]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/ext4/namei.c
ext4: remove "nokey" check from ext4_lookup()
[mirror_ubuntu-artful-kernel.git] / fs / ext4 / namei.c
1 /*
2 * linux/fs/ext4/namei.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
24 * Theodore Ts'o, 2002
25 */
26
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
36 #include "ext4.h"
37 #include "ext4_jbd2.h"
38
39 #include "xattr.h"
40 #include "acl.h"
41
42 #include <trace/events/ext4.h>
43 /*
44 * define how far ahead to read directories while searching them.
45 */
46 #define NAMEI_RA_CHUNKS 2
47 #define NAMEI_RA_BLOCKS 4
48 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49
50 static struct buffer_head *ext4_append(handle_t *handle,
51 struct inode *inode,
52 ext4_lblk_t *block)
53 {
54 struct buffer_head *bh;
55 int err;
56
57 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 ((inode->i_size >> 10) >=
59 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 return ERR_PTR(-ENOSPC);
61
62 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
63
64 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
65 if (IS_ERR(bh))
66 return bh;
67 inode->i_size += inode->i_sb->s_blocksize;
68 EXT4_I(inode)->i_disksize = inode->i_size;
69 BUFFER_TRACE(bh, "get_write_access");
70 err = ext4_journal_get_write_access(handle, bh);
71 if (err) {
72 brelse(bh);
73 ext4_std_error(inode->i_sb, err);
74 return ERR_PTR(err);
75 }
76 return bh;
77 }
78
79 static int ext4_dx_csum_verify(struct inode *inode,
80 struct ext4_dir_entry *dirent);
81
82 typedef enum {
83 EITHER, INDEX, DIRENT
84 } dirblock_type_t;
85
86 #define ext4_read_dirblock(inode, block, type) \
87 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
88
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90 ext4_lblk_t block,
91 dirblock_type_t type,
92 const char *func,
93 unsigned int line)
94 {
95 struct buffer_head *bh;
96 struct ext4_dir_entry *dirent;
97 int is_dx_block = 0;
98
99 bh = ext4_bread(NULL, inode, block, 0);
100 if (IS_ERR(bh)) {
101 __ext4_warning(inode->i_sb, func, line,
102 "inode #%lu: lblock %lu: comm %s: "
103 "error %ld reading directory block",
104 inode->i_ino, (unsigned long)block,
105 current->comm, PTR_ERR(bh));
106
107 return bh;
108 }
109 if (!bh) {
110 ext4_error_inode(inode, func, line, block,
111 "Directory hole found");
112 return ERR_PTR(-EFSCORRUPTED);
113 }
114 dirent = (struct ext4_dir_entry *) bh->b_data;
115 /* Determine whether or not we have an index block */
116 if (is_dx(inode)) {
117 if (block == 0)
118 is_dx_block = 1;
119 else if (ext4_rec_len_from_disk(dirent->rec_len,
120 inode->i_sb->s_blocksize) ==
121 inode->i_sb->s_blocksize)
122 is_dx_block = 1;
123 }
124 if (!is_dx_block && type == INDEX) {
125 ext4_error_inode(inode, func, line, block,
126 "directory leaf block found instead of index block");
127 return ERR_PTR(-EFSCORRUPTED);
128 }
129 if (!ext4_has_metadata_csum(inode->i_sb) ||
130 buffer_verified(bh))
131 return bh;
132
133 /*
134 * An empty leaf block can get mistaken for a index block; for
135 * this reason, we can only check the index checksum when the
136 * caller is sure it should be an index block.
137 */
138 if (is_dx_block && type == INDEX) {
139 if (ext4_dx_csum_verify(inode, dirent))
140 set_buffer_verified(bh);
141 else {
142 ext4_error_inode(inode, func, line, block,
143 "Directory index failed checksum");
144 brelse(bh);
145 return ERR_PTR(-EFSBADCRC);
146 }
147 }
148 if (!is_dx_block) {
149 if (ext4_dirent_csum_verify(inode, dirent))
150 set_buffer_verified(bh);
151 else {
152 ext4_error_inode(inode, func, line, block,
153 "Directory block failed checksum");
154 brelse(bh);
155 return ERR_PTR(-EFSBADCRC);
156 }
157 }
158 return bh;
159 }
160
161 #ifndef assert
162 #define assert(test) J_ASSERT(test)
163 #endif
164
165 #ifdef DX_DEBUG
166 #define dxtrace(command) command
167 #else
168 #define dxtrace(command)
169 #endif
170
171 struct fake_dirent
172 {
173 __le32 inode;
174 __le16 rec_len;
175 u8 name_len;
176 u8 file_type;
177 };
178
179 struct dx_countlimit
180 {
181 __le16 limit;
182 __le16 count;
183 };
184
185 struct dx_entry
186 {
187 __le32 hash;
188 __le32 block;
189 };
190
191 /*
192 * dx_root_info is laid out so that if it should somehow get overlaid by a
193 * dirent the two low bits of the hash version will be zero. Therefore, the
194 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
195 */
196
197 struct dx_root
198 {
199 struct fake_dirent dot;
200 char dot_name[4];
201 struct fake_dirent dotdot;
202 char dotdot_name[4];
203 struct dx_root_info
204 {
205 __le32 reserved_zero;
206 u8 hash_version;
207 u8 info_length; /* 8 */
208 u8 indirect_levels;
209 u8 unused_flags;
210 }
211 info;
212 struct dx_entry entries[0];
213 };
214
215 struct dx_node
216 {
217 struct fake_dirent fake;
218 struct dx_entry entries[0];
219 };
220
221
222 struct dx_frame
223 {
224 struct buffer_head *bh;
225 struct dx_entry *entries;
226 struct dx_entry *at;
227 };
228
229 struct dx_map_entry
230 {
231 u32 hash;
232 u16 offs;
233 u16 size;
234 };
235
236 /*
237 * This goes at the end of each htree block.
238 */
239 struct dx_tail {
240 u32 dt_reserved;
241 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
242 };
243
244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246 static inline unsigned dx_get_hash(struct dx_entry *entry);
247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
248 static unsigned dx_get_count(struct dx_entry *entries);
249 static unsigned dx_get_limit(struct dx_entry *entries);
250 static void dx_set_count(struct dx_entry *entries, unsigned value);
251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253 static unsigned dx_node_limit(struct inode *dir);
254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
255 struct inode *dir,
256 struct dx_hash_info *hinfo,
257 struct dx_frame *frame);
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 unsigned blocksize, struct dx_hash_info *hinfo,
261 struct dx_map_entry map[]);
262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264 struct dx_map_entry *offsets, int count, unsigned blocksize);
265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266 static void dx_insert_block(struct dx_frame *frame,
267 u32 hash, ext4_lblk_t block);
268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269 struct dx_frame *frame,
270 struct dx_frame *frames,
271 __u32 *start_hash);
272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273 struct ext4_filename *fname,
274 struct ext4_dir_entry_2 **res_dir);
275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276 struct inode *dir, struct inode *inode);
277
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 unsigned int blocksize)
281 {
282 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 t->det_rec_len = ext4_rec_len_to_disk(
284 sizeof(struct ext4_dir_entry_tail), blocksize);
285 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
286 }
287
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 struct ext4_dir_entry *de)
291 {
292 struct ext4_dir_entry_tail *t;
293
294 #ifdef PARANOID
295 struct ext4_dir_entry *d, *top;
296
297 d = de;
298 top = (struct ext4_dir_entry *)(((void *)de) +
299 (EXT4_BLOCK_SIZE(inode->i_sb) -
300 sizeof(struct ext4_dir_entry_tail)));
301 while (d < top && d->rec_len)
302 d = (struct ext4_dir_entry *)(((void *)d) +
303 le16_to_cpu(d->rec_len));
304
305 if (d != top)
306 return NULL;
307
308 t = (struct ext4_dir_entry_tail *)d;
309 #else
310 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
311 #endif
312
313 if (t->det_reserved_zero1 ||
314 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 t->det_reserved_zero2 ||
316 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
317 return NULL;
318
319 return t;
320 }
321
322 static __le32 ext4_dirent_csum(struct inode *inode,
323 struct ext4_dir_entry *dirent, int size)
324 {
325 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 struct ext4_inode_info *ei = EXT4_I(inode);
327 __u32 csum;
328
329 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 return cpu_to_le32(csum);
331 }
332
333 #define warn_no_space_for_csum(inode) \
334 __warn_no_space_for_csum((inode), __func__, __LINE__)
335
336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
337 unsigned int line)
338 {
339 __ext4_warning_inode(inode, func, line,
340 "No space for directory leaf checksum. Please run e2fsck -D.");
341 }
342
343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
344 {
345 struct ext4_dir_entry_tail *t;
346
347 if (!ext4_has_metadata_csum(inode->i_sb))
348 return 1;
349
350 t = get_dirent_tail(inode, dirent);
351 if (!t) {
352 warn_no_space_for_csum(inode);
353 return 0;
354 }
355
356 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 (void *)t - (void *)dirent))
358 return 0;
359
360 return 1;
361 }
362
363 static void ext4_dirent_csum_set(struct inode *inode,
364 struct ext4_dir_entry *dirent)
365 {
366 struct ext4_dir_entry_tail *t;
367
368 if (!ext4_has_metadata_csum(inode->i_sb))
369 return;
370
371 t = get_dirent_tail(inode, dirent);
372 if (!t) {
373 warn_no_space_for_csum(inode);
374 return;
375 }
376
377 t->det_checksum = ext4_dirent_csum(inode, dirent,
378 (void *)t - (void *)dirent);
379 }
380
381 int ext4_handle_dirty_dirent_node(handle_t *handle,
382 struct inode *inode,
383 struct buffer_head *bh)
384 {
385 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 return ext4_handle_dirty_metadata(handle, inode, bh);
387 }
388
389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 struct ext4_dir_entry *dirent,
391 int *offset)
392 {
393 struct ext4_dir_entry *dp;
394 struct dx_root_info *root;
395 int count_offset;
396
397 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
398 count_offset = 8;
399 else if (le16_to_cpu(dirent->rec_len) == 12) {
400 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 if (le16_to_cpu(dp->rec_len) !=
402 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
403 return NULL;
404 root = (struct dx_root_info *)(((void *)dp + 12));
405 if (root->reserved_zero ||
406 root->info_length != sizeof(struct dx_root_info))
407 return NULL;
408 count_offset = 32;
409 } else
410 return NULL;
411
412 if (offset)
413 *offset = count_offset;
414 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
415 }
416
417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 int count_offset, int count, struct dx_tail *t)
419 {
420 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 struct ext4_inode_info *ei = EXT4_I(inode);
422 __u32 csum;
423 int size;
424 __u32 dummy_csum = 0;
425 int offset = offsetof(struct dx_tail, dt_checksum);
426
427 size = count_offset + (count * sizeof(struct dx_entry));
428 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
429 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
430 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
431
432 return cpu_to_le32(csum);
433 }
434
435 static int ext4_dx_csum_verify(struct inode *inode,
436 struct ext4_dir_entry *dirent)
437 {
438 struct dx_countlimit *c;
439 struct dx_tail *t;
440 int count_offset, limit, count;
441
442 if (!ext4_has_metadata_csum(inode->i_sb))
443 return 1;
444
445 c = get_dx_countlimit(inode, dirent, &count_offset);
446 if (!c) {
447 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
448 return 0;
449 }
450 limit = le16_to_cpu(c->limit);
451 count = le16_to_cpu(c->count);
452 if (count_offset + (limit * sizeof(struct dx_entry)) >
453 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
454 warn_no_space_for_csum(inode);
455 return 0;
456 }
457 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
458
459 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
460 count, t))
461 return 0;
462 return 1;
463 }
464
465 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
466 {
467 struct dx_countlimit *c;
468 struct dx_tail *t;
469 int count_offset, limit, count;
470
471 if (!ext4_has_metadata_csum(inode->i_sb))
472 return;
473
474 c = get_dx_countlimit(inode, dirent, &count_offset);
475 if (!c) {
476 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
477 return;
478 }
479 limit = le16_to_cpu(c->limit);
480 count = le16_to_cpu(c->count);
481 if (count_offset + (limit * sizeof(struct dx_entry)) >
482 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
483 warn_no_space_for_csum(inode);
484 return;
485 }
486 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
487
488 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
489 }
490
491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
492 struct inode *inode,
493 struct buffer_head *bh)
494 {
495 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496 return ext4_handle_dirty_metadata(handle, inode, bh);
497 }
498
499 /*
500 * p is at least 6 bytes before the end of page
501 */
502 static inline struct ext4_dir_entry_2 *
503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
504 {
505 return (struct ext4_dir_entry_2 *)((char *)p +
506 ext4_rec_len_from_disk(p->rec_len, blocksize));
507 }
508
509 /*
510 * Future: use high four bits of block for coalesce-on-delete flags
511 * Mask them off for now.
512 */
513
514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
515 {
516 return le32_to_cpu(entry->block) & 0x00ffffff;
517 }
518
519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
520 {
521 entry->block = cpu_to_le32(value);
522 }
523
524 static inline unsigned dx_get_hash(struct dx_entry *entry)
525 {
526 return le32_to_cpu(entry->hash);
527 }
528
529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
530 {
531 entry->hash = cpu_to_le32(value);
532 }
533
534 static inline unsigned dx_get_count(struct dx_entry *entries)
535 {
536 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
537 }
538
539 static inline unsigned dx_get_limit(struct dx_entry *entries)
540 {
541 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
542 }
543
544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
545 {
546 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
547 }
548
549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
550 {
551 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
552 }
553
554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
555 {
556 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 EXT4_DIR_REC_LEN(2) - infosize;
558
559 if (ext4_has_metadata_csum(dir->i_sb))
560 entry_space -= sizeof(struct dx_tail);
561 return entry_space / sizeof(struct dx_entry);
562 }
563
564 static inline unsigned dx_node_limit(struct inode *dir)
565 {
566 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
567
568 if (ext4_has_metadata_csum(dir->i_sb))
569 entry_space -= sizeof(struct dx_tail);
570 return entry_space / sizeof(struct dx_entry);
571 }
572
573 /*
574 * Debug
575 */
576 #ifdef DX_DEBUG
577 static void dx_show_index(char * label, struct dx_entry *entries)
578 {
579 int i, n = dx_get_count (entries);
580 printk(KERN_DEBUG "%s index", label);
581 for (i = 0; i < n; i++) {
582 printk(KERN_CONT " %x->%lu",
583 i ? dx_get_hash(entries + i) : 0,
584 (unsigned long)dx_get_block(entries + i));
585 }
586 printk(KERN_CONT "\n");
587 }
588
589 struct stats
590 {
591 unsigned names;
592 unsigned space;
593 unsigned bcount;
594 };
595
596 static struct stats dx_show_leaf(struct inode *dir,
597 struct dx_hash_info *hinfo,
598 struct ext4_dir_entry_2 *de,
599 int size, int show_names)
600 {
601 unsigned names = 0, space = 0;
602 char *base = (char *) de;
603 struct dx_hash_info h = *hinfo;
604
605 printk("names: ");
606 while ((char *) de < base + size)
607 {
608 if (de->inode)
609 {
610 if (show_names)
611 {
612 #ifdef CONFIG_EXT4_FS_ENCRYPTION
613 int len;
614 char *name;
615 struct fscrypt_str fname_crypto_str =
616 FSTR_INIT(NULL, 0);
617 int res = 0;
618
619 name = de->name;
620 len = de->name_len;
621 if (ext4_encrypted_inode(dir))
622 res = fscrypt_get_encryption_info(dir);
623 if (res) {
624 printk(KERN_WARNING "Error setting up"
625 " fname crypto: %d\n", res);
626 }
627 if (!fscrypt_has_encryption_key(dir)) {
628 /* Directory is not encrypted */
629 ext4fs_dirhash(de->name,
630 de->name_len, &h);
631 printk("%*.s:(U)%x.%u ", len,
632 name, h.hash,
633 (unsigned) ((char *) de
634 - base));
635 } else {
636 struct fscrypt_str de_name =
637 FSTR_INIT(name, len);
638
639 /* Directory is encrypted */
640 res = fscrypt_fname_alloc_buffer(
641 dir, len,
642 &fname_crypto_str);
643 if (res)
644 printk(KERN_WARNING "Error "
645 "allocating crypto "
646 "buffer--skipping "
647 "crypto\n");
648 res = fscrypt_fname_disk_to_usr(dir,
649 0, 0, &de_name,
650 &fname_crypto_str);
651 if (res) {
652 printk(KERN_WARNING "Error "
653 "converting filename "
654 "from disk to usr"
655 "\n");
656 name = "??";
657 len = 2;
658 } else {
659 name = fname_crypto_str.name;
660 len = fname_crypto_str.len;
661 }
662 ext4fs_dirhash(de->name, de->name_len,
663 &h);
664 printk("%*.s:(E)%x.%u ", len, name,
665 h.hash, (unsigned) ((char *) de
666 - base));
667 fscrypt_fname_free_buffer(
668 &fname_crypto_str);
669 }
670 #else
671 int len = de->name_len;
672 char *name = de->name;
673 ext4fs_dirhash(de->name, de->name_len, &h);
674 printk("%*.s:%x.%u ", len, name, h.hash,
675 (unsigned) ((char *) de - base));
676 #endif
677 }
678 space += EXT4_DIR_REC_LEN(de->name_len);
679 names++;
680 }
681 de = ext4_next_entry(de, size);
682 }
683 printk(KERN_CONT "(%i)\n", names);
684 return (struct stats) { names, space, 1 };
685 }
686
687 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
688 struct dx_entry *entries, int levels)
689 {
690 unsigned blocksize = dir->i_sb->s_blocksize;
691 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
692 unsigned bcount = 0;
693 struct buffer_head *bh;
694 printk("%i indexed blocks...\n", count);
695 for (i = 0; i < count; i++, entries++)
696 {
697 ext4_lblk_t block = dx_get_block(entries);
698 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
699 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
700 struct stats stats;
701 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
702 bh = ext4_bread(NULL,dir, block, 0);
703 if (!bh || IS_ERR(bh))
704 continue;
705 stats = levels?
706 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
707 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
708 bh->b_data, blocksize, 0);
709 names += stats.names;
710 space += stats.space;
711 bcount += stats.bcount;
712 brelse(bh);
713 }
714 if (bcount)
715 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
716 levels ? "" : " ", names, space/bcount,
717 (space/bcount)*100/blocksize);
718 return (struct stats) { names, space, bcount};
719 }
720 #endif /* DX_DEBUG */
721
722 /*
723 * Probe for a directory leaf block to search.
724 *
725 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
726 * error in the directory index, and the caller should fall back to
727 * searching the directory normally. The callers of dx_probe **MUST**
728 * check for this error code, and make sure it never gets reflected
729 * back to userspace.
730 */
731 static struct dx_frame *
732 dx_probe(struct ext4_filename *fname, struct inode *dir,
733 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
734 {
735 unsigned count, indirect;
736 struct dx_entry *at, *entries, *p, *q, *m;
737 struct dx_root *root;
738 struct dx_frame *frame = frame_in;
739 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
740 u32 hash;
741
742 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
743 if (IS_ERR(frame->bh))
744 return (struct dx_frame *) frame->bh;
745
746 root = (struct dx_root *) frame->bh->b_data;
747 if (root->info.hash_version != DX_HASH_TEA &&
748 root->info.hash_version != DX_HASH_HALF_MD4 &&
749 root->info.hash_version != DX_HASH_LEGACY) {
750 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
751 root->info.hash_version);
752 goto fail;
753 }
754 if (fname)
755 hinfo = &fname->hinfo;
756 hinfo->hash_version = root->info.hash_version;
757 if (hinfo->hash_version <= DX_HASH_TEA)
758 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
759 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
760 if (fname && fname_name(fname))
761 ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
762 hash = hinfo->hash;
763
764 if (root->info.unused_flags & 1) {
765 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
766 root->info.unused_flags);
767 goto fail;
768 }
769
770 indirect = root->info.indirect_levels;
771 if (indirect > 1) {
772 ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
773 root->info.indirect_levels);
774 goto fail;
775 }
776
777 entries = (struct dx_entry *)(((char *)&root->info) +
778 root->info.info_length);
779
780 if (dx_get_limit(entries) != dx_root_limit(dir,
781 root->info.info_length)) {
782 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
783 dx_get_limit(entries),
784 dx_root_limit(dir, root->info.info_length));
785 goto fail;
786 }
787
788 dxtrace(printk("Look up %x", hash));
789 while (1) {
790 count = dx_get_count(entries);
791 if (!count || count > dx_get_limit(entries)) {
792 ext4_warning_inode(dir,
793 "dx entry: count %u beyond limit %u",
794 count, dx_get_limit(entries));
795 goto fail;
796 }
797
798 p = entries + 1;
799 q = entries + count - 1;
800 while (p <= q) {
801 m = p + (q - p) / 2;
802 dxtrace(printk(KERN_CONT "."));
803 if (dx_get_hash(m) > hash)
804 q = m - 1;
805 else
806 p = m + 1;
807 }
808
809 if (0) { // linear search cross check
810 unsigned n = count - 1;
811 at = entries;
812 while (n--)
813 {
814 dxtrace(printk(KERN_CONT ","));
815 if (dx_get_hash(++at) > hash)
816 {
817 at--;
818 break;
819 }
820 }
821 assert (at == p - 1);
822 }
823
824 at = p - 1;
825 dxtrace(printk(KERN_CONT " %x->%u\n",
826 at == entries ? 0 : dx_get_hash(at),
827 dx_get_block(at)));
828 frame->entries = entries;
829 frame->at = at;
830 if (!indirect--)
831 return frame;
832 frame++;
833 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
834 if (IS_ERR(frame->bh)) {
835 ret_err = (struct dx_frame *) frame->bh;
836 frame->bh = NULL;
837 goto fail;
838 }
839 entries = ((struct dx_node *) frame->bh->b_data)->entries;
840
841 if (dx_get_limit(entries) != dx_node_limit(dir)) {
842 ext4_warning_inode(dir,
843 "dx entry: limit %u != node limit %u",
844 dx_get_limit(entries), dx_node_limit(dir));
845 goto fail;
846 }
847 }
848 fail:
849 while (frame >= frame_in) {
850 brelse(frame->bh);
851 frame--;
852 }
853
854 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
855 ext4_warning_inode(dir,
856 "Corrupt directory, running e2fsck is recommended");
857 return ret_err;
858 }
859
860 static void dx_release(struct dx_frame *frames)
861 {
862 if (frames[0].bh == NULL)
863 return;
864
865 if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
866 brelse(frames[1].bh);
867 brelse(frames[0].bh);
868 }
869
870 /*
871 * This function increments the frame pointer to search the next leaf
872 * block, and reads in the necessary intervening nodes if the search
873 * should be necessary. Whether or not the search is necessary is
874 * controlled by the hash parameter. If the hash value is even, then
875 * the search is only continued if the next block starts with that
876 * hash value. This is used if we are searching for a specific file.
877 *
878 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
879 *
880 * This function returns 1 if the caller should continue to search,
881 * or 0 if it should not. If there is an error reading one of the
882 * index blocks, it will a negative error code.
883 *
884 * If start_hash is non-null, it will be filled in with the starting
885 * hash of the next page.
886 */
887 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
888 struct dx_frame *frame,
889 struct dx_frame *frames,
890 __u32 *start_hash)
891 {
892 struct dx_frame *p;
893 struct buffer_head *bh;
894 int num_frames = 0;
895 __u32 bhash;
896
897 p = frame;
898 /*
899 * Find the next leaf page by incrementing the frame pointer.
900 * If we run out of entries in the interior node, loop around and
901 * increment pointer in the parent node. When we break out of
902 * this loop, num_frames indicates the number of interior
903 * nodes need to be read.
904 */
905 while (1) {
906 if (++(p->at) < p->entries + dx_get_count(p->entries))
907 break;
908 if (p == frames)
909 return 0;
910 num_frames++;
911 p--;
912 }
913
914 /*
915 * If the hash is 1, then continue only if the next page has a
916 * continuation hash of any value. This is used for readdir
917 * handling. Otherwise, check to see if the hash matches the
918 * desired contiuation hash. If it doesn't, return since
919 * there's no point to read in the successive index pages.
920 */
921 bhash = dx_get_hash(p->at);
922 if (start_hash)
923 *start_hash = bhash;
924 if ((hash & 1) == 0) {
925 if ((bhash & ~1) != hash)
926 return 0;
927 }
928 /*
929 * If the hash is HASH_NB_ALWAYS, we always go to the next
930 * block so no check is necessary
931 */
932 while (num_frames--) {
933 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
934 if (IS_ERR(bh))
935 return PTR_ERR(bh);
936 p++;
937 brelse(p->bh);
938 p->bh = bh;
939 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
940 }
941 return 1;
942 }
943
944
945 /*
946 * This function fills a red-black tree with information from a
947 * directory block. It returns the number directory entries loaded
948 * into the tree. If there is an error it is returned in err.
949 */
950 static int htree_dirblock_to_tree(struct file *dir_file,
951 struct inode *dir, ext4_lblk_t block,
952 struct dx_hash_info *hinfo,
953 __u32 start_hash, __u32 start_minor_hash)
954 {
955 struct buffer_head *bh;
956 struct ext4_dir_entry_2 *de, *top;
957 int err = 0, count = 0;
958 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
959
960 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
961 (unsigned long)block));
962 bh = ext4_read_dirblock(dir, block, DIRENT);
963 if (IS_ERR(bh))
964 return PTR_ERR(bh);
965
966 de = (struct ext4_dir_entry_2 *) bh->b_data;
967 top = (struct ext4_dir_entry_2 *) ((char *) de +
968 dir->i_sb->s_blocksize -
969 EXT4_DIR_REC_LEN(0));
970 #ifdef CONFIG_EXT4_FS_ENCRYPTION
971 /* Check if the directory is encrypted */
972 if (ext4_encrypted_inode(dir)) {
973 err = fscrypt_get_encryption_info(dir);
974 if (err < 0) {
975 brelse(bh);
976 return err;
977 }
978 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
979 &fname_crypto_str);
980 if (err < 0) {
981 brelse(bh);
982 return err;
983 }
984 }
985 #endif
986 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
987 if (ext4_check_dir_entry(dir, NULL, de, bh,
988 bh->b_data, bh->b_size,
989 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
990 + ((char *)de - bh->b_data))) {
991 /* silently ignore the rest of the block */
992 break;
993 }
994 ext4fs_dirhash(de->name, de->name_len, hinfo);
995 if ((hinfo->hash < start_hash) ||
996 ((hinfo->hash == start_hash) &&
997 (hinfo->minor_hash < start_minor_hash)))
998 continue;
999 if (de->inode == 0)
1000 continue;
1001 if (!ext4_encrypted_inode(dir)) {
1002 tmp_str.name = de->name;
1003 tmp_str.len = de->name_len;
1004 err = ext4_htree_store_dirent(dir_file,
1005 hinfo->hash, hinfo->minor_hash, de,
1006 &tmp_str);
1007 } else {
1008 int save_len = fname_crypto_str.len;
1009 struct fscrypt_str de_name = FSTR_INIT(de->name,
1010 de->name_len);
1011
1012 /* Directory is encrypted */
1013 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1014 hinfo->minor_hash, &de_name,
1015 &fname_crypto_str);
1016 if (err) {
1017 count = err;
1018 goto errout;
1019 }
1020 err = ext4_htree_store_dirent(dir_file,
1021 hinfo->hash, hinfo->minor_hash, de,
1022 &fname_crypto_str);
1023 fname_crypto_str.len = save_len;
1024 }
1025 if (err != 0) {
1026 count = err;
1027 goto errout;
1028 }
1029 count++;
1030 }
1031 errout:
1032 brelse(bh);
1033 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1034 fscrypt_fname_free_buffer(&fname_crypto_str);
1035 #endif
1036 return count;
1037 }
1038
1039
1040 /*
1041 * This function fills a red-black tree with information from a
1042 * directory. We start scanning the directory in hash order, starting
1043 * at start_hash and start_minor_hash.
1044 *
1045 * This function returns the number of entries inserted into the tree,
1046 * or a negative error code.
1047 */
1048 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1049 __u32 start_minor_hash, __u32 *next_hash)
1050 {
1051 struct dx_hash_info hinfo;
1052 struct ext4_dir_entry_2 *de;
1053 struct dx_frame frames[2], *frame;
1054 struct inode *dir;
1055 ext4_lblk_t block;
1056 int count = 0;
1057 int ret, err;
1058 __u32 hashval;
1059 struct fscrypt_str tmp_str;
1060
1061 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1062 start_hash, start_minor_hash));
1063 dir = file_inode(dir_file);
1064 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1065 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1066 if (hinfo.hash_version <= DX_HASH_TEA)
1067 hinfo.hash_version +=
1068 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1069 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1070 if (ext4_has_inline_data(dir)) {
1071 int has_inline_data = 1;
1072 count = htree_inlinedir_to_tree(dir_file, dir, 0,
1073 &hinfo, start_hash,
1074 start_minor_hash,
1075 &has_inline_data);
1076 if (has_inline_data) {
1077 *next_hash = ~0;
1078 return count;
1079 }
1080 }
1081 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1082 start_hash, start_minor_hash);
1083 *next_hash = ~0;
1084 return count;
1085 }
1086 hinfo.hash = start_hash;
1087 hinfo.minor_hash = 0;
1088 frame = dx_probe(NULL, dir, &hinfo, frames);
1089 if (IS_ERR(frame))
1090 return PTR_ERR(frame);
1091
1092 /* Add '.' and '..' from the htree header */
1093 if (!start_hash && !start_minor_hash) {
1094 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1095 tmp_str.name = de->name;
1096 tmp_str.len = de->name_len;
1097 err = ext4_htree_store_dirent(dir_file, 0, 0,
1098 de, &tmp_str);
1099 if (err != 0)
1100 goto errout;
1101 count++;
1102 }
1103 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1104 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1105 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1106 tmp_str.name = de->name;
1107 tmp_str.len = de->name_len;
1108 err = ext4_htree_store_dirent(dir_file, 2, 0,
1109 de, &tmp_str);
1110 if (err != 0)
1111 goto errout;
1112 count++;
1113 }
1114
1115 while (1) {
1116 if (fatal_signal_pending(current)) {
1117 err = -ERESTARTSYS;
1118 goto errout;
1119 }
1120 cond_resched();
1121 block = dx_get_block(frame->at);
1122 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1123 start_hash, start_minor_hash);
1124 if (ret < 0) {
1125 err = ret;
1126 goto errout;
1127 }
1128 count += ret;
1129 hashval = ~0;
1130 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1131 frame, frames, &hashval);
1132 *next_hash = hashval;
1133 if (ret < 0) {
1134 err = ret;
1135 goto errout;
1136 }
1137 /*
1138 * Stop if: (a) there are no more entries, or
1139 * (b) we have inserted at least one entry and the
1140 * next hash value is not a continuation
1141 */
1142 if ((ret == 0) ||
1143 (count && ((hashval & 1) == 0)))
1144 break;
1145 }
1146 dx_release(frames);
1147 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1148 "next hash: %x\n", count, *next_hash));
1149 return count;
1150 errout:
1151 dx_release(frames);
1152 return (err);
1153 }
1154
1155 static inline int search_dirblock(struct buffer_head *bh,
1156 struct inode *dir,
1157 struct ext4_filename *fname,
1158 const struct qstr *d_name,
1159 unsigned int offset,
1160 struct ext4_dir_entry_2 **res_dir)
1161 {
1162 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1163 fname, d_name, offset, res_dir);
1164 }
1165
1166 /*
1167 * Directory block splitting, compacting
1168 */
1169
1170 /*
1171 * Create map of hash values, offsets, and sizes, stored at end of block.
1172 * Returns number of entries mapped.
1173 */
1174 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1175 unsigned blocksize, struct dx_hash_info *hinfo,
1176 struct dx_map_entry *map_tail)
1177 {
1178 int count = 0;
1179 char *base = (char *) de;
1180 struct dx_hash_info h = *hinfo;
1181
1182 while ((char *) de < base + blocksize) {
1183 if (de->name_len && de->inode) {
1184 ext4fs_dirhash(de->name, de->name_len, &h);
1185 map_tail--;
1186 map_tail->hash = h.hash;
1187 map_tail->offs = ((char *) de - base)>>2;
1188 map_tail->size = le16_to_cpu(de->rec_len);
1189 count++;
1190 cond_resched();
1191 }
1192 /* XXX: do we need to check rec_len == 0 case? -Chris */
1193 de = ext4_next_entry(de, blocksize);
1194 }
1195 return count;
1196 }
1197
1198 /* Sort map by hash value */
1199 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1200 {
1201 struct dx_map_entry *p, *q, *top = map + count - 1;
1202 int more;
1203 /* Combsort until bubble sort doesn't suck */
1204 while (count > 2) {
1205 count = count*10/13;
1206 if (count - 9 < 2) /* 9, 10 -> 11 */
1207 count = 11;
1208 for (p = top, q = p - count; q >= map; p--, q--)
1209 if (p->hash < q->hash)
1210 swap(*p, *q);
1211 }
1212 /* Garden variety bubble sort */
1213 do {
1214 more = 0;
1215 q = top;
1216 while (q-- > map) {
1217 if (q[1].hash >= q[0].hash)
1218 continue;
1219 swap(*(q+1), *q);
1220 more = 1;
1221 }
1222 } while(more);
1223 }
1224
1225 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1226 {
1227 struct dx_entry *entries = frame->entries;
1228 struct dx_entry *old = frame->at, *new = old + 1;
1229 int count = dx_get_count(entries);
1230
1231 assert(count < dx_get_limit(entries));
1232 assert(old < entries + count);
1233 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1234 dx_set_hash(new, hash);
1235 dx_set_block(new, block);
1236 dx_set_count(entries, count + 1);
1237 }
1238
1239 /*
1240 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1241 *
1242 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1243 * `de != NULL' is guaranteed by caller.
1244 */
1245 static inline int ext4_match(struct ext4_filename *fname,
1246 struct ext4_dir_entry_2 *de)
1247 {
1248 const void *name = fname_name(fname);
1249 u32 len = fname_len(fname);
1250
1251 if (!de->inode)
1252 return 0;
1253
1254 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1255 if (unlikely(!name)) {
1256 if (fname->usr_fname->name[0] == '_') {
1257 int ret;
1258 if (de->name_len < 16)
1259 return 0;
1260 ret = memcmp(de->name + de->name_len - 16,
1261 fname->crypto_buf.name + 8, 16);
1262 return (ret == 0) ? 1 : 0;
1263 }
1264 name = fname->crypto_buf.name;
1265 len = fname->crypto_buf.len;
1266 }
1267 #endif
1268 if (de->name_len != len)
1269 return 0;
1270 return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1271 }
1272
1273 /*
1274 * Returns 0 if not found, -1 on failure, and 1 on success
1275 */
1276 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1277 struct inode *dir, struct ext4_filename *fname,
1278 const struct qstr *d_name,
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 int res;
1285
1286 de = (struct ext4_dir_entry_2 *)search_buf;
1287 dlimit = search_buf + buf_size;
1288 while ((char *) de < dlimit) {
1289 /* this code is executed quadratically often */
1290 /* do minimal checking `by hand' */
1291 if ((char *) de + de->name_len <= dlimit) {
1292 res = ext4_match(fname, de);
1293 if (res < 0) {
1294 res = -1;
1295 goto return_result;
1296 }
1297 if (res > 0) {
1298 /* found a match - just to be sure, do
1299 * a full check */
1300 if (ext4_check_dir_entry(dir, NULL, de, bh,
1301 bh->b_data,
1302 bh->b_size, offset)) {
1303 res = -1;
1304 goto return_result;
1305 }
1306 *res_dir = de;
1307 res = 1;
1308 goto return_result;
1309 }
1310
1311 }
1312 /* prevent looping on a bad block */
1313 de_len = ext4_rec_len_from_disk(de->rec_len,
1314 dir->i_sb->s_blocksize);
1315 if (de_len <= 0) {
1316 res = -1;
1317 goto return_result;
1318 }
1319 offset += de_len;
1320 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1321 }
1322
1323 res = 0;
1324 return_result:
1325 return res;
1326 }
1327
1328 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1329 struct ext4_dir_entry *de)
1330 {
1331 struct super_block *sb = dir->i_sb;
1332
1333 if (!is_dx(dir))
1334 return 0;
1335 if (block == 0)
1336 return 1;
1337 if (de->inode == 0 &&
1338 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1339 sb->s_blocksize)
1340 return 1;
1341 return 0;
1342 }
1343
1344 /*
1345 * ext4_find_entry()
1346 *
1347 * finds an entry in the specified directory with the wanted name. It
1348 * returns the cache buffer in which the entry was found, and the entry
1349 * itself (as a parameter - res_dir). It does NOT read the inode of the
1350 * entry - you'll have to do that yourself if you want to.
1351 *
1352 * The returned buffer_head has ->b_count elevated. The caller is expected
1353 * to brelse() it when appropriate.
1354 */
1355 static struct buffer_head * ext4_find_entry (struct inode *dir,
1356 const struct qstr *d_name,
1357 struct ext4_dir_entry_2 **res_dir,
1358 int *inlined)
1359 {
1360 struct super_block *sb;
1361 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1362 struct buffer_head *bh, *ret = NULL;
1363 ext4_lblk_t start, block, b;
1364 const u8 *name = d_name->name;
1365 int ra_max = 0; /* Number of bh's in the readahead
1366 buffer, bh_use[] */
1367 int ra_ptr = 0; /* Current index into readahead
1368 buffer */
1369 int num = 0;
1370 ext4_lblk_t nblocks;
1371 int i, namelen, retval;
1372 struct ext4_filename fname;
1373
1374 *res_dir = NULL;
1375 sb = dir->i_sb;
1376 namelen = d_name->len;
1377 if (namelen > EXT4_NAME_LEN)
1378 return NULL;
1379
1380 retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1381 if (retval == -ENOENT)
1382 return NULL;
1383 if (retval)
1384 return ERR_PTR(retval);
1385
1386 if (ext4_has_inline_data(dir)) {
1387 int has_inline_data = 1;
1388 ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1389 &has_inline_data);
1390 if (has_inline_data) {
1391 if (inlined)
1392 *inlined = 1;
1393 goto cleanup_and_exit;
1394 }
1395 }
1396
1397 if ((namelen <= 2) && (name[0] == '.') &&
1398 (name[1] == '.' || name[1] == '\0')) {
1399 /*
1400 * "." or ".." will only be in the first block
1401 * NFS may look up ".."; "." should be handled by the VFS
1402 */
1403 block = start = 0;
1404 nblocks = 1;
1405 goto restart;
1406 }
1407 if (is_dx(dir)) {
1408 ret = ext4_dx_find_entry(dir, &fname, res_dir);
1409 /*
1410 * On success, or if the error was file not found,
1411 * return. Otherwise, fall back to doing a search the
1412 * old fashioned way.
1413 */
1414 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1415 goto cleanup_and_exit;
1416 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1417 "falling back\n"));
1418 }
1419 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1420 start = EXT4_I(dir)->i_dir_start_lookup;
1421 if (start >= nblocks)
1422 start = 0;
1423 block = start;
1424 restart:
1425 do {
1426 /*
1427 * We deal with the read-ahead logic here.
1428 */
1429 if (ra_ptr >= ra_max) {
1430 /* Refill the readahead buffer */
1431 ra_ptr = 0;
1432 b = block;
1433 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1434 /*
1435 * Terminate if we reach the end of the
1436 * directory and must wrap, or if our
1437 * search has finished at this block.
1438 */
1439 if (b >= nblocks || (num && block == start)) {
1440 bh_use[ra_max] = NULL;
1441 break;
1442 }
1443 num++;
1444 bh = ext4_getblk(NULL, dir, b++, 0);
1445 if (IS_ERR(bh)) {
1446 if (ra_max == 0) {
1447 ret = bh;
1448 goto cleanup_and_exit;
1449 }
1450 break;
1451 }
1452 bh_use[ra_max] = bh;
1453 if (bh)
1454 ll_rw_block(REQ_OP_READ,
1455 REQ_META | REQ_PRIO,
1456 1, &bh);
1457 }
1458 }
1459 if ((bh = bh_use[ra_ptr++]) == NULL)
1460 goto next;
1461 wait_on_buffer(bh);
1462 if (!buffer_uptodate(bh)) {
1463 /* read error, skip block & hope for the best */
1464 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1465 (unsigned long) block);
1466 brelse(bh);
1467 goto next;
1468 }
1469 if (!buffer_verified(bh) &&
1470 !is_dx_internal_node(dir, block,
1471 (struct ext4_dir_entry *)bh->b_data) &&
1472 !ext4_dirent_csum_verify(dir,
1473 (struct ext4_dir_entry *)bh->b_data)) {
1474 EXT4_ERROR_INODE(dir, "checksumming directory "
1475 "block %lu", (unsigned long)block);
1476 brelse(bh);
1477 goto next;
1478 }
1479 set_buffer_verified(bh);
1480 i = search_dirblock(bh, dir, &fname, d_name,
1481 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1482 if (i == 1) {
1483 EXT4_I(dir)->i_dir_start_lookup = block;
1484 ret = bh;
1485 goto cleanup_and_exit;
1486 } else {
1487 brelse(bh);
1488 if (i < 0)
1489 goto cleanup_and_exit;
1490 }
1491 next:
1492 if (++block >= nblocks)
1493 block = 0;
1494 } while (block != start);
1495
1496 /*
1497 * If the directory has grown while we were searching, then
1498 * search the last part of the directory before giving up.
1499 */
1500 block = nblocks;
1501 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1502 if (block < nblocks) {
1503 start = 0;
1504 goto restart;
1505 }
1506
1507 cleanup_and_exit:
1508 /* Clean up the read-ahead blocks */
1509 for (; ra_ptr < ra_max; ra_ptr++)
1510 brelse(bh_use[ra_ptr]);
1511 ext4_fname_free_filename(&fname);
1512 return ret;
1513 }
1514
1515 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1516 struct ext4_filename *fname,
1517 struct ext4_dir_entry_2 **res_dir)
1518 {
1519 struct super_block * sb = dir->i_sb;
1520 struct dx_frame frames[2], *frame;
1521 const struct qstr *d_name = fname->usr_fname;
1522 struct buffer_head *bh;
1523 ext4_lblk_t block;
1524 int retval;
1525
1526 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1527 *res_dir = NULL;
1528 #endif
1529 frame = dx_probe(fname, dir, NULL, frames);
1530 if (IS_ERR(frame))
1531 return (struct buffer_head *) frame;
1532 do {
1533 block = dx_get_block(frame->at);
1534 bh = ext4_read_dirblock(dir, block, DIRENT);
1535 if (IS_ERR(bh))
1536 goto errout;
1537
1538 retval = search_dirblock(bh, dir, fname, d_name,
1539 block << EXT4_BLOCK_SIZE_BITS(sb),
1540 res_dir);
1541 if (retval == 1)
1542 goto success;
1543 brelse(bh);
1544 if (retval == -1) {
1545 bh = ERR_PTR(ERR_BAD_DX_DIR);
1546 goto errout;
1547 }
1548
1549 /* Check to see if we should continue to search */
1550 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1551 frames, NULL);
1552 if (retval < 0) {
1553 ext4_warning_inode(dir,
1554 "error %d reading directory index block",
1555 retval);
1556 bh = ERR_PTR(retval);
1557 goto errout;
1558 }
1559 } while (retval == 1);
1560
1561 bh = NULL;
1562 errout:
1563 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1564 success:
1565 dx_release(frames);
1566 return bh;
1567 }
1568
1569 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1570 {
1571 struct inode *inode;
1572 struct ext4_dir_entry_2 *de;
1573 struct buffer_head *bh;
1574
1575 if (ext4_encrypted_inode(dir)) {
1576 int res = fscrypt_get_encryption_info(dir);
1577
1578 /*
1579 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1580 * created while the directory was encrypted and we
1581 * have access to the key.
1582 */
1583 if (fscrypt_has_encryption_key(dir))
1584 fscrypt_set_encrypted_dentry(dentry);
1585 fscrypt_set_d_op(dentry);
1586 if (res && res != -ENOKEY)
1587 return ERR_PTR(res);
1588 }
1589
1590 if (dentry->d_name.len > EXT4_NAME_LEN)
1591 return ERR_PTR(-ENAMETOOLONG);
1592
1593 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1594 if (IS_ERR(bh))
1595 return (struct dentry *) bh;
1596 inode = NULL;
1597 if (bh) {
1598 __u32 ino = le32_to_cpu(de->inode);
1599 brelse(bh);
1600 if (!ext4_valid_inum(dir->i_sb, ino)) {
1601 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1602 return ERR_PTR(-EFSCORRUPTED);
1603 }
1604 if (unlikely(ino == dir->i_ino)) {
1605 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1606 dentry);
1607 return ERR_PTR(-EFSCORRUPTED);
1608 }
1609 inode = ext4_iget_normal(dir->i_sb, ino);
1610 if (inode == ERR_PTR(-ESTALE)) {
1611 EXT4_ERROR_INODE(dir,
1612 "deleted inode referenced: %u",
1613 ino);
1614 return ERR_PTR(-EFSCORRUPTED);
1615 }
1616 if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1617 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1618 !fscrypt_has_permitted_context(dir, inode)) {
1619 ext4_warning(inode->i_sb,
1620 "Inconsistent encryption contexts: %lu/%lu",
1621 dir->i_ino, inode->i_ino);
1622 iput(inode);
1623 return ERR_PTR(-EPERM);
1624 }
1625 }
1626 return d_splice_alias(inode, dentry);
1627 }
1628
1629
1630 struct dentry *ext4_get_parent(struct dentry *child)
1631 {
1632 __u32 ino;
1633 static const struct qstr dotdot = QSTR_INIT("..", 2);
1634 struct ext4_dir_entry_2 * de;
1635 struct buffer_head *bh;
1636
1637 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1638 if (IS_ERR(bh))
1639 return (struct dentry *) bh;
1640 if (!bh)
1641 return ERR_PTR(-ENOENT);
1642 ino = le32_to_cpu(de->inode);
1643 brelse(bh);
1644
1645 if (!ext4_valid_inum(child->d_sb, ino)) {
1646 EXT4_ERROR_INODE(d_inode(child),
1647 "bad parent inode number: %u", ino);
1648 return ERR_PTR(-EFSCORRUPTED);
1649 }
1650
1651 return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1652 }
1653
1654 /*
1655 * Move count entries from end of map between two memory locations.
1656 * Returns pointer to last entry moved.
1657 */
1658 static struct ext4_dir_entry_2 *
1659 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1660 unsigned blocksize)
1661 {
1662 unsigned rec_len = 0;
1663
1664 while (count--) {
1665 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1666 (from + (map->offs<<2));
1667 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1668 memcpy (to, de, rec_len);
1669 ((struct ext4_dir_entry_2 *) to)->rec_len =
1670 ext4_rec_len_to_disk(rec_len, blocksize);
1671 de->inode = 0;
1672 map++;
1673 to += rec_len;
1674 }
1675 return (struct ext4_dir_entry_2 *) (to - rec_len);
1676 }
1677
1678 /*
1679 * Compact each dir entry in the range to the minimal rec_len.
1680 * Returns pointer to last entry in range.
1681 */
1682 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1683 {
1684 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1685 unsigned rec_len = 0;
1686
1687 prev = to = de;
1688 while ((char*)de < base + blocksize) {
1689 next = ext4_next_entry(de, blocksize);
1690 if (de->inode && de->name_len) {
1691 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1692 if (de > to)
1693 memmove(to, de, rec_len);
1694 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1695 prev = to;
1696 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1697 }
1698 de = next;
1699 }
1700 return prev;
1701 }
1702
1703 /*
1704 * Split a full leaf block to make room for a new dir entry.
1705 * Allocate a new block, and move entries so that they are approx. equally full.
1706 * Returns pointer to de in block into which the new entry will be inserted.
1707 */
1708 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1709 struct buffer_head **bh,struct dx_frame *frame,
1710 struct dx_hash_info *hinfo)
1711 {
1712 unsigned blocksize = dir->i_sb->s_blocksize;
1713 unsigned count, continued;
1714 struct buffer_head *bh2;
1715 ext4_lblk_t newblock;
1716 u32 hash2;
1717 struct dx_map_entry *map;
1718 char *data1 = (*bh)->b_data, *data2;
1719 unsigned split, move, size;
1720 struct ext4_dir_entry_2 *de = NULL, *de2;
1721 struct ext4_dir_entry_tail *t;
1722 int csum_size = 0;
1723 int err = 0, i;
1724
1725 if (ext4_has_metadata_csum(dir->i_sb))
1726 csum_size = sizeof(struct ext4_dir_entry_tail);
1727
1728 bh2 = ext4_append(handle, dir, &newblock);
1729 if (IS_ERR(bh2)) {
1730 brelse(*bh);
1731 *bh = NULL;
1732 return (struct ext4_dir_entry_2 *) bh2;
1733 }
1734
1735 BUFFER_TRACE(*bh, "get_write_access");
1736 err = ext4_journal_get_write_access(handle, *bh);
1737 if (err)
1738 goto journal_error;
1739
1740 BUFFER_TRACE(frame->bh, "get_write_access");
1741 err = ext4_journal_get_write_access(handle, frame->bh);
1742 if (err)
1743 goto journal_error;
1744
1745 data2 = bh2->b_data;
1746
1747 /* create map in the end of data2 block */
1748 map = (struct dx_map_entry *) (data2 + blocksize);
1749 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1750 blocksize, hinfo, map);
1751 map -= count;
1752 dx_sort_map(map, count);
1753 /* Split the existing block in the middle, size-wise */
1754 size = 0;
1755 move = 0;
1756 for (i = count-1; i >= 0; i--) {
1757 /* is more than half of this entry in 2nd half of the block? */
1758 if (size + map[i].size/2 > blocksize/2)
1759 break;
1760 size += map[i].size;
1761 move++;
1762 }
1763 /* map index at which we will split */
1764 split = count - move;
1765 hash2 = map[split].hash;
1766 continued = hash2 == map[split - 1].hash;
1767 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1768 (unsigned long)dx_get_block(frame->at),
1769 hash2, split, count-split));
1770
1771 /* Fancy dance to stay within two buffers */
1772 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1773 blocksize);
1774 de = dx_pack_dirents(data1, blocksize);
1775 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1776 (char *) de,
1777 blocksize);
1778 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1779 (char *) de2,
1780 blocksize);
1781 if (csum_size) {
1782 t = EXT4_DIRENT_TAIL(data2, blocksize);
1783 initialize_dirent_tail(t, blocksize);
1784
1785 t = EXT4_DIRENT_TAIL(data1, blocksize);
1786 initialize_dirent_tail(t, blocksize);
1787 }
1788
1789 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1790 blocksize, 1));
1791 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1792 blocksize, 1));
1793
1794 /* Which block gets the new entry? */
1795 if (hinfo->hash >= hash2) {
1796 swap(*bh, bh2);
1797 de = de2;
1798 }
1799 dx_insert_block(frame, hash2 + continued, newblock);
1800 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1801 if (err)
1802 goto journal_error;
1803 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1804 if (err)
1805 goto journal_error;
1806 brelse(bh2);
1807 dxtrace(dx_show_index("frame", frame->entries));
1808 return de;
1809
1810 journal_error:
1811 brelse(*bh);
1812 brelse(bh2);
1813 *bh = NULL;
1814 ext4_std_error(dir->i_sb, err);
1815 return ERR_PTR(err);
1816 }
1817
1818 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1819 struct buffer_head *bh,
1820 void *buf, int buf_size,
1821 struct ext4_filename *fname,
1822 struct ext4_dir_entry_2 **dest_de)
1823 {
1824 struct ext4_dir_entry_2 *de;
1825 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1826 int nlen, rlen;
1827 unsigned int offset = 0;
1828 char *top;
1829 int res;
1830
1831 de = (struct ext4_dir_entry_2 *)buf;
1832 top = buf + buf_size - reclen;
1833 while ((char *) de <= top) {
1834 if (ext4_check_dir_entry(dir, NULL, de, bh,
1835 buf, buf_size, offset)) {
1836 res = -EFSCORRUPTED;
1837 goto return_result;
1838 }
1839 /* Provide crypto context and crypto buffer to ext4 match */
1840 res = ext4_match(fname, de);
1841 if (res < 0)
1842 goto return_result;
1843 if (res > 0) {
1844 res = -EEXIST;
1845 goto return_result;
1846 }
1847 nlen = EXT4_DIR_REC_LEN(de->name_len);
1848 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1849 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1850 break;
1851 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1852 offset += rlen;
1853 }
1854
1855 if ((char *) de > top)
1856 res = -ENOSPC;
1857 else {
1858 *dest_de = de;
1859 res = 0;
1860 }
1861 return_result:
1862 return res;
1863 }
1864
1865 int ext4_insert_dentry(struct inode *dir,
1866 struct inode *inode,
1867 struct ext4_dir_entry_2 *de,
1868 int buf_size,
1869 struct ext4_filename *fname)
1870 {
1871
1872 int nlen, rlen;
1873
1874 nlen = EXT4_DIR_REC_LEN(de->name_len);
1875 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1876 if (de->inode) {
1877 struct ext4_dir_entry_2 *de1 =
1878 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1879 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1880 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1881 de = de1;
1882 }
1883 de->file_type = EXT4_FT_UNKNOWN;
1884 de->inode = cpu_to_le32(inode->i_ino);
1885 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1886 de->name_len = fname_len(fname);
1887 memcpy(de->name, fname_name(fname), fname_len(fname));
1888 return 0;
1889 }
1890
1891 /*
1892 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1893 * it points to a directory entry which is guaranteed to be large
1894 * enough for new directory entry. If de is NULL, then
1895 * add_dirent_to_buf will attempt search the directory block for
1896 * space. It will return -ENOSPC if no space is available, and -EIO
1897 * and -EEXIST if directory entry already exists.
1898 */
1899 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1900 struct inode *dir,
1901 struct inode *inode, struct ext4_dir_entry_2 *de,
1902 struct buffer_head *bh)
1903 {
1904 unsigned int blocksize = dir->i_sb->s_blocksize;
1905 int csum_size = 0;
1906 int err;
1907
1908 if (ext4_has_metadata_csum(inode->i_sb))
1909 csum_size = sizeof(struct ext4_dir_entry_tail);
1910
1911 if (!de) {
1912 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1913 blocksize - csum_size, fname, &de);
1914 if (err)
1915 return err;
1916 }
1917 BUFFER_TRACE(bh, "get_write_access");
1918 err = ext4_journal_get_write_access(handle, bh);
1919 if (err) {
1920 ext4_std_error(dir->i_sb, err);
1921 return err;
1922 }
1923
1924 /* By now the buffer is marked for journaling. Due to crypto operations,
1925 * the following function call may fail */
1926 err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1927 if (err < 0)
1928 return err;
1929
1930 /*
1931 * XXX shouldn't update any times until successful
1932 * completion of syscall, but too many callers depend
1933 * on this.
1934 *
1935 * XXX similarly, too many callers depend on
1936 * ext4_new_inode() setting the times, but error
1937 * recovery deletes the inode, so the worst that can
1938 * happen is that the times are slightly out of date
1939 * and/or different from the directory change time.
1940 */
1941 dir->i_mtime = dir->i_ctime = current_time(dir);
1942 ext4_update_dx_flag(dir);
1943 dir->i_version++;
1944 ext4_mark_inode_dirty(handle, dir);
1945 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1946 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1947 if (err)
1948 ext4_std_error(dir->i_sb, err);
1949 return 0;
1950 }
1951
1952 /*
1953 * This converts a one block unindexed directory to a 3 block indexed
1954 * directory, and adds the dentry to the indexed directory.
1955 */
1956 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1957 struct inode *dir,
1958 struct inode *inode, struct buffer_head *bh)
1959 {
1960 struct buffer_head *bh2;
1961 struct dx_root *root;
1962 struct dx_frame frames[2], *frame;
1963 struct dx_entry *entries;
1964 struct ext4_dir_entry_2 *de, *de2;
1965 struct ext4_dir_entry_tail *t;
1966 char *data1, *top;
1967 unsigned len;
1968 int retval;
1969 unsigned blocksize;
1970 ext4_lblk_t block;
1971 struct fake_dirent *fde;
1972 int csum_size = 0;
1973
1974 if (ext4_has_metadata_csum(inode->i_sb))
1975 csum_size = sizeof(struct ext4_dir_entry_tail);
1976
1977 blocksize = dir->i_sb->s_blocksize;
1978 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1979 BUFFER_TRACE(bh, "get_write_access");
1980 retval = ext4_journal_get_write_access(handle, bh);
1981 if (retval) {
1982 ext4_std_error(dir->i_sb, retval);
1983 brelse(bh);
1984 return retval;
1985 }
1986 root = (struct dx_root *) bh->b_data;
1987
1988 /* The 0th block becomes the root, move the dirents out */
1989 fde = &root->dotdot;
1990 de = (struct ext4_dir_entry_2 *)((char *)fde +
1991 ext4_rec_len_from_disk(fde->rec_len, blocksize));
1992 if ((char *) de >= (((char *) root) + blocksize)) {
1993 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1994 brelse(bh);
1995 return -EFSCORRUPTED;
1996 }
1997 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1998
1999 /* Allocate new block for the 0th block's dirents */
2000 bh2 = ext4_append(handle, dir, &block);
2001 if (IS_ERR(bh2)) {
2002 brelse(bh);
2003 return PTR_ERR(bh2);
2004 }
2005 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2006 data1 = bh2->b_data;
2007
2008 memcpy (data1, de, len);
2009 de = (struct ext4_dir_entry_2 *) data1;
2010 top = data1 + len;
2011 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2012 de = de2;
2013 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2014 (char *) de,
2015 blocksize);
2016
2017 if (csum_size) {
2018 t = EXT4_DIRENT_TAIL(data1, blocksize);
2019 initialize_dirent_tail(t, blocksize);
2020 }
2021
2022 /* Initialize the root; the dot dirents already exist */
2023 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2024 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2025 blocksize);
2026 memset (&root->info, 0, sizeof(root->info));
2027 root->info.info_length = sizeof(root->info);
2028 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2029 entries = root->entries;
2030 dx_set_block(entries, 1);
2031 dx_set_count(entries, 1);
2032 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2033
2034 /* Initialize as for dx_probe */
2035 fname->hinfo.hash_version = root->info.hash_version;
2036 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2037 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2038 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2039 ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2040
2041 memset(frames, 0, sizeof(frames));
2042 frame = frames;
2043 frame->entries = entries;
2044 frame->at = entries;
2045 frame->bh = bh;
2046
2047 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2048 if (retval)
2049 goto out_frames;
2050 retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2051 if (retval)
2052 goto out_frames;
2053
2054 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2055 if (IS_ERR(de)) {
2056 retval = PTR_ERR(de);
2057 goto out_frames;
2058 }
2059
2060 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2061 out_frames:
2062 /*
2063 * Even if the block split failed, we have to properly write
2064 * out all the changes we did so far. Otherwise we can end up
2065 * with corrupted filesystem.
2066 */
2067 if (retval)
2068 ext4_mark_inode_dirty(handle, dir);
2069 dx_release(frames);
2070 brelse(bh2);
2071 return retval;
2072 }
2073
2074 /*
2075 * ext4_add_entry()
2076 *
2077 * adds a file entry to the specified directory, using the same
2078 * semantics as ext4_find_entry(). It returns NULL if it failed.
2079 *
2080 * NOTE!! The inode part of 'de' is left at 0 - which means you
2081 * may not sleep between calling this and putting something into
2082 * the entry, as someone else might have used it while you slept.
2083 */
2084 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2085 struct inode *inode)
2086 {
2087 struct inode *dir = d_inode(dentry->d_parent);
2088 struct buffer_head *bh = NULL;
2089 struct ext4_dir_entry_2 *de;
2090 struct ext4_dir_entry_tail *t;
2091 struct super_block *sb;
2092 struct ext4_filename fname;
2093 int retval;
2094 int dx_fallback=0;
2095 unsigned blocksize;
2096 ext4_lblk_t block, blocks;
2097 int csum_size = 0;
2098
2099 if (ext4_has_metadata_csum(inode->i_sb))
2100 csum_size = sizeof(struct ext4_dir_entry_tail);
2101
2102 sb = dir->i_sb;
2103 blocksize = sb->s_blocksize;
2104 if (!dentry->d_name.len)
2105 return -EINVAL;
2106
2107 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2108 if (retval)
2109 return retval;
2110
2111 if (ext4_has_inline_data(dir)) {
2112 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2113 if (retval < 0)
2114 goto out;
2115 if (retval == 1) {
2116 retval = 0;
2117 goto out;
2118 }
2119 }
2120
2121 if (is_dx(dir)) {
2122 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2123 if (!retval || (retval != ERR_BAD_DX_DIR))
2124 goto out;
2125 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2126 dx_fallback++;
2127 ext4_mark_inode_dirty(handle, dir);
2128 }
2129 blocks = dir->i_size >> sb->s_blocksize_bits;
2130 for (block = 0; block < blocks; block++) {
2131 bh = ext4_read_dirblock(dir, block, DIRENT);
2132 if (IS_ERR(bh)) {
2133 retval = PTR_ERR(bh);
2134 bh = NULL;
2135 goto out;
2136 }
2137 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2138 NULL, bh);
2139 if (retval != -ENOSPC)
2140 goto out;
2141
2142 if (blocks == 1 && !dx_fallback &&
2143 ext4_has_feature_dir_index(sb)) {
2144 retval = make_indexed_dir(handle, &fname, dir,
2145 inode, bh);
2146 bh = NULL; /* make_indexed_dir releases bh */
2147 goto out;
2148 }
2149 brelse(bh);
2150 }
2151 bh = ext4_append(handle, dir, &block);
2152 if (IS_ERR(bh)) {
2153 retval = PTR_ERR(bh);
2154 bh = NULL;
2155 goto out;
2156 }
2157 de = (struct ext4_dir_entry_2 *) bh->b_data;
2158 de->inode = 0;
2159 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2160
2161 if (csum_size) {
2162 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2163 initialize_dirent_tail(t, blocksize);
2164 }
2165
2166 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2167 out:
2168 ext4_fname_free_filename(&fname);
2169 brelse(bh);
2170 if (retval == 0)
2171 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2172 return retval;
2173 }
2174
2175 /*
2176 * Returns 0 for success, or a negative error value
2177 */
2178 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2179 struct inode *dir, struct inode *inode)
2180 {
2181 struct dx_frame frames[2], *frame;
2182 struct dx_entry *entries, *at;
2183 struct buffer_head *bh;
2184 struct super_block *sb = dir->i_sb;
2185 struct ext4_dir_entry_2 *de;
2186 int err;
2187
2188 frame = dx_probe(fname, dir, NULL, frames);
2189 if (IS_ERR(frame))
2190 return PTR_ERR(frame);
2191 entries = frame->entries;
2192 at = frame->at;
2193 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2194 if (IS_ERR(bh)) {
2195 err = PTR_ERR(bh);
2196 bh = NULL;
2197 goto cleanup;
2198 }
2199
2200 BUFFER_TRACE(bh, "get_write_access");
2201 err = ext4_journal_get_write_access(handle, bh);
2202 if (err)
2203 goto journal_error;
2204
2205 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2206 if (err != -ENOSPC)
2207 goto cleanup;
2208
2209 /* Block full, should compress but for now just split */
2210 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2211 dx_get_count(entries), dx_get_limit(entries)));
2212 /* Need to split index? */
2213 if (dx_get_count(entries) == dx_get_limit(entries)) {
2214 ext4_lblk_t newblock;
2215 unsigned icount = dx_get_count(entries);
2216 int levels = frame - frames;
2217 struct dx_entry *entries2;
2218 struct dx_node *node2;
2219 struct buffer_head *bh2;
2220
2221 if (levels && (dx_get_count(frames->entries) ==
2222 dx_get_limit(frames->entries))) {
2223 ext4_warning_inode(dir, "Directory index full!");
2224 err = -ENOSPC;
2225 goto cleanup;
2226 }
2227 bh2 = ext4_append(handle, dir, &newblock);
2228 if (IS_ERR(bh2)) {
2229 err = PTR_ERR(bh2);
2230 goto cleanup;
2231 }
2232 node2 = (struct dx_node *)(bh2->b_data);
2233 entries2 = node2->entries;
2234 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2235 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2236 sb->s_blocksize);
2237 BUFFER_TRACE(frame->bh, "get_write_access");
2238 err = ext4_journal_get_write_access(handle, frame->bh);
2239 if (err)
2240 goto journal_error;
2241 if (levels) {
2242 unsigned icount1 = icount/2, icount2 = icount - icount1;
2243 unsigned hash2 = dx_get_hash(entries + icount1);
2244 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2245 icount1, icount2));
2246
2247 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2248 err = ext4_journal_get_write_access(handle,
2249 frames[0].bh);
2250 if (err)
2251 goto journal_error;
2252
2253 memcpy((char *) entries2, (char *) (entries + icount1),
2254 icount2 * sizeof(struct dx_entry));
2255 dx_set_count(entries, icount1);
2256 dx_set_count(entries2, icount2);
2257 dx_set_limit(entries2, dx_node_limit(dir));
2258
2259 /* Which index block gets the new entry? */
2260 if (at - entries >= icount1) {
2261 frame->at = at = at - entries - icount1 + entries2;
2262 frame->entries = entries = entries2;
2263 swap(frame->bh, bh2);
2264 }
2265 dx_insert_block(frames + 0, hash2, newblock);
2266 dxtrace(dx_show_index("node", frames[1].entries));
2267 dxtrace(dx_show_index("node",
2268 ((struct dx_node *) bh2->b_data)->entries));
2269 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2270 if (err)
2271 goto journal_error;
2272 brelse (bh2);
2273 } else {
2274 dxtrace(printk(KERN_DEBUG
2275 "Creating second level index...\n"));
2276 memcpy((char *) entries2, (char *) entries,
2277 icount * sizeof(struct dx_entry));
2278 dx_set_limit(entries2, dx_node_limit(dir));
2279
2280 /* Set up root */
2281 dx_set_count(entries, 1);
2282 dx_set_block(entries + 0, newblock);
2283 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2284
2285 /* Add new access path frame */
2286 frame = frames + 1;
2287 frame->at = at = at - entries + entries2;
2288 frame->entries = entries = entries2;
2289 frame->bh = bh2;
2290 err = ext4_journal_get_write_access(handle,
2291 frame->bh);
2292 if (err)
2293 goto journal_error;
2294 }
2295 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2296 if (err) {
2297 ext4_std_error(inode->i_sb, err);
2298 goto cleanup;
2299 }
2300 }
2301 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2302 if (IS_ERR(de)) {
2303 err = PTR_ERR(de);
2304 goto cleanup;
2305 }
2306 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2307 goto cleanup;
2308
2309 journal_error:
2310 ext4_std_error(dir->i_sb, err);
2311 cleanup:
2312 brelse(bh);
2313 dx_release(frames);
2314 return err;
2315 }
2316
2317 /*
2318 * ext4_generic_delete_entry deletes a directory entry by merging it
2319 * with the previous entry
2320 */
2321 int ext4_generic_delete_entry(handle_t *handle,
2322 struct inode *dir,
2323 struct ext4_dir_entry_2 *de_del,
2324 struct buffer_head *bh,
2325 void *entry_buf,
2326 int buf_size,
2327 int csum_size)
2328 {
2329 struct ext4_dir_entry_2 *de, *pde;
2330 unsigned int blocksize = dir->i_sb->s_blocksize;
2331 int i;
2332
2333 i = 0;
2334 pde = NULL;
2335 de = (struct ext4_dir_entry_2 *)entry_buf;
2336 while (i < buf_size - csum_size) {
2337 if (ext4_check_dir_entry(dir, NULL, de, bh,
2338 bh->b_data, bh->b_size, i))
2339 return -EFSCORRUPTED;
2340 if (de == de_del) {
2341 if (pde)
2342 pde->rec_len = ext4_rec_len_to_disk(
2343 ext4_rec_len_from_disk(pde->rec_len,
2344 blocksize) +
2345 ext4_rec_len_from_disk(de->rec_len,
2346 blocksize),
2347 blocksize);
2348 else
2349 de->inode = 0;
2350 dir->i_version++;
2351 return 0;
2352 }
2353 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2354 pde = de;
2355 de = ext4_next_entry(de, blocksize);
2356 }
2357 return -ENOENT;
2358 }
2359
2360 static int ext4_delete_entry(handle_t *handle,
2361 struct inode *dir,
2362 struct ext4_dir_entry_2 *de_del,
2363 struct buffer_head *bh)
2364 {
2365 int err, csum_size = 0;
2366
2367 if (ext4_has_inline_data(dir)) {
2368 int has_inline_data = 1;
2369 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2370 &has_inline_data);
2371 if (has_inline_data)
2372 return err;
2373 }
2374
2375 if (ext4_has_metadata_csum(dir->i_sb))
2376 csum_size = sizeof(struct ext4_dir_entry_tail);
2377
2378 BUFFER_TRACE(bh, "get_write_access");
2379 err = ext4_journal_get_write_access(handle, bh);
2380 if (unlikely(err))
2381 goto out;
2382
2383 err = ext4_generic_delete_entry(handle, dir, de_del,
2384 bh, bh->b_data,
2385 dir->i_sb->s_blocksize, csum_size);
2386 if (err)
2387 goto out;
2388
2389 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2390 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2391 if (unlikely(err))
2392 goto out;
2393
2394 return 0;
2395 out:
2396 if (err != -ENOENT)
2397 ext4_std_error(dir->i_sb, err);
2398 return err;
2399 }
2400
2401 /*
2402 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2403 * since this indicates that nlinks count was previously 1.
2404 */
2405 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2406 {
2407 inc_nlink(inode);
2408 if (is_dx(inode) && inode->i_nlink > 1) {
2409 /* limit is 16-bit i_links_count */
2410 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2411 set_nlink(inode, 1);
2412 ext4_set_feature_dir_nlink(inode->i_sb);
2413 }
2414 }
2415 }
2416
2417 /*
2418 * If a directory had nlink == 1, then we should let it be 1. This indicates
2419 * directory has >EXT4_LINK_MAX subdirs.
2420 */
2421 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2422 {
2423 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2424 drop_nlink(inode);
2425 }
2426
2427
2428 static int ext4_add_nondir(handle_t *handle,
2429 struct dentry *dentry, struct inode *inode)
2430 {
2431 int err = ext4_add_entry(handle, dentry, inode);
2432 if (!err) {
2433 ext4_mark_inode_dirty(handle, inode);
2434 unlock_new_inode(inode);
2435 d_instantiate(dentry, inode);
2436 return 0;
2437 }
2438 drop_nlink(inode);
2439 unlock_new_inode(inode);
2440 iput(inode);
2441 return err;
2442 }
2443
2444 /*
2445 * By the time this is called, we already have created
2446 * the directory cache entry for the new file, but it
2447 * is so far negative - it has no inode.
2448 *
2449 * If the create succeeds, we fill in the inode information
2450 * with d_instantiate().
2451 */
2452 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2453 bool excl)
2454 {
2455 handle_t *handle;
2456 struct inode *inode;
2457 int err, credits, retries = 0;
2458
2459 err = dquot_initialize(dir);
2460 if (err)
2461 return err;
2462
2463 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2464 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2465 retry:
2466 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2467 NULL, EXT4_HT_DIR, credits);
2468 handle = ext4_journal_current_handle();
2469 err = PTR_ERR(inode);
2470 if (!IS_ERR(inode)) {
2471 inode->i_op = &ext4_file_inode_operations;
2472 inode->i_fop = &ext4_file_operations;
2473 ext4_set_aops(inode);
2474 err = ext4_add_nondir(handle, dentry, inode);
2475 if (!err && IS_DIRSYNC(dir))
2476 ext4_handle_sync(handle);
2477 }
2478 if (handle)
2479 ext4_journal_stop(handle);
2480 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2481 goto retry;
2482 return err;
2483 }
2484
2485 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2486 umode_t mode, dev_t rdev)
2487 {
2488 handle_t *handle;
2489 struct inode *inode;
2490 int err, credits, retries = 0;
2491
2492 err = dquot_initialize(dir);
2493 if (err)
2494 return err;
2495
2496 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2497 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2498 retry:
2499 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2500 NULL, EXT4_HT_DIR, credits);
2501 handle = ext4_journal_current_handle();
2502 err = PTR_ERR(inode);
2503 if (!IS_ERR(inode)) {
2504 init_special_inode(inode, inode->i_mode, rdev);
2505 inode->i_op = &ext4_special_inode_operations;
2506 err = ext4_add_nondir(handle, dentry, inode);
2507 if (!err && IS_DIRSYNC(dir))
2508 ext4_handle_sync(handle);
2509 }
2510 if (handle)
2511 ext4_journal_stop(handle);
2512 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2513 goto retry;
2514 return err;
2515 }
2516
2517 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2518 {
2519 handle_t *handle;
2520 struct inode *inode;
2521 int err, retries = 0;
2522
2523 err = dquot_initialize(dir);
2524 if (err)
2525 return err;
2526
2527 retry:
2528 inode = ext4_new_inode_start_handle(dir, mode,
2529 NULL, 0, NULL,
2530 EXT4_HT_DIR,
2531 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2532 4 + EXT4_XATTR_TRANS_BLOCKS);
2533 handle = ext4_journal_current_handle();
2534 err = PTR_ERR(inode);
2535 if (!IS_ERR(inode)) {
2536 inode->i_op = &ext4_file_inode_operations;
2537 inode->i_fop = &ext4_file_operations;
2538 ext4_set_aops(inode);
2539 d_tmpfile(dentry, inode);
2540 err = ext4_orphan_add(handle, inode);
2541 if (err)
2542 goto err_unlock_inode;
2543 mark_inode_dirty(inode);
2544 unlock_new_inode(inode);
2545 }
2546 if (handle)
2547 ext4_journal_stop(handle);
2548 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2549 goto retry;
2550 return err;
2551 err_unlock_inode:
2552 ext4_journal_stop(handle);
2553 unlock_new_inode(inode);
2554 return err;
2555 }
2556
2557 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2558 struct ext4_dir_entry_2 *de,
2559 int blocksize, int csum_size,
2560 unsigned int parent_ino, int dotdot_real_len)
2561 {
2562 de->inode = cpu_to_le32(inode->i_ino);
2563 de->name_len = 1;
2564 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2565 blocksize);
2566 strcpy(de->name, ".");
2567 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2568
2569 de = ext4_next_entry(de, blocksize);
2570 de->inode = cpu_to_le32(parent_ino);
2571 de->name_len = 2;
2572 if (!dotdot_real_len)
2573 de->rec_len = ext4_rec_len_to_disk(blocksize -
2574 (csum_size + EXT4_DIR_REC_LEN(1)),
2575 blocksize);
2576 else
2577 de->rec_len = ext4_rec_len_to_disk(
2578 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2579 strcpy(de->name, "..");
2580 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2581
2582 return ext4_next_entry(de, blocksize);
2583 }
2584
2585 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2586 struct inode *inode)
2587 {
2588 struct buffer_head *dir_block = NULL;
2589 struct ext4_dir_entry_2 *de;
2590 struct ext4_dir_entry_tail *t;
2591 ext4_lblk_t block = 0;
2592 unsigned int blocksize = dir->i_sb->s_blocksize;
2593 int csum_size = 0;
2594 int err;
2595
2596 if (ext4_has_metadata_csum(dir->i_sb))
2597 csum_size = sizeof(struct ext4_dir_entry_tail);
2598
2599 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2600 err = ext4_try_create_inline_dir(handle, dir, inode);
2601 if (err < 0 && err != -ENOSPC)
2602 goto out;
2603 if (!err)
2604 goto out;
2605 }
2606
2607 inode->i_size = 0;
2608 dir_block = ext4_append(handle, inode, &block);
2609 if (IS_ERR(dir_block))
2610 return PTR_ERR(dir_block);
2611 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2612 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2613 set_nlink(inode, 2);
2614 if (csum_size) {
2615 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2616 initialize_dirent_tail(t, blocksize);
2617 }
2618
2619 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2620 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2621 if (err)
2622 goto out;
2623 set_buffer_verified(dir_block);
2624 out:
2625 brelse(dir_block);
2626 return err;
2627 }
2628
2629 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2630 {
2631 handle_t *handle;
2632 struct inode *inode;
2633 int err, credits, retries = 0;
2634
2635 if (EXT4_DIR_LINK_MAX(dir))
2636 return -EMLINK;
2637
2638 err = dquot_initialize(dir);
2639 if (err)
2640 return err;
2641
2642 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2643 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2644 retry:
2645 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2646 &dentry->d_name,
2647 0, NULL, EXT4_HT_DIR, credits);
2648 handle = ext4_journal_current_handle();
2649 err = PTR_ERR(inode);
2650 if (IS_ERR(inode))
2651 goto out_stop;
2652
2653 inode->i_op = &ext4_dir_inode_operations;
2654 inode->i_fop = &ext4_dir_operations;
2655 err = ext4_init_new_dir(handle, dir, inode);
2656 if (err)
2657 goto out_clear_inode;
2658 err = ext4_mark_inode_dirty(handle, inode);
2659 if (!err)
2660 err = ext4_add_entry(handle, dentry, inode);
2661 if (err) {
2662 out_clear_inode:
2663 clear_nlink(inode);
2664 unlock_new_inode(inode);
2665 ext4_mark_inode_dirty(handle, inode);
2666 iput(inode);
2667 goto out_stop;
2668 }
2669 ext4_inc_count(handle, dir);
2670 ext4_update_dx_flag(dir);
2671 err = ext4_mark_inode_dirty(handle, dir);
2672 if (err)
2673 goto out_clear_inode;
2674 unlock_new_inode(inode);
2675 d_instantiate(dentry, inode);
2676 if (IS_DIRSYNC(dir))
2677 ext4_handle_sync(handle);
2678
2679 out_stop:
2680 if (handle)
2681 ext4_journal_stop(handle);
2682 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2683 goto retry;
2684 return err;
2685 }
2686
2687 /*
2688 * routine to check that the specified directory is empty (for rmdir)
2689 */
2690 bool ext4_empty_dir(struct inode *inode)
2691 {
2692 unsigned int offset;
2693 struct buffer_head *bh;
2694 struct ext4_dir_entry_2 *de, *de1;
2695 struct super_block *sb;
2696
2697 if (ext4_has_inline_data(inode)) {
2698 int has_inline_data = 1;
2699 int ret;
2700
2701 ret = empty_inline_dir(inode, &has_inline_data);
2702 if (has_inline_data)
2703 return ret;
2704 }
2705
2706 sb = inode->i_sb;
2707 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2708 EXT4_ERROR_INODE(inode, "invalid size");
2709 return true;
2710 }
2711 bh = ext4_read_dirblock(inode, 0, EITHER);
2712 if (IS_ERR(bh))
2713 return true;
2714
2715 de = (struct ext4_dir_entry_2 *) bh->b_data;
2716 de1 = ext4_next_entry(de, sb->s_blocksize);
2717 if (le32_to_cpu(de->inode) != inode->i_ino ||
2718 le32_to_cpu(de1->inode) == 0 ||
2719 strcmp(".", de->name) || strcmp("..", de1->name)) {
2720 ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2721 brelse(bh);
2722 return true;
2723 }
2724 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2725 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2726 de = ext4_next_entry(de1, sb->s_blocksize);
2727 while (offset < inode->i_size) {
2728 if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2729 unsigned int lblock;
2730 brelse(bh);
2731 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2732 bh = ext4_read_dirblock(inode, lblock, EITHER);
2733 if (IS_ERR(bh))
2734 return true;
2735 de = (struct ext4_dir_entry_2 *) bh->b_data;
2736 }
2737 if (ext4_check_dir_entry(inode, NULL, de, bh,
2738 bh->b_data, bh->b_size, offset)) {
2739 de = (struct ext4_dir_entry_2 *)(bh->b_data +
2740 sb->s_blocksize);
2741 offset = (offset | (sb->s_blocksize - 1)) + 1;
2742 continue;
2743 }
2744 if (le32_to_cpu(de->inode)) {
2745 brelse(bh);
2746 return false;
2747 }
2748 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2749 de = ext4_next_entry(de, sb->s_blocksize);
2750 }
2751 brelse(bh);
2752 return true;
2753 }
2754
2755 /*
2756 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2757 * such inodes, starting at the superblock, in case we crash before the
2758 * file is closed/deleted, or in case the inode truncate spans multiple
2759 * transactions and the last transaction is not recovered after a crash.
2760 *
2761 * At filesystem recovery time, we walk this list deleting unlinked
2762 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2763 *
2764 * Orphan list manipulation functions must be called under i_mutex unless
2765 * we are just creating the inode or deleting it.
2766 */
2767 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2768 {
2769 struct super_block *sb = inode->i_sb;
2770 struct ext4_sb_info *sbi = EXT4_SB(sb);
2771 struct ext4_iloc iloc;
2772 int err = 0, rc;
2773 bool dirty = false;
2774
2775 if (!sbi->s_journal || is_bad_inode(inode))
2776 return 0;
2777
2778 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2779 !inode_is_locked(inode));
2780 /*
2781 * Exit early if inode already is on orphan list. This is a big speedup
2782 * since we don't have to contend on the global s_orphan_lock.
2783 */
2784 if (!list_empty(&EXT4_I(inode)->i_orphan))
2785 return 0;
2786
2787 /*
2788 * Orphan handling is only valid for files with data blocks
2789 * being truncated, or files being unlinked. Note that we either
2790 * hold i_mutex, or the inode can not be referenced from outside,
2791 * so i_nlink should not be bumped due to race
2792 */
2793 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2794 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2795
2796 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2797 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2798 if (err)
2799 goto out;
2800
2801 err = ext4_reserve_inode_write(handle, inode, &iloc);
2802 if (err)
2803 goto out;
2804
2805 mutex_lock(&sbi->s_orphan_lock);
2806 /*
2807 * Due to previous errors inode may be already a part of on-disk
2808 * orphan list. If so skip on-disk list modification.
2809 */
2810 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2811 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2812 /* Insert this inode at the head of the on-disk orphan list */
2813 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2814 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2815 dirty = true;
2816 }
2817 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2818 mutex_unlock(&sbi->s_orphan_lock);
2819
2820 if (dirty) {
2821 err = ext4_handle_dirty_super(handle, sb);
2822 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2823 if (!err)
2824 err = rc;
2825 if (err) {
2826 /*
2827 * We have to remove inode from in-memory list if
2828 * addition to on disk orphan list failed. Stray orphan
2829 * list entries can cause panics at unmount time.
2830 */
2831 mutex_lock(&sbi->s_orphan_lock);
2832 list_del_init(&EXT4_I(inode)->i_orphan);
2833 mutex_unlock(&sbi->s_orphan_lock);
2834 }
2835 }
2836 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2837 jbd_debug(4, "orphan inode %lu will point to %d\n",
2838 inode->i_ino, NEXT_ORPHAN(inode));
2839 out:
2840 ext4_std_error(sb, err);
2841 return err;
2842 }
2843
2844 /*
2845 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2846 * of such inodes stored on disk, because it is finally being cleaned up.
2847 */
2848 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2849 {
2850 struct list_head *prev;
2851 struct ext4_inode_info *ei = EXT4_I(inode);
2852 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2853 __u32 ino_next;
2854 struct ext4_iloc iloc;
2855 int err = 0;
2856
2857 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2858 return 0;
2859
2860 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2861 !inode_is_locked(inode));
2862 /* Do this quick check before taking global s_orphan_lock. */
2863 if (list_empty(&ei->i_orphan))
2864 return 0;
2865
2866 if (handle) {
2867 /* Grab inode buffer early before taking global s_orphan_lock */
2868 err = ext4_reserve_inode_write(handle, inode, &iloc);
2869 }
2870
2871 mutex_lock(&sbi->s_orphan_lock);
2872 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2873
2874 prev = ei->i_orphan.prev;
2875 list_del_init(&ei->i_orphan);
2876
2877 /* If we're on an error path, we may not have a valid
2878 * transaction handle with which to update the orphan list on
2879 * disk, but we still need to remove the inode from the linked
2880 * list in memory. */
2881 if (!handle || err) {
2882 mutex_unlock(&sbi->s_orphan_lock);
2883 goto out_err;
2884 }
2885
2886 ino_next = NEXT_ORPHAN(inode);
2887 if (prev == &sbi->s_orphan) {
2888 jbd_debug(4, "superblock will point to %u\n", ino_next);
2889 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2890 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2891 if (err) {
2892 mutex_unlock(&sbi->s_orphan_lock);
2893 goto out_brelse;
2894 }
2895 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2896 mutex_unlock(&sbi->s_orphan_lock);
2897 err = ext4_handle_dirty_super(handle, inode->i_sb);
2898 } else {
2899 struct ext4_iloc iloc2;
2900 struct inode *i_prev =
2901 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2902
2903 jbd_debug(4, "orphan inode %lu will point to %u\n",
2904 i_prev->i_ino, ino_next);
2905 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2906 if (err) {
2907 mutex_unlock(&sbi->s_orphan_lock);
2908 goto out_brelse;
2909 }
2910 NEXT_ORPHAN(i_prev) = ino_next;
2911 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2912 mutex_unlock(&sbi->s_orphan_lock);
2913 }
2914 if (err)
2915 goto out_brelse;
2916 NEXT_ORPHAN(inode) = 0;
2917 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2918 out_err:
2919 ext4_std_error(inode->i_sb, err);
2920 return err;
2921
2922 out_brelse:
2923 brelse(iloc.bh);
2924 goto out_err;
2925 }
2926
2927 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2928 {
2929 int retval;
2930 struct inode *inode;
2931 struct buffer_head *bh;
2932 struct ext4_dir_entry_2 *de;
2933 handle_t *handle = NULL;
2934
2935 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2936 return -EIO;
2937
2938 /* Initialize quotas before so that eventual writes go in
2939 * separate transaction */
2940 retval = dquot_initialize(dir);
2941 if (retval)
2942 return retval;
2943 retval = dquot_initialize(d_inode(dentry));
2944 if (retval)
2945 return retval;
2946
2947 retval = -ENOENT;
2948 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2949 if (IS_ERR(bh))
2950 return PTR_ERR(bh);
2951 if (!bh)
2952 goto end_rmdir;
2953
2954 inode = d_inode(dentry);
2955
2956 retval = -EFSCORRUPTED;
2957 if (le32_to_cpu(de->inode) != inode->i_ino)
2958 goto end_rmdir;
2959
2960 retval = -ENOTEMPTY;
2961 if (!ext4_empty_dir(inode))
2962 goto end_rmdir;
2963
2964 handle = ext4_journal_start(dir, EXT4_HT_DIR,
2965 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2966 if (IS_ERR(handle)) {
2967 retval = PTR_ERR(handle);
2968 handle = NULL;
2969 goto end_rmdir;
2970 }
2971
2972 if (IS_DIRSYNC(dir))
2973 ext4_handle_sync(handle);
2974
2975 retval = ext4_delete_entry(handle, dir, de, bh);
2976 if (retval)
2977 goto end_rmdir;
2978 if (!EXT4_DIR_LINK_EMPTY(inode))
2979 ext4_warning_inode(inode,
2980 "empty directory '%.*s' has too many links (%u)",
2981 dentry->d_name.len, dentry->d_name.name,
2982 inode->i_nlink);
2983 inode->i_version++;
2984 clear_nlink(inode);
2985 /* There's no need to set i_disksize: the fact that i_nlink is
2986 * zero will ensure that the right thing happens during any
2987 * recovery. */
2988 inode->i_size = 0;
2989 ext4_orphan_add(handle, inode);
2990 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2991 ext4_mark_inode_dirty(handle, inode);
2992 ext4_dec_count(handle, dir);
2993 ext4_update_dx_flag(dir);
2994 ext4_mark_inode_dirty(handle, dir);
2995
2996 end_rmdir:
2997 brelse(bh);
2998 if (handle)
2999 ext4_journal_stop(handle);
3000 return retval;
3001 }
3002
3003 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3004 {
3005 int retval;
3006 struct inode *inode;
3007 struct buffer_head *bh;
3008 struct ext4_dir_entry_2 *de;
3009 handle_t *handle = NULL;
3010
3011 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3012 return -EIO;
3013
3014 trace_ext4_unlink_enter(dir, dentry);
3015 /* Initialize quotas before so that eventual writes go
3016 * in separate transaction */
3017 retval = dquot_initialize(dir);
3018 if (retval)
3019 return retval;
3020 retval = dquot_initialize(d_inode(dentry));
3021 if (retval)
3022 return retval;
3023
3024 retval = -ENOENT;
3025 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3026 if (IS_ERR(bh))
3027 return PTR_ERR(bh);
3028 if (!bh)
3029 goto end_unlink;
3030
3031 inode = d_inode(dentry);
3032
3033 retval = -EFSCORRUPTED;
3034 if (le32_to_cpu(de->inode) != inode->i_ino)
3035 goto end_unlink;
3036
3037 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3038 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3039 if (IS_ERR(handle)) {
3040 retval = PTR_ERR(handle);
3041 handle = NULL;
3042 goto end_unlink;
3043 }
3044
3045 if (IS_DIRSYNC(dir))
3046 ext4_handle_sync(handle);
3047
3048 if (inode->i_nlink == 0) {
3049 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3050 dentry->d_name.len, dentry->d_name.name);
3051 set_nlink(inode, 1);
3052 }
3053 retval = ext4_delete_entry(handle, dir, de, bh);
3054 if (retval)
3055 goto end_unlink;
3056 dir->i_ctime = dir->i_mtime = current_time(dir);
3057 ext4_update_dx_flag(dir);
3058 ext4_mark_inode_dirty(handle, dir);
3059 drop_nlink(inode);
3060 if (!inode->i_nlink)
3061 ext4_orphan_add(handle, inode);
3062 inode->i_ctime = current_time(inode);
3063 ext4_mark_inode_dirty(handle, inode);
3064
3065 end_unlink:
3066 brelse(bh);
3067 if (handle)
3068 ext4_journal_stop(handle);
3069 trace_ext4_unlink_exit(dentry, retval);
3070 return retval;
3071 }
3072
3073 static int ext4_symlink(struct inode *dir,
3074 struct dentry *dentry, const char *symname)
3075 {
3076 handle_t *handle;
3077 struct inode *inode;
3078 int err, len = strlen(symname);
3079 int credits;
3080 bool encryption_required;
3081 struct fscrypt_str disk_link;
3082 struct fscrypt_symlink_data *sd = NULL;
3083
3084 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3085 return -EIO;
3086
3087 disk_link.len = len + 1;
3088 disk_link.name = (char *) symname;
3089
3090 encryption_required = (ext4_encrypted_inode(dir) ||
3091 DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3092 if (encryption_required) {
3093 err = fscrypt_get_encryption_info(dir);
3094 if (err)
3095 return err;
3096 if (!fscrypt_has_encryption_key(dir))
3097 return -ENOKEY;
3098 disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3099 sizeof(struct fscrypt_symlink_data));
3100 sd = kzalloc(disk_link.len, GFP_KERNEL);
3101 if (!sd)
3102 return -ENOMEM;
3103 }
3104
3105 if (disk_link.len > dir->i_sb->s_blocksize) {
3106 err = -ENAMETOOLONG;
3107 goto err_free_sd;
3108 }
3109
3110 err = dquot_initialize(dir);
3111 if (err)
3112 goto err_free_sd;
3113
3114 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3115 /*
3116 * For non-fast symlinks, we just allocate inode and put it on
3117 * orphan list in the first transaction => we need bitmap,
3118 * group descriptor, sb, inode block, quota blocks, and
3119 * possibly selinux xattr blocks.
3120 */
3121 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3122 EXT4_XATTR_TRANS_BLOCKS;
3123 } else {
3124 /*
3125 * Fast symlink. We have to add entry to directory
3126 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3127 * allocate new inode (bitmap, group descriptor, inode block,
3128 * quota blocks, sb is already counted in previous macros).
3129 */
3130 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3131 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3132 }
3133
3134 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3135 &dentry->d_name, 0, NULL,
3136 EXT4_HT_DIR, credits);
3137 handle = ext4_journal_current_handle();
3138 if (IS_ERR(inode)) {
3139 if (handle)
3140 ext4_journal_stop(handle);
3141 err = PTR_ERR(inode);
3142 goto err_free_sd;
3143 }
3144
3145 if (encryption_required) {
3146 struct qstr istr;
3147 struct fscrypt_str ostr =
3148 FSTR_INIT(sd->encrypted_path, disk_link.len);
3149
3150 istr.name = (const unsigned char *) symname;
3151 istr.len = len;
3152 err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3153 if (err)
3154 goto err_drop_inode;
3155 sd->len = cpu_to_le16(ostr.len);
3156 disk_link.name = (char *) sd;
3157 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3158 }
3159
3160 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3161 if (!encryption_required)
3162 inode->i_op = &ext4_symlink_inode_operations;
3163 inode_nohighmem(inode);
3164 ext4_set_aops(inode);
3165 /*
3166 * We cannot call page_symlink() with transaction started
3167 * because it calls into ext4_write_begin() which can wait
3168 * for transaction commit if we are running out of space
3169 * and thus we deadlock. So we have to stop transaction now
3170 * and restart it when symlink contents is written.
3171 *
3172 * To keep fs consistent in case of crash, we have to put inode
3173 * to orphan list in the mean time.
3174 */
3175 drop_nlink(inode);
3176 err = ext4_orphan_add(handle, inode);
3177 ext4_journal_stop(handle);
3178 handle = NULL;
3179 if (err)
3180 goto err_drop_inode;
3181 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3182 if (err)
3183 goto err_drop_inode;
3184 /*
3185 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3186 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3187 */
3188 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3189 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3190 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3191 if (IS_ERR(handle)) {
3192 err = PTR_ERR(handle);
3193 handle = NULL;
3194 goto err_drop_inode;
3195 }
3196 set_nlink(inode, 1);
3197 err = ext4_orphan_del(handle, inode);
3198 if (err)
3199 goto err_drop_inode;
3200 } else {
3201 /* clear the extent format for fast symlink */
3202 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3203 if (!encryption_required) {
3204 inode->i_op = &ext4_fast_symlink_inode_operations;
3205 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3206 }
3207 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3208 disk_link.len);
3209 inode->i_size = disk_link.len - 1;
3210 }
3211 EXT4_I(inode)->i_disksize = inode->i_size;
3212 err = ext4_add_nondir(handle, dentry, inode);
3213 if (!err && IS_DIRSYNC(dir))
3214 ext4_handle_sync(handle);
3215
3216 if (handle)
3217 ext4_journal_stop(handle);
3218 kfree(sd);
3219 return err;
3220 err_drop_inode:
3221 if (handle)
3222 ext4_journal_stop(handle);
3223 clear_nlink(inode);
3224 unlock_new_inode(inode);
3225 iput(inode);
3226 err_free_sd:
3227 kfree(sd);
3228 return err;
3229 }
3230
3231 static int ext4_link(struct dentry *old_dentry,
3232 struct inode *dir, struct dentry *dentry)
3233 {
3234 handle_t *handle;
3235 struct inode *inode = d_inode(old_dentry);
3236 int err, retries = 0;
3237
3238 if (inode->i_nlink >= EXT4_LINK_MAX)
3239 return -EMLINK;
3240 if (ext4_encrypted_inode(dir) &&
3241 !fscrypt_has_permitted_context(dir, inode))
3242 return -EPERM;
3243
3244 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3245 (!projid_eq(EXT4_I(dir)->i_projid,
3246 EXT4_I(old_dentry->d_inode)->i_projid)))
3247 return -EXDEV;
3248
3249 err = dquot_initialize(dir);
3250 if (err)
3251 return err;
3252
3253 retry:
3254 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3255 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3256 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3257 if (IS_ERR(handle))
3258 return PTR_ERR(handle);
3259
3260 if (IS_DIRSYNC(dir))
3261 ext4_handle_sync(handle);
3262
3263 inode->i_ctime = current_time(inode);
3264 ext4_inc_count(handle, inode);
3265 ihold(inode);
3266
3267 err = ext4_add_entry(handle, dentry, inode);
3268 if (!err) {
3269 ext4_mark_inode_dirty(handle, inode);
3270 /* this can happen only for tmpfile being
3271 * linked the first time
3272 */
3273 if (inode->i_nlink == 1)
3274 ext4_orphan_del(handle, inode);
3275 d_instantiate(dentry, inode);
3276 } else {
3277 drop_nlink(inode);
3278 iput(inode);
3279 }
3280 ext4_journal_stop(handle);
3281 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3282 goto retry;
3283 return err;
3284 }
3285
3286
3287 /*
3288 * Try to find buffer head where contains the parent block.
3289 * It should be the inode block if it is inlined or the 1st block
3290 * if it is a normal dir.
3291 */
3292 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3293 struct inode *inode,
3294 int *retval,
3295 struct ext4_dir_entry_2 **parent_de,
3296 int *inlined)
3297 {
3298 struct buffer_head *bh;
3299
3300 if (!ext4_has_inline_data(inode)) {
3301 bh = ext4_read_dirblock(inode, 0, EITHER);
3302 if (IS_ERR(bh)) {
3303 *retval = PTR_ERR(bh);
3304 return NULL;
3305 }
3306 *parent_de = ext4_next_entry(
3307 (struct ext4_dir_entry_2 *)bh->b_data,
3308 inode->i_sb->s_blocksize);
3309 return bh;
3310 }
3311
3312 *inlined = 1;
3313 return ext4_get_first_inline_block(inode, parent_de, retval);
3314 }
3315
3316 struct ext4_renament {
3317 struct inode *dir;
3318 struct dentry *dentry;
3319 struct inode *inode;
3320 bool is_dir;
3321 int dir_nlink_delta;
3322
3323 /* entry for "dentry" */
3324 struct buffer_head *bh;
3325 struct ext4_dir_entry_2 *de;
3326 int inlined;
3327
3328 /* entry for ".." in inode if it's a directory */
3329 struct buffer_head *dir_bh;
3330 struct ext4_dir_entry_2 *parent_de;
3331 int dir_inlined;
3332 };
3333
3334 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3335 {
3336 int retval;
3337
3338 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3339 &retval, &ent->parent_de,
3340 &ent->dir_inlined);
3341 if (!ent->dir_bh)
3342 return retval;
3343 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3344 return -EFSCORRUPTED;
3345 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3346 return ext4_journal_get_write_access(handle, ent->dir_bh);
3347 }
3348
3349 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3350 unsigned dir_ino)
3351 {
3352 int retval;
3353
3354 ent->parent_de->inode = cpu_to_le32(dir_ino);
3355 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3356 if (!ent->dir_inlined) {
3357 if (is_dx(ent->inode)) {
3358 retval = ext4_handle_dirty_dx_node(handle,
3359 ent->inode,
3360 ent->dir_bh);
3361 } else {
3362 retval = ext4_handle_dirty_dirent_node(handle,
3363 ent->inode,
3364 ent->dir_bh);
3365 }
3366 } else {
3367 retval = ext4_mark_inode_dirty(handle, ent->inode);
3368 }
3369 if (retval) {
3370 ext4_std_error(ent->dir->i_sb, retval);
3371 return retval;
3372 }
3373 return 0;
3374 }
3375
3376 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3377 unsigned ino, unsigned file_type)
3378 {
3379 int retval;
3380
3381 BUFFER_TRACE(ent->bh, "get write access");
3382 retval = ext4_journal_get_write_access(handle, ent->bh);
3383 if (retval)
3384 return retval;
3385 ent->de->inode = cpu_to_le32(ino);
3386 if (ext4_has_feature_filetype(ent->dir->i_sb))
3387 ent->de->file_type = file_type;
3388 ent->dir->i_version++;
3389 ent->dir->i_ctime = ent->dir->i_mtime =
3390 current_time(ent->dir);
3391 ext4_mark_inode_dirty(handle, ent->dir);
3392 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3393 if (!ent->inlined) {
3394 retval = ext4_handle_dirty_dirent_node(handle,
3395 ent->dir, ent->bh);
3396 if (unlikely(retval)) {
3397 ext4_std_error(ent->dir->i_sb, retval);
3398 return retval;
3399 }
3400 }
3401 brelse(ent->bh);
3402 ent->bh = NULL;
3403
3404 return 0;
3405 }
3406
3407 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3408 const struct qstr *d_name)
3409 {
3410 int retval = -ENOENT;
3411 struct buffer_head *bh;
3412 struct ext4_dir_entry_2 *de;
3413
3414 bh = ext4_find_entry(dir, d_name, &de, NULL);
3415 if (IS_ERR(bh))
3416 return PTR_ERR(bh);
3417 if (bh) {
3418 retval = ext4_delete_entry(handle, dir, de, bh);
3419 brelse(bh);
3420 }
3421 return retval;
3422 }
3423
3424 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3425 int force_reread)
3426 {
3427 int retval;
3428 /*
3429 * ent->de could have moved from under us during htree split, so make
3430 * sure that we are deleting the right entry. We might also be pointing
3431 * to a stale entry in the unused part of ent->bh so just checking inum
3432 * and the name isn't enough.
3433 */
3434 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3435 ent->de->name_len != ent->dentry->d_name.len ||
3436 strncmp(ent->de->name, ent->dentry->d_name.name,
3437 ent->de->name_len) ||
3438 force_reread) {
3439 retval = ext4_find_delete_entry(handle, ent->dir,
3440 &ent->dentry->d_name);
3441 } else {
3442 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3443 if (retval == -ENOENT) {
3444 retval = ext4_find_delete_entry(handle, ent->dir,
3445 &ent->dentry->d_name);
3446 }
3447 }
3448
3449 if (retval) {
3450 ext4_warning_inode(ent->dir,
3451 "Deleting old file: nlink %d, error=%d",
3452 ent->dir->i_nlink, retval);
3453 }
3454 }
3455
3456 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3457 {
3458 if (ent->dir_nlink_delta) {
3459 if (ent->dir_nlink_delta == -1)
3460 ext4_dec_count(handle, ent->dir);
3461 else
3462 ext4_inc_count(handle, ent->dir);
3463 ext4_mark_inode_dirty(handle, ent->dir);
3464 }
3465 }
3466
3467 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3468 int credits, handle_t **h)
3469 {
3470 struct inode *wh;
3471 handle_t *handle;
3472 int retries = 0;
3473
3474 /*
3475 * for inode block, sb block, group summaries,
3476 * and inode bitmap
3477 */
3478 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3479 EXT4_XATTR_TRANS_BLOCKS + 4);
3480 retry:
3481 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3482 &ent->dentry->d_name, 0, NULL,
3483 EXT4_HT_DIR, credits);
3484
3485 handle = ext4_journal_current_handle();
3486 if (IS_ERR(wh)) {
3487 if (handle)
3488 ext4_journal_stop(handle);
3489 if (PTR_ERR(wh) == -ENOSPC &&
3490 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3491 goto retry;
3492 } else {
3493 *h = handle;
3494 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3495 wh->i_op = &ext4_special_inode_operations;
3496 }
3497 return wh;
3498 }
3499
3500 /*
3501 * Anybody can rename anything with this: the permission checks are left to the
3502 * higher-level routines.
3503 *
3504 * n.b. old_{dentry,inode) refers to the source dentry/inode
3505 * while new_{dentry,inode) refers to the destination dentry/inode
3506 * This comes from rename(const char *oldpath, const char *newpath)
3507 */
3508 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3509 struct inode *new_dir, struct dentry *new_dentry,
3510 unsigned int flags)
3511 {
3512 handle_t *handle = NULL;
3513 struct ext4_renament old = {
3514 .dir = old_dir,
3515 .dentry = old_dentry,
3516 .inode = d_inode(old_dentry),
3517 };
3518 struct ext4_renament new = {
3519 .dir = new_dir,
3520 .dentry = new_dentry,
3521 .inode = d_inode(new_dentry),
3522 };
3523 int force_reread;
3524 int retval;
3525 struct inode *whiteout = NULL;
3526 int credits;
3527 u8 old_file_type;
3528
3529 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3530 (!projid_eq(EXT4_I(new_dir)->i_projid,
3531 EXT4_I(old_dentry->d_inode)->i_projid)))
3532 return -EXDEV;
3533
3534 if ((ext4_encrypted_inode(old_dir) &&
3535 !fscrypt_has_encryption_key(old_dir)) ||
3536 (ext4_encrypted_inode(new_dir) &&
3537 !fscrypt_has_encryption_key(new_dir)))
3538 return -ENOKEY;
3539
3540 retval = dquot_initialize(old.dir);
3541 if (retval)
3542 return retval;
3543 retval = dquot_initialize(new.dir);
3544 if (retval)
3545 return retval;
3546
3547 /* Initialize quotas before so that eventual writes go
3548 * in separate transaction */
3549 if (new.inode) {
3550 retval = dquot_initialize(new.inode);
3551 if (retval)
3552 return retval;
3553 }
3554
3555 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3556 if (IS_ERR(old.bh))
3557 return PTR_ERR(old.bh);
3558 /*
3559 * Check for inode number is _not_ due to possible IO errors.
3560 * We might rmdir the source, keep it as pwd of some process
3561 * and merrily kill the link to whatever was created under the
3562 * same name. Goodbye sticky bit ;-<
3563 */
3564 retval = -ENOENT;
3565 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3566 goto end_rename;
3567
3568 if ((old.dir != new.dir) &&
3569 ext4_encrypted_inode(new.dir) &&
3570 !fscrypt_has_permitted_context(new.dir, old.inode)) {
3571 retval = -EPERM;
3572 goto end_rename;
3573 }
3574
3575 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3576 &new.de, &new.inlined);
3577 if (IS_ERR(new.bh)) {
3578 retval = PTR_ERR(new.bh);
3579 new.bh = NULL;
3580 goto end_rename;
3581 }
3582 if (new.bh) {
3583 if (!new.inode) {
3584 brelse(new.bh);
3585 new.bh = NULL;
3586 }
3587 }
3588 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3589 ext4_alloc_da_blocks(old.inode);
3590
3591 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3592 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3593 if (!(flags & RENAME_WHITEOUT)) {
3594 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3595 if (IS_ERR(handle)) {
3596 retval = PTR_ERR(handle);
3597 handle = NULL;
3598 goto end_rename;
3599 }
3600 } else {
3601 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3602 if (IS_ERR(whiteout)) {
3603 retval = PTR_ERR(whiteout);
3604 whiteout = NULL;
3605 goto end_rename;
3606 }
3607 }
3608
3609 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3610 ext4_handle_sync(handle);
3611
3612 if (S_ISDIR(old.inode->i_mode)) {
3613 if (new.inode) {
3614 retval = -ENOTEMPTY;
3615 if (!ext4_empty_dir(new.inode))
3616 goto end_rename;
3617 } else {
3618 retval = -EMLINK;
3619 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3620 goto end_rename;
3621 }
3622 retval = ext4_rename_dir_prepare(handle, &old);
3623 if (retval)
3624 goto end_rename;
3625 }
3626 /*
3627 * If we're renaming a file within an inline_data dir and adding or
3628 * setting the new dirent causes a conversion from inline_data to
3629 * extents/blockmap, we need to force the dirent delete code to
3630 * re-read the directory, or else we end up trying to delete a dirent
3631 * from what is now the extent tree root (or a block map).
3632 */
3633 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3634 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3635
3636 old_file_type = old.de->file_type;
3637 if (whiteout) {
3638 /*
3639 * Do this before adding a new entry, so the old entry is sure
3640 * to be still pointing to the valid old entry.
3641 */
3642 retval = ext4_setent(handle, &old, whiteout->i_ino,
3643 EXT4_FT_CHRDEV);
3644 if (retval)
3645 goto end_rename;
3646 ext4_mark_inode_dirty(handle, whiteout);
3647 }
3648 if (!new.bh) {
3649 retval = ext4_add_entry(handle, new.dentry, old.inode);
3650 if (retval)
3651 goto end_rename;
3652 } else {
3653 retval = ext4_setent(handle, &new,
3654 old.inode->i_ino, old_file_type);
3655 if (retval)
3656 goto end_rename;
3657 }
3658 if (force_reread)
3659 force_reread = !ext4_test_inode_flag(new.dir,
3660 EXT4_INODE_INLINE_DATA);
3661
3662 /*
3663 * Like most other Unix systems, set the ctime for inodes on a
3664 * rename.
3665 */
3666 old.inode->i_ctime = current_time(old.inode);
3667 ext4_mark_inode_dirty(handle, old.inode);
3668
3669 if (!whiteout) {
3670 /*
3671 * ok, that's it
3672 */
3673 ext4_rename_delete(handle, &old, force_reread);
3674 }
3675
3676 if (new.inode) {
3677 ext4_dec_count(handle, new.inode);
3678 new.inode->i_ctime = current_time(new.inode);
3679 }
3680 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3681 ext4_update_dx_flag(old.dir);
3682 if (old.dir_bh) {
3683 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3684 if (retval)
3685 goto end_rename;
3686
3687 ext4_dec_count(handle, old.dir);
3688 if (new.inode) {
3689 /* checked ext4_empty_dir above, can't have another
3690 * parent, ext4_dec_count() won't work for many-linked
3691 * dirs */
3692 clear_nlink(new.inode);
3693 } else {
3694 ext4_inc_count(handle, new.dir);
3695 ext4_update_dx_flag(new.dir);
3696 ext4_mark_inode_dirty(handle, new.dir);
3697 }
3698 }
3699 ext4_mark_inode_dirty(handle, old.dir);
3700 if (new.inode) {
3701 ext4_mark_inode_dirty(handle, new.inode);
3702 if (!new.inode->i_nlink)
3703 ext4_orphan_add(handle, new.inode);
3704 }
3705 retval = 0;
3706
3707 end_rename:
3708 brelse(old.dir_bh);
3709 brelse(old.bh);
3710 brelse(new.bh);
3711 if (whiteout) {
3712 if (retval)
3713 drop_nlink(whiteout);
3714 unlock_new_inode(whiteout);
3715 iput(whiteout);
3716 }
3717 if (handle)
3718 ext4_journal_stop(handle);
3719 return retval;
3720 }
3721
3722 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3723 struct inode *new_dir, struct dentry *new_dentry)
3724 {
3725 handle_t *handle = NULL;
3726 struct ext4_renament old = {
3727 .dir = old_dir,
3728 .dentry = old_dentry,
3729 .inode = d_inode(old_dentry),
3730 };
3731 struct ext4_renament new = {
3732 .dir = new_dir,
3733 .dentry = new_dentry,
3734 .inode = d_inode(new_dentry),
3735 };
3736 u8 new_file_type;
3737 int retval;
3738 struct timespec ctime;
3739
3740 if ((ext4_encrypted_inode(old_dir) &&
3741 !fscrypt_has_encryption_key(old_dir)) ||
3742 (ext4_encrypted_inode(new_dir) &&
3743 !fscrypt_has_encryption_key(new_dir)))
3744 return -ENOKEY;
3745
3746 if ((ext4_encrypted_inode(old_dir) ||
3747 ext4_encrypted_inode(new_dir)) &&
3748 (old_dir != new_dir) &&
3749 (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3750 !fscrypt_has_permitted_context(old_dir, new.inode)))
3751 return -EPERM;
3752
3753 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3754 !projid_eq(EXT4_I(new_dir)->i_projid,
3755 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3756 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3757 !projid_eq(EXT4_I(old_dir)->i_projid,
3758 EXT4_I(new_dentry->d_inode)->i_projid)))
3759 return -EXDEV;
3760
3761 retval = dquot_initialize(old.dir);
3762 if (retval)
3763 return retval;
3764 retval = dquot_initialize(new.dir);
3765 if (retval)
3766 return retval;
3767
3768 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3769 &old.de, &old.inlined);
3770 if (IS_ERR(old.bh))
3771 return PTR_ERR(old.bh);
3772 /*
3773 * Check for inode number is _not_ due to possible IO errors.
3774 * We might rmdir the source, keep it as pwd of some process
3775 * and merrily kill the link to whatever was created under the
3776 * same name. Goodbye sticky bit ;-<
3777 */
3778 retval = -ENOENT;
3779 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3780 goto end_rename;
3781
3782 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3783 &new.de, &new.inlined);
3784 if (IS_ERR(new.bh)) {
3785 retval = PTR_ERR(new.bh);
3786 new.bh = NULL;
3787 goto end_rename;
3788 }
3789
3790 /* RENAME_EXCHANGE case: old *and* new must both exist */
3791 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3792 goto end_rename;
3793
3794 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3795 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3796 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3797 if (IS_ERR(handle)) {
3798 retval = PTR_ERR(handle);
3799 handle = NULL;
3800 goto end_rename;
3801 }
3802
3803 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3804 ext4_handle_sync(handle);
3805
3806 if (S_ISDIR(old.inode->i_mode)) {
3807 old.is_dir = true;
3808 retval = ext4_rename_dir_prepare(handle, &old);
3809 if (retval)
3810 goto end_rename;
3811 }
3812 if (S_ISDIR(new.inode->i_mode)) {
3813 new.is_dir = true;
3814 retval = ext4_rename_dir_prepare(handle, &new);
3815 if (retval)
3816 goto end_rename;
3817 }
3818
3819 /*
3820 * Other than the special case of overwriting a directory, parents'
3821 * nlink only needs to be modified if this is a cross directory rename.
3822 */
3823 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3824 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3825 new.dir_nlink_delta = -old.dir_nlink_delta;
3826 retval = -EMLINK;
3827 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3828 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3829 goto end_rename;
3830 }
3831
3832 new_file_type = new.de->file_type;
3833 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3834 if (retval)
3835 goto end_rename;
3836
3837 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3838 if (retval)
3839 goto end_rename;
3840
3841 /*
3842 * Like most other Unix systems, set the ctime for inodes on a
3843 * rename.
3844 */
3845 ctime = current_time(old.inode);
3846 old.inode->i_ctime = ctime;
3847 new.inode->i_ctime = ctime;
3848 ext4_mark_inode_dirty(handle, old.inode);
3849 ext4_mark_inode_dirty(handle, new.inode);
3850
3851 if (old.dir_bh) {
3852 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3853 if (retval)
3854 goto end_rename;
3855 }
3856 if (new.dir_bh) {
3857 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3858 if (retval)
3859 goto end_rename;
3860 }
3861 ext4_update_dir_count(handle, &old);
3862 ext4_update_dir_count(handle, &new);
3863 retval = 0;
3864
3865 end_rename:
3866 brelse(old.dir_bh);
3867 brelse(new.dir_bh);
3868 brelse(old.bh);
3869 brelse(new.bh);
3870 if (handle)
3871 ext4_journal_stop(handle);
3872 return retval;
3873 }
3874
3875 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3876 struct inode *new_dir, struct dentry *new_dentry,
3877 unsigned int flags)
3878 {
3879 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3880 return -EIO;
3881
3882 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3883 return -EINVAL;
3884
3885 if (flags & RENAME_EXCHANGE) {
3886 return ext4_cross_rename(old_dir, old_dentry,
3887 new_dir, new_dentry);
3888 }
3889
3890 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3891 }
3892
3893 /*
3894 * directories can handle most operations...
3895 */
3896 const struct inode_operations ext4_dir_inode_operations = {
3897 .create = ext4_create,
3898 .lookup = ext4_lookup,
3899 .link = ext4_link,
3900 .unlink = ext4_unlink,
3901 .symlink = ext4_symlink,
3902 .mkdir = ext4_mkdir,
3903 .rmdir = ext4_rmdir,
3904 .mknod = ext4_mknod,
3905 .tmpfile = ext4_tmpfile,
3906 .rename = ext4_rename2,
3907 .setattr = ext4_setattr,
3908 .getattr = ext4_getattr,
3909 .listxattr = ext4_listxattr,
3910 .get_acl = ext4_get_acl,
3911 .set_acl = ext4_set_acl,
3912 .fiemap = ext4_fiemap,
3913 };
3914
3915 const struct inode_operations ext4_special_inode_operations = {
3916 .setattr = ext4_setattr,
3917 .getattr = ext4_getattr,
3918 .listxattr = ext4_listxattr,
3919 .get_acl = ext4_get_acl,
3920 .set_acl = ext4_set_acl,
3921 };