projects / GitHub / moto-9609 / android_kernel_motorola_exynos9610.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge 4.14.73 into android-4.14-p
[GitHub/moto-9609/android_kernel_motorola_exynos9610.git] / fs / ext4 / namei.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext4/namei.c
4 *
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)
9 *
10 * from
11 *
12 * linux/fs/minix/namei.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
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
26 */
27
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 "ext4.h"
38 #include "ext4_jbd2.h"
39
40 #include "xattr.h"
41 #include "acl.h"
42
43 #include <trace/events/ext4.h>
44 /*
45 * define how far ahead to read directories while searching them.
46 */
47 #define NAMEI_RA_CHUNKS 2
48 #define NAMEI_RA_BLOCKS 4
49 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50
51 static struct buffer_head *ext4_append(handle_t *handle,
52 struct inode *inode,
53 ext4_lblk_t *block)
54 {
55 struct buffer_head *bh;
56 int err;
57
58 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59 ((inode->i_size >> 10) >=
60 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
61 return ERR_PTR(-ENOSPC);
62
63 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64
65 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
66 if (IS_ERR(bh))
67 return bh;
68 inode->i_size += inode->i_sb->s_blocksize;
69 EXT4_I(inode)->i_disksize = inode->i_size;
70 BUFFER_TRACE(bh, "get_write_access");
71 err = ext4_journal_get_write_access(handle, bh);
72 if (err) {
73 brelse(bh);
74 ext4_std_error(inode->i_sb, err);
75 return ERR_PTR(err);
76 }
77 return bh;
78 }
79
80 static int ext4_dx_csum_verify(struct inode *inode,
81 struct ext4_dir_entry *dirent);
82
83 typedef enum {
84 EITHER, INDEX, DIRENT
85 } dirblock_type_t;
86
87 #define ext4_read_dirblock(inode, block, type) \
88 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
89
90 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
91 ext4_lblk_t block,
92 dirblock_type_t type,
93 const char *func,
94 unsigned int line)
95 {
96 struct buffer_head *bh;
97 struct ext4_dir_entry *dirent;
98 int is_dx_block = 0;
99
100 bh = ext4_bread(NULL, inode, block, 0);
101 if (IS_ERR(bh)) {
102 __ext4_warning(inode->i_sb, func, line,
103 "inode #%lu: lblock %lu: comm %s: "
104 "error %ld reading directory block",
105 inode->i_ino, (unsigned long)block,
106 current->comm, PTR_ERR(bh));
107
108 return bh;
109 }
110 if (!bh) {
111 ext4_error_inode(inode, func, line, block,
112 "Directory hole found");
113 return ERR_PTR(-EFSCORRUPTED);
114 }
115 dirent = (struct ext4_dir_entry *) bh->b_data;
116 /* Determine whether or not we have an index block */
117 if (is_dx(inode)) {
118 if (block == 0)
119 is_dx_block = 1;
120 else if (ext4_rec_len_from_disk(dirent->rec_len,
121 inode->i_sb->s_blocksize) ==
122 inode->i_sb->s_blocksize)
123 is_dx_block = 1;
124 }
125 if (!is_dx_block && type == INDEX) {
126 ext4_error_inode(inode, func, line, block,
127 "directory leaf block found instead of index block");
128 return ERR_PTR(-EFSCORRUPTED);
129 }
130 if (!ext4_has_metadata_csum(inode->i_sb) ||
131 buffer_verified(bh))
132 return bh;
133
134 /*
135 * An empty leaf block can get mistaken for a index block; for
136 * this reason, we can only check the index checksum when the
137 * caller is sure it should be an index block.
138 */
139 if (is_dx_block && type == INDEX) {
140 if (ext4_dx_csum_verify(inode, dirent))
141 set_buffer_verified(bh);
142 else {
143 ext4_error_inode(inode, func, line, block,
144 "Directory index failed checksum");
145 brelse(bh);
146 return ERR_PTR(-EFSBADCRC);
147 }
148 }
149 if (!is_dx_block) {
150 if (ext4_dirent_csum_verify(inode, dirent))
151 set_buffer_verified(bh);
152 else {
153 ext4_error_inode(inode, func, line, block,
154 "Directory block failed checksum");
155 brelse(bh);
156 return ERR_PTR(-EFSBADCRC);
157 }
158 }
159 return bh;
160 }
161
162 #ifndef assert
163 #define assert(test) J_ASSERT(test)
164 #endif
165
166 #ifdef DX_DEBUG
167 #define dxtrace(command) command
168 #else
169 #define dxtrace(command)
170 #endif
171
172 struct fake_dirent
173 {
174 __le32 inode;
175 __le16 rec_len;
176 u8 name_len;
177 u8 file_type;
178 };
179
180 struct dx_countlimit
181 {
182 __le16 limit;
183 __le16 count;
184 };
185
186 struct dx_entry
187 {
188 __le32 hash;
189 __le32 block;
190 };
191
192 /*
193 * dx_root_info is laid out so that if it should somehow get overlaid by a
194 * dirent the two low bits of the hash version will be zero. Therefore, the
195 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
196 */
197
198 struct dx_root
199 {
200 struct fake_dirent dot;
201 char dot_name[4];
202 struct fake_dirent dotdot;
203 char dotdot_name[4];
204 struct dx_root_info
205 {
206 __le32 reserved_zero;
207 u8 hash_version;
208 u8 info_length; /* 8 */
209 u8 indirect_levels;
210 u8 unused_flags;
211 }
212 info;
213 struct dx_entry entries[0];
214 };
215
216 struct dx_node
217 {
218 struct fake_dirent fake;
219 struct dx_entry entries[0];
220 };
221
222
223 struct dx_frame
224 {
225 struct buffer_head *bh;
226 struct dx_entry *entries;
227 struct dx_entry *at;
228 };
229
230 struct dx_map_entry
231 {
232 u32 hash;
233 u16 offs;
234 u16 size;
235 };
236
237 /*
238 * This goes at the end of each htree block.
239 */
240 struct dx_tail {
241 u32 dt_reserved;
242 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
243 };
244
245 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
246 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
247 static inline unsigned dx_get_hash(struct dx_entry *entry);
248 static void dx_set_hash(struct dx_entry *entry, unsigned value);
249 static unsigned dx_get_count(struct dx_entry *entries);
250 static unsigned dx_get_limit(struct dx_entry *entries);
251 static void dx_set_count(struct dx_entry *entries, unsigned value);
252 static void dx_set_limit(struct dx_entry *entries, unsigned value);
253 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
254 static unsigned dx_node_limit(struct inode *dir);
255 static struct dx_frame *dx_probe(struct ext4_filename *fname,
256 struct inode *dir,
257 struct dx_hash_info *hinfo,
258 struct dx_frame *frame);
259 static void dx_release(struct dx_frame *frames);
260 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
261 unsigned blocksize, struct dx_hash_info *hinfo,
262 struct dx_map_entry map[]);
263 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
264 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
265 struct dx_map_entry *offsets, int count, unsigned blocksize);
266 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
267 static void dx_insert_block(struct dx_frame *frame,
268 u32 hash, ext4_lblk_t block);
269 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
270 struct dx_frame *frame,
271 struct dx_frame *frames,
272 __u32 *start_hash);
273 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
274 struct ext4_filename *fname,
275 struct ext4_dir_entry_2 **res_dir);
276 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
277 struct inode *dir, struct inode *inode);
278
279 /* checksumming functions */
280 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
281 unsigned int blocksize)
282 {
283 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
284 t->det_rec_len = ext4_rec_len_to_disk(
285 sizeof(struct ext4_dir_entry_tail), blocksize);
286 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
287 }
288
289 /* Walk through a dirent block to find a checksum "dirent" at the tail */
290 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
291 struct ext4_dir_entry *de)
292 {
293 struct ext4_dir_entry_tail *t;
294
295 #ifdef PARANOID
296 struct ext4_dir_entry *d, *top;
297
298 d = de;
299 top = (struct ext4_dir_entry *)(((void *)de) +
300 (EXT4_BLOCK_SIZE(inode->i_sb) -
301 sizeof(struct ext4_dir_entry_tail)));
302 while (d < top && d->rec_len)
303 d = (struct ext4_dir_entry *)(((void *)d) +
304 le16_to_cpu(d->rec_len));
305
306 if (d != top)
307 return NULL;
308
309 t = (struct ext4_dir_entry_tail *)d;
310 #else
311 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
312 #endif
313
314 if (t->det_reserved_zero1 ||
315 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
316 t->det_reserved_zero2 ||
317 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
318 return NULL;
319
320 return t;
321 }
322
323 static __le32 ext4_dirent_csum(struct inode *inode,
324 struct ext4_dir_entry *dirent, int size)
325 {
326 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
327 struct ext4_inode_info *ei = EXT4_I(inode);
328 __u32 csum;
329
330 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
331 return cpu_to_le32(csum);
332 }
333
334 #define warn_no_space_for_csum(inode) \
335 __warn_no_space_for_csum((inode), __func__, __LINE__)
336
337 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
338 unsigned int line)
339 {
340 __ext4_warning_inode(inode, func, line,
341 "No space for directory leaf checksum. Please run e2fsck -D.");
342 }
343
344 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
345 {
346 struct ext4_dir_entry_tail *t;
347
348 if (!ext4_has_metadata_csum(inode->i_sb))
349 return 1;
350
351 t = get_dirent_tail(inode, dirent);
352 if (!t) {
353 warn_no_space_for_csum(inode);
354 return 0;
355 }
356
357 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
358 (void *)t - (void *)dirent))
359 return 0;
360
361 return 1;
362 }
363
364 static void ext4_dirent_csum_set(struct inode *inode,
365 struct ext4_dir_entry *dirent)
366 {
367 struct ext4_dir_entry_tail *t;
368
369 if (!ext4_has_metadata_csum(inode->i_sb))
370 return;
371
372 t = get_dirent_tail(inode, dirent);
373 if (!t) {
374 warn_no_space_for_csum(inode);
375 return;
376 }
377
378 t->det_checksum = ext4_dirent_csum(inode, dirent,
379 (void *)t - (void *)dirent);
380 }
381
382 int ext4_handle_dirty_dirent_node(handle_t *handle,
383 struct inode *inode,
384 struct buffer_head *bh)
385 {
386 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
387 return ext4_handle_dirty_metadata(handle, inode, bh);
388 }
389
390 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
391 struct ext4_dir_entry *dirent,
392 int *offset)
393 {
394 struct ext4_dir_entry *dp;
395 struct dx_root_info *root;
396 int count_offset;
397
398 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
399 count_offset = 8;
400 else if (le16_to_cpu(dirent->rec_len) == 12) {
401 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
402 if (le16_to_cpu(dp->rec_len) !=
403 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
404 return NULL;
405 root = (struct dx_root_info *)(((void *)dp + 12));
406 if (root->reserved_zero ||
407 root->info_length != sizeof(struct dx_root_info))
408 return NULL;
409 count_offset = 32;
410 } else
411 return NULL;
412
413 if (offset)
414 *offset = count_offset;
415 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
416 }
417
418 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
419 int count_offset, int count, struct dx_tail *t)
420 {
421 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
422 struct ext4_inode_info *ei = EXT4_I(inode);
423 __u32 csum;
424 int size;
425 __u32 dummy_csum = 0;
426 int offset = offsetof(struct dx_tail, dt_checksum);
427
428 size = count_offset + (count * sizeof(struct dx_entry));
429 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
430 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
431 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
432
433 return cpu_to_le32(csum);
434 }
435
436 static int ext4_dx_csum_verify(struct inode *inode,
437 struct ext4_dir_entry *dirent)
438 {
439 struct dx_countlimit *c;
440 struct dx_tail *t;
441 int count_offset, limit, count;
442
443 if (!ext4_has_metadata_csum(inode->i_sb))
444 return 1;
445
446 c = get_dx_countlimit(inode, dirent, &count_offset);
447 if (!c) {
448 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
449 return 0;
450 }
451 limit = le16_to_cpu(c->limit);
452 count = le16_to_cpu(c->count);
453 if (count_offset + (limit * sizeof(struct dx_entry)) >
454 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
455 warn_no_space_for_csum(inode);
456 return 0;
457 }
458 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
459
460 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
461 count, t))
462 return 0;
463 return 1;
464 }
465
466 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
467 {
468 struct dx_countlimit *c;
469 struct dx_tail *t;
470 int count_offset, limit, count;
471
472 if (!ext4_has_metadata_csum(inode->i_sb))
473 return;
474
475 c = get_dx_countlimit(inode, dirent, &count_offset);
476 if (!c) {
477 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
478 return;
479 }
480 limit = le16_to_cpu(c->limit);
481 count = le16_to_cpu(c->count);
482 if (count_offset + (limit * sizeof(struct dx_entry)) >
483 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
484 warn_no_space_for_csum(inode);
485 return;
486 }
487 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
488
489 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
490 }
491
492 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
493 struct inode *inode,
494 struct buffer_head *bh)
495 {
496 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
497 return ext4_handle_dirty_metadata(handle, inode, bh);
498 }
499
500 /*
501 * p is at least 6 bytes before the end of page
502 */
503 static inline struct ext4_dir_entry_2 *
504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
505 {
506 return (struct ext4_dir_entry_2 *)((char *)p +
507 ext4_rec_len_from_disk(p->rec_len, blocksize));
508 }
509
510 /*
511 * Future: use high four bits of block for coalesce-on-delete flags
512 * Mask them off for now.
513 */
514
515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
516 {
517 return le32_to_cpu(entry->block) & 0x0fffffff;
518 }
519
520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
521 {
522 entry->block = cpu_to_le32(value);
523 }
524
525 static inline unsigned dx_get_hash(struct dx_entry *entry)
526 {
527 return le32_to_cpu(entry->hash);
528 }
529
530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
531 {
532 entry->hash = cpu_to_le32(value);
533 }
534
535 static inline unsigned dx_get_count(struct dx_entry *entries)
536 {
537 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
538 }
539
540 static inline unsigned dx_get_limit(struct dx_entry *entries)
541 {
542 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
543 }
544
545 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
546 {
547 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
548 }
549
550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
551 {
552 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
553 }
554
555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
556 {
557 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
558 EXT4_DIR_REC_LEN(2) - infosize;
559
560 if (ext4_has_metadata_csum(dir->i_sb))
561 entry_space -= sizeof(struct dx_tail);
562 return entry_space / sizeof(struct dx_entry);
563 }
564
565 static inline unsigned dx_node_limit(struct inode *dir)
566 {
567 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568
569 if (ext4_has_metadata_csum(dir->i_sb))
570 entry_space -= sizeof(struct dx_tail);
571 return entry_space / sizeof(struct dx_entry);
572 }
573
574 /*
575 * Debug
576 */
577 #ifdef DX_DEBUG
578 static void dx_show_index(char * label, struct dx_entry *entries)
579 {
580 int i, n = dx_get_count (entries);
581 printk(KERN_DEBUG "%s index", label);
582 for (i = 0; i < n; i++) {
583 printk(KERN_CONT " %x->%lu",
584 i ? dx_get_hash(entries + i) : 0,
585 (unsigned long)dx_get_block(entries + i));
586 }
587 printk(KERN_CONT "\n");
588 }
589
590 struct stats
591 {
592 unsigned names;
593 unsigned space;
594 unsigned bcount;
595 };
596
597 static struct stats dx_show_leaf(struct inode *dir,
598 struct dx_hash_info *hinfo,
599 struct ext4_dir_entry_2 *de,
600 int size, int show_names)
601 {
602 unsigned names = 0, space = 0;
603 char *base = (char *) de;
604 struct dx_hash_info h = *hinfo;
605
606 printk("names: ");
607 while ((char *) de < base + size)
608 {
609 if (de->inode)
610 {
611 if (show_names)
612 {
613 #ifdef CONFIG_EXT4_FS_ENCRYPTION
614 int len;
615 char *name;
616 struct fscrypt_str fname_crypto_str =
617 FSTR_INIT(NULL, 0);
618 int res = 0;
619
620 name = de->name;
621 len = de->name_len;
622 if (ext4_encrypted_inode(dir))
623 res = fscrypt_get_encryption_info(dir);
624 if (res) {
625 printk(KERN_WARNING "Error setting up"
626 " fname crypto: %d\n", res);
627 }
628 if (!fscrypt_has_encryption_key(dir)) {
629 /* Directory is not encrypted */
630 ext4fs_dirhash(de->name,
631 de->name_len, &h);
632 printk("%*.s:(U)%x.%u ", len,
633 name, h.hash,
634 (unsigned) ((char *) de
635 - base));
636 } else {
637 struct fscrypt_str de_name =
638 FSTR_INIT(name, len);
639
640 /* Directory is encrypted */
641 res = fscrypt_fname_alloc_buffer(
642 dir, len,
643 &fname_crypto_str);
644 if (res)
645 printk(KERN_WARNING "Error "
646 "allocating crypto "
647 "buffer--skipping "
648 "crypto\n");
649 res = fscrypt_fname_disk_to_usr(dir,
650 0, 0, &de_name,
651 &fname_crypto_str);
652 if (res) {
653 printk(KERN_WARNING "Error "
654 "converting filename "
655 "from disk to usr"
656 "\n");
657 name = "??";
658 len = 2;
659 } else {
660 name = fname_crypto_str.name;
661 len = fname_crypto_str.len;
662 }
663 ext4fs_dirhash(de->name, de->name_len,
664 &h);
665 printk("%*.s:(E)%x.%u ", len, name,
666 h.hash, (unsigned) ((char *) de
667 - base));
668 fscrypt_fname_free_buffer(
669 &fname_crypto_str);
670 }
671 #else
672 int len = de->name_len;
673 char *name = de->name;
674 ext4fs_dirhash(de->name, de->name_len, &h);
675 printk("%*.s:%x.%u ", len, name, h.hash,
676 (unsigned) ((char *) de - base));
677 #endif
678 }
679 space += EXT4_DIR_REC_LEN(de->name_len);
680 names++;
681 }
682 de = ext4_next_entry(de, size);
683 }
684 printk(KERN_CONT "(%i)\n", names);
685 return (struct stats) { names, space, 1 };
686 }
687
688 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
689 struct dx_entry *entries, int levels)
690 {
691 unsigned blocksize = dir->i_sb->s_blocksize;
692 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
693 unsigned bcount = 0;
694 struct buffer_head *bh;
695 printk("%i indexed blocks...\n", count);
696 for (i = 0; i < count; i++, entries++)
697 {
698 ext4_lblk_t block = dx_get_block(entries);
699 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
700 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
701 struct stats stats;
702 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
703 bh = ext4_bread(NULL,dir, block, 0);
704 if (!bh || IS_ERR(bh))
705 continue;
706 stats = levels?
707 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
708 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
709 bh->b_data, blocksize, 0);
710 names += stats.names;
711 space += stats.space;
712 bcount += stats.bcount;
713 brelse(bh);
714 }
715 if (bcount)
716 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
717 levels ? "" : " ", names, space/bcount,
718 (space/bcount)*100/blocksize);
719 return (struct stats) { names, space, bcount};
720 }
721 #endif /* DX_DEBUG */
722
723 /*
724 * Probe for a directory leaf block to search.
725 *
726 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
727 * error in the directory index, and the caller should fall back to
728 * searching the directory normally. The callers of dx_probe **MUST**
729 * check for this error code, and make sure it never gets reflected
730 * back to userspace.
731 */
732 static struct dx_frame *
733 dx_probe(struct ext4_filename *fname, struct inode *dir,
734 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
735 {
736 unsigned count, indirect;
737 struct dx_entry *at, *entries, *p, *q, *m;
738 struct dx_root *root;
739 struct dx_frame *frame = frame_in;
740 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
741 u32 hash;
742
743 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
744 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
745 if (IS_ERR(frame->bh))
746 return (struct dx_frame *) frame->bh;
747
748 root = (struct dx_root *) frame->bh->b_data;
749 if (root->info.hash_version != DX_HASH_TEA &&
750 root->info.hash_version != DX_HASH_HALF_MD4 &&
751 root->info.hash_version != DX_HASH_LEGACY) {
752 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
753 root->info.hash_version);
754 goto fail;
755 }
756 if (fname)
757 hinfo = &fname->hinfo;
758 hinfo->hash_version = root->info.hash_version;
759 if (hinfo->hash_version <= DX_HASH_TEA)
760 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
761 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
762 if (fname && fname_name(fname))
763 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
764 hash = hinfo->hash;
765
766 if (root->info.unused_flags & 1) {
767 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
768 root->info.unused_flags);
769 goto fail;
770 }
771
772 indirect = root->info.indirect_levels;
773 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
774 ext4_warning(dir->i_sb,
775 "Directory (ino: %lu) htree depth %#06x exceed"
776 "supported value", dir->i_ino,
777 ext4_dir_htree_level(dir->i_sb));
778 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
779 ext4_warning(dir->i_sb, "Enable large directory "
780 "feature to access it");
781 }
782 goto fail;
783 }
784
785 entries = (struct dx_entry *)(((char *)&root->info) +
786 root->info.info_length);
787
788 if (dx_get_limit(entries) != dx_root_limit(dir,
789 root->info.info_length)) {
790 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
791 dx_get_limit(entries),
792 dx_root_limit(dir, root->info.info_length));
793 goto fail;
794 }
795
796 dxtrace(printk("Look up %x", hash));
797 while (1) {
798 count = dx_get_count(entries);
799 if (!count || count > dx_get_limit(entries)) {
800 ext4_warning_inode(dir,
801 "dx entry: count %u beyond limit %u",
802 count, dx_get_limit(entries));
803 goto fail;
804 }
805
806 p = entries + 1;
807 q = entries + count - 1;
808 while (p <= q) {
809 m = p + (q - p) / 2;
810 dxtrace(printk(KERN_CONT "."));
811 if (dx_get_hash(m) > hash)
812 q = m - 1;
813 else
814 p = m + 1;
815 }
816
817 if (0) { // linear search cross check
818 unsigned n = count - 1;
819 at = entries;
820 while (n--)
821 {
822 dxtrace(printk(KERN_CONT ","));
823 if (dx_get_hash(++at) > hash)
824 {
825 at--;
826 break;
827 }
828 }
829 assert (at == p - 1);
830 }
831
832 at = p - 1;
833 dxtrace(printk(KERN_CONT " %x->%u\n",
834 at == entries ? 0 : dx_get_hash(at),
835 dx_get_block(at)));
836 frame->entries = entries;
837 frame->at = at;
838 if (!indirect--)
839 return frame;
840 frame++;
841 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
842 if (IS_ERR(frame->bh)) {
843 ret_err = (struct dx_frame *) frame->bh;
844 frame->bh = NULL;
845 goto fail;
846 }
847 entries = ((struct dx_node *) frame->bh->b_data)->entries;
848
849 if (dx_get_limit(entries) != dx_node_limit(dir)) {
850 ext4_warning_inode(dir,
851 "dx entry: limit %u != node limit %u",
852 dx_get_limit(entries), dx_node_limit(dir));
853 goto fail;
854 }
855 }
856 fail:
857 while (frame >= frame_in) {
858 brelse(frame->bh);
859 frame--;
860 }
861
862 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
863 ext4_warning_inode(dir,
864 "Corrupt directory, running e2fsck is recommended");
865 return ret_err;
866 }
867
868 static void dx_release(struct dx_frame *frames)
869 {
870 struct dx_root_info *info;
871 int i;
872
873 if (frames[0].bh == NULL)
874 return;
875
876 info = &((struct dx_root *)frames[0].bh->b_data)->info;
877 for (i = 0; i <= info->indirect_levels; i++) {
878 if (frames[i].bh == NULL)
879 break;
880 brelse(frames[i].bh);
881 frames[i].bh = NULL;
882 }
883 }
884
885 /*
886 * This function increments the frame pointer to search the next leaf
887 * block, and reads in the necessary intervening nodes if the search
888 * should be necessary. Whether or not the search is necessary is
889 * controlled by the hash parameter. If the hash value is even, then
890 * the search is only continued if the next block starts with that
891 * hash value. This is used if we are searching for a specific file.
892 *
893 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
894 *
895 * This function returns 1 if the caller should continue to search,
896 * or 0 if it should not. If there is an error reading one of the
897 * index blocks, it will a negative error code.
898 *
899 * If start_hash is non-null, it will be filled in with the starting
900 * hash of the next page.
901 */
902 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
903 struct dx_frame *frame,
904 struct dx_frame *frames,
905 __u32 *start_hash)
906 {
907 struct dx_frame *p;
908 struct buffer_head *bh;
909 int num_frames = 0;
910 __u32 bhash;
911
912 p = frame;
913 /*
914 * Find the next leaf page by incrementing the frame pointer.
915 * If we run out of entries in the interior node, loop around and
916 * increment pointer in the parent node. When we break out of
917 * this loop, num_frames indicates the number of interior
918 * nodes need to be read.
919 */
920 while (1) {
921 if (++(p->at) < p->entries + dx_get_count(p->entries))
922 break;
923 if (p == frames)
924 return 0;
925 num_frames++;
926 p--;
927 }
928
929 /*
930 * If the hash is 1, then continue only if the next page has a
931 * continuation hash of any value. This is used for readdir
932 * handling. Otherwise, check to see if the hash matches the
933 * desired contiuation hash. If it doesn't, return since
934 * there's no point to read in the successive index pages.
935 */
936 bhash = dx_get_hash(p->at);
937 if (start_hash)
938 *start_hash = bhash;
939 if ((hash & 1) == 0) {
940 if ((bhash & ~1) != hash)
941 return 0;
942 }
943 /*
944 * If the hash is HASH_NB_ALWAYS, we always go to the next
945 * block so no check is necessary
946 */
947 while (num_frames--) {
948 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
949 if (IS_ERR(bh))
950 return PTR_ERR(bh);
951 p++;
952 brelse(p->bh);
953 p->bh = bh;
954 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
955 }
956 return 1;
957 }
958
959
960 /*
961 * This function fills a red-black tree with information from a
962 * directory block. It returns the number directory entries loaded
963 * into the tree. If there is an error it is returned in err.
964 */
965 static int htree_dirblock_to_tree(struct file *dir_file,
966 struct inode *dir, ext4_lblk_t block,
967 struct dx_hash_info *hinfo,
968 __u32 start_hash, __u32 start_minor_hash)
969 {
970 struct buffer_head *bh;
971 struct ext4_dir_entry_2 *de, *top;
972 int err = 0, count = 0;
973 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
974
975 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
976 (unsigned long)block));
977 bh = ext4_read_dirblock(dir, block, DIRENT);
978 if (IS_ERR(bh))
979 return PTR_ERR(bh);
980
981 de = (struct ext4_dir_entry_2 *) bh->b_data;
982 top = (struct ext4_dir_entry_2 *) ((char *) de +
983 dir->i_sb->s_blocksize -
984 EXT4_DIR_REC_LEN(0));
985 #ifdef CONFIG_EXT4_FS_ENCRYPTION
986 /* Check if the directory is encrypted */
987 if (ext4_encrypted_inode(dir)) {
988 err = fscrypt_get_encryption_info(dir);
989 if (err < 0) {
990 brelse(bh);
991 return err;
992 }
993 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
994 &fname_crypto_str);
995 if (err < 0) {
996 brelse(bh);
997 return err;
998 }
999 }
1000 #endif
1001 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1002 if (ext4_check_dir_entry(dir, NULL, de, bh,
1003 bh->b_data, bh->b_size,
1004 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1005 + ((char *)de - bh->b_data))) {
1006 /* silently ignore the rest of the block */
1007 break;
1008 }
1009 ext4fs_dirhash(de->name, de->name_len, hinfo);
1010 if ((hinfo->hash < start_hash) ||
1011 ((hinfo->hash == start_hash) &&
1012 (hinfo->minor_hash < start_minor_hash)))
1013 continue;
1014 if (de->inode == 0)
1015 continue;
1016 if (!ext4_encrypted_inode(dir)) {
1017 tmp_str.name = de->name;
1018 tmp_str.len = de->name_len;
1019 err = ext4_htree_store_dirent(dir_file,
1020 hinfo->hash, hinfo->minor_hash, de,
1021 &tmp_str);
1022 } else {
1023 int save_len = fname_crypto_str.len;
1024 struct fscrypt_str de_name = FSTR_INIT(de->name,
1025 de->name_len);
1026
1027 /* Directory is encrypted */
1028 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1029 hinfo->minor_hash, &de_name,
1030 &fname_crypto_str);
1031 if (err) {
1032 count = err;
1033 goto errout;
1034 }
1035 err = ext4_htree_store_dirent(dir_file,
1036 hinfo->hash, hinfo->minor_hash, de,
1037 &fname_crypto_str);
1038 fname_crypto_str.len = save_len;
1039 }
1040 if (err != 0) {
1041 count = err;
1042 goto errout;
1043 }
1044 count++;
1045 }
1046 errout:
1047 brelse(bh);
1048 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1049 fscrypt_fname_free_buffer(&fname_crypto_str);
1050 #endif
1051 return count;
1052 }
1053
1054
1055 /*
1056 * This function fills a red-black tree with information from a
1057 * directory. We start scanning the directory in hash order, starting
1058 * at start_hash and start_minor_hash.
1059 *
1060 * This function returns the number of entries inserted into the tree,
1061 * or a negative error code.
1062 */
1063 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1064 __u32 start_minor_hash, __u32 *next_hash)
1065 {
1066 struct dx_hash_info hinfo;
1067 struct ext4_dir_entry_2 *de;
1068 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1069 struct inode *dir;
1070 ext4_lblk_t block;
1071 int count = 0;
1072 int ret, err;
1073 __u32 hashval;
1074 struct fscrypt_str tmp_str;
1075
1076 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1077 start_hash, start_minor_hash));
1078 dir = file_inode(dir_file);
1079 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1080 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1081 if (hinfo.hash_version <= DX_HASH_TEA)
1082 hinfo.hash_version +=
1083 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1084 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1085 if (ext4_has_inline_data(dir)) {
1086 int has_inline_data = 1;
1087 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1088 &hinfo, start_hash,
1089 start_minor_hash,
1090 &has_inline_data);
1091 if (has_inline_data) {
1092 *next_hash = ~0;
1093 return count;
1094 }
1095 }
1096 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1097 start_hash, start_minor_hash);
1098 *next_hash = ~0;
1099 return count;
1100 }
1101 hinfo.hash = start_hash;
1102 hinfo.minor_hash = 0;
1103 frame = dx_probe(NULL, dir, &hinfo, frames);
1104 if (IS_ERR(frame))
1105 return PTR_ERR(frame);
1106
1107 /* Add '.' and '..' from the htree header */
1108 if (!start_hash && !start_minor_hash) {
1109 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1110 tmp_str.name = de->name;
1111 tmp_str.len = de->name_len;
1112 err = ext4_htree_store_dirent(dir_file, 0, 0,
1113 de, &tmp_str);
1114 if (err != 0)
1115 goto errout;
1116 count++;
1117 }
1118 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1119 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1120 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1121 tmp_str.name = de->name;
1122 tmp_str.len = de->name_len;
1123 err = ext4_htree_store_dirent(dir_file, 2, 0,
1124 de, &tmp_str);
1125 if (err != 0)
1126 goto errout;
1127 count++;
1128 }
1129
1130 while (1) {
1131 if (fatal_signal_pending(current)) {
1132 err = -ERESTARTSYS;
1133 goto errout;
1134 }
1135 cond_resched();
1136 block = dx_get_block(frame->at);
1137 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1138 start_hash, start_minor_hash);
1139 if (ret < 0) {
1140 err = ret;
1141 goto errout;
1142 }
1143 count += ret;
1144 hashval = ~0;
1145 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1146 frame, frames, &hashval);
1147 *next_hash = hashval;
1148 if (ret < 0) {
1149 err = ret;
1150 goto errout;
1151 }
1152 /*
1153 * Stop if: (a) there are no more entries, or
1154 * (b) we have inserted at least one entry and the
1155 * next hash value is not a continuation
1156 */
1157 if ((ret == 0) ||
1158 (count && ((hashval & 1) == 0)))
1159 break;
1160 }
1161 dx_release(frames);
1162 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1163 "next hash: %x\n", count, *next_hash));
1164 return count;
1165 errout:
1166 dx_release(frames);
1167 return (err);
1168 }
1169
1170 static inline int search_dirblock(struct buffer_head *bh,
1171 struct inode *dir,
1172 struct ext4_filename *fname,
1173 unsigned int offset,
1174 struct ext4_dir_entry_2 **res_dir)
1175 {
1176 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1177 fname, offset, res_dir);
1178 }
1179
1180 /*
1181 * Directory block splitting, compacting
1182 */
1183
1184 /*
1185 * Create map of hash values, offsets, and sizes, stored at end of block.
1186 * Returns number of entries mapped.
1187 */
1188 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1189 unsigned blocksize, struct dx_hash_info *hinfo,
1190 struct dx_map_entry *map_tail)
1191 {
1192 int count = 0;
1193 char *base = (char *) de;
1194 struct dx_hash_info h = *hinfo;
1195
1196 while ((char *) de < base + blocksize) {
1197 if (de->name_len && de->inode) {
1198 ext4fs_dirhash(de->name, de->name_len, &h);
1199 map_tail--;
1200 map_tail->hash = h.hash;
1201 map_tail->offs = ((char *) de - base)>>2;
1202 map_tail->size = le16_to_cpu(de->rec_len);
1203 count++;
1204 cond_resched();
1205 }
1206 /* XXX: do we need to check rec_len == 0 case? -Chris */
1207 de = ext4_next_entry(de, blocksize);
1208 }
1209 return count;
1210 }
1211
1212 /* Sort map by hash value */
1213 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1214 {
1215 struct dx_map_entry *p, *q, *top = map + count - 1;
1216 int more;
1217 /* Combsort until bubble sort doesn't suck */
1218 while (count > 2) {
1219 count = count*10/13;
1220 if (count - 9 < 2) /* 9, 10 -> 11 */
1221 count = 11;
1222 for (p = top, q = p - count; q >= map; p--, q--)
1223 if (p->hash < q->hash)
1224 swap(*p, *q);
1225 }
1226 /* Garden variety bubble sort */
1227 do {
1228 more = 0;
1229 q = top;
1230 while (q-- > map) {
1231 if (q[1].hash >= q[0].hash)
1232 continue;
1233 swap(*(q+1), *q);
1234 more = 1;
1235 }
1236 } while(more);
1237 }
1238
1239 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1240 {
1241 struct dx_entry *entries = frame->entries;
1242 struct dx_entry *old = frame->at, *new = old + 1;
1243 int count = dx_get_count(entries);
1244
1245 assert(count < dx_get_limit(entries));
1246 assert(old < entries + count);
1247 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1248 dx_set_hash(new, hash);
1249 dx_set_block(new, block);
1250 dx_set_count(entries, count + 1);
1251 }
1252
1253 /*
1254 * Test whether a directory entry matches the filename being searched for.
1255 *
1256 * Return: %true if the directory entry matches, otherwise %false.
1257 */
1258 static inline bool ext4_match(const struct ext4_filename *fname,
1259 const struct ext4_dir_entry_2 *de)
1260 {
1261 struct fscrypt_name f;
1262
1263 if (!de->inode)
1264 return false;
1265
1266 f.usr_fname = fname->usr_fname;
1267 f.disk_name = fname->disk_name;
1268 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1269 f.crypto_buf = fname->crypto_buf;
1270 #endif
1271 return fscrypt_match_name(&f, de->name, de->name_len);
1272 }
1273
1274 /*
1275 * Returns 0 if not found, -1 on failure, and 1 on success
1276 */
1277 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1278 struct inode *dir, struct ext4_filename *fname,
1279 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1280 {
1281 struct ext4_dir_entry_2 * de;
1282 char * dlimit;
1283 int de_len;
1284
1285 de = (struct ext4_dir_entry_2 *)search_buf;
1286 dlimit = search_buf + buf_size;
1287 while ((char *) de < dlimit) {
1288 /* this code is executed quadratically often */
1289 /* do minimal checking `by hand' */
1290 if ((char *) de + de->name_len <= dlimit &&
1291 ext4_match(fname, de)) {
1292 /* found a match - just to be sure, do
1293 * a full check */
1294 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1295 bh->b_size, offset))
1296 return -1;
1297 *res_dir = de;
1298 return 1;
1299 }
1300 /* prevent looping on a bad block */
1301 de_len = ext4_rec_len_from_disk(de->rec_len,
1302 dir->i_sb->s_blocksize);
1303 if (de_len <= 0)
1304 return -1;
1305 offset += de_len;
1306 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1307 }
1308 return 0;
1309 }
1310
1311 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1312 struct ext4_dir_entry *de)
1313 {
1314 struct super_block *sb = dir->i_sb;
1315
1316 if (!is_dx(dir))
1317 return 0;
1318 if (block == 0)
1319 return 1;
1320 if (de->inode == 0 &&
1321 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1322 sb->s_blocksize)
1323 return 1;
1324 return 0;
1325 }
1326
1327 /*
1328 * ext4_find_entry()
1329 *
1330 * finds an entry in the specified directory with the wanted name. It
1331 * returns the cache buffer in which the entry was found, and the entry
1332 * itself (as a parameter - res_dir). It does NOT read the inode of the
1333 * entry - you'll have to do that yourself if you want to.
1334 *
1335 * The returned buffer_head has ->b_count elevated. The caller is expected
1336 * to brelse() it when appropriate.
1337 */
1338 static struct buffer_head * ext4_find_entry (struct inode *dir,
1339 const struct qstr *d_name,
1340 struct ext4_dir_entry_2 **res_dir,
1341 int *inlined)
1342 {
1343 struct super_block *sb;
1344 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1345 struct buffer_head *bh, *ret = NULL;
1346 ext4_lblk_t start, block;
1347 const u8 *name = d_name->name;
1348 size_t ra_max = 0; /* Number of bh's in the readahead
1349 buffer, bh_use[] */
1350 size_t ra_ptr = 0; /* Current index into readahead
1351 buffer */
1352 ext4_lblk_t nblocks;
1353 int i, namelen, retval;
1354 struct ext4_filename fname;
1355
1356 *res_dir = NULL;
1357 sb = dir->i_sb;
1358 namelen = d_name->len;
1359 if (namelen > EXT4_NAME_LEN)
1360 return NULL;
1361
1362 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1363 if (retval == -ENOENT)
1364 return NULL;
1365 if (retval)
1366 return ERR_PTR(retval);
1367
1368 if (ext4_has_inline_data(dir)) {
1369 int has_inline_data = 1;
1370 ret = ext4_find_inline_entry(dir, &fname, res_dir,
1371 &has_inline_data);
1372 if (has_inline_data) {
1373 if (inlined)
1374 *inlined = 1;
1375 goto cleanup_and_exit;
1376 }
1377 }
1378
1379 if ((namelen <= 2) && (name[0] == '.') &&
1380 (name[1] == '.' || name[1] == '\0')) {
1381 /*
1382 * "." or ".." will only be in the first block
1383 * NFS may look up ".."; "." should be handled by the VFS
1384 */
1385 block = start = 0;
1386 nblocks = 1;
1387 goto restart;
1388 }
1389 if (is_dx(dir)) {
1390 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1391 /*
1392 * On success, or if the error was file not found,
1393 * return. Otherwise, fall back to doing a search the
1394 * old fashioned way.
1395 */
1396 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1397 goto cleanup_and_exit;
1398 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1399 "falling back\n"));
1400 ret = NULL;
1401 }
1402 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1403 if (!nblocks) {
1404 ret = NULL;
1405 goto cleanup_and_exit;
1406 }
1407 start = EXT4_I(dir)->i_dir_start_lookup;
1408 if (start >= nblocks)
1409 start = 0;
1410 block = start;
1411 restart:
1412 do {
1413 /*
1414 * We deal with the read-ahead logic here.
1415 */
1416 if (ra_ptr >= ra_max) {
1417 /* Refill the readahead buffer */
1418 ra_ptr = 0;
1419 if (block < start)
1420 ra_max = start - block;
1421 else
1422 ra_max = nblocks - block;
1423 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1424 retval = ext4_bread_batch(dir, block, ra_max,
1425 false /* wait */, bh_use);
1426 if (retval) {
1427 ret = ERR_PTR(retval);
1428 ra_max = 0;
1429 goto cleanup_and_exit;
1430 }
1431 }
1432 if ((bh = bh_use[ra_ptr++]) == NULL)
1433 goto next;
1434 wait_on_buffer(bh);
1435 if (!buffer_uptodate(bh)) {
1436 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1437 (unsigned long) block);
1438 brelse(bh);
1439 ret = ERR_PTR(-EIO);
1440 goto cleanup_and_exit;
1441 }
1442 if (!buffer_verified(bh) &&
1443 !is_dx_internal_node(dir, block,
1444 (struct ext4_dir_entry *)bh->b_data) &&
1445 !ext4_dirent_csum_verify(dir,
1446 (struct ext4_dir_entry *)bh->b_data)) {
1447 EXT4_ERROR_INODE(dir, "checksumming directory "
1448 "block %lu", (unsigned long)block);
1449 brelse(bh);
1450 ret = ERR_PTR(-EFSBADCRC);
1451 goto cleanup_and_exit;
1452 }
1453 set_buffer_verified(bh);
1454 i = search_dirblock(bh, dir, &fname,
1455 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1456 if (i == 1) {
1457 EXT4_I(dir)->i_dir_start_lookup = block;
1458 ret = bh;
1459 goto cleanup_and_exit;
1460 } else {
1461 brelse(bh);
1462 if (i < 0)
1463 goto cleanup_and_exit;
1464 }
1465 next:
1466 if (++block >= nblocks)
1467 block = 0;
1468 } while (block != start);
1469
1470 /*
1471 * If the directory has grown while we were searching, then
1472 * search the last part of the directory before giving up.
1473 */
1474 block = nblocks;
1475 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1476 if (block < nblocks) {
1477 start = 0;
1478 goto restart;
1479 }
1480
1481 cleanup_and_exit:
1482 /* Clean up the read-ahead blocks */
1483 for (; ra_ptr < ra_max; ra_ptr++)
1484 brelse(bh_use[ra_ptr]);
1485 ext4_fname_free_filename(&fname);
1486 return ret;
1487 }
1488
1489 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1490 struct ext4_filename *fname,
1491 struct ext4_dir_entry_2 **res_dir)
1492 {
1493 struct super_block * sb = dir->i_sb;
1494 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1495 struct buffer_head *bh;
1496 ext4_lblk_t block;
1497 int retval;
1498
1499 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1500 *res_dir = NULL;
1501 #endif
1502 frame = dx_probe(fname, dir, NULL, frames);
1503 if (IS_ERR(frame))
1504 return (struct buffer_head *) frame;
1505 do {
1506 block = dx_get_block(frame->at);
1507 bh = ext4_read_dirblock(dir, block, DIRENT);
1508 if (IS_ERR(bh))
1509 goto errout;
1510
1511 retval = search_dirblock(bh, dir, fname,
1512 block << EXT4_BLOCK_SIZE_BITS(sb),
1513 res_dir);
1514 if (retval == 1)
1515 goto success;
1516 brelse(bh);
1517 if (retval == -1) {
1518 bh = ERR_PTR(ERR_BAD_DX_DIR);
1519 goto errout;
1520 }
1521
1522 /* Check to see if we should continue to search */
1523 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1524 frames, NULL);
1525 if (retval < 0) {
1526 ext4_warning_inode(dir,
1527 "error %d reading directory index block",
1528 retval);
1529 bh = ERR_PTR(retval);
1530 goto errout;
1531 }
1532 } while (retval == 1);
1533
1534 bh = NULL;
1535 errout:
1536 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1537 success:
1538 dx_release(frames);
1539 return bh;
1540 }
1541
1542 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1543 {
1544 struct inode *inode;
1545 struct ext4_dir_entry_2 *de;
1546 struct buffer_head *bh;
1547
1548 if (ext4_encrypted_inode(dir)) {
1549 int res = fscrypt_get_encryption_info(dir);
1550
1551 /*
1552 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1553 * created while the directory was encrypted and we
1554 * have access to the key.
1555 */
1556 if (fscrypt_has_encryption_key(dir))
1557 fscrypt_set_encrypted_dentry(dentry);
1558 fscrypt_set_d_op(dentry);
1559 if (res && res != -ENOKEY)
1560 return ERR_PTR(res);
1561 }
1562
1563 if (dentry->d_name.len > EXT4_NAME_LEN)
1564 return ERR_PTR(-ENAMETOOLONG);
1565
1566 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1567 if (IS_ERR(bh))
1568 return (struct dentry *) bh;
1569 inode = NULL;
1570 if (bh) {
1571 __u32 ino = le32_to_cpu(de->inode);
1572 brelse(bh);
1573 if (!ext4_valid_inum(dir->i_sb, ino)) {
1574 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1575 return ERR_PTR(-EFSCORRUPTED);
1576 }
1577 if (unlikely(ino == dir->i_ino)) {
1578 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1579 dentry);
1580 return ERR_PTR(-EFSCORRUPTED);
1581 }
1582 inode = ext4_iget_normal(dir->i_sb, ino);
1583 if (inode == ERR_PTR(-ESTALE)) {
1584 EXT4_ERROR_INODE(dir,
1585 "deleted inode referenced: %u",
1586 ino);
1587 return ERR_PTR(-EFSCORRUPTED);
1588 }
1589 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1590 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1591 !fscrypt_has_permitted_context(dir, inode)) {
1592 ext4_warning(inode->i_sb,
1593 "Inconsistent encryption contexts: %lu/%lu",
1594 dir->i_ino, inode->i_ino);
1595 iput(inode);
1596 return ERR_PTR(-EPERM);
1597 }
1598 }
1599 return d_splice_alias(inode, dentry);
1600 }
1601
1602
1603 struct dentry *ext4_get_parent(struct dentry *child)
1604 {
1605 __u32 ino;
1606 static const struct qstr dotdot = QSTR_INIT("..", 2);
1607 struct ext4_dir_entry_2 * de;
1608 struct buffer_head *bh;
1609
1610 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1611 if (IS_ERR(bh))
1612 return (struct dentry *) bh;
1613 if (!bh)
1614 return ERR_PTR(-ENOENT);
1615 ino = le32_to_cpu(de->inode);
1616 brelse(bh);
1617
1618 if (!ext4_valid_inum(child->d_sb, ino)) {
1619 EXT4_ERROR_INODE(d_inode(child),
1620 "bad parent inode number: %u", ino);
1621 return ERR_PTR(-EFSCORRUPTED);
1622 }
1623
1624 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1625 }
1626
1627 /*
1628 * Move count entries from end of map between two memory locations.
1629 * Returns pointer to last entry moved.
1630 */
1631 static struct ext4_dir_entry_2 *
1632 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1633 unsigned blocksize)
1634 {
1635 unsigned rec_len = 0;
1636
1637 while (count--) {
1638 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1639 (from + (map->offs<<2));
1640 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1641 memcpy (to, de, rec_len);
1642 ((struct ext4_dir_entry_2 *) to)->rec_len =
1643 ext4_rec_len_to_disk(rec_len, blocksize);
1644 de->inode = 0;
1645 map++;
1646 to += rec_len;
1647 }
1648 return (struct ext4_dir_entry_2 *) (to - rec_len);
1649 }
1650
1651 /*
1652 * Compact each dir entry in the range to the minimal rec_len.
1653 * Returns pointer to last entry in range.
1654 */
1655 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1656 {
1657 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1658 unsigned rec_len = 0;
1659
1660 prev = to = de;
1661 while ((char*)de < base + blocksize) {
1662 next = ext4_next_entry(de, blocksize);
1663 if (de->inode && de->name_len) {
1664 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1665 if (de > to)
1666 memmove(to, de, rec_len);
1667 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1668 prev = to;
1669 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1670 }
1671 de = next;
1672 }
1673 return prev;
1674 }
1675
1676 /*
1677 * Split a full leaf block to make room for a new dir entry.
1678 * Allocate a new block, and move entries so that they are approx. equally full.
1679 * Returns pointer to de in block into which the new entry will be inserted.
1680 */
1681 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1682 struct buffer_head **bh,struct dx_frame *frame,
1683 struct dx_hash_info *hinfo)
1684 {
1685 unsigned blocksize = dir->i_sb->s_blocksize;
1686 unsigned count, continued;
1687 struct buffer_head *bh2;
1688 ext4_lblk_t newblock;
1689 u32 hash2;
1690 struct dx_map_entry *map;
1691 char *data1 = (*bh)->b_data, *data2;
1692 unsigned split, move, size;
1693 struct ext4_dir_entry_2 *de = NULL, *de2;
1694 struct ext4_dir_entry_tail *t;
1695 int csum_size = 0;
1696 int err = 0, i;
1697
1698 if (ext4_has_metadata_csum(dir->i_sb))
1699 csum_size = sizeof(struct ext4_dir_entry_tail);
1700
1701 bh2 = ext4_append(handle, dir, &newblock);
1702 if (IS_ERR(bh2)) {
1703 brelse(*bh);
1704 *bh = NULL;
1705 return (struct ext4_dir_entry_2 *) bh2;
1706 }
1707
1708 BUFFER_TRACE(*bh, "get_write_access");
1709 err = ext4_journal_get_write_access(handle, *bh);
1710 if (err)
1711 goto journal_error;
1712
1713 BUFFER_TRACE(frame->bh, "get_write_access");
1714 err = ext4_journal_get_write_access(handle, frame->bh);
1715 if (err)
1716 goto journal_error;
1717
1718 data2 = bh2->b_data;
1719
1720 /* create map in the end of data2 block */
1721 map = (struct dx_map_entry *) (data2 + blocksize);
1722 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1723 blocksize, hinfo, map);
1724 map -= count;
1725 dx_sort_map(map, count);
1726 /* Split the existing block in the middle, size-wise */
1727 size = 0;
1728 move = 0;
1729 for (i = count-1; i >= 0; i--) {
1730 /* is more than half of this entry in 2nd half of the block? */
1731 if (size + map[i].size/2 > blocksize/2)
1732 break;
1733 size += map[i].size;
1734 move++;
1735 }
1736 /* map index at which we will split */
1737 split = count - move;
1738 hash2 = map[split].hash;
1739 continued = hash2 == map[split - 1].hash;
1740 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1741 (unsigned long)dx_get_block(frame->at),
1742 hash2, split, count-split));
1743
1744 /* Fancy dance to stay within two buffers */
1745 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1746 blocksize);
1747 de = dx_pack_dirents(data1, blocksize);
1748 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1749 (char *) de,
1750 blocksize);
1751 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1752 (char *) de2,
1753 blocksize);
1754 if (csum_size) {
1755 t = EXT4_DIRENT_TAIL(data2, blocksize);
1756 initialize_dirent_tail(t, blocksize);
1757
1758 t = EXT4_DIRENT_TAIL(data1, blocksize);
1759 initialize_dirent_tail(t, blocksize);
1760 }
1761
1762 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1763 blocksize, 1));
1764 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1765 blocksize, 1));
1766
1767 /* Which block gets the new entry? */
1768 if (hinfo->hash >= hash2) {
1769 swap(*bh, bh2);
1770 de = de2;
1771 }
1772 dx_insert_block(frame, hash2 + continued, newblock);
1773 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1774 if (err)
1775 goto journal_error;
1776 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1777 if (err)
1778 goto journal_error;
1779 brelse(bh2);
1780 dxtrace(dx_show_index("frame", frame->entries));
1781 return de;
1782
1783 journal_error:
1784 brelse(*bh);
1785 brelse(bh2);
1786 *bh = NULL;
1787 ext4_std_error(dir->i_sb, err);
1788 return ERR_PTR(err);
1789 }
1790
1791 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1792 struct buffer_head *bh,
1793 void *buf, int buf_size,
1794 struct ext4_filename *fname,
1795 struct ext4_dir_entry_2 **dest_de)
1796 {
1797 struct ext4_dir_entry_2 *de;
1798 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1799 int nlen, rlen;
1800 unsigned int offset = 0;
1801 char *top;
1802
1803 de = (struct ext4_dir_entry_2 *)buf;
1804 top = buf + buf_size - reclen;
1805 while ((char *) de <= top) {
1806 if (ext4_check_dir_entry(dir, NULL, de, bh,
1807 buf, buf_size, offset))
1808 return -EFSCORRUPTED;
1809 if (ext4_match(fname, de))
1810 return -EEXIST;
1811 nlen = EXT4_DIR_REC_LEN(de->name_len);
1812 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1813 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1814 break;
1815 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1816 offset += rlen;
1817 }
1818 if ((char *) de > top)
1819 return -ENOSPC;
1820
1821 *dest_de = de;
1822 return 0;
1823 }
1824
1825 void ext4_insert_dentry(struct inode *inode,
1826 struct ext4_dir_entry_2 *de,
1827 int buf_size,
1828 struct ext4_filename *fname)
1829 {
1830
1831 int nlen, rlen;
1832
1833 nlen = EXT4_DIR_REC_LEN(de->name_len);
1834 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1835 if (de->inode) {
1836 struct ext4_dir_entry_2 *de1 =
1837 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1838 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1839 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1840 de = de1;
1841 }
1842 de->file_type = EXT4_FT_UNKNOWN;
1843 de->inode = cpu_to_le32(inode->i_ino);
1844 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1845 de->name_len = fname_len(fname);
1846 memcpy(de->name, fname_name(fname), fname_len(fname));
1847 }
1848
1849 /*
1850 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1851 * it points to a directory entry which is guaranteed to be large
1852 * enough for new directory entry. If de is NULL, then
1853 * add_dirent_to_buf will attempt search the directory block for
1854 * space. It will return -ENOSPC if no space is available, and -EIO
1855 * and -EEXIST if directory entry already exists.
1856 */
1857 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1858 struct inode *dir,
1859 struct inode *inode, struct ext4_dir_entry_2 *de,
1860 struct buffer_head *bh)
1861 {
1862 unsigned int blocksize = dir->i_sb->s_blocksize;
1863 int csum_size = 0;
1864 int err;
1865
1866 if (ext4_has_metadata_csum(inode->i_sb))
1867 csum_size = sizeof(struct ext4_dir_entry_tail);
1868
1869 if (!de) {
1870 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1871 blocksize - csum_size, fname, &de);
1872 if (err)
1873 return err;
1874 }
1875 BUFFER_TRACE(bh, "get_write_access");
1876 err = ext4_journal_get_write_access(handle, bh);
1877 if (err) {
1878 ext4_std_error(dir->i_sb, err);
1879 return err;
1880 }
1881
1882 /* By now the buffer is marked for journaling */
1883 ext4_insert_dentry(inode, de, blocksize, fname);
1884
1885 /*
1886 * XXX shouldn't update any times until successful
1887 * completion of syscall, but too many callers depend
1888 * on this.
1889 *
1890 * XXX similarly, too many callers depend on
1891 * ext4_new_inode() setting the times, but error
1892 * recovery deletes the inode, so the worst that can
1893 * happen is that the times are slightly out of date
1894 * and/or different from the directory change time.
1895 */
1896 dir->i_mtime = dir->i_ctime = current_time(dir);
1897 ext4_update_dx_flag(dir);
1898 inode_inc_iversion(dir);
1899 ext4_mark_inode_dirty(handle, dir);
1900 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1901 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1902 if (err)
1903 ext4_std_error(dir->i_sb, err);
1904 return 0;
1905 }
1906
1907 /*
1908 * This converts a one block unindexed directory to a 3 block indexed
1909 * directory, and adds the dentry to the indexed directory.
1910 */
1911 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1912 struct inode *dir,
1913 struct inode *inode, struct buffer_head *bh)
1914 {
1915 struct buffer_head *bh2;
1916 struct dx_root *root;
1917 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1918 struct dx_entry *entries;
1919 struct ext4_dir_entry_2 *de, *de2;
1920 struct ext4_dir_entry_tail *t;
1921 char *data1, *top;
1922 unsigned len;
1923 int retval;
1924 unsigned blocksize;
1925 ext4_lblk_t block;
1926 struct fake_dirent *fde;
1927 int csum_size = 0;
1928
1929 if (ext4_has_metadata_csum(inode->i_sb))
1930 csum_size = sizeof(struct ext4_dir_entry_tail);
1931
1932 blocksize = dir->i_sb->s_blocksize;
1933 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1934 BUFFER_TRACE(bh, "get_write_access");
1935 retval = ext4_journal_get_write_access(handle, bh);
1936 if (retval) {
1937 ext4_std_error(dir->i_sb, retval);
1938 brelse(bh);
1939 return retval;
1940 }
1941 root = (struct dx_root *) bh->b_data;
1942
1943 /* The 0th block becomes the root, move the dirents out */
1944 fde = &root->dotdot;
1945 de = (struct ext4_dir_entry_2 *)((char *)fde +
1946 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1947 if ((char *) de >= (((char *) root) + blocksize)) {
1948 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1949 brelse(bh);
1950 return -EFSCORRUPTED;
1951 }
1952 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1953
1954 /* Allocate new block for the 0th block's dirents */
1955 bh2 = ext4_append(handle, dir, &block);
1956 if (IS_ERR(bh2)) {
1957 brelse(bh);
1958 return PTR_ERR(bh2);
1959 }
1960 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1961 data1 = bh2->b_data;
1962
1963 memcpy (data1, de, len);
1964 de = (struct ext4_dir_entry_2 *) data1;
1965 top = data1 + len;
1966 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1967 de = de2;
1968 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1969 (char *) de,
1970 blocksize);
1971
1972 if (csum_size) {
1973 t = EXT4_DIRENT_TAIL(data1, blocksize);
1974 initialize_dirent_tail(t, blocksize);
1975 }
1976
1977 /* Initialize the root; the dot dirents already exist */
1978 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1979 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1980 blocksize);
1981 memset (&root->info, 0, sizeof(root->info));
1982 root->info.info_length = sizeof(root->info);
1983 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1984 entries = root->entries;
1985 dx_set_block(entries, 1);
1986 dx_set_count(entries, 1);
1987 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1988
1989 /* Initialize as for dx_probe */
1990 fname->hinfo.hash_version = root->info.hash_version;
1991 if (fname->hinfo.hash_version <= DX_HASH_TEA)
1992 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1993 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1994 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1995
1996 memset(frames, 0, sizeof(frames));
1997 frame = frames;
1998 frame->entries = entries;
1999 frame->at = entries;
2000 frame->bh = bh;
2001
2002 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2003 if (retval)
2004 goto out_frames;
2005 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2006 if (retval)
2007 goto out_frames;
2008
2009 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2010 if (IS_ERR(de)) {
2011 retval = PTR_ERR(de);
2012 goto out_frames;
2013 }
2014
2015 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2016 out_frames:
2017 /*
2018 * Even if the block split failed, we have to properly write
2019 * out all the changes we did so far. Otherwise we can end up
2020 * with corrupted filesystem.
2021 */
2022 if (retval)
2023 ext4_mark_inode_dirty(handle, dir);
2024 dx_release(frames);
2025 brelse(bh2);
2026 return retval;
2027 }
2028
2029 /*
2030 * ext4_add_entry()
2031 *
2032 * adds a file entry to the specified directory, using the same
2033 * semantics as ext4_find_entry(). It returns NULL if it failed.
2034 *
2035 * NOTE!! The inode part of 'de' is left at 0 - which means you
2036 * may not sleep between calling this and putting something into
2037 * the entry, as someone else might have used it while you slept.
2038 */
2039 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2040 struct inode *inode)
2041 {
2042 struct inode *dir = d_inode(dentry->d_parent);
2043 struct buffer_head *bh = NULL;
2044 struct ext4_dir_entry_2 *de;
2045 struct ext4_dir_entry_tail *t;
2046 struct super_block *sb;
2047 struct ext4_filename fname;
2048 int retval;
2049 int dx_fallback=0;
2050 unsigned blocksize;
2051 ext4_lblk_t block, blocks;
2052 int csum_size = 0;
2053
2054 if (ext4_has_metadata_csum(inode->i_sb))
2055 csum_size = sizeof(struct ext4_dir_entry_tail);
2056
2057 sb = dir->i_sb;
2058 blocksize = sb->s_blocksize;
2059 if (!dentry->d_name.len)
2060 return -EINVAL;
2061
2062 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2063 if (retval)
2064 return retval;
2065
2066 if (ext4_has_inline_data(dir)) {
2067 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2068 if (retval < 0)
2069 goto out;
2070 if (retval == 1) {
2071 retval = 0;
2072 goto out;
2073 }
2074 }
2075
2076 if (is_dx(dir)) {
2077 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2078 if (!retval || (retval != ERR_BAD_DX_DIR))
2079 goto out;
2080 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2081 dx_fallback++;
2082 ext4_mark_inode_dirty(handle, dir);
2083 }
2084 blocks = dir->i_size >> sb->s_blocksize_bits;
2085 for (block = 0; block < blocks; block++) {
2086 bh = ext4_read_dirblock(dir, block, DIRENT);
2087 if (IS_ERR(bh)) {
2088 retval = PTR_ERR(bh);
2089 bh = NULL;
2090 goto out;
2091 }
2092 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2093 NULL, bh);
2094 if (retval != -ENOSPC)
2095 goto out;
2096
2097 if (blocks == 1 && !dx_fallback &&
2098 ext4_has_feature_dir_index(sb)) {
2099 retval = make_indexed_dir(handle, &fname, dir,
2100 inode, bh);
2101 bh = NULL; /* make_indexed_dir releases bh */
2102 goto out;
2103 }
2104 brelse(bh);
2105 }
2106 bh = ext4_append(handle, dir, &block);
2107 if (IS_ERR(bh)) {
2108 retval = PTR_ERR(bh);
2109 bh = NULL;
2110 goto out;
2111 }
2112 de = (struct ext4_dir_entry_2 *) bh->b_data;
2113 de->inode = 0;
2114 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2115
2116 if (csum_size) {
2117 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2118 initialize_dirent_tail(t, blocksize);
2119 }
2120
2121 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2122 out:
2123 ext4_fname_free_filename(&fname);
2124 brelse(bh);
2125 if (retval == 0)
2126 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2127 return retval;
2128 }
2129
2130 /*
2131 * Returns 0 for success, or a negative error value
2132 */
2133 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2134 struct inode *dir, struct inode *inode)
2135 {
2136 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2137 struct dx_entry *entries, *at;
2138 struct buffer_head *bh;
2139 struct super_block *sb = dir->i_sb;
2140 struct ext4_dir_entry_2 *de;
2141 int restart;
2142 int err;
2143
2144 again:
2145 restart = 0;
2146 frame = dx_probe(fname, dir, NULL, frames);
2147 if (IS_ERR(frame))
2148 return PTR_ERR(frame);
2149 entries = frame->entries;
2150 at = frame->at;
2151 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2152 if (IS_ERR(bh)) {
2153 err = PTR_ERR(bh);
2154 bh = NULL;
2155 goto cleanup;
2156 }
2157
2158 BUFFER_TRACE(bh, "get_write_access");
2159 err = ext4_journal_get_write_access(handle, bh);
2160 if (err)
2161 goto journal_error;
2162
2163 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2164 if (err != -ENOSPC)
2165 goto cleanup;
2166
2167 err = 0;
2168 /* Block full, should compress but for now just split */
2169 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2170 dx_get_count(entries), dx_get_limit(entries)));
2171 /* Need to split index? */
2172 if (dx_get_count(entries) == dx_get_limit(entries)) {
2173 ext4_lblk_t newblock;
2174 int levels = frame - frames + 1;
2175 unsigned int icount;
2176 int add_level = 1;
2177 struct dx_entry *entries2;
2178 struct dx_node *node2;
2179 struct buffer_head *bh2;
2180
2181 while (frame > frames) {
2182 if (dx_get_count((frame - 1)->entries) <
2183 dx_get_limit((frame - 1)->entries)) {
2184 add_level = 0;
2185 break;
2186 }
2187 frame--; /* split higher index block */
2188 at = frame->at;
2189 entries = frame->entries;
2190 restart = 1;
2191 }
2192 if (add_level && levels == ext4_dir_htree_level(sb)) {
2193 ext4_warning(sb, "Directory (ino: %lu) index full, "
2194 "reach max htree level :%d",
2195 dir->i_ino, levels);
2196 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2197 ext4_warning(sb, "Large directory feature is "
2198 "not enabled on this "
2199 "filesystem");
2200 }
2201 err = -ENOSPC;
2202 goto cleanup;
2203 }
2204 icount = dx_get_count(entries);
2205 bh2 = ext4_append(handle, dir, &newblock);
2206 if (IS_ERR(bh2)) {
2207 err = PTR_ERR(bh2);
2208 goto cleanup;
2209 }
2210 node2 = (struct dx_node *)(bh2->b_data);
2211 entries2 = node2->entries;
2212 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2213 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2214 sb->s_blocksize);
2215 BUFFER_TRACE(frame->bh, "get_write_access");
2216 err = ext4_journal_get_write_access(handle, frame->bh);
2217 if (err)
2218 goto journal_error;
2219 if (!add_level) {
2220 unsigned icount1 = icount/2, icount2 = icount - icount1;
2221 unsigned hash2 = dx_get_hash(entries + icount1);
2222 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2223 icount1, icount2));
2224
2225 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2226 err = ext4_journal_get_write_access(handle,
2227 (frame - 1)->bh);
2228 if (err)
2229 goto journal_error;
2230
2231 memcpy((char *) entries2, (char *) (entries + icount1),
2232 icount2 * sizeof(struct dx_entry));
2233 dx_set_count(entries, icount1);
2234 dx_set_count(entries2, icount2);
2235 dx_set_limit(entries2, dx_node_limit(dir));
2236
2237 /* Which index block gets the new entry? */
2238 if (at - entries >= icount1) {
2239 frame->at = at = at - entries - icount1 + entries2;
2240 frame->entries = entries = entries2;
2241 swap(frame->bh, bh2);
2242 }
2243 dx_insert_block((frame - 1), hash2, newblock);
2244 dxtrace(dx_show_index("node", frame->entries));
2245 dxtrace(dx_show_index("node",
2246 ((struct dx_node *) bh2->b_data)->entries));
2247 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2248 if (err)
2249 goto journal_error;
2250 brelse (bh2);
2251 err = ext4_handle_dirty_dx_node(handle, dir,
2252 (frame - 1)->bh);
2253 if (err)
2254 goto journal_error;
2255 if (restart) {
2256 err = ext4_handle_dirty_dx_node(handle, dir,
2257 frame->bh);
2258 goto journal_error;
2259 }
2260 } else {
2261 struct dx_root *dxroot;
2262 memcpy((char *) entries2, (char *) entries,
2263 icount * sizeof(struct dx_entry));
2264 dx_set_limit(entries2, dx_node_limit(dir));
2265
2266 /* Set up root */
2267 dx_set_count(entries, 1);
2268 dx_set_block(entries + 0, newblock);
2269 dxroot = (struct dx_root *)frames[0].bh->b_data;
2270 dxroot->info.indirect_levels += 1;
2271 dxtrace(printk(KERN_DEBUG
2272 "Creating %d level index...\n",
2273 info->indirect_levels));
2274 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2275 if (err)
2276 goto journal_error;
2277 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2278 brelse(bh2);
2279 restart = 1;
2280 goto journal_error;
2281 }
2282 }
2283 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2284 if (IS_ERR(de)) {
2285 err = PTR_ERR(de);
2286 goto cleanup;
2287 }
2288 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2289 goto cleanup;
2290
2291 journal_error:
2292 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2293 cleanup:
2294 brelse(bh);
2295 dx_release(frames);
2296 /* @restart is true means htree-path has been changed, we need to
2297 * repeat dx_probe() to find out valid htree-path
2298 */
2299 if (restart && err == 0)
2300 goto again;
2301 return err;
2302 }
2303
2304 /*
2305 * ext4_generic_delete_entry deletes a directory entry by merging it
2306 * with the previous entry
2307 */
2308 int ext4_generic_delete_entry(handle_t *handle,
2309 struct inode *dir,
2310 struct ext4_dir_entry_2 *de_del,
2311 struct buffer_head *bh,
2312 void *entry_buf,
2313 int buf_size,
2314 int csum_size)
2315 {
2316 struct ext4_dir_entry_2 *de, *pde;
2317 unsigned int blocksize = dir->i_sb->s_blocksize;
2318 int i;
2319
2320 i = 0;
2321 pde = NULL;
2322 de = (struct ext4_dir_entry_2 *)entry_buf;
2323 while (i < buf_size - csum_size) {
2324 if (ext4_check_dir_entry(dir, NULL, de, bh,
2325 bh->b_data, bh->b_size, i))
2326 return -EFSCORRUPTED;
2327 if (de == de_del) {
2328 if (pde)
2329 pde->rec_len = ext4_rec_len_to_disk(
2330 ext4_rec_len_from_disk(pde->rec_len,
2331 blocksize) +
2332 ext4_rec_len_from_disk(de->rec_len,
2333 blocksize),
2334 blocksize);
2335 else
2336 de->inode = 0;
2337 inode_inc_iversion(dir);
2338 return 0;
2339 }
2340 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2341 pde = de;
2342 de = ext4_next_entry(de, blocksize);
2343 }
2344 return -ENOENT;
2345 }
2346
2347 static int ext4_delete_entry(handle_t *handle,
2348 struct inode *dir,
2349 struct ext4_dir_entry_2 *de_del,
2350 struct buffer_head *bh)
2351 {
2352 int err, csum_size = 0;
2353
2354 if (ext4_has_inline_data(dir)) {
2355 int has_inline_data = 1;
2356 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2357 &has_inline_data);
2358 if (has_inline_data)
2359 return err;
2360 }
2361
2362 if (ext4_has_metadata_csum(dir->i_sb))
2363 csum_size = sizeof(struct ext4_dir_entry_tail);
2364
2365 BUFFER_TRACE(bh, "get_write_access");
2366 err = ext4_journal_get_write_access(handle, bh);
2367 if (unlikely(err))
2368 goto out;
2369
2370 err = ext4_generic_delete_entry(handle, dir, de_del,
2371 bh, bh->b_data,
2372 dir->i_sb->s_blocksize, csum_size);
2373 if (err)
2374 goto out;
2375
2376 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2377 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2378 if (unlikely(err))
2379 goto out;
2380
2381 return 0;
2382 out:
2383 if (err != -ENOENT)
2384 ext4_std_error(dir->i_sb, err);
2385 return err;
2386 }
2387
2388 /*
2389 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2390 * since this indicates that nlinks count was previously 1 to avoid overflowing
2391 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2392 * that subdirectory link counts are not being maintained accurately.
2393 *
2394 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2395 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2396 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2397 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2398 */
2399 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2400 {
2401 inc_nlink(inode);
2402 if (is_dx(inode) &&
2403 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2404 set_nlink(inode, 1);
2405 }
2406
2407 /*
2408 * If a directory had nlink == 1, then we should let it be 1. This indicates
2409 * directory has >EXT4_LINK_MAX subdirs.
2410 */
2411 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2412 {
2413 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2414 drop_nlink(inode);
2415 }
2416
2417
2418 static int ext4_add_nondir(handle_t *handle,
2419 struct dentry *dentry, struct inode *inode)
2420 {
2421 int err = ext4_add_entry(handle, dentry, inode);
2422 if (!err) {
2423 ext4_mark_inode_dirty(handle, inode);
2424 d_instantiate_new(dentry, inode);
2425 return 0;
2426 }
2427 drop_nlink(inode);
2428 unlock_new_inode(inode);
2429 iput(inode);
2430 return err;
2431 }
2432
2433 /*
2434 * By the time this is called, we already have created
2435 * the directory cache entry for the new file, but it
2436 * is so far negative - it has no inode.
2437 *
2438 * If the create succeeds, we fill in the inode information
2439 * with d_instantiate().
2440 */
2441 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2442 bool excl)
2443 {
2444 handle_t *handle;
2445 struct inode *inode;
2446 int err, credits, retries = 0;
2447
2448 err = dquot_initialize(dir);
2449 if (err)
2450 return err;
2451
2452 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2453 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2454 retry:
2455 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2456 NULL, EXT4_HT_DIR, credits);
2457 handle = ext4_journal_current_handle();
2458 err = PTR_ERR(inode);
2459 if (!IS_ERR(inode)) {
2460 inode->i_op = &ext4_file_inode_operations;
2461 inode->i_fop = &ext4_file_operations;
2462 ext4_set_aops(inode);
2463 err = ext4_add_nondir(handle, dentry, inode);
2464 if (!err && IS_DIRSYNC(dir))
2465 ext4_handle_sync(handle);
2466 }
2467 if (handle)
2468 ext4_journal_stop(handle);
2469 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2470 goto retry;
2471 return err;
2472 }
2473
2474 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2475 umode_t mode, dev_t rdev)
2476 {
2477 handle_t *handle;
2478 struct inode *inode;
2479 int err, credits, retries = 0;
2480
2481 err = dquot_initialize(dir);
2482 if (err)
2483 return err;
2484
2485 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2486 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2487 retry:
2488 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2489 NULL, EXT4_HT_DIR, credits);
2490 handle = ext4_journal_current_handle();
2491 err = PTR_ERR(inode);
2492 if (!IS_ERR(inode)) {
2493 init_special_inode(inode, inode->i_mode, rdev);
2494 inode->i_op = &ext4_special_inode_operations;
2495 err = ext4_add_nondir(handle, dentry, inode);
2496 if (!err && IS_DIRSYNC(dir))
2497 ext4_handle_sync(handle);
2498 }
2499 if (handle)
2500 ext4_journal_stop(handle);
2501 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2502 goto retry;
2503 return err;
2504 }
2505
2506 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2507 {
2508 handle_t *handle;
2509 struct inode *inode;
2510 int err, retries = 0;
2511
2512 err = dquot_initialize(dir);
2513 if (err)
2514 return err;
2515
2516 retry:
2517 inode = ext4_new_inode_start_handle(dir, mode,
2518 NULL, 0, NULL,
2519 EXT4_HT_DIR,
2520 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2521 4 + EXT4_XATTR_TRANS_BLOCKS);
2522 handle = ext4_journal_current_handle();
2523 err = PTR_ERR(inode);
2524 if (!IS_ERR(inode)) {
2525 inode->i_op = &ext4_file_inode_operations;
2526 inode->i_fop = &ext4_file_operations;
2527 ext4_set_aops(inode);
2528 d_tmpfile(dentry, inode);
2529 err = ext4_orphan_add(handle, inode);
2530 if (err)
2531 goto err_unlock_inode;
2532 mark_inode_dirty(inode);
2533 unlock_new_inode(inode);
2534 }
2535 if (handle)
2536 ext4_journal_stop(handle);
2537 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2538 goto retry;
2539 return err;
2540 err_unlock_inode:
2541 ext4_journal_stop(handle);
2542 unlock_new_inode(inode);
2543 return err;
2544 }
2545
2546 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2547 struct ext4_dir_entry_2 *de,
2548 int blocksize, int csum_size,
2549 unsigned int parent_ino, int dotdot_real_len)
2550 {
2551 de->inode = cpu_to_le32(inode->i_ino);
2552 de->name_len = 1;
2553 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2554 blocksize);
2555 strcpy(de->name, ".");
2556 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2557
2558 de = ext4_next_entry(de, blocksize);
2559 de->inode = cpu_to_le32(parent_ino);
2560 de->name_len = 2;
2561 if (!dotdot_real_len)
2562 de->rec_len = ext4_rec_len_to_disk(blocksize -
2563 (csum_size + EXT4_DIR_REC_LEN(1)),
2564 blocksize);
2565 else
2566 de->rec_len = ext4_rec_len_to_disk(
2567 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2568 strcpy(de->name, "..");
2569 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2570
2571 return ext4_next_entry(de, blocksize);
2572 }
2573
2574 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2575 struct inode *inode)
2576 {
2577 struct buffer_head *dir_block = NULL;
2578 struct ext4_dir_entry_2 *de;
2579 struct ext4_dir_entry_tail *t;
2580 ext4_lblk_t block = 0;
2581 unsigned int blocksize = dir->i_sb->s_blocksize;
2582 int csum_size = 0;
2583 int err;
2584
2585 if (ext4_has_metadata_csum(dir->i_sb))
2586 csum_size = sizeof(struct ext4_dir_entry_tail);
2587
2588 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2589 err = ext4_try_create_inline_dir(handle, dir, inode);
2590 if (err < 0 && err != -ENOSPC)
2591 goto out;
2592 if (!err)
2593 goto out;
2594 }
2595
2596 inode->i_size = 0;
2597 dir_block = ext4_append(handle, inode, &block);
2598 if (IS_ERR(dir_block))
2599 return PTR_ERR(dir_block);
2600 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2601 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2602 set_nlink(inode, 2);
2603 if (csum_size) {
2604 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2605 initialize_dirent_tail(t, blocksize);
2606 }
2607
2608 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2609 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2610 if (err)
2611 goto out;
2612 set_buffer_verified(dir_block);
2613 out:
2614 brelse(dir_block);
2615 return err;
2616 }
2617
2618 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2619 {
2620 handle_t *handle;
2621 struct inode *inode;
2622 int err, credits, retries = 0;
2623
2624 if (EXT4_DIR_LINK_MAX(dir))
2625 return -EMLINK;
2626
2627 err = dquot_initialize(dir);
2628 if (err)
2629 return err;
2630
2631 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2632 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2633 retry:
2634 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2635 &dentry->d_name,
2636 0, NULL, EXT4_HT_DIR, credits);
2637 handle = ext4_journal_current_handle();
2638 err = PTR_ERR(inode);
2639 if (IS_ERR(inode))
2640 goto out_stop;
2641
2642 inode->i_op = &ext4_dir_inode_operations;
2643 inode->i_fop = &ext4_dir_operations;
2644 err = ext4_init_new_dir(handle, dir, inode);
2645 if (err)
2646 goto out_clear_inode;
2647 err = ext4_mark_inode_dirty(handle, inode);
2648 if (!err)
2649 err = ext4_add_entry(handle, dentry, inode);
2650 if (err) {
2651 out_clear_inode:
2652 clear_nlink(inode);
2653 unlock_new_inode(inode);
2654 ext4_mark_inode_dirty(handle, inode);
2655 iput(inode);
2656 goto out_stop;
2657 }
2658 ext4_inc_count(handle, dir);
2659 ext4_update_dx_flag(dir);
2660 err = ext4_mark_inode_dirty(handle, dir);
2661 if (err)
2662 goto out_clear_inode;
2663 d_instantiate_new(dentry, inode);
2664 if (IS_DIRSYNC(dir))
2665 ext4_handle_sync(handle);
2666
2667 out_stop:
2668 if (handle)
2669 ext4_journal_stop(handle);
2670 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2671 goto retry;
2672 return err;
2673 }
2674
2675 /*
2676 * routine to check that the specified directory is empty (for rmdir)
2677 */
2678 bool ext4_empty_dir(struct inode *inode)
2679 {
2680 unsigned int offset;
2681 struct buffer_head *bh;
2682 struct ext4_dir_entry_2 *de, *de1;
2683 struct super_block *sb;
2684
2685 if (ext4_has_inline_data(inode)) {
2686 int has_inline_data = 1;
2687 int ret;
2688
2689 ret = empty_inline_dir(inode, &has_inline_data);
2690 if (has_inline_data)
2691 return ret;
2692 }
2693
2694 sb = inode->i_sb;
2695 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2696 EXT4_ERROR_INODE(inode, "invalid size");
2697 return true;
2698 }
2699 bh = ext4_read_dirblock(inode, 0, EITHER);
2700 if (IS_ERR(bh))
2701 return true;
2702
2703 de = (struct ext4_dir_entry_2 *) bh->b_data;
2704 de1 = ext4_next_entry(de, sb->s_blocksize);
2705 if (le32_to_cpu(de->inode) != inode->i_ino ||
2706 le32_to_cpu(de1->inode) == 0 ||
2707 strcmp(".", de->name) || strcmp("..", de1->name)) {
2708 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2709 brelse(bh);
2710 return true;
2711 }
2712 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2713 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2714 de = ext4_next_entry(de1, sb->s_blocksize);
2715 while (offset < inode->i_size) {
2716 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2717 unsigned int lblock;
2718 brelse(bh);
2719 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2720 bh = ext4_read_dirblock(inode, lblock, EITHER);
2721 if (IS_ERR(bh))
2722 return true;
2723 de = (struct ext4_dir_entry_2 *) bh->b_data;
2724 }
2725 if (ext4_check_dir_entry(inode, NULL, de, bh,
2726 bh->b_data, bh->b_size, offset)) {
2727 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2728 sb->s_blocksize);
2729 offset = (offset | (sb->s_blocksize - 1)) + 1;
2730 continue;
2731 }
2732 if (le32_to_cpu(de->inode)) {
2733 brelse(bh);
2734 return false;
2735 }
2736 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2737 de = ext4_next_entry(de, sb->s_blocksize);
2738 }
2739 brelse(bh);
2740 return true;
2741 }
2742
2743 /*
2744 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2745 * such inodes, starting at the superblock, in case we crash before the
2746 * file is closed/deleted, or in case the inode truncate spans multiple
2747 * transactions and the last transaction is not recovered after a crash.
2748 *
2749 * At filesystem recovery time, we walk this list deleting unlinked
2750 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2751 *
2752 * Orphan list manipulation functions must be called under i_mutex unless
2753 * we are just creating the inode or deleting it.
2754 */
2755 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2756 {
2757 struct super_block *sb = inode->i_sb;
2758 struct ext4_sb_info *sbi = EXT4_SB(sb);
2759 struct ext4_iloc iloc;
2760 int err = 0, rc;
2761 bool dirty = false;
2762
2763 if (!sbi->s_journal || is_bad_inode(inode))
2764 return 0;
2765
2766 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2767 !inode_is_locked(inode));
2768 /*
2769 * Exit early if inode already is on orphan list. This is a big speedup
2770 * since we don't have to contend on the global s_orphan_lock.
2771 */
2772 if (!list_empty(&EXT4_I(inode)->i_orphan))
2773 return 0;
2774
2775 /*
2776 * Orphan handling is only valid for files with data blocks
2777 * being truncated, or files being unlinked. Note that we either
2778 * hold i_mutex, or the inode can not be referenced from outside,
2779 * so i_nlink should not be bumped due to race
2780 */
2781 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2782 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2783
2784 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2785 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2786 if (err)
2787 goto out;
2788
2789 err = ext4_reserve_inode_write(handle, inode, &iloc);
2790 if (err)
2791 goto out;
2792
2793 mutex_lock(&sbi->s_orphan_lock);
2794 /*
2795 * Due to previous errors inode may be already a part of on-disk
2796 * orphan list. If so skip on-disk list modification.
2797 */
2798 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2799 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2800 /* Insert this inode at the head of the on-disk orphan list */
2801 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2802 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2803 dirty = true;
2804 }
2805 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2806 mutex_unlock(&sbi->s_orphan_lock);
2807
2808 if (dirty) {
2809 err = ext4_handle_dirty_super(handle, sb);
2810 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2811 if (!err)
2812 err = rc;
2813 if (err) {
2814 /*
2815 * We have to remove inode from in-memory list if
2816 * addition to on disk orphan list failed. Stray orphan
2817 * list entries can cause panics at unmount time.
2818 */
2819 mutex_lock(&sbi->s_orphan_lock);
2820 list_del_init(&EXT4_I(inode)->i_orphan);
2821 mutex_unlock(&sbi->s_orphan_lock);
2822 }
2823 }
2824 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2825 jbd_debug(4, "orphan inode %lu will point to %d\n",
2826 inode->i_ino, NEXT_ORPHAN(inode));
2827 out:
2828 ext4_std_error(sb, err);
2829 return err;
2830 }
2831
2832 /*
2833 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2834 * of such inodes stored on disk, because it is finally being cleaned up.
2835 */
2836 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2837 {
2838 struct list_head *prev;
2839 struct ext4_inode_info *ei = EXT4_I(inode);
2840 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2841 __u32 ino_next;
2842 struct ext4_iloc iloc;
2843 int err = 0;
2844
2845 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2846 return 0;
2847
2848 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2849 !inode_is_locked(inode));
2850 /* Do this quick check before taking global s_orphan_lock. */
2851 if (list_empty(&ei->i_orphan))
2852 return 0;
2853
2854 if (handle) {
2855 /* Grab inode buffer early before taking global s_orphan_lock */
2856 err = ext4_reserve_inode_write(handle, inode, &iloc);
2857 }
2858
2859 mutex_lock(&sbi->s_orphan_lock);
2860 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2861
2862 prev = ei->i_orphan.prev;
2863 list_del_init(&ei->i_orphan);
2864
2865 /* If we're on an error path, we may not have a valid
2866 * transaction handle with which to update the orphan list on
2867 * disk, but we still need to remove the inode from the linked
2868 * list in memory. */
2869 if (!handle || err) {
2870 mutex_unlock(&sbi->s_orphan_lock);
2871 goto out_err;
2872 }
2873
2874 ino_next = NEXT_ORPHAN(inode);
2875 if (prev == &sbi->s_orphan) {
2876 jbd_debug(4, "superblock will point to %u\n", ino_next);
2877 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2878 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2879 if (err) {
2880 mutex_unlock(&sbi->s_orphan_lock);
2881 goto out_brelse;
2882 }
2883 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2884 mutex_unlock(&sbi->s_orphan_lock);
2885 err = ext4_handle_dirty_super(handle, inode->i_sb);
2886 } else {
2887 struct ext4_iloc iloc2;
2888 struct inode *i_prev =
2889 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2890
2891 jbd_debug(4, "orphan inode %lu will point to %u\n",
2892 i_prev->i_ino, ino_next);
2893 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2894 if (err) {
2895 mutex_unlock(&sbi->s_orphan_lock);
2896 goto out_brelse;
2897 }
2898 NEXT_ORPHAN(i_prev) = ino_next;
2899 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2900 mutex_unlock(&sbi->s_orphan_lock);
2901 }
2902 if (err)
2903 goto out_brelse;
2904 NEXT_ORPHAN(inode) = 0;
2905 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2906 out_err:
2907 ext4_std_error(inode->i_sb, err);
2908 return err;
2909
2910 out_brelse:
2911 brelse(iloc.bh);
2912 goto out_err;
2913 }
2914
2915 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2916 {
2917 int retval;
2918 struct inode *inode;
2919 struct buffer_head *bh;
2920 struct ext4_dir_entry_2 *de;
2921 handle_t *handle = NULL;
2922
2923 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2924 return -EIO;
2925
2926 /* Initialize quotas before so that eventual writes go in
2927 * separate transaction */
2928 retval = dquot_initialize(dir);
2929 if (retval)
2930 return retval;
2931 retval = dquot_initialize(d_inode(dentry));
2932 if (retval)
2933 return retval;
2934
2935 retval = -ENOENT;
2936 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2937 if (IS_ERR(bh))
2938 return PTR_ERR(bh);
2939 if (!bh)
2940 goto end_rmdir;
2941
2942 inode = d_inode(dentry);
2943
2944 retval = -EFSCORRUPTED;
2945 if (le32_to_cpu(de->inode) != inode->i_ino)
2946 goto end_rmdir;
2947
2948 retval = -ENOTEMPTY;
2949 if (!ext4_empty_dir(inode))
2950 goto end_rmdir;
2951
2952 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2953 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2954 if (IS_ERR(handle)) {
2955 retval = PTR_ERR(handle);
2956 handle = NULL;
2957 goto end_rmdir;
2958 }
2959
2960 if (IS_DIRSYNC(dir))
2961 ext4_handle_sync(handle);
2962
2963 retval = ext4_delete_entry(handle, dir, de, bh);
2964 if (retval)
2965 goto end_rmdir;
2966 if (!EXT4_DIR_LINK_EMPTY(inode))
2967 ext4_warning_inode(inode,
2968 "empty directory '%.*s' has too many links (%u)",
2969 dentry->d_name.len, dentry->d_name.name,
2970 inode->i_nlink);
2971 inode->i_version++;
2972 clear_nlink(inode);
2973 /* There's no need to set i_disksize: the fact that i_nlink is
2974 * zero will ensure that the right thing happens during any
2975 * recovery. */
2976 inode->i_size = 0;
2977 ext4_orphan_add(handle, inode);
2978 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2979 ext4_mark_inode_dirty(handle, inode);
2980 ext4_dec_count(handle, dir);
2981 ext4_update_dx_flag(dir);
2982 ext4_mark_inode_dirty(handle, dir);
2983
2984 end_rmdir:
2985 brelse(bh);
2986 if (handle)
2987 ext4_journal_stop(handle);
2988 return retval;
2989 }
2990
2991 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2992 {
2993 int retval;
2994 struct inode *inode;
2995 struct buffer_head *bh;
2996 struct ext4_dir_entry_2 *de;
2997 handle_t *handle = NULL;
2998
2999 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3000 return -EIO;
3001
3002 trace_ext4_unlink_enter(dir, dentry);
3003 /* Initialize quotas before so that eventual writes go
3004 * in separate transaction */
3005 retval = dquot_initialize(dir);
3006 if (retval)
3007 return retval;
3008 retval = dquot_initialize(d_inode(dentry));
3009 if (retval)
3010 return retval;
3011
3012 retval = -ENOENT;
3013 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3014 if (IS_ERR(bh))
3015 return PTR_ERR(bh);
3016 if (!bh)
3017 goto end_unlink;
3018
3019 inode = d_inode(dentry);
3020
3021 retval = -EFSCORRUPTED;
3022 if (le32_to_cpu(de->inode) != inode->i_ino)
3023 goto end_unlink;
3024
3025 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3026 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3027 if (IS_ERR(handle)) {
3028 retval = PTR_ERR(handle);
3029 handle = NULL;
3030 goto end_unlink;
3031 }
3032
3033 if (IS_DIRSYNC(dir))
3034 ext4_handle_sync(handle);
3035
3036 if (inode->i_nlink == 0) {
3037 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3038 dentry->d_name.len, dentry->d_name.name);
3039 set_nlink(inode, 1);
3040 }
3041 retval = ext4_delete_entry(handle, dir, de, bh);
3042 if (retval)
3043 goto end_unlink;
3044 dir->i_ctime = dir->i_mtime = current_time(dir);
3045 ext4_update_dx_flag(dir);
3046 ext4_mark_inode_dirty(handle, dir);
3047 drop_nlink(inode);
3048 if (!inode->i_nlink)
3049 ext4_orphan_add(handle, inode);
3050 inode->i_ctime = current_time(inode);
3051 ext4_mark_inode_dirty(handle, inode);
3052
3053 end_unlink:
3054 brelse(bh);
3055 if (handle)
3056 ext4_journal_stop(handle);
3057 trace_ext4_unlink_exit(dentry, retval);
3058 return retval;
3059 }
3060
3061 static int ext4_symlink(struct inode *dir,
3062 struct dentry *dentry, const char *symname)
3063 {
3064 handle_t *handle;
3065 struct inode *inode;
3066 int err, len = strlen(symname);
3067 int credits;
3068 struct fscrypt_str disk_link;
3069
3070 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3071 return -EIO;
3072
3073 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3074 &disk_link);
3075 if (err)
3076 return err;
3077
3078 err = dquot_initialize(dir);
3079 if (err)
3080 return err;
3081
3082 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3083 /*
3084 * For non-fast symlinks, we just allocate inode and put it on
3085 * orphan list in the first transaction => we need bitmap,
3086 * group descriptor, sb, inode block, quota blocks, and
3087 * possibly selinux xattr blocks.
3088 */
3089 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3090 EXT4_XATTR_TRANS_BLOCKS;
3091 } else {
3092 /*
3093 * Fast symlink. We have to add entry to directory
3094 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3095 * allocate new inode (bitmap, group descriptor, inode block,
3096 * quota blocks, sb is already counted in previous macros).
3097 */
3098 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3099 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3100 }
3101
3102 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3103 &dentry->d_name, 0, NULL,
3104 EXT4_HT_DIR, credits);
3105 handle = ext4_journal_current_handle();
3106 if (IS_ERR(inode)) {
3107 if (handle)
3108 ext4_journal_stop(handle);
3109 return PTR_ERR(inode);
3110 }
3111
3112 if (IS_ENCRYPTED(inode)) {
3113 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3114 if (err)
3115 goto err_drop_inode;
3116 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3117 }
3118
3119 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3120 if (!IS_ENCRYPTED(inode))
3121 inode->i_op = &ext4_symlink_inode_operations;
3122 inode_nohighmem(inode);
3123 ext4_set_aops(inode);
3124 /*
3125 * We cannot call page_symlink() with transaction started
3126 * because it calls into ext4_write_begin() which can wait
3127 * for transaction commit if we are running out of space
3128 * and thus we deadlock. So we have to stop transaction now
3129 * and restart it when symlink contents is written.
3130 *
3131 * To keep fs consistent in case of crash, we have to put inode
3132 * to orphan list in the mean time.
3133 */
3134 drop_nlink(inode);
3135 err = ext4_orphan_add(handle, inode);
3136 ext4_journal_stop(handle);
3137 handle = NULL;
3138 if (err)
3139 goto err_drop_inode;
3140 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3141 if (err)
3142 goto err_drop_inode;
3143 /*
3144 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3145 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3146 */
3147 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3148 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3149 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3150 if (IS_ERR(handle)) {
3151 err = PTR_ERR(handle);
3152 handle = NULL;
3153 goto err_drop_inode;
3154 }
3155 set_nlink(inode, 1);
3156 err = ext4_orphan_del(handle, inode);
3157 if (err)
3158 goto err_drop_inode;
3159 } else {
3160 /* clear the extent format for fast symlink */
3161 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3162 if (!IS_ENCRYPTED(inode)) {
3163 inode->i_op = &ext4_fast_symlink_inode_operations;
3164 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3165 }
3166 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3167 disk_link.len);
3168 inode->i_size = disk_link.len - 1;
3169 }
3170 EXT4_I(inode)->i_disksize = inode->i_size;
3171 err = ext4_add_nondir(handle, dentry, inode);
3172 if (!err && IS_DIRSYNC(dir))
3173 ext4_handle_sync(handle);
3174
3175 if (handle)
3176 ext4_journal_stop(handle);
3177 goto out_free_encrypted_link;
3178
3179 err_drop_inode:
3180 if (handle)
3181 ext4_journal_stop(handle);
3182 clear_nlink(inode);
3183 unlock_new_inode(inode);
3184 iput(inode);
3185 out_free_encrypted_link:
3186 if (disk_link.name != (unsigned char *)symname)
3187 kfree(disk_link.name);
3188 return err;
3189 }
3190
3191 static int ext4_link(struct dentry *old_dentry,
3192 struct inode *dir, struct dentry *dentry)
3193 {
3194 handle_t *handle;
3195 struct inode *inode = d_inode(old_dentry);
3196 int err, retries = 0;
3197
3198 if (inode->i_nlink >= EXT4_LINK_MAX)
3199 return -EMLINK;
3200 if (ext4_encrypted_inode(dir) &&
3201 !fscrypt_has_permitted_context(dir, inode))
3202 return -EPERM;
3203
3204 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3205 (!projid_eq(EXT4_I(dir)->i_projid,
3206 EXT4_I(old_dentry->d_inode)->i_projid)))
3207 return -EXDEV;
3208
3209 err = dquot_initialize(dir);
3210 if (err)
3211 return err;
3212
3213 retry:
3214 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3215 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3216 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3217 if (IS_ERR(handle))
3218 return PTR_ERR(handle);
3219
3220 if (IS_DIRSYNC(dir))
3221 ext4_handle_sync(handle);
3222
3223 inode->i_ctime = current_time(inode);
3224 ext4_inc_count(handle, inode);
3225 ihold(inode);
3226
3227 err = ext4_add_entry(handle, dentry, inode);
3228 if (!err) {
3229 ext4_mark_inode_dirty(handle, inode);
3230 /* this can happen only for tmpfile being
3231 * linked the first time
3232 */
3233 if (inode->i_nlink == 1)
3234 ext4_orphan_del(handle, inode);
3235 d_instantiate(dentry, inode);
3236 } else {
3237 drop_nlink(inode);
3238 iput(inode);
3239 }
3240 ext4_journal_stop(handle);
3241 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3242 goto retry;
3243 return err;
3244 }
3245
3246
3247 /*
3248 * Try to find buffer head where contains the parent block.
3249 * It should be the inode block if it is inlined or the 1st block
3250 * if it is a normal dir.
3251 */
3252 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3253 struct inode *inode,
3254 int *retval,
3255 struct ext4_dir_entry_2 **parent_de,
3256 int *inlined)
3257 {
3258 struct buffer_head *bh;
3259
3260 if (!ext4_has_inline_data(inode)) {
3261 bh = ext4_read_dirblock(inode, 0, EITHER);
3262 if (IS_ERR(bh)) {
3263 *retval = PTR_ERR(bh);
3264 return NULL;
3265 }
3266 *parent_de = ext4_next_entry(
3267 (struct ext4_dir_entry_2 *)bh->b_data,
3268 inode->i_sb->s_blocksize);
3269 return bh;
3270 }
3271
3272 *inlined = 1;
3273 return ext4_get_first_inline_block(inode, parent_de, retval);
3274 }
3275
3276 struct ext4_renament {
3277 struct inode *dir;
3278 struct dentry *dentry;
3279 struct inode *inode;
3280 bool is_dir;
3281 int dir_nlink_delta;
3282
3283 /* entry for "dentry" */
3284 struct buffer_head *bh;
3285 struct ext4_dir_entry_2 *de;
3286 int inlined;
3287
3288 /* entry for ".." in inode if it's a directory */
3289 struct buffer_head *dir_bh;
3290 struct ext4_dir_entry_2 *parent_de;
3291 int dir_inlined;
3292 };
3293
3294 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3295 {
3296 int retval;
3297
3298 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3299 &retval, &ent->parent_de,
3300 &ent->dir_inlined);
3301 if (!ent->dir_bh)
3302 return retval;
3303 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3304 return -EFSCORRUPTED;
3305 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3306 return ext4_journal_get_write_access(handle, ent->dir_bh);
3307 }
3308
3309 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3310 unsigned dir_ino)
3311 {
3312 int retval;
3313
3314 ent->parent_de->inode = cpu_to_le32(dir_ino);
3315 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3316 if (!ent->dir_inlined) {
3317 if (is_dx(ent->inode)) {
3318 retval = ext4_handle_dirty_dx_node(handle,
3319 ent->inode,
3320 ent->dir_bh);
3321 } else {
3322 retval = ext4_handle_dirty_dirent_node(handle,
3323 ent->inode,
3324 ent->dir_bh);
3325 }
3326 } else {
3327 retval = ext4_mark_inode_dirty(handle, ent->inode);
3328 }
3329 if (retval) {
3330 ext4_std_error(ent->dir->i_sb, retval);
3331 return retval;
3332 }
3333 return 0;
3334 }
3335
3336 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3337 unsigned ino, unsigned file_type)
3338 {
3339 int retval;
3340
3341 BUFFER_TRACE(ent->bh, "get write access");
3342 retval = ext4_journal_get_write_access(handle, ent->bh);
3343 if (retval)
3344 return retval;
3345 ent->de->inode = cpu_to_le32(ino);
3346 if (ext4_has_feature_filetype(ent->dir->i_sb))
3347 ent->de->file_type = file_type;
3348 ent->dir->i_version++;
3349 ent->dir->i_ctime = ent->dir->i_mtime =
3350 current_time(ent->dir);
3351 ext4_mark_inode_dirty(handle, ent->dir);
3352 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3353 if (!ent->inlined) {
3354 retval = ext4_handle_dirty_dirent_node(handle,
3355 ent->dir, ent->bh);
3356 if (unlikely(retval)) {
3357 ext4_std_error(ent->dir->i_sb, retval);
3358 return retval;
3359 }
3360 }
3361 brelse(ent->bh);
3362 ent->bh = NULL;
3363
3364 return 0;
3365 }
3366
3367 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3368 const struct qstr *d_name)
3369 {
3370 int retval = -ENOENT;
3371 struct buffer_head *bh;
3372 struct ext4_dir_entry_2 *de;
3373
3374 bh = ext4_find_entry(dir, d_name, &de, NULL);
3375 if (IS_ERR(bh))
3376 return PTR_ERR(bh);
3377 if (bh) {
3378 retval = ext4_delete_entry(handle, dir, de, bh);
3379 brelse(bh);
3380 }
3381 return retval;
3382 }
3383
3384 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3385 int force_reread)
3386 {
3387 int retval;
3388 /*
3389 * ent->de could have moved from under us during htree split, so make
3390 * sure that we are deleting the right entry. We might also be pointing
3391 * to a stale entry in the unused part of ent->bh so just checking inum
3392 * and the name isn't enough.
3393 */
3394 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3395 ent->de->name_len != ent->dentry->d_name.len ||
3396 strncmp(ent->de->name, ent->dentry->d_name.name,
3397 ent->de->name_len) ||
3398 force_reread) {
3399 retval = ext4_find_delete_entry(handle, ent->dir,
3400 &ent->dentry->d_name);
3401 } else {
3402 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3403 if (retval == -ENOENT) {
3404 retval = ext4_find_delete_entry(handle, ent->dir,
3405 &ent->dentry->d_name);
3406 }
3407 }
3408
3409 if (retval) {
3410 ext4_warning_inode(ent->dir,
3411 "Deleting old file: nlink %d, error=%d",
3412 ent->dir->i_nlink, retval);
3413 }
3414 }
3415
3416 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3417 {
3418 if (ent->dir_nlink_delta) {
3419 if (ent->dir_nlink_delta == -1)
3420 ext4_dec_count(handle, ent->dir);
3421 else
3422 ext4_inc_count(handle, ent->dir);
3423 ext4_mark_inode_dirty(handle, ent->dir);
3424 }
3425 }
3426
3427 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3428 int credits, handle_t **h)
3429 {
3430 struct inode *wh;
3431 handle_t *handle;
3432 int retries = 0;
3433
3434 /*
3435 * for inode block, sb block, group summaries,
3436 * and inode bitmap
3437 */
3438 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3439 EXT4_XATTR_TRANS_BLOCKS + 4);
3440 retry:
3441 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3442 &ent->dentry->d_name, 0, NULL,
3443 EXT4_HT_DIR, credits);
3444
3445 handle = ext4_journal_current_handle();
3446 if (IS_ERR(wh)) {
3447 if (handle)
3448 ext4_journal_stop(handle);
3449 if (PTR_ERR(wh) == -ENOSPC &&
3450 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3451 goto retry;
3452 } else {
3453 *h = handle;
3454 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3455 wh->i_op = &ext4_special_inode_operations;
3456 }
3457 return wh;
3458 }
3459
3460 /*
3461 * Anybody can rename anything with this: the permission checks are left to the
3462 * higher-level routines.
3463 *
3464 * n.b. old_{dentry,inode) refers to the source dentry/inode
3465 * while new_{dentry,inode) refers to the destination dentry/inode
3466 * This comes from rename(const char *oldpath, const char *newpath)
3467 */
3468 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3469 struct inode *new_dir, struct dentry *new_dentry,
3470 unsigned int flags)
3471 {
3472 handle_t *handle = NULL;
3473 struct ext4_renament old = {
3474 .dir = old_dir,
3475 .dentry = old_dentry,
3476 .inode = d_inode(old_dentry),
3477 };
3478 struct ext4_renament new = {
3479 .dir = new_dir,
3480 .dentry = new_dentry,
3481 .inode = d_inode(new_dentry),
3482 };
3483 int force_reread;
3484 int retval;
3485 struct inode *whiteout = NULL;
3486 int credits;
3487 u8 old_file_type;
3488
3489 if (new.inode && new.inode->i_nlink == 0) {
3490 EXT4_ERROR_INODE(new.inode,
3491 "target of rename is already freed");
3492 return -EFSCORRUPTED;
3493 }
3494
3495 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3496 (!projid_eq(EXT4_I(new_dir)->i_projid,
3497 EXT4_I(old_dentry->d_inode)->i_projid)))
3498 return -EXDEV;
3499
3500 if ((ext4_encrypted_inode(old_dir) &&
3501 !fscrypt_has_encryption_key(old_dir)) ||
3502 (ext4_encrypted_inode(new_dir) &&
3503 !fscrypt_has_encryption_key(new_dir)))
3504 return -ENOKEY;
3505
3506 retval = dquot_initialize(old.dir);
3507 if (retval)
3508 return retval;
3509 retval = dquot_initialize(new.dir);
3510 if (retval)
3511 return retval;
3512
3513 /* Initialize quotas before so that eventual writes go
3514 * in separate transaction */
3515 if (new.inode) {
3516 retval = dquot_initialize(new.inode);
3517 if (retval)
3518 return retval;
3519 }
3520
3521 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3522 if (IS_ERR(old.bh))
3523 return PTR_ERR(old.bh);
3524 /*
3525 * Check for inode number is _not_ due to possible IO errors.
3526 * We might rmdir the source, keep it as pwd of some process
3527 * and merrily kill the link to whatever was created under the
3528 * same name. Goodbye sticky bit ;-<
3529 */
3530 retval = -ENOENT;
3531 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3532 goto end_rename;
3533
3534 if ((old.dir != new.dir) &&
3535 ext4_encrypted_inode(new.dir) &&
3536 !fscrypt_has_permitted_context(new.dir, old.inode)) {
3537 retval = -EPERM;
3538 goto end_rename;
3539 }
3540
3541 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3542 &new.de, &new.inlined);
3543 if (IS_ERR(new.bh)) {
3544 retval = PTR_ERR(new.bh);
3545 new.bh = NULL;
3546 goto end_rename;
3547 }
3548 if (new.bh) {
3549 if (!new.inode) {
3550 brelse(new.bh);
3551 new.bh = NULL;
3552 }
3553 }
3554 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3555 ext4_alloc_da_blocks(old.inode);
3556
3557 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3558 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3559 if (!(flags & RENAME_WHITEOUT)) {
3560 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3561 if (IS_ERR(handle)) {
3562 retval = PTR_ERR(handle);
3563 handle = NULL;
3564 goto end_rename;
3565 }
3566 } else {
3567 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3568 if (IS_ERR(whiteout)) {
3569 retval = PTR_ERR(whiteout);
3570 whiteout = NULL;
3571 goto end_rename;
3572 }
3573 }
3574
3575 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3576 ext4_handle_sync(handle);
3577
3578 if (S_ISDIR(old.inode->i_mode)) {
3579 if (new.inode) {
3580 retval = -ENOTEMPTY;
3581 if (!ext4_empty_dir(new.inode))
3582 goto end_rename;
3583 } else {
3584 retval = -EMLINK;
3585 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3586 goto end_rename;
3587 }
3588 retval = ext4_rename_dir_prepare(handle, &old);
3589 if (retval)
3590 goto end_rename;
3591 }
3592 /*
3593 * If we're renaming a file within an inline_data dir and adding or
3594 * setting the new dirent causes a conversion from inline_data to
3595 * extents/blockmap, we need to force the dirent delete code to
3596 * re-read the directory, or else we end up trying to delete a dirent
3597 * from what is now the extent tree root (or a block map).
3598 */
3599 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3600 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3601
3602 old_file_type = old.de->file_type;
3603 if (whiteout) {
3604 /*
3605 * Do this before adding a new entry, so the old entry is sure
3606 * to be still pointing to the valid old entry.
3607 */
3608 retval = ext4_setent(handle, &old, whiteout->i_ino,
3609 EXT4_FT_CHRDEV);
3610 if (retval)
3611 goto end_rename;
3612 ext4_mark_inode_dirty(handle, whiteout);
3613 }
3614 if (!new.bh) {
3615 retval = ext4_add_entry(handle, new.dentry, old.inode);
3616 if (retval)
3617 goto end_rename;
3618 } else {
3619 retval = ext4_setent(handle, &new,
3620 old.inode->i_ino, old_file_type);
3621 if (retval)
3622 goto end_rename;
3623 }
3624 if (force_reread)
3625 force_reread = !ext4_test_inode_flag(new.dir,
3626 EXT4_INODE_INLINE_DATA);
3627
3628 /*
3629 * Like most other Unix systems, set the ctime for inodes on a
3630 * rename.
3631 */
3632 old.inode->i_ctime = current_time(old.inode);
3633 ext4_mark_inode_dirty(handle, old.inode);
3634
3635 if (!whiteout) {
3636 /*
3637 * ok, that's it
3638 */
3639 ext4_rename_delete(handle, &old, force_reread);
3640 }
3641
3642 if (new.inode) {
3643 ext4_dec_count(handle, new.inode);
3644 new.inode->i_ctime = current_time(new.inode);
3645 }
3646 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3647 ext4_update_dx_flag(old.dir);
3648 if (old.dir_bh) {
3649 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3650 if (retval)
3651 goto end_rename;
3652
3653 ext4_dec_count(handle, old.dir);
3654 if (new.inode) {
3655 /* checked ext4_empty_dir above, can't have another
3656 * parent, ext4_dec_count() won't work for many-linked
3657 * dirs */
3658 clear_nlink(new.inode);
3659 } else {
3660 ext4_inc_count(handle, new.dir);
3661 ext4_update_dx_flag(new.dir);
3662 ext4_mark_inode_dirty(handle, new.dir);
3663 }
3664 }
3665 ext4_mark_inode_dirty(handle, old.dir);
3666 if (new.inode) {
3667 ext4_mark_inode_dirty(handle, new.inode);
3668 if (!new.inode->i_nlink)
3669 ext4_orphan_add(handle, new.inode);
3670 }
3671 retval = 0;
3672
3673 end_rename:
3674 brelse(old.dir_bh);
3675 brelse(old.bh);
3676 brelse(new.bh);
3677 if (whiteout) {
3678 if (retval)
3679 drop_nlink(whiteout);
3680 unlock_new_inode(whiteout);
3681 iput(whiteout);
3682 }
3683 if (handle)
3684 ext4_journal_stop(handle);
3685 return retval;
3686 }
3687
3688 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3689 struct inode *new_dir, struct dentry *new_dentry)
3690 {
3691 handle_t *handle = NULL;
3692 struct ext4_renament old = {
3693 .dir = old_dir,
3694 .dentry = old_dentry,
3695 .inode = d_inode(old_dentry),
3696 };
3697 struct ext4_renament new = {
3698 .dir = new_dir,
3699 .dentry = new_dentry,
3700 .inode = d_inode(new_dentry),
3701 };
3702 u8 new_file_type;
3703 int retval;
3704 struct timespec ctime;
3705
3706 if ((ext4_encrypted_inode(old_dir) &&
3707 !fscrypt_has_encryption_key(old_dir)) ||
3708 (ext4_encrypted_inode(new_dir) &&
3709 !fscrypt_has_encryption_key(new_dir)))
3710 return -ENOKEY;
3711
3712 if ((ext4_encrypted_inode(old_dir) ||
3713 ext4_encrypted_inode(new_dir)) &&
3714 (old_dir != new_dir) &&
3715 (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3716 !fscrypt_has_permitted_context(old_dir, new.inode)))
3717 return -EPERM;
3718
3719 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3720 !projid_eq(EXT4_I(new_dir)->i_projid,
3721 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3722 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3723 !projid_eq(EXT4_I(old_dir)->i_projid,
3724 EXT4_I(new_dentry->d_inode)->i_projid)))
3725 return -EXDEV;
3726
3727 retval = dquot_initialize(old.dir);
3728 if (retval)
3729 return retval;
3730 retval = dquot_initialize(new.dir);
3731 if (retval)
3732 return retval;
3733
3734 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3735 &old.de, &old.inlined);
3736 if (IS_ERR(old.bh))
3737 return PTR_ERR(old.bh);
3738 /*
3739 * Check for inode number is _not_ due to possible IO errors.
3740 * We might rmdir the source, keep it as pwd of some process
3741 * and merrily kill the link to whatever was created under the
3742 * same name. Goodbye sticky bit ;-<
3743 */
3744 retval = -ENOENT;
3745 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3746 goto end_rename;
3747
3748 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3749 &new.de, &new.inlined);
3750 if (IS_ERR(new.bh)) {
3751 retval = PTR_ERR(new.bh);
3752 new.bh = NULL;
3753 goto end_rename;
3754 }
3755
3756 /* RENAME_EXCHANGE case: old *and* new must both exist */
3757 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3758 goto end_rename;
3759
3760 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3761 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3762 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3763 if (IS_ERR(handle)) {
3764 retval = PTR_ERR(handle);
3765 handle = NULL;
3766 goto end_rename;
3767 }
3768
3769 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3770 ext4_handle_sync(handle);
3771
3772 if (S_ISDIR(old.inode->i_mode)) {
3773 old.is_dir = true;
3774 retval = ext4_rename_dir_prepare(handle, &old);
3775 if (retval)
3776 goto end_rename;
3777 }
3778 if (S_ISDIR(new.inode->i_mode)) {
3779 new.is_dir = true;
3780 retval = ext4_rename_dir_prepare(handle, &new);
3781 if (retval)
3782 goto end_rename;
3783 }
3784
3785 /*
3786 * Other than the special case of overwriting a directory, parents'
3787 * nlink only needs to be modified if this is a cross directory rename.
3788 */
3789 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3790 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3791 new.dir_nlink_delta = -old.dir_nlink_delta;
3792 retval = -EMLINK;
3793 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3794 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3795 goto end_rename;
3796 }
3797
3798 new_file_type = new.de->file_type;
3799 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3800 if (retval)
3801 goto end_rename;
3802
3803 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3804 if (retval)
3805 goto end_rename;
3806
3807 /*
3808 * Like most other Unix systems, set the ctime for inodes on a
3809 * rename.
3810 */
3811 ctime = current_time(old.inode);
3812 old.inode->i_ctime = ctime;
3813 new.inode->i_ctime = ctime;
3814 ext4_mark_inode_dirty(handle, old.inode);
3815 ext4_mark_inode_dirty(handle, new.inode);
3816
3817 if (old.dir_bh) {
3818 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3819 if (retval)
3820 goto end_rename;
3821 }
3822 if (new.dir_bh) {
3823 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3824 if (retval)
3825 goto end_rename;
3826 }
3827 ext4_update_dir_count(handle, &old);
3828 ext4_update_dir_count(handle, &new);
3829 retval = 0;
3830
3831 end_rename:
3832 brelse(old.dir_bh);
3833 brelse(new.dir_bh);
3834 brelse(old.bh);
3835 brelse(new.bh);
3836 if (handle)
3837 ext4_journal_stop(handle);
3838 return retval;
3839 }
3840
3841 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3842 struct inode *new_dir, struct dentry *new_dentry,
3843 unsigned int flags)
3844 {
3845 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3846 return -EIO;
3847
3848 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3849 return -EINVAL;
3850
3851 if (flags & RENAME_EXCHANGE) {
3852 return ext4_cross_rename(old_dir, old_dentry,
3853 new_dir, new_dentry);
3854 }
3855
3856 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3857 }
3858
3859 /*
3860 * directories can handle most operations...
3861 */
3862 const struct inode_operations ext4_dir_inode_operations = {
3863 .create = ext4_create,
3864 .lookup = ext4_lookup,
3865 .link = ext4_link,
3866 .unlink = ext4_unlink,
3867 .symlink = ext4_symlink,
3868 .mkdir = ext4_mkdir,
3869 .rmdir = ext4_rmdir,
3870 .mknod = ext4_mknod,
3871 .tmpfile = ext4_tmpfile,
3872 .rename = ext4_rename2,
3873 .setattr = ext4_setattr,
3874 .getattr = ext4_getattr,
3875 .listxattr = ext4_listxattr,
3876 .get_acl = ext4_get_acl,
3877 .set_acl = ext4_set_acl,
3878 .fiemap = ext4_fiemap,
3879 };
3880
3881 const struct inode_operations ext4_special_inode_operations = {
3882 .setattr = ext4_setattr,
3883 .getattr = ext4_getattr,
3884 .listxattr = ext4_listxattr,
3885 .get_acl = ext4_get_acl,
3886 .set_acl = ext4_set_acl,
3887 };