 *offset = count_offset;
435 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
438 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
439 int count_offset, int count, struct dx_tail *t)
441 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
442 struct ext4_inode_info *ei = EXT4_I(inode);
443 __u32 csum;
444 int size;
445 __u32 dummy_csum = 0;
446 int offset = offsetof(struct dx_tail, dt_checksum);
448 size = count_offset + (count * sizeof(struct dx_entry));
449 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
450 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
451 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
453 return cpu_to_le32(csum);
456 static int ext4_dx_csum_verify(struct inode *inode,
457 struct ext4_dir_entry *dirent)
459 struct dx_countlimit *c;
460 struct dx_tail *t;
461 int count_offset, limit, count;
463 if (!ext4_has_metadata_csum(inode->i_sb))
464 return 1;
466 c = get_dx_countlimit(inode, dirent, &count_offset);
467 if (!c) {
468 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
469 return 0;
471 limit = le16_to_cpu(c->limit);
472 count = le16_to_cpu(c->count);
473 if (count_offset + (limit * sizeof(struct dx_entry)) >
474 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
475 warn_no_space_for_csum(inode);
476 return 0;
478 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
480 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
481 count, t))
482 return 0;
483 return 1;
486 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
488 struct dx_countlimit *c;
489 struct dx_tail *t;
490 int count_offset, limit, count;
492 if (!ext4_has_metadata_csum(inode->i_sb))
493 return;
495 c = get_dx_countlimit(inode, dirent, &count_offset);
496 if (!c) {
497 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
498 return;
500 limit = le16_to_cpu(c->limit);
501 count = le16_to_cpu(c->count);
502 if (count_offset + (limit * sizeof(struct dx_entry)) >
503 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
504 warn_no_space_for_csum(inode);
505 return;
507 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
509 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
512 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
513 struct inode *inode,
514 struct buffer_head *bh)
516 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
517 return ext4_handle_dirty_metadata(handle, inode, bh);
521 * p is at least 6 bytes before the end of page
523 static inline struct ext4_dir_entry_2 *
524 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
526 return (struct ext4_dir_entry_2 *)((char *)p +
527 ext4_rec_len_from_disk(p->rec_len, blocksize));
531 * Future: use high four bits of block for coalesce-on-delete flags
532 * Mask them off for now.
535 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
537 return le32_to_cpu(entry->block) & 0x0fffffff;
540 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
542 entry->block = cpu_to_le32(value);
545 static inline unsigned dx_get_hash(struct dx_entry *entry)
547 return le32_to_cpu(entry->hash);
550 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
552 entry->hash = cpu_to_le32(value);
555 static inline unsigned dx_get_count(struct dx_entry *entries)
557 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
560 static inline unsigned dx_get_limit(struct dx_entry *entries)
562 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
565 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
567 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
570 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
572 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
575 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
577 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
578 EXT4_DIR_REC_LEN(2) - infosize;
580 if (ext4_has_metadata_csum(dir->i_sb))
581 entry_space -= sizeof(struct dx_tail);
582 return entry_space / sizeof(struct dx_entry);
585 static inline unsigned dx_node_limit(struct inode *dir)
587 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
589 if (ext4_has_metadata_csum(dir->i_sb))
590 entry_space -= sizeof(struct dx_tail);
591 return entry_space / sizeof(struct dx_entry);
595 * Debug
597 #ifdef DX_DEBUG
598 static void dx_show_index(char * label, struct dx_entry *entries)
600 int i, n = dx_get_count (entries);
601 printk(KERN_DEBUG "%s index", label);
602 for (i = 0; i < n; i++) {
603 printk(KERN_CONT " %x->%lu",
604 i ? dx_get_hash(entries + i) : 0,
605 (unsigned long)dx_get_block(entries + i));
607 printk(KERN_CONT "\n");
610 struct stats
612 unsigned names;
613 unsigned space;
614 unsigned bcount;
617 static struct stats dx_show_leaf(struct inode *dir,
618 struct dx_hash_info *hinfo,
619 struct ext4_dir_entry_2 *de,
620 int size, int show_names)
622 unsigned names = 0, space = 0;
623 char *base = (char *) de;
624 struct dx_hash_info h = *hinfo;
626 printk("names: ");
627 while ((char *) de < base + size)
629 if (de->inode)
631 if (show_names)
633 #ifdef CONFIG_FS_ENCRYPTION
634 int len;
635 char *name;
636 struct fscrypt_str fname_crypto_str =
637 FSTR_INIT(NULL, 0);
638 int res = 0;
640 name = de->name;
641 len = de->name_len;
642 if (!IS_ENCRYPTED(dir)) {
643 /* Directory is not encrypted */
644 ext4fs_dirhash(dir, de->name,
645 de->name_len, &h);
646 printk("%*.s:(U)%x.%u ", len,
647 name, h.hash,
648 (unsigned) ((char *) de
649 - base));
650 } else {
651 struct fscrypt_str de_name =
652 FSTR_INIT(name, len);
654 /* Directory is encrypted */
655 res = fscrypt_fname_alloc_buffer(
656 len, &fname_crypto_str);
657 if (res)
658 printk(KERN_WARNING "Error "
659 "allocating crypto "
660 "buffer--skipping "
661 "crypto\n");
662 res = fscrypt_fname_disk_to_usr(dir,
663 0, 0, &de_name,
664 &fname_crypto_str);
665 if (res) {
666 printk(KERN_WARNING "Error "
667 "converting filename "
668 "from disk to usr"
669 "\n");
670 name = "??";
671 len = 2;
672 } else {
673 name = fname_crypto_str.name;
674 len = fname_crypto_str.len;
676 ext4fs_dirhash(dir, de->name,
677 de->name_len, &h);
678 printk("%*.s:(E)%x.%u ", len, name,
679 h.hash, (unsigned) ((char *) de
680 - base));
681 fscrypt_fname_free_buffer(
682 &fname_crypto_str);
684 #else
685 int len = de->name_len;
686 char *name = de->name;
687 ext4fs_dirhash(dir, de->name, de->name_len, &h);
688 printk("%*.s:%x.%u ", len, name, h.hash,
689 (unsigned) ((char *) de - base));
690 #endif
692 space += EXT4_DIR_REC_LEN(de->name_len);
693 names++;
695 de = ext4_next_entry(de, size);
697 printk(KERN_CONT "(%i)\n", names);
698 return (struct stats) { names, space, 1 };
701 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
702 struct dx_entry *entries, int levels)
704 unsigned blocksize = dir->i_sb->s_blocksize;
705 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
706 unsigned bcount = 0;
707 struct buffer_head *bh;
708 printk("%i indexed blocks...\n", count);
709 for (i = 0; i < count; i++, entries++)
711 ext4_lblk_t block = dx_get_block(entries);
712 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
713 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
714 struct stats stats;
715 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
716 bh = ext4_bread(NULL,dir, block, 0);
717 if (!bh || IS_ERR(bh))
718 continue;
719 stats = levels?
720 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
721 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
722 bh->b_data, blocksize, 0);
723 names += stats.names;
724 space += stats.space;
725 bcount += stats.bcount;
726 brelse(bh);
728 if (bcount)
729 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
730 levels ? "" : " ", names, space/bcount,
731 (space/bcount)*100/blocksize);
732 return (struct stats) { names, space, bcount};
734 #endif /* DX_DEBUG */
737 * Probe for a directory leaf block to search.
739 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
740 * error in the directory index, and the caller should fall back to
741 * searching the directory normally. The callers of dx_probe **MUST**
742 * check for this error code, and make sure it never gets reflected
743 * back to userspace.
745 static struct dx_frame *
746 dx_probe(struct ext4_filename *fname, struct inode *dir,
747 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
749 unsigned count, indirect;
750 struct dx_entry *at, *entries, *p, *q, *m;
751 struct dx_root *root;
752 struct dx_frame *frame = frame_in;
753 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
754 u32 hash;
756 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
757 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
758 if (IS_ERR(frame->bh))
759 return (struct dx_frame *) frame->bh;
761 root = (struct dx_root *) frame->bh->b_data;
762 if (root->info.hash_version != DX_HASH_TEA &&
763 root->info.hash_version != DX_HASH_HALF_MD4 &&
764 root->info.hash_version != DX_HASH_LEGACY) {
765 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
766 root->info.hash_version);
767 goto fail;
769 if (fname)
770 hinfo = &fname->hinfo;
771 hinfo->hash_version = root->info.hash_version;
772 if (hinfo->hash_version <= DX_HASH_TEA)
773 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
774 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
775 if (fname && fname_name(fname))
776 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
777 hash = hinfo->hash;
779 if (root->info.unused_flags & 1) {
780 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
781 root->info.unused_flags);
782 goto fail;
785 indirect = root->info.indirect_levels;
786 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
787 ext4_warning(dir->i_sb,
788 "Directory (ino: %lu) htree depth %#06x exceed"
789 "supported value", dir->i_ino,
790 ext4_dir_htree_level(dir->i_sb));
791 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
792 ext4_warning(dir->i_sb, "Enable large directory "
793 "feature to access it");
795 goto fail;
798 entries = (struct dx_entry *)(((char *)&root->info) +
799 root->info.info_length);
801 if (dx_get_limit(entries) != dx_root_limit(dir,
802 root->info.info_length)) {
803 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
804 dx_get_limit(entries),
805 dx_root_limit(dir, root->info.info_length));
806 goto fail;
809 dxtrace(printk("Look up %x", hash));
810 while (1) {
811 count = dx_get_count(entries);
812 if (!count || count > dx_get_limit(entries)) {
813 ext4_warning_inode(dir,
814 "dx entry: count %u beyond limit %u",
815 count, dx_get_limit(entries));
816 goto fail;
819 p = entries + 1;
820 q = entries + count - 1;
821 while (p <= q) {
822 m = p + (q - p) / 2;
823 dxtrace(printk(KERN_CONT "."));
824 if (dx_get_hash(m) > hash)
825 q = m - 1;
826 else
827 p = m + 1;
830 if (0) { // linear search cross check
831 unsigned n = count - 1;
832 at = entries;
833 while (n--)
835 dxtrace(printk(KERN_CONT ","));
836 if (dx_get_hash(++at) > hash)
838 at--;
839 break;
842 ASSERT(at == p - 1);
845 at = p - 1;
846 dxtrace(printk(KERN_CONT " %x->%u\n",
847 at == entries ? 0 : dx_get_hash(at),
848 dx_get_block(at)));
849 frame->entries = entries;
850 frame->at = at;
851 if (!indirect--)
852 return frame;
853 frame++;
854 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
855 if (IS_ERR(frame->bh)) {
856 ret_err = (struct dx_frame *) frame->bh;
857 frame->bh = NULL;
858 goto fail;
860 entries = ((struct dx_node *) frame->bh->b_data)->entries;
862 if (dx_get_limit(entries) != dx_node_limit(dir)) {
863 ext4_warning_inode(dir,
864 "dx entry: limit %u != node limit %u",
865 dx_get_limit(entries), dx_node_limit(dir));
866 goto fail;
869 fail:
870 while (frame >= frame_in) {
871 brelse(frame->bh);
872 frame--;
875 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
876 ext4_warning_inode(dir,
877 "Corrupt directory, running e2fsck is recommended");
878 return ret_err;
881 static void dx_release(struct dx_frame *frames)
883 struct dx_root_info *info;
884 int i;
885 unsigned int indirect_levels;
887 if (frames[0].bh == NULL)
888 return;
890 info = &((struct dx_root *)frames[0].bh->b_data)->info;
891 /* save local copy, "info" may be freed after brelse() */
892 indirect_levels = info->indirect_levels;
893 for (i = 0; i <= indirect_levels; i++) {
894 if (frames[i].bh == NULL)
895 break;
896 brelse(frames[i].bh);
897 frames[i].bh = NULL;
902 * This function increments the frame pointer to search the next leaf
903 * block, and reads in the necessary intervening nodes if the search
904 * should be necessary. Whether or not the search is necessary is
905 * controlled by the hash parameter. If the hash value is even, then
906 * the search is only continued if the next block starts with that
907 * hash value. This is used if we are searching for a specific file.
909 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
911 * This function returns 1 if the caller should continue to search,
912 * or 0 if it should not. If there is an error reading one of the
913 * index blocks, it will a negative error code.
915 * If start_hash is non-null, it will be filled in with the starting
916 * hash of the next page.
918 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
919 struct dx_frame *frame,
920 struct dx_frame *frames,
921 __u32 *start_hash)
923 struct dx_frame *p;
924 struct buffer_head *bh;
925 int num_frames = 0;
926 __u32 bhash;
928 p = frame;
930 * Find the next leaf page by incrementing the frame pointer.
931 * If we run out of entries in the interior node, loop around and
932 * increment pointer in the parent node. When we break out of
933 * this loop, num_frames indicates the number of interior
934 * nodes need to be read.
936 while (1) {
937 if (++(p->at) < p->entries + dx_get_count(p->entries))
938 break;
939 if (p == frames)
940 return 0;
941 num_frames++;
942 p--;
946 * If the hash is 1, then continue only if the next page has a
947 * continuation hash of any value. This is used for readdir
948 * handling. Otherwise, check to see if the hash matches the
949 * desired contiuation hash. If it doesn't, return since
950 * there's no point to read in the successive index pages.
952 bhash = dx_get_hash(p->at);
953 if (start_hash)
954 *start_hash = bhash;
955 if ((hash & 1) == 0) {
956 if ((bhash & ~1) != hash)
957 return 0;
960 * If the hash is HASH_NB_ALWAYS, we always go to the next
961 * block so no check is necessary
963 while (num_frames--) {
964 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
965 if (IS_ERR(bh))
966 return PTR_ERR(bh);
967 p++;
968 brelse(p->bh);
969 p->bh = bh;
970 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
972 return 1;
977 * This function fills a red-black tree with information from a
978 * directory block. It returns the number directory entries loaded
979 * into the tree. If there is an error it is returned in err.
981 static int htree_dirblock_to_tree(struct file *dir_file,
982 struct inode *dir, ext4_lblk_t block,
983 struct dx_hash_info *hinfo,
984 __u32 start_hash, __u32 start_minor_hash)
986 struct buffer_head *bh;
987 struct ext4_dir_entry_2 *de, *top;
988 int err = 0, count = 0;
989 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
991 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
992 (unsigned long)block));
993 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
994 if (IS_ERR(bh))
995 return PTR_ERR(bh);
997 de = (struct ext4_dir_entry_2 *) bh->b_data;
998 top = (struct ext4_dir_entry_2 *) ((char *) de +
999 dir->i_sb->s_blocksize -
1000 EXT4_DIR_REC_LEN(0));
1001 /* Check if the directory is encrypted */
1002 if (IS_ENCRYPTED(dir)) {
1003 err = fscrypt_prepare_readdir(dir);
1004 if (err < 0) {
1005 brelse(bh);
1006 return err;
1008 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1009 &fname_crypto_str);
1010 if (err < 0) {
1011 brelse(bh);
1012 return err;
1016 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1017 if (ext4_check_dir_entry(dir, NULL, de, bh,
1018 bh->b_data, bh->b_size,
1019 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1020 + ((char *)de - bh->b_data))) {
1021 /* silently ignore the rest of the block */
1022 break;
1024 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1025 if ((hinfo->hash < start_hash) ||
1026 ((hinfo->hash == start_hash) &&
1027 (hinfo->minor_hash < start_minor_hash)))
1028 continue;
1029 if (de->inode == 0)
1030 continue;
1031 if (!IS_ENCRYPTED(dir)) {
1032 tmp_str.name = de->name;
1033 tmp_str.len = de->name_len;
1034 err = ext4_htree_store_dirent(dir_file,
1035 hinfo->hash, hinfo->minor_hash, de,
1036 &tmp_str);
1037 } else {
1038 int save_len = fname_crypto_str.len;
1039 struct fscrypt_str de_name = FSTR_INIT(de->name,
1040 de->name_len);
1042 /* Directory is encrypted */
1043 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1044 hinfo->minor_hash, &de_name,
1045 &fname_crypto_str);
1046 if (err) {
1047 count = err;
1048 goto errout;
1050 err = ext4_htree_store_dirent(dir_file,
1051 hinfo->hash, hinfo->minor_hash, de,
1052 &fname_crypto_str);
1053 fname_crypto_str.len = save_len;
1055 if (err != 0) {
1056 count = err;
1057 goto errout;
1059 count++;
1061 errout:
1062 brelse(bh);
1063 fscrypt_fname_free_buffer(&fname_crypto_str);
1064 return count;
1069 * This function fills a red-black tree with information from a
1070 * directory. We start scanning the directory in hash order, starting
1071 * at start_hash and start_minor_hash.
1073 * This function returns the number of entries inserted into the tree,
1074 * or a negative error code.
1076 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1077 __u32 start_minor_hash, __u32 *next_hash)
1079 struct dx_hash_info hinfo;
1080 struct ext4_dir_entry_2 *de;
1081 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1082 struct inode *dir;
1083 ext4_lblk_t block;
1084 int count = 0;
1085 int ret, err;
1086 __u32 hashval;
1087 struct fscrypt_str tmp_str;
1089 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1090 start_hash, start_minor_hash));
1091 dir = file_inode(dir_file);
1092 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1093 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1094 if (hinfo.hash_version <= DX_HASH_TEA)
1095 hinfo.hash_version +=
1096 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1097 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1098 if (ext4_has_inline_data(dir)) {
1099 int has_inline_data = 1;
1100 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1101 &hinfo, start_hash,
1102 start_minor_hash,
1103 &has_inline_data);
1104 if (has_inline_data) {
1105 *next_hash = ~0;
1106 return count;
1109 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1110 start_hash, start_minor_hash);
1111 *next_hash = ~0;
1112 return count;
1114 hinfo.hash = start_hash;
1115 hinfo.minor_hash = 0;
1116 frame = dx_probe(NULL, dir, &hinfo, frames);
1117 if (IS_ERR(frame))
1118 return PTR_ERR(frame);
1120 /* Add '.' and '..' from the htree header */
1121 if (!start_hash && !start_minor_hash) {
1122 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1123 tmp_str.name = de->name;
1124 tmp_str.len = de->name_len;
1125 err = ext4_htree_store_dirent(dir_file, 0, 0,
1126 de, &tmp_str);
1127 if (err != 0)
1128 goto errout;
1129 count++;
1131 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1132 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1133 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1134 tmp_str.name = de->name;
1135 tmp_str.len = de->name_len;
1136 err = ext4_htree_store_dirent(dir_file, 2, 0,
1137 de, &tmp_str);
1138 if (err != 0)
1139 goto errout;
1140 count++;
1143 while (1) {
1144 if (fatal_signal_pending(current)) {
1145 err = -ERESTARTSYS;
1146 goto errout;
1148 cond_resched();
1149 block = dx_get_block(frame->at);
1150 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1151 start_hash, start_minor_hash);
1152 if (ret < 0) {
1153 err = ret;
1154 goto errout;
1156 count += ret;
1157 hashval = ~0;
1158 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1159 frame, frames, &hashval);
1160 *next_hash = hashval;
1161 if (ret < 0) {
1162 err = ret;
1163 goto errout;
1166 * Stop if: (a) there are no more entries, or
1167 * (b) we have inserted at least one entry and the
1168 * next hash value is not a continuation
1170 if ((ret == 0) ||
1171 (count && ((hashval & 1) == 0)))
1172 break;
1174 dx_release(frames);
1175 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1176 "next hash: %x\n", count, *next_hash));
1177 return count;
1178 errout:
1179 dx_release(frames);
1180 return (err);
1183 static inline int search_dirblock(struct buffer_head *bh,
1184 struct inode *dir,
1185 struct ext4_filename *fname,
1186 unsigned int offset,
1187 struct ext4_dir_entry_2 **res_dir)
1189 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1190 fname, offset, res_dir);
1194 * Directory block splitting, compacting
1198 * Create map of hash values, offsets, and sizes, stored at end of block.
1199 * Returns number of entries mapped.
1201 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1202 unsigned blocksize, struct dx_hash_info *hinfo,
1203 struct dx_map_entry *map_tail)
1205 int count = 0;
1206 char *base = (char *) de;
1207 struct dx_hash_info h = *hinfo;
1209 while ((char *) de < base + blocksize) {
1210 if (de->name_len && de->inode) {
1211 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1212 map_tail--;
1213 map_tail->hash = h.hash;
1214 map_tail->offs = ((char *) de - base)>>2;
1215 map_tail->size = le16_to_cpu(de->rec_len);
1216 count++;
1217 cond_resched();
1219 /* XXX: do we need to check rec_len == 0 case? -Chris */
1220 de = ext4_next_entry(de, blocksize);
1222 return count;
1225 /* Sort map by hash value */
1226 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1228 struct dx_map_entry *p, *q, *top = map + count - 1;
1229 int more;
1230 /* Combsort until bubble sort doesn't suck */
1231 while (count > 2) {
1232 count = count*10/13;
1233 if (count - 9 < 2) /* 9, 10 -> 11 */
1234 count = 11;
1235 for (p = top, q = p - count; q >= map; p--, q--)
1236 if (p->hash < q->hash)
1237 swap(*p, *q);
1239 /* Garden variety bubble sort */
1240 do {
1241 more = 0;
1242 q = top;
1243 while (q-- > map) {
1244 if (q[1].hash >= q[0].hash)
1245 continue;
1246 swap(*(q+1), *q);
1247 more = 1;
1249 } while(more);
1252 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1254 struct dx_entry *entries = frame->entries;
1255 struct dx_entry *old = frame->at, *new = old + 1;
1256 int count = dx_get_count(entries);
1258 ASSERT(count < dx_get_limit(entries));
1259 ASSERT(old < entries + count);
1260 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1261 dx_set_hash(new, hash);
1262 dx_set_block(new, block);
1263 dx_set_count(entries, count + 1);
1266 #ifdef CONFIG_UNICODE
1268 * Test whether a case-insensitive directory entry matches the filename
1269 * being searched for. If quick is set, assume the name being looked up
1270 * is already in the casefolded form.
1272 * Returns: 0 if the directory entry matches, more than 0 if it
1273 * doesn't match or less than zero on error.
1275 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1276 const struct qstr *entry, bool quick)
1278 const struct super_block *sb = parent->i_sb;
1279 const struct unicode_map *um = sb->s_encoding;
1280 int ret;
1282 if (quick)
1283 ret = utf8_strncasecmp_folded(um, name, entry);
1284 else
1285 ret = utf8_strncasecmp(um, name, entry);
1287 if (ret < 0) {
1288 /* Handle invalid character sequence as either an error
1289 * or as an opaque byte sequence.
1291 if (sb_has_strict_encoding(sb))
1292 return -EINVAL;
1294 if (name->len != entry->len)
1295 return 1;
1297 return !!memcmp(name->name, entry->name, name->len);
1300 return ret;
1303 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1304 struct fscrypt_str *cf_name)
1306 int len;
1308 if (!IS_CASEFOLDED(dir) || !dir->i_sb->s_encoding) {
1309 cf_name->name = NULL;
1310 return;
1313 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1314 if (!cf_name->name)
1315 return;
1317 len = utf8_casefold(dir->i_sb->s_encoding,
1318 iname, cf_name->name,
1319 EXT4_NAME_LEN);
1320 if (len <= 0) {
1321 kfree(cf_name->name);
1322 cf_name->name = NULL;
1323 return;
1325 cf_name->len = (unsigned) len;
1328 #endif
1331 * Test whether a directory entry matches the filename being searched for.
1333 * Return: %true if the directory entry matches, otherwise %false.
1335 static inline bool ext4_match(const struct inode *parent,
1336 const struct ext4_filename *fname,
1337 const struct ext4_dir_entry_2 *de)
1339 struct fscrypt_name f;
1340 #ifdef CONFIG_UNICODE
1341 const struct qstr entry = {.name = de->name, .len = de->name_len};
1342 #endif
1344 if (!de->inode)
1345 return false;
1347 f.usr_fname = fname->usr_fname;
1348 f.disk_name = fname->disk_name;
1349 #ifdef CONFIG_FS_ENCRYPTION
1350 f.crypto_buf = fname->crypto_buf;
1351 #endif
1353 #ifdef CONFIG_UNICODE
1354 if (parent->i_sb->s_encoding && IS_CASEFOLDED(parent)) {
1355 if (fname->cf_name.name) {
1356 struct qstr cf = {.name = fname->cf_name.name,
1357 .len = fname->cf_name.len};
1358 return !ext4_ci_compare(parent, &cf, &entry, true);
1360 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1361 false);
1363 #endif
1365 return fscrypt_match_name(&f, de->name, de->name_len);
1369 * Returns 0 if not found, -1 on failure, and 1 on success
1371 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1372 struct inode *dir, struct ext4_filename *fname,
1373 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1375 struct ext4_dir_entry_2 * de;
1376 char * dlimit;
1377 int de_len;
1379 de = (struct ext4_dir_entry_2 *)search_buf;
1380 dlimit = search_buf + buf_size;
1381 while ((char *) de < dlimit) {
1382 /* this code is executed quadratically often */
1383 /* do minimal checking `by hand' */
1384 if ((char *) de + de->name_len <= dlimit &&
1385 ext4_match(dir, fname, de)) {
1386 /* found a match - just to be sure, do
1387 * a full check */
1388 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1389 buf_size, offset))
1390 return -1;
1391 *res_dir = de;
1392 return 1;
1394 /* prevent looping on a bad block */
1395 de_len = ext4_rec_len_from_disk(de->rec_len,
1396 dir->i_sb->s_blocksize);
1397 if (de_len <= 0)
1398 return -1;
1399 offset += de_len;
1400 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1402 return 0;
1405 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1406 struct ext4_dir_entry *de)
1408 struct super_block *sb = dir->i_sb;
1410 if (!is_dx(dir))
1411 return 0;
1412 if (block == 0)
1413 return 1;
1414 if (de->inode == 0 &&
1415 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1416 sb->s_blocksize)
1417 return 1;
1418 return 0;
1422 * __ext4_find_entry()
1424 * finds an entry in the specified directory with the wanted name. It
1425 * returns the cache buffer in which the entry was found, and the entry
1426 * itself (as a parameter - res_dir). It does NOT read the inode of the
1427 * entry - you'll have to do that yourself if you want to.
1429 * The returned buffer_head has ->b_count elevated. The caller is expected
1430 * to brelse() it when appropriate.
1432 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1433 struct ext4_filename *fname,
1434 struct ext4_dir_entry_2 **res_dir,
1435 int *inlined)
1437 struct super_block *sb;
1438 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1439 struct buffer_head *bh, *ret = NULL;
1440 ext4_lblk_t start, block;
1441 const u8 *name = fname->usr_fname->name;
1442 size_t ra_max = 0; /* Number of bh's in the readahead
1443 buffer, bh_use[] */
1444 size_t ra_ptr = 0; /* Current index into readahead
1445 buffer */
1446 ext4_lblk_t nblocks;
1447 int i, namelen, retval;
1449 *res_dir = NULL;
1450 sb = dir->i_sb;
1451 namelen = fname->usr_fname->len;
1452 if (namelen > EXT4_NAME_LEN)
1453 return NULL;
1455 if (ext4_has_inline_data(dir)) {
1456 int has_inline_data = 1;
1457 ret = ext4_find_inline_entry(dir, fname, res_dir,
1458 &has_inline_data);
1459 if (has_inline_data) {
1460 if (inlined)
1461 *inlined = 1;
1462 goto cleanup_and_exit;
1466 if ((namelen <= 2) && (name[0] == '.') &&
1467 (name[1] == '.' || name[1] == '\0')) {
1469 * "." or ".." will only be in the first block
1470 * NFS may look up ".."; "." should be handled by the VFS
1472 block = start = 0;
1473 nblocks = 1;
1474 goto restart;
1476 if (is_dx(dir)) {
1477 ret = ext4_dx_find_entry(dir, fname, res_dir);
1479 * On success, or if the error was file not found,
1480 * return. Otherwise, fall back to doing a search the
1481 * old fashioned way.
1483 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1484 goto cleanup_and_exit;
1485 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1486 "falling back\n"));
1487 ret = NULL;
1489 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1490 if (!nblocks) {
1491 ret = NULL;
1492 goto cleanup_and_exit;
1494 start = EXT4_I(dir)->i_dir_start_lookup;
1495 if (start >= nblocks)
1496 start = 0;
1497 block = start;
1498 restart:
1499 do {
1501 * We deal with the read-ahead logic here.
1503 cond_resched();
1504 if (ra_ptr >= ra_max) {
1505 /* Refill the readahead buffer */
1506 ra_ptr = 0;
1507 if (block < start)
1508 ra_max = start - block;
1509 else
1510 ra_max = nblocks - block;
1511 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1512 retval = ext4_bread_batch(dir, block, ra_max,
1513 false /* wait */, bh_use);
1514 if (retval) {
1515 ret = ERR_PTR(retval);
1516 ra_max = 0;
1517 goto cleanup_and_exit;
1520 if ((bh = bh_use[ra_ptr++]) == NULL)
1521 goto next;
1522 wait_on_buffer(bh);
1523 if (!buffer_uptodate(bh)) {
1524 EXT4_ERROR_INODE_ERR(dir, EIO,
1525 "reading directory lblock %lu",
1526 (unsigned long) block);
1527 brelse(bh);
1528 ret = ERR_PTR(-EIO);
1529 goto cleanup_and_exit;
1531 if (!buffer_verified(bh) &&
1532 !is_dx_internal_node(dir, block,
1533 (struct ext4_dir_entry *)bh->b_data) &&
1534 !ext4_dirblock_csum_verify(dir, bh)) {
1535 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1536 "checksumming directory "
1537 "block %lu", (unsigned long)block);
1538 brelse(bh);
1539 ret = ERR_PTR(-EFSBADCRC);
1540 goto cleanup_and_exit;
1542 set_buffer_verified(bh);
1543 i = search_dirblock(bh, dir, fname,
1544 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1545 if (i == 1) {
1546 EXT4_I(dir)->i_dir_start_lookup = block;
1547 ret = bh;
1548 goto cleanup_and_exit;
1549 } else {
1550 brelse(bh);
1551 if (i < 0)
1552 goto cleanup_and_exit;
1554 next:
1555 if (++block >= nblocks)
1556 block = 0;
1557 } while (block != start);
1560 * If the directory has grown while we were searching, then
1561 * search the last part of the directory before giving up.
1563 block = nblocks;
1564 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1565 if (block < nblocks) {
1566 start = 0;
1567 goto restart;
1570 cleanup_and_exit:
1571 /* Clean up the read-ahead blocks */
1572 for (; ra_ptr < ra_max; ra_ptr++)
1573 brelse(bh_use[ra_ptr]);
1574 return ret;
1577 static struct buffer_head *ext4_find_entry(struct inode *dir,
1578 const struct qstr *d_name,
1579 struct ext4_dir_entry_2 **res_dir,
1580 int *inlined)
1582 int err;
1583 struct ext4_filename fname;
1584 struct buffer_head *bh;
1586 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1587 if (err == -ENOENT)
1588 return NULL;
1589 if (err)
1590 return ERR_PTR(err);
1592 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1594 ext4_fname_free_filename(&fname);
1595 return bh;
1598 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1599 struct dentry *dentry,
1600 struct ext4_dir_entry_2 **res_dir)
1602 int err;
1603 struct ext4_filename fname;
1604 struct buffer_head *bh;
1606 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1607 generic_set_encrypted_ci_d_ops(dentry);
1608 if (err == -ENOENT)
1609 return NULL;
1610 if (err)
1611 return ERR_PTR(err);
1613 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1615 ext4_fname_free_filename(&fname);
1616 return bh;
1619 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1620 struct ext4_filename *fname,
1621 struct ext4_dir_entry_2 **res_dir)
1623 struct super_block * sb = dir->i_sb;
1624 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1625 struct buffer_head *bh;
1626 ext4_lblk_t block;
1627 int retval;
1629 #ifdef CONFIG_FS_ENCRYPTION
1630 *res_dir = NULL;
1631 #endif
1632 frame = dx_probe(fname, dir, NULL, frames);
1633 if (IS_ERR(frame))
1634 return (struct buffer_head *) frame;
1635 do {
1636 block = dx_get_block(frame->at);
1637 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1638 if (IS_ERR(bh))
1639 goto errout;
1641 retval = search_dirblock(bh, dir, fname,
1642 block << EXT4_BLOCK_SIZE_BITS(sb),
1643 res_dir);
1644 if (retval == 1)
1645 goto success;
1646 brelse(bh);
1647 if (retval == -1) {
1648 bh = ERR_PTR(ERR_BAD_DX_DIR);
1649 goto errout;
1652 /* Check to see if we should continue to search */
1653 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1654 frames, NULL);
1655 if (retval < 0) {
1656 ext4_warning_inode(dir,
1657 "error %d reading directory index block",
1658 retval);
1659 bh = ERR_PTR(retval);
1660 goto errout;
1662 } while (retval == 1);
1664 bh = NULL;
1665 errout:
1666 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1667 success:
1668 dx_release(frames);
1669 return bh;
1672 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1674 struct inode *inode;
1675 struct ext4_dir_entry_2 *de;
1676 struct buffer_head *bh;
1678 if (dentry->d_name.len > EXT4_NAME_LEN)
1679 return ERR_PTR(-ENAMETOOLONG);
1681 bh = ext4_lookup_entry(dir, dentry, &de);
1682 if (IS_ERR(bh))
1683 return ERR_CAST(bh);
1684 inode = NULL;
1685 if (bh) {
1686 __u32 ino = le32_to_cpu(de->inode);
1687 brelse(bh);
1688 if (!ext4_valid_inum(dir->i_sb, ino)) {
1689 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1690 return ERR_PTR(-EFSCORRUPTED);
1692 if (unlikely(ino == dir->i_ino)) {
1693 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1694 dentry);
1695 return ERR_PTR(-EFSCORRUPTED);
1697 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1698 if (inode == ERR_PTR(-ESTALE)) {
1699 EXT4_ERROR_INODE(dir,
1700 "deleted inode referenced: %u",
1701 ino);
1702 return ERR_PTR(-EFSCORRUPTED);
1704 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1705 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1706 !fscrypt_has_permitted_context(dir, inode)) {
1707 ext4_warning(inode->i_sb,
1708 "Inconsistent encryption contexts: %lu/%lu",
1709 dir->i_ino, inode->i_ino);
1710 iput(inode);
1711 return ERR_PTR(-EPERM);
1715 #ifdef CONFIG_UNICODE
1716 if (!inode && IS_CASEFOLDED(dir)) {
1717 /* Eventually we want to call d_add_ci(dentry, NULL)
1718 * for negative dentries in the encoding case as
1719 * well. For now, prevent the negative dentry
1720 * from being cached.
1722 return NULL;
1724 #endif
1725 return d_splice_alias(inode, dentry);
1729 struct dentry *ext4_get_parent(struct dentry *child)
1731 __u32 ino;
1732 static const struct qstr dotdot = QSTR_INIT("..", 2);
1733 struct ext4_dir_entry_2 * de;
1734 struct buffer_head *bh;
1736 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1737 if (IS_ERR(bh))
1738 return ERR_CAST(bh);
1739 if (!bh)
1740 return ERR_PTR(-ENOENT);
1741 ino = le32_to_cpu(de->inode);
1742 brelse(bh);
1744 if (!ext4_valid_inum(child->d_sb, ino)) {
1745 EXT4_ERROR_INODE(d_inode(child),
1746 "bad parent inode number: %u", ino);
1747 return ERR_PTR(-EFSCORRUPTED);
1750 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1754 * Move count entries from end of map between two memory locations.
1755 * Returns pointer to last entry moved.
1757 static struct ext4_dir_entry_2 *
1758 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1759 unsigned blocksize)
1761 unsigned rec_len = 0;
1763 while (count--) {
1764 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1765 (from + (map->offs<<2));
1766 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1767 memcpy (to, de, rec_len);
1768 ((struct ext4_dir_entry_2 *) to)->rec_len =
1769 ext4_rec_len_to_disk(rec_len, blocksize);
1770 de->inode = 0;
1771 map++;
1772 to += rec_len;
1774 return (struct ext4_dir_entry_2 *) (to - rec_len);
1778 * Compact each dir entry in the range to the minimal rec_len.
1779 * Returns pointer to last entry in range.
1781 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1783 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1784 unsigned rec_len = 0;
1786 prev = to = de;
1787 while ((char*)de < base + blocksize) {
1788 next = ext4_next_entry(de, blocksize);
1789 if (de->inode && de->name_len) {
1790 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1791 if (de > to)
1792 memmove(to, de, rec_len);
1793 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1794 prev = to;
1795 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1797 de = next;
1799 return prev;
1803 * Split a full leaf block to make room for a new dir entry.
1804 * Allocate a new block, and move entries so that they are approx. equally full.
1805 * Returns pointer to de in block into which the new entry will be inserted.
1807 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1808 struct buffer_head **bh,struct dx_frame *frame,
1809 struct dx_hash_info *hinfo)
1811 unsigned blocksize = dir->i_sb->s_blocksize;
1812 unsigned count, continued;
1813 struct buffer_head *bh2;
1814 ext4_lblk_t newblock;
1815 u32 hash2;
1816 struct dx_map_entry *map;
1817 char *data1 = (*bh)->b_data, *data2;
1818 unsigned split, move, size;
1819 struct ext4_dir_entry_2 *de = NULL, *de2;
1820 int csum_size = 0;
1821 int err = 0, i;
1823 if (ext4_has_metadata_csum(dir->i_sb))
1824 csum_size = sizeof(struct ext4_dir_entry_tail);
1826 bh2 = ext4_append(handle, dir, &newblock);
1827 if (IS_ERR(bh2)) {
1828 brelse(*bh);
1829 *bh = NULL;
1830 return (struct ext4_dir_entry_2 *) bh2;
1833 BUFFER_TRACE(*bh, "get_write_access");
1834 err = ext4_journal_get_write_access(handle, *bh);
1835 if (err)
1836 goto journal_error;
1838 BUFFER_TRACE(frame->bh, "get_write_access");
1839 err = ext4_journal_get_write_access(handle, frame->bh);
1840 if (err)
1841 goto journal_error;
1843 data2 = bh2->b_data;
1845 /* create map in the end of data2 block */
1846 map = (struct dx_map_entry *) (data2 + blocksize);
1847 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1848 blocksize, hinfo, map);
1849 map -= count;
1850 dx_sort_map(map, count);
1851 /* Ensure that neither split block is over half full */
1852 size = 0;
1853 move = 0;
1854 for (i = count-1; i >= 0; i--) {
1855 /* is more than half of this entry in 2nd half of the block? */
1856 if (size + map[i].size/2 > blocksize/2)
1857 break;
1858 size += map[i].size;
1859 move++;
1862 * map index at which we will split
1864 * If the sum of active entries didn't exceed half the block size, just
1865 * split it in half by count; each resulting block will have at least
1866 * half the space free.
1868 if (i > 0)
1869 split = count - move;
1870 else
1871 split = count/2;
1873 hash2 = map[split].hash;
1874 continued = hash2 == map[split - 1].hash;
1875 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1876 (unsigned long)dx_get_block(frame->at),
1877 hash2, split, count-split));
1879 /* Fancy dance to stay within two buffers */
1880 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1881 blocksize);
1882 de = dx_pack_dirents(data1, blocksize);
1883 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1884 (char *) de,
1885 blocksize);
1886 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1887 (char *) de2,
1888 blocksize);
1889 if (csum_size) {
1890 ext4_initialize_dirent_tail(*bh, blocksize);
1891 ext4_initialize_dirent_tail(bh2, blocksize);
1894 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1895 blocksize, 1));
1896 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1897 blocksize, 1));
1899 /* Which block gets the new entry? */
1900 if (hinfo->hash >= hash2) {
1901 swap(*bh, bh2);
1902 de = de2;
1904 dx_insert_block(frame, hash2 + continued, newblock);
1905 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1906 if (err)
1907 goto journal_error;
1908 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1909 if (err)
1910 goto journal_error;
1911 brelse(bh2);
1912 dxtrace(dx_show_index("frame", frame->entries));
1913 return de;
1915 journal_error:
1916 brelse(*bh);
1917 brelse(bh2);
1918 *bh = NULL;
1919 ext4_std_error(dir->i_sb, err);
1920 return ERR_PTR(err);
1923 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1924 struct buffer_head *bh,
1925 void *buf, int buf_size,
1926 struct ext4_filename *fname,
1927 struct ext4_dir_entry_2 **dest_de)
1929 struct ext4_dir_entry_2 *de;
1930 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1931 int nlen, rlen;
1932 unsigned int offset = 0;
1933 char *top;
1935 de = (struct ext4_dir_entry_2 *)buf;
1936 top = buf + buf_size - reclen;
1937 while ((char *) de <= top) {
1938 if (ext4_check_dir_entry(dir, NULL, de, bh,
1939 buf, buf_size, offset))
1940 return -EFSCORRUPTED;
1941 if (ext4_match(dir, fname, de))
1942 return -EEXIST;
1943 nlen = EXT4_DIR_REC_LEN(de->name_len);
1944 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1945 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1946 break;
1947 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1948 offset += rlen;
1950 if ((char *) de > top)
1951 return -ENOSPC;
1953 *dest_de = de;
1954 return 0;
1957 void ext4_insert_dentry(struct inode *inode,
1958 struct ext4_dir_entry_2 *de,
1959 int buf_size,
1960 struct ext4_filename *fname)
1963 int nlen, rlen;
1965 nlen = EXT4_DIR_REC_LEN(de->name_len);
1966 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1967 if (de->inode) {
1968 struct ext4_dir_entry_2 *de1 =
1969 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1970 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1971 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1972 de = de1;
1974 de->file_type = EXT4_FT_UNKNOWN;
1975 de->inode = cpu_to_le32(inode->i_ino);
1976 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1977 de->name_len = fname_len(fname);
1978 memcpy(de->name, fname_name(fname), fname_len(fname));
1982 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1983 * it points to a directory entry which is guaranteed to be large
1984 * enough for new directory entry. If de is NULL, then
1985 * add_dirent_to_buf will attempt search the directory block for
1986 * space. It will return -ENOSPC if no space is available, and -EIO
1987 * and -EEXIST if directory entry already exists.
1989 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1990 struct inode *dir,
1991 struct inode *inode, struct ext4_dir_entry_2 *de,
1992 struct buffer_head *bh)
1994 unsigned int blocksize = dir->i_sb->s_blocksize;
1995 int csum_size = 0;
1996 int err, err2;
1998 if (ext4_has_metadata_csum(inode->i_sb))
1999 csum_size = sizeof(struct ext4_dir_entry_tail);
2001 if (!de) {
2002 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2003 blocksize - csum_size, fname, &de);
2004 if (err)
2005 return err;
2007 BUFFER_TRACE(bh, "get_write_access");
2008 err = ext4_journal_get_write_access(handle, bh);
2009 if (err) {
2010 ext4_std_error(dir->i_sb, err);
2011 return err;
2014 /* By now the buffer is marked for journaling */
2015 ext4_insert_dentry(inode, de, blocksize, fname);
2018 * XXX shouldn't update any times until successful
2019 * completion of syscall, but too many callers depend
2020 * on this.
2022 * XXX similarly, too many callers depend on
2023 * ext4_new_inode() setting the times, but error
2024 * recovery deletes the inode, so the worst that can
2025 * happen is that the times are slightly out of date
2026 * and/or different from the directory change time.
2028 dir->i_mtime = dir->i_ctime = current_time(dir);
2029 ext4_update_dx_flag(dir);
2030 inode_inc_iversion(dir);
2031 err2 = ext4_mark_inode_dirty(handle, dir);
2032 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2033 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2034 if (err)
2035 ext4_std_error(dir->i_sb, err);
2036 return err ? err : err2;
2040 * This converts a one block unindexed directory to a 3 block indexed
2041 * directory, and adds the dentry to the indexed directory.
2043 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2044 struct inode *dir,
2045 struct inode *inode, struct buffer_head *bh)
2047 struct buffer_head *bh2;
2048 struct dx_root *root;
2049 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2050 struct dx_entry *entries;
2051 struct ext4_dir_entry_2 *de, *de2;
2052 char *data2, *top;
2053 unsigned len;
2054 int retval;
2055 unsigned blocksize;
2056 ext4_lblk_t block;
2057 struct fake_dirent *fde;
2058 int csum_size = 0;
2060 if (ext4_has_metadata_csum(inode->i_sb))
2061 csum_size = sizeof(struct ext4_dir_entry_tail);
2063 blocksize = dir->i_sb->s_blocksize;
2064 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2065 BUFFER_TRACE(bh, "get_write_access");
2066 retval = ext4_journal_get_write_access(handle, bh);
2067 if (retval) {
2068 ext4_std_error(dir->i_sb, retval);
2069 brelse(bh);
2070 return retval;
2072 root = (struct dx_root *) bh->b_data;
2074 /* The 0th block becomes the root, move the dirents out */
2075 fde = &root->dotdot;
2076 de = (struct ext4_dir_entry_2 *)((char *)fde +
2077 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2078 if ((char *) de >= (((char *) root) + blocksize)) {
2079 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2080 brelse(bh);
2081 return -EFSCORRUPTED;
2083 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2085 /* Allocate new block for the 0th block's dirents */
2086 bh2 = ext4_append(handle, dir, &block);
2087 if (IS_ERR(bh2)) {
2088 brelse(bh);
2089 return PTR_ERR(bh2);
2091 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2092 data2 = bh2->b_data;
2094 memcpy(data2, de, len);
2095 de = (struct ext4_dir_entry_2 *) data2;
2096 top = data2 + len;
2097 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2098 de = de2;
2099 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2100 (char *) de, blocksize);
2102 if (csum_size)
2103 ext4_initialize_dirent_tail(bh2, blocksize);
2105 /* Initialize the root; the dot dirents already exist */
2106 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2107 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2108 blocksize);
2109 memset (&root->info, 0, sizeof(root->info));
2110 root->info.info_length = sizeof(root->info);
2111 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2112 entries = root->entries;
2113 dx_set_block(entries, 1);
2114 dx_set_count(entries, 1);
2115 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2117 /* Initialize as for dx_probe */
2118 fname->hinfo.hash_version = root->info.hash_version;
2119 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2120 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2121 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2122 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2124 memset(frames, 0, sizeof(frames));
2125 frame = frames;
2126 frame->entries = entries;
2127 frame->at = entries;
2128 frame->bh = bh;
2130 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2131 if (retval)
2132 goto out_frames;
2133 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2134 if (retval)
2135 goto out_frames;
2137 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2138 if (IS_ERR(de)) {
2139 retval = PTR_ERR(de);
2140 goto out_frames;
2143 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2144 out_frames:
2146 * Even if the block split failed, we have to properly write
2147 * out all the changes we did so far. Otherwise we can end up
2148 * with corrupted filesystem.
2150 if (retval)
2151 ext4_mark_inode_dirty(handle, dir);
2152 dx_release(frames);
2153 brelse(bh2);
2154 return retval;
2158 * ext4_add_entry()
2160 * adds a file entry to the specified directory, using the same
2161 * semantics as ext4_find_entry(). It returns NULL if it failed.
2163 * NOTE!! The inode part of 'de' is left at 0 - which means you
2164 * may not sleep between calling this and putting something into
2165 * the entry, as someone else might have used it while you slept.
2167 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2168 struct inode *inode)
2170 struct inode *dir = d_inode(dentry->d_parent);
2171 struct buffer_head *bh = NULL;
2172 struct ext4_dir_entry_2 *de;
2173 struct super_block *sb;
2174 struct ext4_filename fname;
2175 int retval;
2176 int dx_fallback=0;
2177 unsigned blocksize;
2178 ext4_lblk_t block, blocks;
2179 int csum_size = 0;
2181 if (ext4_has_metadata_csum(inode->i_sb))
2182 csum_size = sizeof(struct ext4_dir_entry_tail);
2184 sb = dir->i_sb;
2185 blocksize = sb->s_blocksize;
2186 if (!dentry->d_name.len)
2187 return -EINVAL;
2189 if (fscrypt_is_nokey_name(dentry))
2190 return -ENOKEY;
2192 #ifdef CONFIG_UNICODE
2193 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2194 sb->s_encoding && utf8_validate(sb->s_encoding, &dentry->d_name))
2195 return -EINVAL;
2196 #endif
2198 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2199 if (retval)
2200 return retval;
2202 if (ext4_has_inline_data(dir)) {
2203 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2204 if (retval < 0)
2205 goto out;
2206 if (retval == 1) {
2207 retval = 0;
2208 goto out;
2212 if (is_dx(dir)) {
2213 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2214 if (!retval || (retval != ERR_BAD_DX_DIR))
2215 goto out;
2216 /* Can we just ignore htree data? */
2217 if (ext4_has_metadata_csum(sb)) {
2218 EXT4_ERROR_INODE(dir,
2219 "Directory has corrupted htree index.");
2220 retval = -EFSCORRUPTED;
2221 goto out;
2223 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2224 dx_fallback++;
2225 retval = ext4_mark_inode_dirty(handle, dir);
2226 if (unlikely(retval))
2227 goto out;
2229 blocks = dir->i_size >> sb->s_blocksize_bits;
2230 for (block = 0; block < blocks; block++) {
2231 bh = ext4_read_dirblock(dir, block, DIRENT);
2232 if (bh == NULL) {
2233 bh = ext4_bread(handle, dir, block,
2234 EXT4_GET_BLOCKS_CREATE);
2235 goto add_to_new_block;
2237 if (IS_ERR(bh)) {
2238 retval = PTR_ERR(bh);
2239 bh = NULL;
2240 goto out;
2242 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2243 NULL, bh);
2244 if (retval != -ENOSPC)
2245 goto out;
2247 if (blocks == 1 && !dx_fallback &&
2248 ext4_has_feature_dir_index(sb)) {
2249 retval = make_indexed_dir(handle, &fname, dir,
2250 inode, bh);
2251 bh = NULL; /* make_indexed_dir releases bh */
2252 goto out;
2254 brelse(bh);
2256 bh = ext4_append(handle, dir, &block);
2257 add_to_new_block:
2258 if (IS_ERR(bh)) {
2259 retval = PTR_ERR(bh);
2260 bh = NULL;
2261 goto out;
2263 de = (struct ext4_dir_entry_2 *) bh->b_data;
2264 de->inode = 0;
2265 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2267 if (csum_size)
2268 ext4_initialize_dirent_tail(bh, blocksize);
2270 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2271 out:
2272 ext4_fname_free_filename(&fname);
2273 brelse(bh);
2274 if (retval == 0)
2275 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2276 return retval;
2280 * Returns 0 for success, or a negative error value
2282 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2283 struct inode *dir, struct inode *inode)
2285 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2286 struct dx_entry *entries, *at;
2287 struct buffer_head *bh;
2288 struct super_block *sb = dir->i_sb;
2289 struct ext4_dir_entry_2 *de;
2290 int restart;
2291 int err;
2293 again:
2294 restart = 0;
2295 frame = dx_probe(fname, dir, NULL, frames);
2296 if (IS_ERR(frame))
2297 return PTR_ERR(frame);
2298 entries = frame->entries;
2299 at = frame->at;
2300 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2301 if (IS_ERR(bh)) {
2302 err = PTR_ERR(bh);
2303 bh = NULL;
2304 goto cleanup;
2307 BUFFER_TRACE(bh, "get_write_access");
2308 err = ext4_journal_get_write_access(handle, bh);
2309 if (err)
2310 goto journal_error;
2312 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2313 if (err != -ENOSPC)
2314 goto cleanup;
2316 err = 0;
2317 /* Block full, should compress but for now just split */
2318 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2319 dx_get_count(entries), dx_get_limit(entries)));
2320 /* Need to split index? */
2321 if (dx_get_count(entries) == dx_get_limit(entries)) {
2322 ext4_lblk_t newblock;
2323 int levels = frame - frames + 1;
2324 unsigned int icount;
2325 int add_level = 1;
2326 struct dx_entry *entries2;
2327 struct dx_node *node2;
2328 struct buffer_head *bh2;
2330 while (frame > frames) {
2331 if (dx_get_count((frame - 1)->entries) <
2332 dx_get_limit((frame - 1)->entries)) {
2333 add_level = 0;
2334 break;
2336 frame--; /* split higher index block */
2337 at = frame->at;
2338 entries = frame->entries;
2339 restart = 1;
2341 if (add_level && levels == ext4_dir_htree_level(sb)) {
2342 ext4_warning(sb, "Directory (ino: %lu) index full, "
2343 "reach max htree level :%d",
2344 dir->i_ino, levels);
2345 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2346 ext4_warning(sb, "Large directory feature is "
2347 "not enabled on this "
2348 "filesystem");
2350 err = -ENOSPC;
2351 goto cleanup;
2353 icount = dx_get_count(entries);
2354 bh2 = ext4_append(handle, dir, &newblock);
2355 if (IS_ERR(bh2)) {
2356 err = PTR_ERR(bh2);
2357 goto cleanup;
2359 node2 = (struct dx_node *)(bh2->b_data);
2360 entries2 = node2->entries;
2361 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2362 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2363 sb->s_blocksize);
2364 BUFFER_TRACE(frame->bh, "get_write_access");
2365 err = ext4_journal_get_write_access(handle, frame->bh);
2366 if (err)
2367 goto journal_error;
2368 if (!add_level) {
2369 unsigned icount1 = icount/2, icount2 = icount - icount1;
2370 unsigned hash2 = dx_get_hash(entries + icount1);
2371 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2372 icount1, icount2));
2374 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2375 err = ext4_journal_get_write_access(handle,
2376 (frame - 1)->bh);
2377 if (err)
2378 goto journal_error;
2380 memcpy((char *) entries2, (char *) (entries + icount1),
2381 icount2 * sizeof(struct dx_entry));
2382 dx_set_count(entries, icount1);
2383 dx_set_count(entries2, icount2);
2384 dx_set_limit(entries2, dx_node_limit(dir));
2386 /* Which index block gets the new entry? */
2387 if (at - entries >= icount1) {
2388 frame->at = at = at - entries - icount1 + entries2;
2389 frame->entries = entries = entries2;
2390 swap(frame->bh, bh2);
2392 dx_insert_block((frame - 1), hash2, newblock);
2393 dxtrace(dx_show_index("node", frame->entries));
2394 dxtrace(dx_show_index("node",
2395 ((struct dx_node *) bh2->b_data)->entries));
2396 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2397 if (err)
2398 goto journal_error;
2399 brelse (bh2);
2400 err = ext4_handle_dirty_dx_node(handle, dir,
2401 (frame - 1)->bh);
2402 if (err)
2403 goto journal_error;
2404 if (restart) {
2405 err = ext4_handle_dirty_dx_node(handle, dir,
2406 frame->bh);
2407 goto journal_error;
2409 } else {
2410 struct dx_root *dxroot;
2411 memcpy((char *) entries2, (char *) entries,
2412 icount * sizeof(struct dx_entry));
2413 dx_set_limit(entries2, dx_node_limit(dir));
2415 /* Set up root */
2416 dx_set_count(entries, 1);
2417 dx_set_block(entries + 0, newblock);
2418 dxroot = (struct dx_root *)frames[0].bh->b_data;
2419 dxroot->info.indirect_levels += 1;
2420 dxtrace(printk(KERN_DEBUG
2421 "Creating %d level index...\n",
2422 dxroot->info.indirect_levels));
2423 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2424 if (err)
2425 goto journal_error;
2426 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2427 brelse(bh2);
2428 restart = 1;
2429 goto journal_error;
2432 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2433 if (IS_ERR(de)) {
2434 err = PTR_ERR(de);
2435 goto cleanup;
2437 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2438 goto cleanup;
2440 journal_error:
2441 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2442 cleanup:
2443 brelse(bh);
2444 dx_release(frames);
2445 /* @restart is true means htree-path has been changed, we need to
2446 * repeat dx_probe() to find out valid htree-path
2448 if (restart && err == 0)
2449 goto again;
2450 return err;
2454 * ext4_generic_delete_entry deletes a directory entry by merging it
2455 * with the previous entry
2457 int ext4_generic_delete_entry(struct inode *dir,
2458 struct ext4_dir_entry_2 *de_del,
2459 struct buffer_head *bh,
2460 void *entry_buf,
2461 int buf_size,
2462 int csum_size)
2464 struct ext4_dir_entry_2 *de, *pde;
2465 unsigned int blocksize = dir->i_sb->s_blocksize;
2466 int i;
2468 i = 0;
2469 pde = NULL;
2470 de = (struct ext4_dir_entry_2 *)entry_buf;
2471 while (i < buf_size - csum_size) {
2472 if (ext4_check_dir_entry(dir, NULL, de, bh,
2473 entry_buf, buf_size, i))
2474 return -EFSCORRUPTED;
2475 if (de == de_del) {
2476 if (pde)
2477 pde->rec_len = ext4_rec_len_to_disk(
2478 ext4_rec_len_from_disk(pde->rec_len,
2479 blocksize) +
2480 ext4_rec_len_from_disk(de->rec_len,
2481 blocksize),
2482 blocksize);
2483 else
2484 de->inode = 0;
2485 inode_inc_iversion(dir);
2486 return 0;
2488 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2489 pde = de;
2490 de = ext4_next_entry(de, blocksize);
2492 return -ENOENT;
2495 static int ext4_delete_entry(handle_t *handle,
2496 struct inode *dir,
2497 struct ext4_dir_entry_2 *de_del,
2498 struct buffer_head *bh)
2500 int err, csum_size = 0;
2502 if (ext4_has_inline_data(dir)) {
2503 int has_inline_data = 1;
2504 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2505 &has_inline_data);
2506 if (has_inline_data)
2507 return err;
2510 if (ext4_has_metadata_csum(dir->i_sb))
2511 csum_size = sizeof(struct ext4_dir_entry_tail);
2513 BUFFER_TRACE(bh, "get_write_access");
2514 err = ext4_journal_get_write_access(handle, bh);
2515 if (unlikely(err))
2516 goto out;
2518 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2519 dir->i_sb->s_blocksize, csum_size);
2520 if (err)
2521 goto out;
2523 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2524 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2525 if (unlikely(err))
2526 goto out;
2528 return 0;
2529 out:
2530 if (err != -ENOENT)
2531 ext4_std_error(dir->i_sb, err);
2532 return err;
2536 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2537 * since this indicates that nlinks count was previously 1 to avoid overflowing
2538 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2539 * that subdirectory link counts are not being maintained accurately.
2541 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2542 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2543 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2544 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2546 static void ext4_inc_count(struct inode *inode)
2548 inc_nlink(inode);
2549 if (is_dx(inode) &&
2550 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2551 set_nlink(inode, 1);
2555 * If a directory had nlink == 1, then we should let it be 1. This indicates
2556 * directory has >EXT4_LINK_MAX subdirs.
2558 static void ext4_dec_count(struct inode *inode)
2560 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2561 drop_nlink(inode);
2566 * Add non-directory inode to a directory. On success, the inode reference is
2567 * consumed by dentry is instantiation. This is also indicated by clearing of
2568 * *inodep pointer. On failure, the caller is responsible for dropping the
2569 * inode reference in the safe context.
2571 static int ext4_add_nondir(handle_t *handle,
2572 struct dentry *dentry, struct inode **inodep)
2574 struct inode *dir = d_inode(dentry->d_parent);
2575 struct inode *inode = *inodep;
2576 int err = ext4_add_entry(handle, dentry, inode);
2577 if (!err) {
2578 err = ext4_mark_inode_dirty(handle, inode);
2579 if (IS_DIRSYNC(dir))
2580 ext4_handle_sync(handle);
2581 d_instantiate_new(dentry, inode);
2582 *inodep = NULL;
2583 return err;
2585 drop_nlink(inode);
2586 ext4_orphan_add(handle, inode);
2587 unlock_new_inode(inode);
2588 return err;
2592 * By the time this is called, we already have created
2593 * the directory cache entry for the new file, but it
2594 * is so far negative - it has no inode.
2596 * If the create succeeds, we fill in the inode information
2597 * with d_instantiate().
2599 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2600 bool excl)
2602 handle_t *handle;
2603 struct inode *inode;
2604 int err, credits, retries = 0;
2606 err = dquot_initialize(dir);
2607 if (err)
2608 return err;
2610 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2611 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2612 retry:
2613 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2614 NULL, EXT4_HT_DIR, credits);
2615 handle = ext4_journal_current_handle();
2616 err = PTR_ERR(inode);
2617 if (!IS_ERR(inode)) {
2618 inode->i_op = &ext4_file_inode_operations;
2619 inode->i_fop = &ext4_file_operations;
2620 ext4_set_aops(inode);
2621 err = ext4_add_nondir(handle, dentry, &inode);
2622 if (!err)
2623 ext4_fc_track_create(handle, dentry);
2625 if (handle)
2626 ext4_journal_stop(handle);
2627 if (!IS_ERR_OR_NULL(inode))
2628 iput(inode);
2629 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2630 goto retry;
2631 return err;
2634 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2635 umode_t mode, dev_t rdev)
2637 handle_t *handle;
2638 struct inode *inode;
2639 int err, credits, retries = 0;
2641 err = dquot_initialize(dir);
2642 if (err)
2643 return err;
2645 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2646 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2647 retry:
2648 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2649 NULL, EXT4_HT_DIR, credits);
2650 handle = ext4_journal_current_handle();
2651 err = PTR_ERR(inode);
2652 if (!IS_ERR(inode)) {
2653 init_special_inode(inode, inode->i_mode, rdev);
2654 inode->i_op = &ext4_special_inode_operations;
2655 err = ext4_add_nondir(handle, dentry, &inode);
2656 if (!err)
2657 ext4_fc_track_create(handle, dentry);
2659 if (handle)
2660 ext4_journal_stop(handle);
2661 if (!IS_ERR_OR_NULL(inode))
2662 iput(inode);
2663 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2664 goto retry;
2665 return err;
2668 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2670 handle_t *handle;
2671 struct inode *inode;
2672 int err, retries = 0;
2674 err = dquot_initialize(dir);
2675 if (err)
2676 return err;
2678 retry:
2679 inode = ext4_new_inode_start_handle(dir, mode,
2680 NULL, 0, NULL,
2681 EXT4_HT_DIR,
2682 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2683 4 + EXT4_XATTR_TRANS_BLOCKS);
2684 handle = ext4_journal_current_handle();
2685 err = PTR_ERR(inode);
2686 if (!IS_ERR(inode)) {
2687 inode->i_op = &ext4_file_inode_operations;
2688 inode->i_fop = &ext4_file_operations;
2689 ext4_set_aops(inode);
2690 d_tmpfile(dentry, inode);
2691 err = ext4_orphan_add(handle, inode);
2692 if (err)
2693 goto err_unlock_inode;
2694 mark_inode_dirty(inode);
2695 unlock_new_inode(inode);
2697 if (handle)
2698 ext4_journal_stop(handle);
2699 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2700 goto retry;
2701 return err;
2702 err_unlock_inode:
2703 ext4_journal_stop(handle);
2704 unlock_new_inode(inode);
2705 return err;
2708 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2709 struct ext4_dir_entry_2 *de,
2710 int blocksize, int csum_size,
2711 unsigned int parent_ino, int dotdot_real_len)
2713 de->inode = cpu_to_le32(inode->i_ino);
2714 de->name_len = 1;
2715 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2716 blocksize);
2717 strcpy(de->name, ".");
2718 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2720 de = ext4_next_entry(de, blocksize);
2721 de->inode = cpu_to_le32(parent_ino);
2722 de->name_len = 2;
2723 if (!dotdot_real_len)
2724 de->rec_len = ext4_rec_len_to_disk(blocksize -
2725 (csum_size + EXT4_DIR_REC_LEN(1)),
2726 blocksize);
2727 else
2728 de->rec_len = ext4_rec_len_to_disk(
2729 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2730 strcpy(de->name, "..");
2731 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2733 return ext4_next_entry(de, blocksize);
2736 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2737 struct inode *inode)
2739 struct buffer_head *dir_block = NULL;
2740 struct ext4_dir_entry_2 *de;
2741 ext4_lblk_t block = 0;
2742 unsigned int blocksize = dir->i_sb->s_blocksize;
2743 int csum_size = 0;
2744 int err;
2746 if (ext4_has_metadata_csum(dir->i_sb))
2747 csum_size = sizeof(struct ext4_dir_entry_tail);
2749 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2750 err = ext4_try_create_inline_dir(handle, dir, inode);
2751 if (err < 0 && err != -ENOSPC)
2752 goto out;
2753 if (!err)
2754 goto out;
2757 inode->i_size = 0;
2758 dir_block = ext4_append(handle, inode, &block);
2759 if (IS_ERR(dir_block))
2760 return PTR_ERR(dir_block);
2761 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2762 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2763 set_nlink(inode, 2);
2764 if (csum_size)
2765 ext4_initialize_dirent_tail(dir_block, blocksize);
2767 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2768 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2769 if (err)
2770 goto out;
2771 set_buffer_verified(dir_block);
2772 out:
2773 brelse(dir_block);
2774 return err;
2777 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2779 handle_t *handle;
2780 struct inode *inode;
2781 int err, err2 = 0, credits, retries = 0;
2783 if (EXT4_DIR_LINK_MAX(dir))
2784 return -EMLINK;
2786 err = dquot_initialize(dir);
2787 if (err)
2788 return err;
2790 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2791 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2792 retry:
2793 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2794 &dentry->d_name,
2795 0, NULL, EXT4_HT_DIR, credits);
2796 handle = ext4_journal_current_handle();
2797 err = PTR_ERR(inode);
2798 if (IS_ERR(inode))
2799 goto out_stop;
2801 inode->i_op = &ext4_dir_inode_operations;
2802 inode->i_fop = &ext4_dir_operations;
2803 err = ext4_init_new_dir(handle, dir, inode);
2804 if (err)
2805 goto out_clear_inode;
2806 err = ext4_mark_inode_dirty(handle, inode);
2807 if (!err)
2808 err = ext4_add_entry(handle, dentry, inode);
2809 if (err) {
2810 out_clear_inode:
2811 clear_nlink(inode);
2812 ext4_orphan_add(handle, inode);
2813 unlock_new_inode(inode);
2814 err2 = ext4_mark_inode_dirty(handle, inode);
2815 if (unlikely(err2))
2816 err = err2;
2817 ext4_journal_stop(handle);
2818 iput(inode);
2819 goto out_retry;
2821 ext4_inc_count(dir);
2823 ext4_update_dx_flag(dir);
2824 err = ext4_mark_inode_dirty(handle, dir);
2825 if (err)
2826 goto out_clear_inode;
2827 d_instantiate_new(dentry, inode);
2828 ext4_fc_track_create(handle, dentry);
2829 if (IS_DIRSYNC(dir))
2830 ext4_handle_sync(handle);
2832 out_stop:
2833 if (handle)
2834 ext4_journal_stop(handle);
2835 out_retry:
2836 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2837 goto retry;
2838 return err;
2842 * routine to check that the specified directory is empty (for rmdir)
2844 bool ext4_empty_dir(struct inode *inode)
2846 unsigned int offset;
2847 struct buffer_head *bh;
2848 struct ext4_dir_entry_2 *de;
2849 struct super_block *sb;
2851 if (ext4_has_inline_data(inode)) {
2852 int has_inline_data = 1;
2853 int ret;
2855 ret = empty_inline_dir(inode, &has_inline_data);
2856 if (has_inline_data)
2857 return ret;
2860 sb = inode->i_sb;
2861 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2862 EXT4_ERROR_INODE(inode, "invalid size");
2863 return true;
2865 /* The first directory block must not be a hole,
2866 * so treat it as DIRENT_HTREE
2868 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2869 if (IS_ERR(bh))
2870 return true;
2872 de = (struct ext4_dir_entry_2 *) bh->b_data;
2873 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2874 0) ||
2875 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2876 ext4_warning_inode(inode, "directory missing '.'");
2877 brelse(bh);
2878 return true;
2880 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2881 de = ext4_next_entry(de, sb->s_blocksize);
2882 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2883 offset) ||
2884 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2885 ext4_warning_inode(inode, "directory missing '..'");
2886 brelse(bh);
2887 return true;
2889 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2890 while (offset < inode->i_size) {
2891 if (!(offset & (sb->s_blocksize - 1))) {
2892 unsigned int lblock;
2893 brelse(bh);
2894 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2895 bh = ext4_read_dirblock(inode, lblock, EITHER);
2896 if (bh == NULL) {
2897 offset += sb->s_blocksize;
2898 continue;
2900 if (IS_ERR(bh))
2901 return true;
2903 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2904 (offset & (sb->s_blocksize - 1)));
2905 if (ext4_check_dir_entry(inode, NULL, de, bh,
2906 bh->b_data, bh->b_size, offset)) {
2907 offset = (offset | (sb->s_blocksize - 1)) + 1;
2908 continue;
2910 if (le32_to_cpu(de->inode)) {
2911 brelse(bh);
2912 return false;
2914 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2916 brelse(bh);
2917 return true;
2921 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2922 * such inodes, starting at the superblock, in case we crash before the
2923 * file is closed/deleted, or in case the inode truncate spans multiple
2924 * transactions and the last transaction is not recovered after a crash.
2926 * At filesystem recovery time, we walk this list deleting unlinked
2927 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2929 * Orphan list manipulation functions must be called under i_mutex unless
2930 * we are just creating the inode or deleting it.
2932 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2934 struct super_block *sb = inode->i_sb;
2935 struct ext4_sb_info *sbi = EXT4_SB(sb);
2936 struct ext4_iloc iloc;
2937 int err = 0, rc;
2938 bool dirty = false;
2940 if (!sbi->s_journal || is_bad_inode(inode))
2941 return 0;
2943 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2944 !inode_is_locked(inode));
2946 * Exit early if inode already is on orphan list. This is a big speedup
2947 * since we don't have to contend on the global s_orphan_lock.
2949 if (!list_empty(&EXT4_I(inode)->i_orphan))
2950 return 0;
2953 * Orphan handling is only valid for files with data blocks
2954 * being truncated, or files being unlinked. Note that we either
2955 * hold i_mutex, or the inode can not be referenced from outside,
2956 * so i_nlink should not be bumped due to race
2958 ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2959 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2961 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2962 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2963 if (err)
2964 goto out;
2966 err = ext4_reserve_inode_write(handle, inode, &iloc);
2967 if (err)
2968 goto out;
2970 mutex_lock(&sbi->s_orphan_lock);
2972 * Due to previous errors inode may be already a part of on-disk
2973 * orphan list. If so skip on-disk list modification.
2975 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2976 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2977 /* Insert this inode at the head of the on-disk orphan list */
2978 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2979 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2980 dirty = true;
2982 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2983 mutex_unlock(&sbi->s_orphan_lock);
2985 if (dirty) {
2986 err = ext4_handle_dirty_super(handle, sb);
2987 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2988 if (!err)
2989 err = rc;
2990 if (err) {
2992 * We have to remove inode from in-memory list if
2993 * addition to on disk orphan list failed. Stray orphan
2994 * list entries can cause panics at unmount time.
2996 mutex_lock(&sbi->s_orphan_lock);
2997 list_del_init(&EXT4_I(inode)->i_orphan);
2998 mutex_unlock(&sbi->s_orphan_lock);
3000 } else
3001 brelse(iloc.bh);
3003 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3004 jbd_debug(4, "orphan inode %lu will point to %d\n",
3005 inode->i_ino, NEXT_ORPHAN(inode));
3006 out:
3007 ext4_std_error(sb, err);
3008 return err;
3012 * ext4_orphan_del() removes an unlinked or truncated inode from the list
3013 * of such inodes stored on disk, because it is finally being cleaned up.
3015 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3017 struct list_head *prev;
3018 struct ext4_inode_info *ei = EXT4_I(inode);
3019 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3020 __u32 ino_next;
3021 struct ext4_iloc iloc;
3022 int err = 0;
3024 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3025 return 0;
3027 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3028 !inode_is_locked(inode));
3029 /* Do this quick check before taking global s_orphan_lock. */
3030 if (list_empty(&ei->i_orphan))
3031 return 0;
3033 if (handle) {
3034 /* Grab inode buffer early before taking global s_orphan_lock */
3035 err = ext4_reserve_inode_write(handle, inode, &iloc);
3038 mutex_lock(&sbi->s_orphan_lock);
3039 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3041 prev = ei->i_orphan.prev;
3042 list_del_init(&ei->i_orphan);
3044 /* If we're on an error path, we may not have a valid
3045 * transaction handle with which to update the orphan list on
3046 * disk, but we still need to remove the inode from the linked
3047 * list in memory. */
3048 if (!handle || err) {
3049 mutex_unlock(&sbi->s_orphan_lock);
3050 goto out_err;
3053 ino_next = NEXT_ORPHAN(inode);
3054 if (prev == &sbi->s_orphan) {
3055 jbd_debug(4, "superblock will point to %u\n", ino_next);
3056 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3057 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3058 if (err) {
3059 mutex_unlock(&sbi->s_orphan_lock);
3060 goto out_brelse;
3062 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3063 mutex_unlock(&sbi->s_orphan_lock);
3064 err = ext4_handle_dirty_super(handle, inode->i_sb);
3065 } else {
3066 struct ext4_iloc iloc2;
3067 struct inode *i_prev =
3068 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3070 jbd_debug(4, "orphan inode %lu will point to %u\n",
3071 i_prev->i_ino, ino_next);
3072 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3073 if (err) {
3074 mutex_unlock(&sbi->s_orphan_lock);
3075 goto out_brelse;
3077 NEXT_ORPHAN(i_prev) = ino_next;
3078 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3079 mutex_unlock(&sbi->s_orphan_lock);
3081 if (err)
3082 goto out_brelse;
3083 NEXT_ORPHAN(inode) = 0;
3084 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3085 out_err:
3086 ext4_std_error(inode->i_sb, err);
3087 return err;
3089 out_brelse:
3090 brelse(iloc.bh);
3091 goto out_err;
3094 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3096 int retval;
3097 struct inode *inode;
3098 struct buffer_head *bh;
3099 struct ext4_dir_entry_2 *de;
3100 handle_t *handle = NULL;
3102 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3103 return -EIO;
3105 /* Initialize quotas before so that eventual writes go in
3106 * separate transaction */
3107 retval = dquot_initialize(dir);
3108 if (retval)
3109 return retval;
3110 retval = dquot_initialize(d_inode(dentry));
3111 if (retval)
3112 return retval;
3114 retval = -ENOENT;
3115 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3116 if (IS_ERR(bh))
3117 return PTR_ERR(bh);
3118 if (!bh)
3119 goto end_rmdir;
3121 inode = d_inode(dentry);
3123 retval = -EFSCORRUPTED;
3124 if (le32_to_cpu(de->inode) != inode->i_ino)
3125 goto end_rmdir;
3127 retval = -ENOTEMPTY;
3128 if (!ext4_empty_dir(inode))
3129 goto end_rmdir;
3131 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3132 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3133 if (IS_ERR(handle)) {
3134 retval = PTR_ERR(handle);
3135 handle = NULL;
3136 goto end_rmdir;
3139 if (IS_DIRSYNC(dir))
3140 ext4_handle_sync(handle);
3142 retval = ext4_delete_entry(handle, dir, de, bh);
3143 if (retval)
3144 goto end_rmdir;
3145 if (!EXT4_DIR_LINK_EMPTY(inode))
3146 ext4_warning_inode(inode,
3147 "empty directory '%.*s' has too many links (%u)",
3148 dentry->d_name.len, dentry->d_name.name,
3149 inode->i_nlink);
3150 inode_inc_iversion(inode);
3151 clear_nlink(inode);
3152 /* There's no need to set i_disksize: the fact that i_nlink is
3153 * zero will ensure that the right thing happens during any
3154 * recovery. */
3155 inode->i_size = 0;
3156 ext4_orphan_add(handle, inode);
3157 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3158 retval = ext4_mark_inode_dirty(handle, inode);
3159 if (retval)
3160 goto end_rmdir;
3161 ext4_dec_count(dir);
3162 ext4_update_dx_flag(dir);
3163 ext4_fc_track_unlink(handle, dentry);
3164 retval = ext4_mark_inode_dirty(handle, dir);
3166 #ifdef CONFIG_UNICODE
3167 /* VFS negative dentries are incompatible with Encoding and
3168 * Case-insensitiveness. Eventually we'll want avoid
3169 * invalidating the dentries here, alongside with returning the
3170 * negative dentries at ext4_lookup(), when it is better
3171 * supported by the VFS for the CI case.
3173 if (IS_CASEFOLDED(dir))
3174 d_invalidate(dentry);
3175 #endif
3177 end_rmdir:
3178 brelse(bh);
3179 if (handle)
3180 ext4_journal_stop(handle);
3181 return retval;
3184 int __ext4_unlink(handle_t *handle, struct inode *dir, const struct qstr *d_name,
3185 struct inode *inode)
3187 int retval = -ENOENT;
3188 struct buffer_head *bh;
3189 struct ext4_dir_entry_2 *de;
3190 int skip_remove_dentry = 0;
3192 bh = ext4_find_entry(dir, d_name, &de, NULL);
3193 if (IS_ERR(bh))
3194 return PTR_ERR(bh);
3196 if (!bh)
3197 return -ENOENT;
3199 if (le32_to_cpu(de->inode) != inode->i_ino) {
3201 * It's okay if we find dont find dentry which matches
3202 * the inode. That's because it might have gotten
3203 * renamed to a different inode number
3205 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3206 skip_remove_dentry = 1;
3207 else
3208 goto out;
3211 if (IS_DIRSYNC(dir))
3212 ext4_handle_sync(handle);
3214 if (!skip_remove_dentry) {
3215 retval = ext4_delete_entry(handle, dir, de, bh);
3216 if (retval)
3217 goto out;
3218 dir->i_ctime = dir->i_mtime = current_time(dir);
3219 ext4_update_dx_flag(dir);
3220 retval = ext4_mark_inode_dirty(handle, dir);
3221 if (retval)
3222 goto out;
3223 } else {
3224 retval = 0;
3226 if (inode->i_nlink == 0)
3227 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3228 d_name->len, d_name->name);
3229 else
3230 drop_nlink(inode);
3231 if (!inode->i_nlink)
3232 ext4_orphan_add(handle, inode);
3233 inode->i_ctime = current_time(inode);
3234 retval = ext4_mark_inode_dirty(handle, inode);
3236 out:
3237 brelse(bh);
3238 return retval;
3241 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3243 handle_t *handle;
3244 int retval;
3246 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3247 return -EIO;
3249 trace_ext4_unlink_enter(dir, dentry);
3251 * Initialize quotas before so that eventual writes go
3252 * in separate transaction
3254 retval = dquot_initialize(dir);
3255 if (retval)
3256 goto out_trace;
3257 retval = dquot_initialize(d_inode(dentry));
3258 if (retval)
3259 goto out_trace;
3261 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3262 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3263 if (IS_ERR(handle)) {
3264 retval = PTR_ERR(handle);
3265 goto out_trace;
3268 retval = __ext4_unlink(handle, dir, &dentry->d_name, d_inode(dentry));
3269 if (!retval)
3270 ext4_fc_track_unlink(handle, dentry);
3271 #ifdef CONFIG_UNICODE
3272 /* VFS negative dentries are incompatible with Encoding and
3273 * Case-insensitiveness. Eventually we'll want avoid
3274 * invalidating the dentries here, alongside with returning the
3275 * negative dentries at ext4_lookup(), when it is better
3276 * supported by the VFS for the CI case.
3278 if (IS_CASEFOLDED(dir))
3279 d_invalidate(dentry);
3280 #endif
3281 if (handle)
3282 ext4_journal_stop(handle);
3284 out_trace:
3285 trace_ext4_unlink_exit(dentry, retval);
3286 return retval;
3289 static int ext4_symlink(struct inode *dir,
3290 struct dentry *dentry, const char *symname)
3292 handle_t *handle;
3293 struct inode *inode;
3294 int err, len = strlen(symname);
3295 int credits;
3296 struct fscrypt_str disk_link;
3298 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3299 return -EIO;
3301 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3302 &disk_link);
3303 if (err)
3304 return err;
3306 err = dquot_initialize(dir);
3307 if (err)
3308 return err;
3310 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3312 * For non-fast symlinks, we just allocate inode and put it on
3313 * orphan list in the first transaction => we need bitmap,
3314 * group descriptor, sb, inode block, quota blocks, and
3315 * possibly selinux xattr blocks.
3317 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3318 EXT4_XATTR_TRANS_BLOCKS;
3319 } else {
3321 * Fast symlink. We have to add entry to directory
3322 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3323 * allocate new inode (bitmap, group descriptor, inode block,
3324 * quota blocks, sb is already counted in previous macros).
3326 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3327 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3330 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3331 &dentry->d_name, 0, NULL,
3332 EXT4_HT_DIR, credits);
3333 handle = ext4_journal_current_handle();
3334 if (IS_ERR(inode)) {
3335 if (handle)
3336 ext4_journal_stop(handle);
3337 return PTR_ERR(inode);
3340 if (IS_ENCRYPTED(inode)) {
3341 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3342 if (err)
3343 goto err_drop_inode;
3344 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3347 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3348 if (!IS_ENCRYPTED(inode))
3349 inode->i_op = &ext4_symlink_inode_operations;
3350 inode_nohighmem(inode);
3351 ext4_set_aops(inode);
3353 * We cannot call page_symlink() with transaction started
3354 * because it calls into ext4_write_begin() which can wait
3355 * for transaction commit if we are running out of space
3356 * and thus we deadlock. So we have to stop transaction now
3357 * and restart it when symlink contents is written.
3359 * To keep fs consistent in case of crash, we have to put inode
3360 * to orphan list in the mean time.
3362 drop_nlink(inode);
3363 err = ext4_orphan_add(handle, inode);
3364 if (handle)
3365 ext4_journal_stop(handle);
3366 handle = NULL;
3367 if (err)
3368 goto err_drop_inode;
3369 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3370 if (err)
3371 goto err_drop_inode;
3373 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3374 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3376 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3377 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3378 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3379 if (IS_ERR(handle)) {
3380 err = PTR_ERR(handle);
3381 handle = NULL;
3382 goto err_drop_inode;
3384 set_nlink(inode, 1);
3385 err = ext4_orphan_del(handle, inode);
3386 if (err)
3387 goto err_drop_inode;
3388 } else {
3389 /* clear the extent format for fast symlink */
3390 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3391 if (!IS_ENCRYPTED(inode)) {
3392 inode->i_op = &ext4_fast_symlink_inode_operations;
3393 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3395 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3396 disk_link.len);
3397 inode->i_size = disk_link.len - 1;
3399 EXT4_I(inode)->i_disksize = inode->i_size;
3400 err = ext4_add_nondir(handle, dentry, &inode);
3401 if (handle)
3402 ext4_journal_stop(handle);
3403 if (inode)
3404 iput(inode);
3405 goto out_free_encrypted_link;
3407 err_drop_inode:
3408 if (handle)
3409 ext4_journal_stop(handle);
3410 clear_nlink(inode);
3411 unlock_new_inode(inode);
3412 iput(inode);
3413 out_free_encrypted_link:
3414 if (disk_link.name != (unsigned char *)symname)
3415 kfree(disk_link.name);
3416 return err;
3419 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3421 handle_t *handle;
3422 int err, retries = 0;
3423 retry:
3424 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3425 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3426 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3427 if (IS_ERR(handle))
3428 return PTR_ERR(handle);
3430 if (IS_DIRSYNC(dir))
3431 ext4_handle_sync(handle);
3433 inode->i_ctime = current_time(inode);
3434 ext4_inc_count(inode);
3435 ihold(inode);
3437 err = ext4_add_entry(handle, dentry, inode);
3438 if (!err) {
3439 err = ext4_mark_inode_dirty(handle, inode);
3440 /* this can happen only for tmpfile being
3441 * linked the first time
3443 if (inode->i_nlink == 1)
3444 ext4_orphan_del(handle, inode);
3445 d_instantiate(dentry, inode);
3446 ext4_fc_track_link(handle, dentry);
3447 } else {
3448 drop_nlink(inode);
3449 iput(inode);
3451 ext4_journal_stop(handle);
3452 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3453 goto retry;
3454 return err;
3457 static int ext4_link(struct dentry *old_dentry,
3458 struct inode *dir, struct dentry *dentry)
3460 struct inode *inode = d_inode(old_dentry);
3461 int err;
3463 if (inode->i_nlink >= EXT4_LINK_MAX)
3464 return -EMLINK;
3466 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3467 if (err)
3468 return err;
3470 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3471 (!projid_eq(EXT4_I(dir)->i_projid,
3472 EXT4_I(old_dentry->d_inode)->i_projid)))
3473 return -EXDEV;
3475 err = dquot_initialize(dir);
3476 if (err)
3477 return err;
3478 return __ext4_link(dir, inode, dentry);
3482 * Try to find buffer head where contains the parent block.
3483 * It should be the inode block if it is inlined or the 1st block
3484 * if it is a normal dir.
3486 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3487 struct inode *inode,
3488 int *retval,
3489 struct ext4_dir_entry_2 **parent_de,
3490 int *inlined)
3492 struct buffer_head *bh;
3494 if (!ext4_has_inline_data(inode)) {
3495 /* The first directory block must not be a hole, so
3496 * treat it as DIRENT_HTREE
3498 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3499 if (IS_ERR(bh)) {
3500 *retval = PTR_ERR(bh);
3501 return NULL;
3503 *parent_de = ext4_next_entry(
3504 (struct ext4_dir_entry_2 *)bh->b_data,
3505 inode->i_sb->s_blocksize);
3506 return bh;
3509 *inlined = 1;
3510 return ext4_get_first_inline_block(inode, parent_de, retval);
3513 struct ext4_renament {
3514 struct inode *dir;
3515 struct dentry *dentry;
3516 struct inode *inode;
3517 bool is_dir;
3518 int dir_nlink_delta;
3520 /* entry for "dentry" */
3521 struct buffer_head *bh;
3522 struct ext4_dir_entry_2 *de;
3523 int inlined;
3525 /* entry for ".." in inode if it's a directory */
3526 struct buffer_head *dir_bh;
3527 struct ext4_dir_entry_2 *parent_de;
3528 int dir_inlined;
3531 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3533 int retval;
3535 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3536 &retval, &ent->parent_de,
3537 &ent->dir_inlined);
3538 if (!ent->dir_bh)
3539 return retval;
3540 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3541 return -EFSCORRUPTED;
3542 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3543 return ext4_journal_get_write_access(handle, ent->dir_bh);
3546 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3547 unsigned dir_ino)
3549 int retval;
3551 ent->parent_de->inode = cpu_to_le32(dir_ino);
3552 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3553 if (!ent->dir_inlined) {
3554 if (is_dx(ent->inode)) {
3555 retval = ext4_handle_dirty_dx_node(handle,
3556 ent->inode,
3557 ent->dir_bh);
3558 } else {
3559 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3560 ent->dir_bh);
3562 } else {
3563 retval = ext4_mark_inode_dirty(handle, ent->inode);
3565 if (retval) {
3566 ext4_std_error(ent->dir->i_sb, retval);
3567 return retval;
3569 return 0;
3572 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3573 unsigned ino, unsigned file_type)
3575 int retval, retval2;
3577 BUFFER_TRACE(ent->bh, "get write access");
3578 retval = ext4_journal_get_write_access(handle, ent->bh);
3579 if (retval)
3580 return retval;
3581 ent->de->inode = cpu_to_le32(ino);
3582 if (ext4_has_feature_filetype(ent->dir->i_sb))
3583 ent->de->file_type = file_type;
3584 inode_inc_iversion(ent->dir);
3585 ent->dir->i_ctime = ent->dir->i_mtime =
3586 current_time(ent->dir);
3587 retval = ext4_mark_inode_dirty(handle, ent->dir);
3588 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3589 if (!ent->inlined) {
3590 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3591 if (unlikely(retval2)) {
3592 ext4_std_error(ent->dir->i_sb, retval2);
3593 return retval2;
3596 brelse(ent->bh);
3597 ent->bh = NULL;
3599 return retval;
3602 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3603 const struct qstr *d_name)
3605 int retval = -ENOENT;
3606 struct buffer_head *bh;
3607 struct ext4_dir_entry_2 *de;
3609 bh = ext4_find_entry(dir, d_name, &de, NULL);
3610 if (IS_ERR(bh))
3611 return PTR_ERR(bh);
3612 if (bh) {
3613 retval = ext4_delete_entry(handle, dir, de, bh);
3614 brelse(bh);
3616 return retval;
3619 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3620 int force_reread)
3622 int retval;
3624 * ent->de could have moved from under us during htree split, so make
3625 * sure that we are deleting the right entry. We might also be pointing
3626 * to a stale entry in the unused part of ent->bh so just checking inum
3627 * and the name isn't enough.
3629 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3630 ent->de->name_len != ent->dentry->d_name.len ||
3631 strncmp(ent->de->name, ent->dentry->d_name.name,
3632 ent->de->name_len) ||
3633 force_reread) {
3634 retval = ext4_find_delete_entry(handle, ent->dir,
3635 &ent->dentry->d_name);
3636 } else {
3637 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3638 if (retval == -ENOENT) {
3639 retval = ext4_find_delete_entry(handle, ent->dir,
3640 &ent->dentry->d_name);
3644 if (retval) {
3645 ext4_warning_inode(ent->dir,
3646 "Deleting old file: nlink %d, error=%d",
3647 ent->dir->i_nlink, retval);
3651 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3653 if (ent->dir_nlink_delta) {
3654 if (ent->dir_nlink_delta == -1)
3655 ext4_dec_count(ent->dir);
3656 else
3657 ext4_inc_count(ent->dir);
3658 ext4_mark_inode_dirty(handle, ent->dir);
3662 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3663 int credits, handle_t **h)
3665 struct inode *wh;
3666 handle_t *handle;
3667 int retries = 0;
3670 * for inode block, sb block, group summaries,
3671 * and inode bitmap
3673 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3674 EXT4_XATTR_TRANS_BLOCKS + 4);
3675 retry:
3676 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3677 &ent->dentry->d_name, 0, NULL,
3678 EXT4_HT_DIR, credits);
3680 handle = ext4_journal_current_handle();
3681 if (IS_ERR(wh)) {
3682 if (handle)
3683 ext4_journal_stop(handle);
3684 if (PTR_ERR(wh) == -ENOSPC &&
3685 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3686 goto retry;
3687 } else {
3688 *h = handle;
3689 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3690 wh->i_op = &ext4_special_inode_operations;
3692 return wh;
3696 * Anybody can rename anything with this: the permission checks are left to the
3697 * higher-level routines.
3699 * n.b. old_{dentry,inode) refers to the source dentry/inode
3700 * while new_{dentry,inode) refers to the destination dentry/inode
3701 * This comes from rename(const char *oldpath, const char *newpath)
3703 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3704 struct inode *new_dir, struct dentry *new_dentry,
3705 unsigned int flags)
3707 handle_t *handle = NULL;
3708 struct ext4_renament old = {
3709 .dir = old_dir,
3710 .dentry = old_dentry,
3711 .inode = d_inode(old_dentry),
3713 struct ext4_renament new = {
3714 .dir = new_dir,
3715 .dentry = new_dentry,
3716 .inode = d_inode(new_dentry),
3718 int force_reread;
3719 int retval;
3720 struct inode *whiteout = NULL;
3721 int credits;
3722 u8 old_file_type;
3724 if (new.inode && new.inode->i_nlink == 0) {
3725 EXT4_ERROR_INODE(new.inode,
3726 "target of rename is already freed");
3727 return -EFSCORRUPTED;
3730 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3731 (!projid_eq(EXT4_I(new_dir)->i_projid,
3732 EXT4_I(old_dentry->d_inode)->i_projid)))
3733 return -EXDEV;
3735 retval = dquot_initialize(old.dir);
3736 if (retval)
3737 return retval;
3738 retval = dquot_initialize(new.dir);
3739 if (retval)
3740 return retval;
3742 /* Initialize quotas before so that eventual writes go
3743 * in separate transaction */
3744 if (new.inode) {
3745 retval = dquot_initialize(new.inode);
3746 if (retval)
3747 return retval;
3750 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3751 if (IS_ERR(old.bh))
3752 return PTR_ERR(old.bh);
3754 * Check for inode number is _not_ due to possible IO errors.
3755 * We might rmdir the source, keep it as pwd of some process
3756 * and merrily kill the link to whatever was created under the
3757 * same name. Goodbye sticky bit ;-<
3759 retval = -ENOENT;
3760 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3761 goto end_rename;
3763 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3764 &new.de, &new.inlined);
3765 if (IS_ERR(new.bh)) {
3766 retval = PTR_ERR(new.bh);
3767 new.bh = NULL;
3768 goto end_rename;
3770 if (new.bh) {
3771 if (!new.inode) {
3772 brelse(new.bh);
3773 new.bh = NULL;
3776 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3777 ext4_alloc_da_blocks(old.inode);
3779 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3780 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3781 if (!(flags & RENAME_WHITEOUT)) {
3782 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3783 if (IS_ERR(handle)) {
3784 retval = PTR_ERR(handle);
3785 handle = NULL;
3786 goto end_rename;
3788 } else {
3789 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3790 if (IS_ERR(whiteout)) {
3791 retval = PTR_ERR(whiteout);
3792 whiteout = NULL;
3793 goto end_rename;
3797 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3798 ext4_handle_sync(handle);
3800 if (S_ISDIR(old.inode->i_mode)) {
3801 if (new.inode) {
3802 retval = -ENOTEMPTY;
3803 if (!ext4_empty_dir(new.inode))
3804 goto end_rename;
3805 } else {
3806 retval = -EMLINK;
3807 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3808 goto end_rename;
3810 retval = ext4_rename_dir_prepare(handle, &old);
3811 if (retval)
3812 goto end_rename;
3815 * If we're renaming a file within an inline_data dir and adding or
3816 * setting the new dirent causes a conversion from inline_data to
3817 * extents/blockmap, we need to force the dirent delete code to
3818 * re-read the directory, or else we end up trying to delete a dirent
3819 * from what is now the extent tree root (or a block map).
3821 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3822 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3824 old_file_type = old.de->file_type;
3825 if (whiteout) {
3827 * Do this before adding a new entry, so the old entry is sure
3828 * to be still pointing to the valid old entry.
3830 retval = ext4_setent(handle, &old, whiteout->i_ino,
3831 EXT4_FT_CHRDEV);
3832 if (retval)
3833 goto end_rename;
3834 retval = ext4_mark_inode_dirty(handle, whiteout);
3835 if (unlikely(retval))
3836 goto end_rename;
3838 if (!new.bh) {
3839 retval = ext4_add_entry(handle, new.dentry, old.inode);
3840 if (retval)
3841 goto end_rename;
3842 } else {
3843 retval = ext4_setent(handle, &new,
3844 old.inode->i_ino, old_file_type);
3845 if (retval)
3846 goto end_rename;
3848 if (force_reread)
3849 force_reread = !ext4_test_inode_flag(new.dir,
3850 EXT4_INODE_INLINE_DATA);
3853 * Like most other Unix systems, set the ctime for inodes on a
3854 * rename.
3856 old.inode->i_ctime = current_time(old.inode);
3857 retval = ext4_mark_inode_dirty(handle, old.inode);
3858 if (unlikely(retval))
3859 goto end_rename;
3861 if (!whiteout) {
3863 * ok, that's it
3865 ext4_rename_delete(handle, &old, force_reread);
3868 if (new.inode) {
3869 ext4_dec_count(new.inode);
3870 new.inode->i_ctime = current_time(new.inode);
3872 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3873 ext4_update_dx_flag(old.dir);
3874 if (old.dir_bh) {
3875 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3876 if (retval)
3877 goto end_rename;
3879 ext4_dec_count(old.dir);
3880 if (new.inode) {
3881 /* checked ext4_empty_dir above, can't have another
3882 * parent, ext4_dec_count() won't work for many-linked
3883 * dirs */
3884 clear_nlink(new.inode);
3885 } else {
3886 ext4_inc_count(new.dir);
3887 ext4_update_dx_flag(new.dir);
3888 retval = ext4_mark_inode_dirty(handle, new.dir);
3889 if (unlikely(retval))
3890 goto end_rename;
3893 retval = ext4_mark_inode_dirty(handle, old.dir);
3894 if (unlikely(retval))
3895 goto end_rename;
3897 if (S_ISDIR(old.inode->i_mode)) {
3899 * We disable fast commits here that's because the
3900 * replay code is not yet capable of changing dot dot
3901 * dirents in directories.
3903 ext4_fc_mark_ineligible(old.inode->i_sb,
3904 EXT4_FC_REASON_RENAME_DIR);
3905 } else {
3906 if (new.inode)
3907 ext4_fc_track_unlink(handle, new.dentry);
3908 __ext4_fc_track_link(handle, old.inode, new.dentry);
3909 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
3912 if (new.inode) {
3913 retval = ext4_mark_inode_dirty(handle, new.inode);
3914 if (unlikely(retval))
3915 goto end_rename;
3916 if (!new.inode->i_nlink)
3917 ext4_orphan_add(handle, new.inode);
3919 retval = 0;
3921 end_rename:
3922 brelse(old.dir_bh);
3923 brelse(old.bh);
3924 brelse(new.bh);
3925 if (whiteout) {
3926 if (retval)
3927 drop_nlink(whiteout);
3928 unlock_new_inode(whiteout);
3929 iput(whiteout);
3931 if (handle)
3932 ext4_journal_stop(handle);
3933 return retval;
3936 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3937 struct inode *new_dir, struct dentry *new_dentry)
3939 handle_t *handle = NULL;
3940 struct ext4_renament old = {
3941 .dir = old_dir,
3942 .dentry = old_dentry,
3943 .inode = d_inode(old_dentry),
3945 struct ext4_renament new = {
3946 .dir = new_dir,
3947 .dentry = new_dentry,
3948 .inode = d_inode(new_dentry),
3950 u8 new_file_type;
3951 int retval;
3952 struct timespec64 ctime;
3954 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3955 !projid_eq(EXT4_I(new_dir)->i_projid,
3956 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3957 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3958 !projid_eq(EXT4_I(old_dir)->i_projid,
3959 EXT4_I(new_dentry->d_inode)->i_projid)))
3960 return -EXDEV;
3962 retval = dquot_initialize(old.dir);
3963 if (retval)
3964 return retval;
3965 retval = dquot_initialize(new.dir);
3966 if (retval)
3967 return retval;
3969 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3970 &old.de, &old.inlined);
3971 if (IS_ERR(old.bh))
3972 return PTR_ERR(old.bh);
3974 * Check for inode number is _not_ due to possible IO errors.
3975 * We might rmdir the source, keep it as pwd of some process
3976 * and merrily kill the link to whatever was created under the
3977 * same name. Goodbye sticky bit ;-<
3979 retval = -ENOENT;
3980 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3981 goto end_rename;
3983 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3984 &new.de, &new.inlined);
3985 if (IS_ERR(new.bh)) {
3986 retval = PTR_ERR(new.bh);
3987 new.bh = NULL;
3988 goto end_rename;
3991 /* RENAME_EXCHANGE case: old *and* new must both exist */
3992 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3993 goto end_rename;
3995 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3996 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3997 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3998 if (IS_ERR(handle)) {
3999 retval = PTR_ERR(handle);
4000 handle = NULL;
4001 goto end_rename;
4004 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4005 ext4_handle_sync(handle);
4007 if (S_ISDIR(old.inode->i_mode)) {
4008 old.is_dir = true;
4009 retval = ext4_rename_dir_prepare(handle, &old);
4010 if (retval)
4011 goto end_rename;
4013 if (S_ISDIR(new.inode->i_mode)) {
4014 new.is_dir = true;
4015 retval = ext4_rename_dir_prepare(handle, &new);
4016 if (retval)
4017 goto end_rename;
4021 * Other than the special case of overwriting a directory, parents'
4022 * nlink only needs to be modified if this is a cross directory rename.
4024 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4025 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4026 new.dir_nlink_delta = -old.dir_nlink_delta;
4027 retval = -EMLINK;
4028 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4029 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4030 goto end_rename;
4033 new_file_type = new.de->file_type;
4034 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4035 if (retval)
4036 goto end_rename;
4038 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4039 if (retval)
4040 goto end_rename;
4043 * Like most other Unix systems, set the ctime for inodes on a
4044 * rename.
4046 ctime = current_time(old.inode);
4047 old.inode->i_ctime = ctime;
4048 new.inode->i_ctime = ctime;
4049 retval = ext4_mark_inode_dirty(handle, old.inode);
4050 if (unlikely(retval))
4051 goto end_rename;
4052 retval = ext4_mark_inode_dirty(handle, new.inode);
4053 if (unlikely(retval))
4054 goto end_rename;
4055 ext4_fc_mark_ineligible(new.inode->i_sb,
4056 EXT4_FC_REASON_CROSS_RENAME);
4057 if (old.dir_bh) {
4058 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4059 if (retval)
4060 goto end_rename;
4062 if (new.dir_bh) {
4063 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4064 if (retval)
4065 goto end_rename;
4067 ext4_update_dir_count(handle, &old);
4068 ext4_update_dir_count(handle, &new);
4069 retval = 0;
4071 end_rename:
4072 brelse(old.dir_bh);
4073 brelse(new.dir_bh);
4074 brelse(old.bh);
4075 brelse(new.bh);
4076 if (handle)
4077 ext4_journal_stop(handle);
4078 return retval;
4081 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4082 struct inode *new_dir, struct dentry *new_dentry,
4083 unsigned int flags)
4085 int err;
4087 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4088 return -EIO;
4090 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4091 return -EINVAL;
4093 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4094 flags);
4095 if (err)
4096 return err;
4098 if (flags & RENAME_EXCHANGE) {
4099 return ext4_cross_rename(old_dir, old_dentry,
4100 new_dir, new_dentry);
4103 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4107 * directories can handle most operations...
4109 const struct inode_operations ext4_dir_inode_operations = {
4110 .create = ext4_create,
4111 .lookup = ext4_lookup,
4112 .link = ext4_link,
4113 .unlink = ext4_unlink,
4114 .symlink = ext4_symlink,
4115 .mkdir = ext4_mkdir,
4116 .rmdir = ext4_rmdir,
4117 .mknod = ext4_mknod,
4118 .tmpfile = ext4_tmpfile,
4119 .rename = ext4_rename2,
4120 .setattr = ext4_setattr,
4121 .getattr = ext4_getattr,
4122 .listxattr = ext4_listxattr,
4123 .get_acl = ext4_get_acl,
4124 .set_acl = ext4_set_acl,
4125 .fiemap = ext4_fiemap,
4128 const struct inode_operations ext4_special_inode_operations = {
4129 .setattr = ext4_setattr,
4130 .getattr = ext4_getattr,
4131 .listxattr = ext4_listxattr,
4132 .get_acl = ext4_get_acl,
4133 .set_acl = ext4_set_acl,