?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux/fpc-iii.git] / fs / ext4 / namei.c
blob56738b538ddf4c066df33a59a06020352313c749
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext4/namei.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/namei.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * Big-endian to little-endian byte-swapping/bitmaps by
17 * David S. Miller (davem@caip.rutgers.edu), 1995
18 * Directory entry file type support and forward compatibility hooks
19 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20 * Hash Tree Directory indexing (c)
21 * Daniel Phillips, 2001
22 * Hash Tree Directory indexing porting
23 * Christopher Li, 2002
24 * Hash Tree Directory indexing cleanup
25 * Theodore Ts'o, 2002
26 */
27
28 #include <linux/fs.h>
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include <linux/iversion.h>
38 #include <linux/unicode.h>
39 #include "ext4.h"
40 #include "ext4_jbd2.h"
41
42 #include "xattr.h"
43 #include "acl.h"
44
45 #include <trace/events/ext4.h>
46 /*
47 * define how far ahead to read directories while searching them.
48 */
49 #define NAMEI_RA_CHUNKS 2
50 #define NAMEI_RA_BLOCKS 4
51 #define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52
53 static struct buffer_head *ext4_append(handle_t *handle,
54 struct inode *inode,
55 ext4_lblk_t *block)
56 {
57 struct buffer_head *bh;
58 int err;
59
60 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
61 ((inode->i_size >> 10) >=
62 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
63 return ERR_PTR(-ENOSPC);
64
65 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
66
67 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
68 if (IS_ERR(bh))
69 return bh;
70 inode->i_size += inode->i_sb->s_blocksize;
71 EXT4_I(inode)->i_disksize = inode->i_size;
72 BUFFER_TRACE(bh, "get_write_access");
73 err = ext4_journal_get_write_access(handle, bh);
74 if (err) {
75 brelse(bh);
76 ext4_std_error(inode->i_sb, err);
77 return ERR_PTR(err);
78 }
79 return bh;
80 }
81
82 static int ext4_dx_csum_verify(struct inode *inode,
83 struct ext4_dir_entry *dirent);
84
85 /*
86 * Hints to ext4_read_dirblock regarding whether we expect a directory
87 * block being read to be an index block, or a block containing
88 * directory entries (and if the latter, whether it was found via a
89 * logical block in an htree index block). This is used to control
90 * what sort of sanity checkinig ext4_read_dirblock() will do on the
91 * directory block read from the storage device. EITHER will means
92 * the caller doesn't know what kind of directory block will be read,
93 * so no specific verification will be done.
94 */
95 typedef enum {
96 EITHER, INDEX, DIRENT, DIRENT_HTREE
97 } dirblock_type_t;
98
99 #define ext4_read_dirblock(inode, block, type) \
100 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
101
102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
103 ext4_lblk_t block,
104 dirblock_type_t type,
105 const char *func,
106 unsigned int line)
107 {
108 struct buffer_head *bh;
109 struct ext4_dir_entry *dirent;
110 int is_dx_block = 0;
111
112 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
113 bh = ERR_PTR(-EIO);
114 else
115 bh = ext4_bread(NULL, inode, block, 0);
116 if (IS_ERR(bh)) {
117 __ext4_warning(inode->i_sb, func, line,
118 "inode #%lu: lblock %lu: comm %s: "
119 "error %ld reading directory block",
120 inode->i_ino, (unsigned long)block,
121 current->comm, PTR_ERR(bh));
122
123 return bh;
124 }
125 if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
126 ext4_error_inode(inode, func, line, block,
127 "Directory hole found for htree %s block",
128 (type == INDEX) ? "index" : "leaf");
129 return ERR_PTR(-EFSCORRUPTED);
130 }
131 if (!bh)
132 return NULL;
133 dirent = (struct ext4_dir_entry *) bh->b_data;
134 /* Determine whether or not we have an index block */
135 if (is_dx(inode)) {
136 if (block == 0)
137 is_dx_block = 1;
138 else if (ext4_rec_len_from_disk(dirent->rec_len,
139 inode->i_sb->s_blocksize) ==
140 inode->i_sb->s_blocksize)
141 is_dx_block = 1;
142 }
143 if (!is_dx_block && type == INDEX) {
144 ext4_error_inode(inode, func, line, block,
145 "directory leaf block found instead of index block");
146 brelse(bh);
147 return ERR_PTR(-EFSCORRUPTED);
148 }
149 if (!ext4_has_metadata_csum(inode->i_sb) ||
150 buffer_verified(bh))
151 return bh;
152
153 /*
154 * An empty leaf block can get mistaken for a index block; for
155 * this reason, we can only check the index checksum when the
156 * caller is sure it should be an index block.
157 */
158 if (is_dx_block && type == INDEX) {
159 if (ext4_dx_csum_verify(inode, dirent) &&
160 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
161 set_buffer_verified(bh);
162 else {
163 ext4_error_inode_err(inode, func, line, block,
164 EFSBADCRC,
165 "Directory index failed checksum");
166 brelse(bh);
167 return ERR_PTR(-EFSBADCRC);
168 }
169 }
170 if (!is_dx_block) {
171 if (ext4_dirblock_csum_verify(inode, bh) &&
172 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
173 set_buffer_verified(bh);
174 else {
175 ext4_error_inode_err(inode, func, line, block,
176 EFSBADCRC,
177 "Directory block failed checksum");
178 brelse(bh);
179 return ERR_PTR(-EFSBADCRC);
180 }
181 }
182 return bh;
183 }
184
185 #ifndef assert
186 #define assert(test) J_ASSERT(test)
187 #endif
188
189 #ifdef DX_DEBUG
190 #define dxtrace(command) command
191 #else
192 #define dxtrace(command)
193 #endif
194
195 struct fake_dirent
196 {
197 __le32 inode;
198 __le16 rec_len;
199 u8 name_len;
200 u8 file_type;
201 };
202
203 struct dx_countlimit
204 {
205 __le16 limit;
206 __le16 count;
207 };
208
209 struct dx_entry
210 {
211 __le32 hash;
212 __le32 block;
213 };
214
215 /*
216 * dx_root_info is laid out so that if it should somehow get overlaid by a
217 * dirent the two low bits of the hash version will be zero. Therefore, the
218 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
219 */
220
221 struct dx_root
222 {
223 struct fake_dirent dot;
224 char dot_name[4];
225 struct fake_dirent dotdot;
226 char dotdot_name[4];
227 struct dx_root_info
228 {
229 __le32 reserved_zero;
230 u8 hash_version;
231 u8 info_length; /* 8 */
232 u8 indirect_levels;
233 u8 unused_flags;
234 }
235 info;
236 struct dx_entry entries[];
237 };
238
239 struct dx_node
240 {
241 struct fake_dirent fake;
242 struct dx_entry entries[];
243 };
244
245
246 struct dx_frame
247 {
248 struct buffer_head *bh;
249 struct dx_entry *entries;
250 struct dx_entry *at;
251 };
252
253 struct dx_map_entry
254 {
255 u32 hash;
256 u16 offs;
257 u16 size;
258 };
259
260 /*
261 * This goes at the end of each htree block.
262 */
263 struct dx_tail {
264 u32 dt_reserved;
265 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
266 };
267
268 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
269 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
270 static inline unsigned dx_get_hash(struct dx_entry *entry);
271 static void dx_set_hash(struct dx_entry *entry, unsigned value);
272 static unsigned dx_get_count(struct dx_entry *entries);
273 static unsigned dx_get_limit(struct dx_entry *entries);
274 static void dx_set_count(struct dx_entry *entries, unsigned value);
275 static void dx_set_limit(struct dx_entry *entries, unsigned value);
276 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
277 static unsigned dx_node_limit(struct inode *dir);
278 static struct dx_frame *dx_probe(struct ext4_filename *fname,
279 struct inode *dir,
280 struct dx_hash_info *hinfo,
281 struct dx_frame *frame);
282 static void dx_release(struct dx_frame *frames);
283 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
284 unsigned blocksize, struct dx_hash_info *hinfo,
285 struct dx_map_entry map[]);
286 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
287 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
288 struct dx_map_entry *offsets, int count, unsigned blocksize);
289 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
290 static void dx_insert_block(struct dx_frame *frame,
291 u32 hash, ext4_lblk_t block);
292 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
293 struct dx_frame *frame,
294 struct dx_frame *frames,
295 __u32 *start_hash);
296 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
297 struct ext4_filename *fname,
298 struct ext4_dir_entry_2 **res_dir);
299 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
300 struct inode *dir, struct inode *inode);
301
302 /* checksumming functions */
303 void ext4_initialize_dirent_tail(struct buffer_head *bh,
304 unsigned int blocksize)
305 {
306 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
307
308 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
309 t->det_rec_len = ext4_rec_len_to_disk(
310 sizeof(struct ext4_dir_entry_tail), blocksize);
311 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
312 }
313
314 /* Walk through a dirent block to find a checksum "dirent" at the tail */
315 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
316 struct buffer_head *bh)
317 {
318 struct ext4_dir_entry_tail *t;
319
320 #ifdef PARANOID
321 struct ext4_dir_entry *d, *top;
322
323 d = (struct ext4_dir_entry *)bh->b_data;
324 top = (struct ext4_dir_entry *)(bh->b_data +
325 (EXT4_BLOCK_SIZE(inode->i_sb) -
326 sizeof(struct ext4_dir_entry_tail)));
327 while (d < top && d->rec_len)
328 d = (struct ext4_dir_entry *)(((void *)d) +
329 le16_to_cpu(d->rec_len));
330
331 if (d != top)
332 return NULL;
333
334 t = (struct ext4_dir_entry_tail *)d;
335 #else
336 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
337 #endif
338
339 if (t->det_reserved_zero1 ||
340 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
341 t->det_reserved_zero2 ||
342 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
343 return NULL;
344
345 return t;
346 }
347
348 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
349 {
350 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
351 struct ext4_inode_info *ei = EXT4_I(inode);
352 __u32 csum;
353
354 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
355 return cpu_to_le32(csum);
356 }
357
358 #define warn_no_space_for_csum(inode) \
359 __warn_no_space_for_csum((inode), __func__, __LINE__)
360
361 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
362 unsigned int line)
363 {
364 __ext4_warning_inode(inode, func, line,
365 "No space for directory leaf checksum. Please run e2fsck -D.");
366 }
367
368 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
369 {
370 struct ext4_dir_entry_tail *t;
371
372 if (!ext4_has_metadata_csum(inode->i_sb))
373 return 1;
374
375 t = get_dirent_tail(inode, bh);
376 if (!t) {
377 warn_no_space_for_csum(inode);
378 return 0;
379 }
380
381 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
382 (char *)t - bh->b_data))
383 return 0;
384
385 return 1;
386 }
387
388 static void ext4_dirblock_csum_set(struct inode *inode,
389 struct buffer_head *bh)
390 {
391 struct ext4_dir_entry_tail *t;
392
393 if (!ext4_has_metadata_csum(inode->i_sb))
394 return;
395
396 t = get_dirent_tail(inode, bh);
397 if (!t) {
398 warn_no_space_for_csum(inode);
399 return;
400 }
401
402 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
403 (char *)t - bh->b_data);
404 }
405
406 int ext4_handle_dirty_dirblock(handle_t *handle,
407 struct inode *inode,
408 struct buffer_head *bh)
409 {
410 ext4_dirblock_csum_set(inode, bh);
411 return ext4_handle_dirty_metadata(handle, inode, bh);
412 }
413
414 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
415 struct ext4_dir_entry *dirent,
416 int *offset)
417 {
418 struct ext4_dir_entry *dp;
419 struct dx_root_info *root;
420 int count_offset;
421
422 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
423 count_offset = 8;
424 else if (le16_to_cpu(dirent->rec_len) == 12) {
425 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
426 if (le16_to_cpu(dp->rec_len) !=
427 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
428 return NULL;
429 root = (struct dx_root_info *)(((void *)dp + 12));
430 if (root->reserved_zero ||
431 root->info_length != sizeof(struct dx_root_info))
432 return NULL;
433 count_offset = 32;
434 } else
435 return NULL;
436
437 if (offset)
438 *offset = count_offset;
439 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
440 }
441
442 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
443 int count_offset, int count, struct dx_tail *t)
444 {
445 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
446 struct ext4_inode_info *ei = EXT4_I(inode);
447 __u32 csum;
448 int size;
449 __u32 dummy_csum = 0;
450 int offset = offsetof(struct dx_tail, dt_checksum);
451
452 size = count_offset + (count * sizeof(struct dx_entry));
453 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
454 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
455 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
456
457 return cpu_to_le32(csum);
458 }
459
460 static int ext4_dx_csum_verify(struct inode *inode,
461 struct ext4_dir_entry *dirent)
462 {
463 struct dx_countlimit *c;
464 struct dx_tail *t;
465 int count_offset, limit, count;
466
467 if (!ext4_has_metadata_csum(inode->i_sb))
468 return 1;
469
470 c = get_dx_countlimit(inode, dirent, &count_offset);
471 if (!c) {
472 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
473 return 0;
474 }
475 limit = le16_to_cpu(c->limit);
476 count = le16_to_cpu(c->count);
477 if (count_offset + (limit * sizeof(struct dx_entry)) >
478 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
479 warn_no_space_for_csum(inode);
480 return 0;
481 }
482 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
483
484 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
485 count, t))
486 return 0;
487 return 1;
488 }
489
490 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
491 {
492 struct dx_countlimit *c;
493 struct dx_tail *t;
494 int count_offset, limit, count;
495
496 if (!ext4_has_metadata_csum(inode->i_sb))
497 return;
498
499 c = get_dx_countlimit(inode, dirent, &count_offset);
500 if (!c) {
501 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
502 return;
503 }
504 limit = le16_to_cpu(c->limit);
505 count = le16_to_cpu(c->count);
506 if (count_offset + (limit * sizeof(struct dx_entry)) >
507 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
508 warn_no_space_for_csum(inode);
509 return;
510 }
511 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
512
513 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
514 }
515
516 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
517 struct inode *inode,
518 struct buffer_head *bh)
519 {
520 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
521 return ext4_handle_dirty_metadata(handle, inode, bh);
522 }
523
524 /*
525 * p is at least 6 bytes before the end of page
526 */
527 static inline struct ext4_dir_entry_2 *
528 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
529 {
530 return (struct ext4_dir_entry_2 *)((char *)p +
531 ext4_rec_len_from_disk(p->rec_len, blocksize));
532 }
533
534 /*
535 * Future: use high four bits of block for coalesce-on-delete flags
536 * Mask them off for now.
537 */
538
539 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
540 {
541 return le32_to_cpu(entry->block) & 0x0fffffff;
542 }
543
544 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
545 {
546 entry->block = cpu_to_le32(value);
547 }
548
549 static inline unsigned dx_get_hash(struct dx_entry *entry)
550 {
551 return le32_to_cpu(entry->hash);
552 }
553
554 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
555 {
556 entry->hash = cpu_to_le32(value);
557 }
558
559 static inline unsigned dx_get_count(struct dx_entry *entries)
560 {
561 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
562 }
563
564 static inline unsigned dx_get_limit(struct dx_entry *entries)
565 {
566 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
567 }
568
569 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
570 {
571 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
572 }
573
574 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
575 {
576 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
577 }
578
579 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
580 {
581 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
582 EXT4_DIR_REC_LEN(2) - infosize;
583
584 if (ext4_has_metadata_csum(dir->i_sb))
585 entry_space -= sizeof(struct dx_tail);
586 return entry_space / sizeof(struct dx_entry);
587 }
588
589 static inline unsigned dx_node_limit(struct inode *dir)
590 {
591 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
592
593 if (ext4_has_metadata_csum(dir->i_sb))
594 entry_space -= sizeof(struct dx_tail);
595 return entry_space / sizeof(struct dx_entry);
596 }
597
598 /*
599 * Debug
600 */
601 #ifdef DX_DEBUG
602 static void dx_show_index(char * label, struct dx_entry *entries)
603 {
604 int i, n = dx_get_count (entries);
605 printk(KERN_DEBUG "%s index", label);
606 for (i = 0; i < n; i++) {
607 printk(KERN_CONT " %x->%lu",
608 i ? dx_get_hash(entries + i) : 0,
609 (unsigned long)dx_get_block(entries + i));
610 }
611 printk(KERN_CONT "\n");
612 }
613
614 struct stats
615 {
616 unsigned names;
617 unsigned space;
618 unsigned bcount;
619 };
620
621 static struct stats dx_show_leaf(struct inode *dir,
622 struct dx_hash_info *hinfo,
623 struct ext4_dir_entry_2 *de,
624 int size, int show_names)
625 {
626 unsigned names = 0, space = 0;
627 char *base = (char *) de;
628 struct dx_hash_info h = *hinfo;
629
630 printk("names: ");
631 while ((char *) de < base + size)
632 {
633 if (de->inode)
634 {
635 if (show_names)
636 {
637 #ifdef CONFIG_FS_ENCRYPTION
638 int len;
639 char *name;
640 struct fscrypt_str fname_crypto_str =
641 FSTR_INIT(NULL, 0);
642 int res = 0;
643
644 name = de->name;
645 len = de->name_len;
646 if (IS_ENCRYPTED(dir))
647 res = fscrypt_get_encryption_info(dir);
648 if (res) {
649 printk(KERN_WARNING "Error setting up"
650 " fname crypto: %d\n", res);
651 }
652 if (!fscrypt_has_encryption_key(dir)) {
653 /* Directory is not encrypted */
654 ext4fs_dirhash(dir, de->name,
655 de->name_len, &h);
656 printk("%*.s:(U)%x.%u ", len,
657 name, h.hash,
658 (unsigned) ((char *) de
659 - base));
660 } else {
661 struct fscrypt_str de_name =
662 FSTR_INIT(name, len);
663
664 /* Directory is encrypted */
665 res = fscrypt_fname_alloc_buffer(
666 dir, len,
667 &fname_crypto_str);
668 if (res)
669 printk(KERN_WARNING "Error "
670 "allocating crypto "
671 "buffer--skipping "
672 "crypto\n");
673 res = fscrypt_fname_disk_to_usr(dir,
674 0, 0, &de_name,
675 &fname_crypto_str);
676 if (res) {
677 printk(KERN_WARNING "Error "
678 "converting filename "
679 "from disk to usr"
680 "\n");
681 name = "??";
682 len = 2;
683 } else {
684 name = fname_crypto_str.name;
685 len = fname_crypto_str.len;
686 }
687 ext4fs_dirhash(dir, de->name,
688 de->name_len, &h);
689 printk("%*.s:(E)%x.%u ", len, name,
690 h.hash, (unsigned) ((char *) de
691 - base));
692 fscrypt_fname_free_buffer(
693 &fname_crypto_str);
694 }
695 #else
696 int len = de->name_len;
697 char *name = de->name;
698 ext4fs_dirhash(dir, de->name, de->name_len, &h);
699 printk("%*.s:%x.%u ", len, name, h.hash,
700 (unsigned) ((char *) de - base));
701 #endif
702 }
703 space += EXT4_DIR_REC_LEN(de->name_len);
704 names++;
705 }
706 de = ext4_next_entry(de, size);
707 }
708 printk(KERN_CONT "(%i)\n", names);
709 return (struct stats) { names, space, 1 };
710 }
711
712 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
713 struct dx_entry *entries, int levels)
714 {
715 unsigned blocksize = dir->i_sb->s_blocksize;
716 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
717 unsigned bcount = 0;
718 struct buffer_head *bh;
719 printk("%i indexed blocks...\n", count);
720 for (i = 0; i < count; i++, entries++)
721 {
722 ext4_lblk_t block = dx_get_block(entries);
723 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
724 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
725 struct stats stats;
726 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
727 bh = ext4_bread(NULL,dir, block, 0);
728 if (!bh || IS_ERR(bh))
729 continue;
730 stats = levels?
731 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
732 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
733 bh->b_data, blocksize, 0);
734 names += stats.names;
735 space += stats.space;
736 bcount += stats.bcount;
737 brelse(bh);
738 }
739 if (bcount)
740 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
741 levels ? "" : " ", names, space/bcount,
742 (space/bcount)*100/blocksize);
743 return (struct stats) { names, space, bcount};
744 }
745 #endif /* DX_DEBUG */
746
747 /*
748 * Probe for a directory leaf block to search.
749 *
750 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
751 * error in the directory index, and the caller should fall back to
752 * searching the directory normally. The callers of dx_probe **MUST**
753 * check for this error code, and make sure it never gets reflected
754 * back to userspace.
755 */
756 static struct dx_frame *
757 dx_probe(struct ext4_filename *fname, struct inode *dir,
758 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
759 {
760 unsigned count, indirect;
761 struct dx_entry *at, *entries, *p, *q, *m;
762 struct dx_root *root;
763 struct dx_frame *frame = frame_in;
764 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
765 u32 hash;
766
767 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
768 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
769 if (IS_ERR(frame->bh))
770 return (struct dx_frame *) frame->bh;
771
772 root = (struct dx_root *) frame->bh->b_data;
773 if (root->info.hash_version != DX_HASH_TEA &&
774 root->info.hash_version != DX_HASH_HALF_MD4 &&
775 root->info.hash_version != DX_HASH_LEGACY) {
776 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
777 root->info.hash_version);
778 goto fail;
779 }
780 if (fname)
781 hinfo = &fname->hinfo;
782 hinfo->hash_version = root->info.hash_version;
783 if (hinfo->hash_version <= DX_HASH_TEA)
784 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
785 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
786 if (fname && fname_name(fname))
787 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
788 hash = hinfo->hash;
789
790 if (root->info.unused_flags & 1) {
791 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
792 root->info.unused_flags);
793 goto fail;
794 }
795
796 indirect = root->info.indirect_levels;
797 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
798 ext4_warning(dir->i_sb,
799 "Directory (ino: %lu) htree depth %#06x exceed"
800 "supported value", dir->i_ino,
801 ext4_dir_htree_level(dir->i_sb));
802 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
803 ext4_warning(dir->i_sb, "Enable large directory "
804 "feature to access it");
805 }
806 goto fail;
807 }
808
809 entries = (struct dx_entry *)(((char *)&root->info) +
810 root->info.info_length);
811
812 if (dx_get_limit(entries) != dx_root_limit(dir,
813 root->info.info_length)) {
814 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
815 dx_get_limit(entries),
816 dx_root_limit(dir, root->info.info_length));
817 goto fail;
818 }
819
820 dxtrace(printk("Look up %x", hash));
821 while (1) {
822 count = dx_get_count(entries);
823 if (!count || count > dx_get_limit(entries)) {
824 ext4_warning_inode(dir,
825 "dx entry: count %u beyond limit %u",
826 count, dx_get_limit(entries));
827 goto fail;
828 }
829
830 p = entries + 1;
831 q = entries + count - 1;
832 while (p <= q) {
833 m = p + (q - p) / 2;
834 dxtrace(printk(KERN_CONT "."));
835 if (dx_get_hash(m) > hash)
836 q = m - 1;
837 else
838 p = m + 1;
839 }
840
841 if (0) { // linear search cross check
842 unsigned n = count - 1;
843 at = entries;
844 while (n--)
845 {
846 dxtrace(printk(KERN_CONT ","));
847 if (dx_get_hash(++at) > hash)
848 {
849 at--;
850 break;
851 }
852 }
853 assert (at == p - 1);
854 }
855
856 at = p - 1;
857 dxtrace(printk(KERN_CONT " %x->%u\n",
858 at == entries ? 0 : dx_get_hash(at),
859 dx_get_block(at)));
860 frame->entries = entries;
861 frame->at = at;
862 if (!indirect--)
863 return frame;
864 frame++;
865 frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
866 if (IS_ERR(frame->bh)) {
867 ret_err = (struct dx_frame *) frame->bh;
868 frame->bh = NULL;
869 goto fail;
870 }
871 entries = ((struct dx_node *) frame->bh->b_data)->entries;
872
873 if (dx_get_limit(entries) != dx_node_limit(dir)) {
874 ext4_warning_inode(dir,
875 "dx entry: limit %u != node limit %u",
876 dx_get_limit(entries), dx_node_limit(dir));
877 goto fail;
878 }
879 }
880 fail:
881 while (frame >= frame_in) {
882 brelse(frame->bh);
883 frame--;
884 }
885
886 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
887 ext4_warning_inode(dir,
888 "Corrupt directory, running e2fsck is recommended");
889 return ret_err;
890 }
891
892 static void dx_release(struct dx_frame *frames)
893 {
894 struct dx_root_info *info;
895 int i;
896 unsigned int indirect_levels;
897
898 if (frames[0].bh == NULL)
899 return;
900
901 info = &((struct dx_root *)frames[0].bh->b_data)->info;
902 /* save local copy, "info" may be freed after brelse() */
903 indirect_levels = info->indirect_levels;
904 for (i = 0; i <= indirect_levels; i++) {
905 if (frames[i].bh == NULL)
906 break;
907 brelse(frames[i].bh);
908 frames[i].bh = NULL;
909 }
910 }
911
912 /*
913 * This function increments the frame pointer to search the next leaf
914 * block, and reads in the necessary intervening nodes if the search
915 * should be necessary. Whether or not the search is necessary is
916 * controlled by the hash parameter. If the hash value is even, then
917 * the search is only continued if the next block starts with that
918 * hash value. This is used if we are searching for a specific file.
919 *
920 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
921 *
922 * This function returns 1 if the caller should continue to search,
923 * or 0 if it should not. If there is an error reading one of the
924 * index blocks, it will a negative error code.
925 *
926 * If start_hash is non-null, it will be filled in with the starting
927 * hash of the next page.
928 */
929 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
930 struct dx_frame *frame,
931 struct dx_frame *frames,
932 __u32 *start_hash)
933 {
934 struct dx_frame *p;
935 struct buffer_head *bh;
936 int num_frames = 0;
937 __u32 bhash;
938
939 p = frame;
940 /*
941 * Find the next leaf page by incrementing the frame pointer.
942 * If we run out of entries in the interior node, loop around and
943 * increment pointer in the parent node. When we break out of
944 * this loop, num_frames indicates the number of interior
945 * nodes need to be read.
946 */
947 while (1) {
948 if (++(p->at) < p->entries + dx_get_count(p->entries))
949 break;
950 if (p == frames)
951 return 0;
952 num_frames++;
953 p--;
954 }
955
956 /*
957 * If the hash is 1, then continue only if the next page has a
958 * continuation hash of any value. This is used for readdir
959 * handling. Otherwise, check to see if the hash matches the
960 * desired contiuation hash. If it doesn't, return since
961 * there's no point to read in the successive index pages.
962 */
963 bhash = dx_get_hash(p->at);
964 if (start_hash)
965 *start_hash = bhash;
966 if ((hash & 1) == 0) {
967 if ((bhash & ~1) != hash)
968 return 0;
969 }
970 /*
971 * If the hash is HASH_NB_ALWAYS, we always go to the next
972 * block so no check is necessary
973 */
974 while (num_frames--) {
975 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
976 if (IS_ERR(bh))
977 return PTR_ERR(bh);
978 p++;
979 brelse(p->bh);
980 p->bh = bh;
981 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
982 }
983 return 1;
984 }
985
986
987 /*
988 * This function fills a red-black tree with information from a
989 * directory block. It returns the number directory entries loaded
990 * into the tree. If there is an error it is returned in err.
991 */
992 static int htree_dirblock_to_tree(struct file *dir_file,
993 struct inode *dir, ext4_lblk_t block,
994 struct dx_hash_info *hinfo,
995 __u32 start_hash, __u32 start_minor_hash)
996 {
997 struct buffer_head *bh;
998 struct ext4_dir_entry_2 *de, *top;
999 int err = 0, count = 0;
1000 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1001
1002 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1003 (unsigned long)block));
1004 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1005 if (IS_ERR(bh))
1006 return PTR_ERR(bh);
1007
1008 de = (struct ext4_dir_entry_2 *) bh->b_data;
1009 top = (struct ext4_dir_entry_2 *) ((char *) de +
1010 dir->i_sb->s_blocksize -
1011 EXT4_DIR_REC_LEN(0));
1012 /* Check if the directory is encrypted */
1013 if (IS_ENCRYPTED(dir)) {
1014 err = fscrypt_get_encryption_info(dir);
1015 if (err < 0) {
1016 brelse(bh);
1017 return err;
1018 }
1019 err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1020 &fname_crypto_str);
1021 if (err < 0) {
1022 brelse(bh);
1023 return err;
1024 }
1025 }
1026
1027 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1028 if (ext4_check_dir_entry(dir, NULL, de, bh,
1029 bh->b_data, bh->b_size,
1030 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1031 + ((char *)de - bh->b_data))) {
1032 /* silently ignore the rest of the block */
1033 break;
1034 }
1035 ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1036 if ((hinfo->hash < start_hash) ||
1037 ((hinfo->hash == start_hash) &&
1038 (hinfo->minor_hash < start_minor_hash)))
1039 continue;
1040 if (de->inode == 0)
1041 continue;
1042 if (!IS_ENCRYPTED(dir)) {
1043 tmp_str.name = de->name;
1044 tmp_str.len = de->name_len;
1045 err = ext4_htree_store_dirent(dir_file,
1046 hinfo->hash, hinfo->minor_hash, de,
1047 &tmp_str);
1048 } else {
1049 int save_len = fname_crypto_str.len;
1050 struct fscrypt_str de_name = FSTR_INIT(de->name,
1051 de->name_len);
1052
1053 /* Directory is encrypted */
1054 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1055 hinfo->minor_hash, &de_name,
1056 &fname_crypto_str);
1057 if (err) {
1058 count = err;
1059 goto errout;
1060 }
1061 err = ext4_htree_store_dirent(dir_file,
1062 hinfo->hash, hinfo->minor_hash, de,
1063 &fname_crypto_str);
1064 fname_crypto_str.len = save_len;
1065 }
1066 if (err != 0) {
1067 count = err;
1068 goto errout;
1069 }
1070 count++;
1071 }
1072 errout:
1073 brelse(bh);
1074 fscrypt_fname_free_buffer(&fname_crypto_str);
1075 return count;
1076 }
1077
1078
1079 /*
1080 * This function fills a red-black tree with information from a
1081 * directory. We start scanning the directory in hash order, starting
1082 * at start_hash and start_minor_hash.
1083 *
1084 * This function returns the number of entries inserted into the tree,
1085 * or a negative error code.
1086 */
1087 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1088 __u32 start_minor_hash, __u32 *next_hash)
1089 {
1090 struct dx_hash_info hinfo;
1091 struct ext4_dir_entry_2 *de;
1092 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1093 struct inode *dir;
1094 ext4_lblk_t block;
1095 int count = 0;
1096 int ret, err;
1097 __u32 hashval;
1098 struct fscrypt_str tmp_str;
1099
1100 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1101 start_hash, start_minor_hash));
1102 dir = file_inode(dir_file);
1103 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1104 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1105 if (hinfo.hash_version <= DX_HASH_TEA)
1106 hinfo.hash_version +=
1107 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1108 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1109 if (ext4_has_inline_data(dir)) {
1110 int has_inline_data = 1;
1111 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1112 &hinfo, start_hash,
1113 start_minor_hash,
1114 &has_inline_data);
1115 if (has_inline_data) {
1116 *next_hash = ~0;
1117 return count;
1118 }
1119 }
1120 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1121 start_hash, start_minor_hash);
1122 *next_hash = ~0;
1123 return count;
1124 }
1125 hinfo.hash = start_hash;
1126 hinfo.minor_hash = 0;
1127 frame = dx_probe(NULL, dir, &hinfo, frames);
1128 if (IS_ERR(frame))
1129 return PTR_ERR(frame);
1130
1131 /* Add '.' and '..' from the htree header */
1132 if (!start_hash && !start_minor_hash) {
1133 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1134 tmp_str.name = de->name;
1135 tmp_str.len = de->name_len;
1136 err = ext4_htree_store_dirent(dir_file, 0, 0,
1137 de, &tmp_str);
1138 if (err != 0)
1139 goto errout;
1140 count++;
1141 }
1142 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1143 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1144 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1145 tmp_str.name = de->name;
1146 tmp_str.len = de->name_len;
1147 err = ext4_htree_store_dirent(dir_file, 2, 0,
1148 de, &tmp_str);
1149 if (err != 0)
1150 goto errout;
1151 count++;
1152 }
1153
1154 while (1) {
1155 if (fatal_signal_pending(current)) {
1156 err = -ERESTARTSYS;
1157 goto errout;
1158 }
1159 cond_resched();
1160 block = dx_get_block(frame->at);
1161 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1162 start_hash, start_minor_hash);
1163 if (ret < 0) {
1164 err = ret;
1165 goto errout;
1166 }
1167 count += ret;
1168 hashval = ~0;
1169 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1170 frame, frames, &hashval);
1171 *next_hash = hashval;
1172 if (ret < 0) {
1173 err = ret;
1174 goto errout;
1175 }
1176 /*
1177 * Stop if: (a) there are no more entries, or
1178 * (b) we have inserted at least one entry and the
1179 * next hash value is not a continuation
1180 */
1181 if ((ret == 0) ||
1182 (count && ((hashval & 1) == 0)))
1183 break;
1184 }
1185 dx_release(frames);
1186 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1187 "next hash: %x\n", count, *next_hash));
1188 return count;
1189 errout:
1190 dx_release(frames);
1191 return (err);
1192 }
1193
1194 static inline int search_dirblock(struct buffer_head *bh,
1195 struct inode *dir,
1196 struct ext4_filename *fname,
1197 unsigned int offset,
1198 struct ext4_dir_entry_2 **res_dir)
1199 {
1200 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1201 fname, offset, res_dir);
1202 }
1203
1204 /*
1205 * Directory block splitting, compacting
1206 */
1207
1208 /*
1209 * Create map of hash values, offsets, and sizes, stored at end of block.
1210 * Returns number of entries mapped.
1211 */
1212 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1213 unsigned blocksize, struct dx_hash_info *hinfo,
1214 struct dx_map_entry *map_tail)
1215 {
1216 int count = 0;
1217 char *base = (char *) de;
1218 struct dx_hash_info h = *hinfo;
1219
1220 while ((char *) de < base + blocksize) {
1221 if (de->name_len && de->inode) {
1222 ext4fs_dirhash(dir, de->name, de->name_len, &h);
1223 map_tail--;
1224 map_tail->hash = h.hash;
1225 map_tail->offs = ((char *) de - base)>>2;
1226 map_tail->size = le16_to_cpu(de->rec_len);
1227 count++;
1228 cond_resched();
1229 }
1230 /* XXX: do we need to check rec_len == 0 case? -Chris */
1231 de = ext4_next_entry(de, blocksize);
1232 }
1233 return count;
1234 }
1235
1236 /* Sort map by hash value */
1237 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1238 {
1239 struct dx_map_entry *p, *q, *top = map + count - 1;
1240 int more;
1241 /* Combsort until bubble sort doesn't suck */
1242 while (count > 2) {
1243 count = count*10/13;
1244 if (count - 9 < 2) /* 9, 10 -> 11 */
1245 count = 11;
1246 for (p = top, q = p - count; q >= map; p--, q--)
1247 if (p->hash < q->hash)
1248 swap(*p, *q);
1249 }
1250 /* Garden variety bubble sort */
1251 do {
1252 more = 0;
1253 q = top;
1254 while (q-- > map) {
1255 if (q[1].hash >= q[0].hash)
1256 continue;
1257 swap(*(q+1), *q);
1258 more = 1;
1259 }
1260 } while(more);
1261 }
1262
1263 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1264 {
1265 struct dx_entry *entries = frame->entries;
1266 struct dx_entry *old = frame->at, *new = old + 1;
1267 int count = dx_get_count(entries);
1268
1269 assert(count < dx_get_limit(entries));
1270 assert(old < entries + count);
1271 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1272 dx_set_hash(new, hash);
1273 dx_set_block(new, block);
1274 dx_set_count(entries, count + 1);
1275 }
1276
1277 #ifdef CONFIG_UNICODE
1278 /*
1279 * Test whether a case-insensitive directory entry matches the filename
1280 * being searched for. If quick is set, assume the name being looked up
1281 * is already in the casefolded form.
1282 *
1283 * Returns: 0 if the directory entry matches, more than 0 if it
1284 * doesn't match or less than zero on error.
1285 */
1286 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1287 const struct qstr *entry, bool quick)
1288 {
1289 const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1290 const struct unicode_map *um = sbi->s_encoding;
1291 int ret;
1292
1293 if (quick)
1294 ret = utf8_strncasecmp_folded(um, name, entry);
1295 else
1296 ret = utf8_strncasecmp(um, name, entry);
1297
1298 if (ret < 0) {
1299 /* Handle invalid character sequence as either an error
1300 * or as an opaque byte sequence.
1301 */
1302 if (ext4_has_strict_mode(sbi))
1303 return -EINVAL;
1304
1305 if (name->len != entry->len)
1306 return 1;
1307
1308 return !!memcmp(name->name, entry->name, name->len);
1309 }
1310
1311 return ret;
1312 }
1313
1314 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1315 struct fscrypt_str *cf_name)
1316 {
1317 int len;
1318
1319 if (!IS_CASEFOLDED(dir) || !EXT4_SB(dir->i_sb)->s_encoding) {
1320 cf_name->name = NULL;
1321 return;
1322 }
1323
1324 cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1325 if (!cf_name->name)
1326 return;
1327
1328 len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1329 iname, cf_name->name,
1330 EXT4_NAME_LEN);
1331 if (len <= 0) {
1332 kfree(cf_name->name);
1333 cf_name->name = NULL;
1334 return;
1335 }
1336 cf_name->len = (unsigned) len;
1337
1338 }
1339 #endif
1340
1341 /*
1342 * Test whether a directory entry matches the filename being searched for.
1343 *
1344 * Return: %true if the directory entry matches, otherwise %false.
1345 */
1346 static inline bool ext4_match(const struct inode *parent,
1347 const struct ext4_filename *fname,
1348 const struct ext4_dir_entry_2 *de)
1349 {
1350 struct fscrypt_name f;
1351 #ifdef CONFIG_UNICODE
1352 const struct qstr entry = {.name = de->name, .len = de->name_len};
1353 #endif
1354
1355 if (!de->inode)
1356 return false;
1357
1358 f.usr_fname = fname->usr_fname;
1359 f.disk_name = fname->disk_name;
1360 #ifdef CONFIG_FS_ENCRYPTION
1361 f.crypto_buf = fname->crypto_buf;
1362 #endif
1363
1364 #ifdef CONFIG_UNICODE
1365 if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1366 if (fname->cf_name.name) {
1367 struct qstr cf = {.name = fname->cf_name.name,
1368 .len = fname->cf_name.len};
1369 return !ext4_ci_compare(parent, &cf, &entry, true);
1370 }
1371 return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1372 false);
1373 }
1374 #endif
1375
1376 return fscrypt_match_name(&f, de->name, de->name_len);
1377 }
1378
1379 /*
1380 * Returns 0 if not found, -1 on failure, and 1 on success
1381 */
1382 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1383 struct inode *dir, struct ext4_filename *fname,
1384 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1385 {
1386 struct ext4_dir_entry_2 * de;
1387 char * dlimit;
1388 int de_len;
1389
1390 de = (struct ext4_dir_entry_2 *)search_buf;
1391 dlimit = search_buf + buf_size;
1392 while ((char *) de < dlimit) {
1393 /* this code is executed quadratically often */
1394 /* do minimal checking `by hand' */
1395 if ((char *) de + de->name_len <= dlimit &&
1396 ext4_match(dir, fname, de)) {
1397 /* found a match - just to be sure, do
1398 * a full check */
1399 if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1400 bh->b_size, offset))
1401 return -1;
1402 *res_dir = de;
1403 return 1;
1404 }
1405 /* prevent looping on a bad block */
1406 de_len = ext4_rec_len_from_disk(de->rec_len,
1407 dir->i_sb->s_blocksize);
1408 if (de_len <= 0)
1409 return -1;
1410 offset += de_len;
1411 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1412 }
1413 return 0;
1414 }
1415
1416 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1417 struct ext4_dir_entry *de)
1418 {
1419 struct super_block *sb = dir->i_sb;
1420
1421 if (!is_dx(dir))
1422 return 0;
1423 if (block == 0)
1424 return 1;
1425 if (de->inode == 0 &&
1426 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1427 sb->s_blocksize)
1428 return 1;
1429 return 0;
1430 }
1431
1432 /*
1433 * __ext4_find_entry()
1434 *
1435 * finds an entry in the specified directory with the wanted name. It
1436 * returns the cache buffer in which the entry was found, and the entry
1437 * itself (as a parameter - res_dir). It does NOT read the inode of the
1438 * entry - you'll have to do that yourself if you want to.
1439 *
1440 * The returned buffer_head has ->b_count elevated. The caller is expected
1441 * to brelse() it when appropriate.
1442 */
1443 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1444 struct ext4_filename *fname,
1445 struct ext4_dir_entry_2 **res_dir,
1446 int *inlined)
1447 {
1448 struct super_block *sb;
1449 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1450 struct buffer_head *bh, *ret = NULL;
1451 ext4_lblk_t start, block;
1452 const u8 *name = fname->usr_fname->name;
1453 size_t ra_max = 0; /* Number of bh's in the readahead
1454 buffer, bh_use[] */
1455 size_t ra_ptr = 0; /* Current index into readahead
1456 buffer */
1457 ext4_lblk_t nblocks;
1458 int i, namelen, retval;
1459
1460 *res_dir = NULL;
1461 sb = dir->i_sb;
1462 namelen = fname->usr_fname->len;
1463 if (namelen > EXT4_NAME_LEN)
1464 return NULL;
1465
1466 if (ext4_has_inline_data(dir)) {
1467 int has_inline_data = 1;
1468 ret = ext4_find_inline_entry(dir, fname, res_dir,
1469 &has_inline_data);
1470 if (has_inline_data) {
1471 if (inlined)
1472 *inlined = 1;
1473 goto cleanup_and_exit;
1474 }
1475 }
1476
1477 if ((namelen <= 2) && (name[0] == '.') &&
1478 (name[1] == '.' || name[1] == '\0')) {
1479 /*
1480 * "." or ".." will only be in the first block
1481 * NFS may look up ".."; "." should be handled by the VFS
1482 */
1483 block = start = 0;
1484 nblocks = 1;
1485 goto restart;
1486 }
1487 if (is_dx(dir)) {
1488 ret = ext4_dx_find_entry(dir, fname, res_dir);
1489 /*
1490 * On success, or if the error was file not found,
1491 * return. Otherwise, fall back to doing a search the
1492 * old fashioned way.
1493 */
1494 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1495 goto cleanup_and_exit;
1496 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1497 "falling back\n"));
1498 ret = NULL;
1499 }
1500 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1501 if (!nblocks) {
1502 ret = NULL;
1503 goto cleanup_and_exit;
1504 }
1505 start = EXT4_I(dir)->i_dir_start_lookup;
1506 if (start >= nblocks)
1507 start = 0;
1508 block = start;
1509 restart:
1510 do {
1511 /*
1512 * We deal with the read-ahead logic here.
1513 */
1514 cond_resched();
1515 if (ra_ptr >= ra_max) {
1516 /* Refill the readahead buffer */
1517 ra_ptr = 0;
1518 if (block < start)
1519 ra_max = start - block;
1520 else
1521 ra_max = nblocks - block;
1522 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1523 retval = ext4_bread_batch(dir, block, ra_max,
1524 false /* wait */, bh_use);
1525 if (retval) {
1526 ret = ERR_PTR(retval);
1527 ra_max = 0;
1528 goto cleanup_and_exit;
1529 }
1530 }
1531 if ((bh = bh_use[ra_ptr++]) == NULL)
1532 goto next;
1533 wait_on_buffer(bh);
1534 if (!buffer_uptodate(bh)) {
1535 EXT4_ERROR_INODE_ERR(dir, EIO,
1536 "reading directory lblock %lu",
1537 (unsigned long) block);
1538 brelse(bh);
1539 ret = ERR_PTR(-EIO);
1540 goto cleanup_and_exit;
1541 }
1542 if (!buffer_verified(bh) &&
1543 !is_dx_internal_node(dir, block,
1544 (struct ext4_dir_entry *)bh->b_data) &&
1545 !ext4_dirblock_csum_verify(dir, bh)) {
1546 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1547 "checksumming directory "
1548 "block %lu", (unsigned long)block);
1549 brelse(bh);
1550 ret = ERR_PTR(-EFSBADCRC);
1551 goto cleanup_and_exit;
1552 }
1553 set_buffer_verified(bh);
1554 i = search_dirblock(bh, dir, fname,
1555 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1556 if (i == 1) {
1557 EXT4_I(dir)->i_dir_start_lookup = block;
1558 ret = bh;
1559 goto cleanup_and_exit;
1560 } else {
1561 brelse(bh);
1562 if (i < 0)
1563 goto cleanup_and_exit;
1564 }
1565 next:
1566 if (++block >= nblocks)
1567 block = 0;
1568 } while (block != start);
1569
1570 /*
1571 * If the directory has grown while we were searching, then
1572 * search the last part of the directory before giving up.
1573 */
1574 block = nblocks;
1575 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1576 if (block < nblocks) {
1577 start = 0;
1578 goto restart;
1579 }
1580
1581 cleanup_and_exit:
1582 /* Clean up the read-ahead blocks */
1583 for (; ra_ptr < ra_max; ra_ptr++)
1584 brelse(bh_use[ra_ptr]);
1585 return ret;
1586 }
1587
1588 static struct buffer_head *ext4_find_entry(struct inode *dir,
1589 const struct qstr *d_name,
1590 struct ext4_dir_entry_2 **res_dir,
1591 int *inlined)
1592 {
1593 int err;
1594 struct ext4_filename fname;
1595 struct buffer_head *bh;
1596
1597 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1598 if (err == -ENOENT)
1599 return NULL;
1600 if (err)
1601 return ERR_PTR(err);
1602
1603 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1604
1605 ext4_fname_free_filename(&fname);
1606 return bh;
1607 }
1608
1609 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1610 struct dentry *dentry,
1611 struct ext4_dir_entry_2 **res_dir)
1612 {
1613 int err;
1614 struct ext4_filename fname;
1615 struct buffer_head *bh;
1616
1617 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1618 if (err == -ENOENT)
1619 return NULL;
1620 if (err)
1621 return ERR_PTR(err);
1622
1623 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1624
1625 ext4_fname_free_filename(&fname);
1626 return bh;
1627 }
1628
1629 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1630 struct ext4_filename *fname,
1631 struct ext4_dir_entry_2 **res_dir)
1632 {
1633 struct super_block * sb = dir->i_sb;
1634 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1635 struct buffer_head *bh;
1636 ext4_lblk_t block;
1637 int retval;
1638
1639 #ifdef CONFIG_FS_ENCRYPTION
1640 *res_dir = NULL;
1641 #endif
1642 frame = dx_probe(fname, dir, NULL, frames);
1643 if (IS_ERR(frame))
1644 return (struct buffer_head *) frame;
1645 do {
1646 block = dx_get_block(frame->at);
1647 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1648 if (IS_ERR(bh))
1649 goto errout;
1650
1651 retval = search_dirblock(bh, dir, fname,
1652 block << EXT4_BLOCK_SIZE_BITS(sb),
1653 res_dir);
1654 if (retval == 1)
1655 goto success;
1656 brelse(bh);
1657 if (retval == -1) {
1658 bh = ERR_PTR(ERR_BAD_DX_DIR);
1659 goto errout;
1660 }
1661
1662 /* Check to see if we should continue to search */
1663 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1664 frames, NULL);
1665 if (retval < 0) {
1666 ext4_warning_inode(dir,
1667 "error %d reading directory index block",
1668 retval);
1669 bh = ERR_PTR(retval);
1670 goto errout;
1671 }
1672 } while (retval == 1);
1673
1674 bh = NULL;
1675 errout:
1676 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1677 success:
1678 dx_release(frames);
1679 return bh;
1680 }
1681
1682 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1683 {
1684 struct inode *inode;
1685 struct ext4_dir_entry_2 *de;
1686 struct buffer_head *bh;
1687
1688 if (dentry->d_name.len > EXT4_NAME_LEN)
1689 return ERR_PTR(-ENAMETOOLONG);
1690
1691 bh = ext4_lookup_entry(dir, dentry, &de);
1692 if (IS_ERR(bh))
1693 return ERR_CAST(bh);
1694 inode = NULL;
1695 if (bh) {
1696 __u32 ino = le32_to_cpu(de->inode);
1697 brelse(bh);
1698 if (!ext4_valid_inum(dir->i_sb, ino)) {
1699 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1700 return ERR_PTR(-EFSCORRUPTED);
1701 }
1702 if (unlikely(ino == dir->i_ino)) {
1703 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1704 dentry);
1705 return ERR_PTR(-EFSCORRUPTED);
1706 }
1707 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1708 if (inode == ERR_PTR(-ESTALE)) {
1709 EXT4_ERROR_INODE(dir,
1710 "deleted inode referenced: %u",
1711 ino);
1712 return ERR_PTR(-EFSCORRUPTED);
1713 }
1714 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1715 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1716 !fscrypt_has_permitted_context(dir, inode)) {
1717 ext4_warning(inode->i_sb,
1718 "Inconsistent encryption contexts: %lu/%lu",
1719 dir->i_ino, inode->i_ino);
1720 iput(inode);
1721 return ERR_PTR(-EPERM);
1722 }
1723 }
1724
1725 #ifdef CONFIG_UNICODE
1726 if (!inode && IS_CASEFOLDED(dir)) {
1727 /* Eventually we want to call d_add_ci(dentry, NULL)
1728 * for negative dentries in the encoding case as
1729 * well. For now, prevent the negative dentry
1730 * from being cached.
1731 */
1732 return NULL;
1733 }
1734 #endif
1735 return d_splice_alias(inode, dentry);
1736 }
1737
1738
1739 struct dentry *ext4_get_parent(struct dentry *child)
1740 {
1741 __u32 ino;
1742 static const struct qstr dotdot = QSTR_INIT("..", 2);
1743 struct ext4_dir_entry_2 * de;
1744 struct buffer_head *bh;
1745
1746 bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1747 if (IS_ERR(bh))
1748 return ERR_CAST(bh);
1749 if (!bh)
1750 return ERR_PTR(-ENOENT);
1751 ino = le32_to_cpu(de->inode);
1752 brelse(bh);
1753
1754 if (!ext4_valid_inum(child->d_sb, ino)) {
1755 EXT4_ERROR_INODE(d_inode(child),
1756 "bad parent inode number: %u", ino);
1757 return ERR_PTR(-EFSCORRUPTED);
1758 }
1759
1760 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1761 }
1762
1763 /*
1764 * Move count entries from end of map between two memory locations.
1765 * Returns pointer to last entry moved.
1766 */
1767 static struct ext4_dir_entry_2 *
1768 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1769 unsigned blocksize)
1770 {
1771 unsigned rec_len = 0;
1772
1773 while (count--) {
1774 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1775 (from + (map->offs<<2));
1776 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1777 memcpy (to, de, rec_len);
1778 ((struct ext4_dir_entry_2 *) to)->rec_len =
1779 ext4_rec_len_to_disk(rec_len, blocksize);
1780 de->inode = 0;
1781 map++;
1782 to += rec_len;
1783 }
1784 return (struct ext4_dir_entry_2 *) (to - rec_len);
1785 }
1786
1787 /*
1788 * Compact each dir entry in the range to the minimal rec_len.
1789 * Returns pointer to last entry in range.
1790 */
1791 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1792 {
1793 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1794 unsigned rec_len = 0;
1795
1796 prev = to = de;
1797 while ((char*)de < base + blocksize) {
1798 next = ext4_next_entry(de, blocksize);
1799 if (de->inode && de->name_len) {
1800 rec_len = EXT4_DIR_REC_LEN(de->name_len);
1801 if (de > to)
1802 memmove(to, de, rec_len);
1803 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1804 prev = to;
1805 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1806 }
1807 de = next;
1808 }
1809 return prev;
1810 }
1811
1812 /*
1813 * Split a full leaf block to make room for a new dir entry.
1814 * Allocate a new block, and move entries so that they are approx. equally full.
1815 * Returns pointer to de in block into which the new entry will be inserted.
1816 */
1817 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1818 struct buffer_head **bh,struct dx_frame *frame,
1819 struct dx_hash_info *hinfo)
1820 {
1821 unsigned blocksize = dir->i_sb->s_blocksize;
1822 unsigned count, continued;
1823 struct buffer_head *bh2;
1824 ext4_lblk_t newblock;
1825 u32 hash2;
1826 struct dx_map_entry *map;
1827 char *data1 = (*bh)->b_data, *data2;
1828 unsigned split, move, size;
1829 struct ext4_dir_entry_2 *de = NULL, *de2;
1830 int csum_size = 0;
1831 int err = 0, i;
1832
1833 if (ext4_has_metadata_csum(dir->i_sb))
1834 csum_size = sizeof(struct ext4_dir_entry_tail);
1835
1836 bh2 = ext4_append(handle, dir, &newblock);
1837 if (IS_ERR(bh2)) {
1838 brelse(*bh);
1839 *bh = NULL;
1840 return (struct ext4_dir_entry_2 *) bh2;
1841 }
1842
1843 BUFFER_TRACE(*bh, "get_write_access");
1844 err = ext4_journal_get_write_access(handle, *bh);
1845 if (err)
1846 goto journal_error;
1847
1848 BUFFER_TRACE(frame->bh, "get_write_access");
1849 err = ext4_journal_get_write_access(handle, frame->bh);
1850 if (err)
1851 goto journal_error;
1852
1853 data2 = bh2->b_data;
1854
1855 /* create map in the end of data2 block */
1856 map = (struct dx_map_entry *) (data2 + blocksize);
1857 count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1858 blocksize, hinfo, map);
1859 map -= count;
1860 dx_sort_map(map, count);
1861 /* Split the existing block in the middle, size-wise */
1862 size = 0;
1863 move = 0;
1864 for (i = count-1; i >= 0; i--) {
1865 /* is more than half of this entry in 2nd half of the block? */
1866 if (size + map[i].size/2 > blocksize/2)
1867 break;
1868 size += map[i].size;
1869 move++;
1870 }
1871 /* map index at which we will split */
1872 split = count - move;
1873 hash2 = map[split].hash;
1874 continued = hash2 == map[split - 1].hash;
1875 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1876 (unsigned long)dx_get_block(frame->at),
1877 hash2, split, count-split));
1878
1879 /* Fancy dance to stay within two buffers */
1880 de2 = dx_move_dirents(data1, data2, map + split, count - split,
1881 blocksize);
1882 de = dx_pack_dirents(data1, blocksize);
1883 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1884 (char *) de,
1885 blocksize);
1886 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1887 (char *) de2,
1888 blocksize);
1889 if (csum_size) {
1890 ext4_initialize_dirent_tail(*bh, blocksize);
1891 ext4_initialize_dirent_tail(bh2, blocksize);
1892 }
1893
1894 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1895 blocksize, 1));
1896 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1897 blocksize, 1));
1898
1899 /* Which block gets the new entry? */
1900 if (hinfo->hash >= hash2) {
1901 swap(*bh, bh2);
1902 de = de2;
1903 }
1904 dx_insert_block(frame, hash2 + continued, newblock);
1905 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1906 if (err)
1907 goto journal_error;
1908 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1909 if (err)
1910 goto journal_error;
1911 brelse(bh2);
1912 dxtrace(dx_show_index("frame", frame->entries));
1913 return de;
1914
1915 journal_error:
1916 brelse(*bh);
1917 brelse(bh2);
1918 *bh = NULL;
1919 ext4_std_error(dir->i_sb, err);
1920 return ERR_PTR(err);
1921 }
1922
1923 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1924 struct buffer_head *bh,
1925 void *buf, int buf_size,
1926 struct ext4_filename *fname,
1927 struct ext4_dir_entry_2 **dest_de)
1928 {
1929 struct ext4_dir_entry_2 *de;
1930 unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1931 int nlen, rlen;
1932 unsigned int offset = 0;
1933 char *top;
1934
1935 de = (struct ext4_dir_entry_2 *)buf;
1936 top = buf + buf_size - reclen;
1937 while ((char *) de <= top) {
1938 if (ext4_check_dir_entry(dir, NULL, de, bh,
1939 buf, buf_size, offset))
1940 return -EFSCORRUPTED;
1941 if (ext4_match(dir, fname, de))
1942 return -EEXIST;
1943 nlen = EXT4_DIR_REC_LEN(de->name_len);
1944 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1945 if ((de->inode ? rlen - nlen : rlen) >= reclen)
1946 break;
1947 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1948 offset += rlen;
1949 }
1950 if ((char *) de > top)
1951 return -ENOSPC;
1952
1953 *dest_de = de;
1954 return 0;
1955 }
1956
1957 void ext4_insert_dentry(struct inode *inode,
1958 struct ext4_dir_entry_2 *de,
1959 int buf_size,
1960 struct ext4_filename *fname)
1961 {
1962
1963 int nlen, rlen;
1964
1965 nlen = EXT4_DIR_REC_LEN(de->name_len);
1966 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1967 if (de->inode) {
1968 struct ext4_dir_entry_2 *de1 =
1969 (struct ext4_dir_entry_2 *)((char *)de + nlen);
1970 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1971 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1972 de = de1;
1973 }
1974 de->file_type = EXT4_FT_UNKNOWN;
1975 de->inode = cpu_to_le32(inode->i_ino);
1976 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1977 de->name_len = fname_len(fname);
1978 memcpy(de->name, fname_name(fname), fname_len(fname));
1979 }
1980
1981 /*
1982 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1983 * it points to a directory entry which is guaranteed to be large
1984 * enough for new directory entry. If de is NULL, then
1985 * add_dirent_to_buf will attempt search the directory block for
1986 * space. It will return -ENOSPC if no space is available, and -EIO
1987 * and -EEXIST if directory entry already exists.
1988 */
1989 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1990 struct inode *dir,
1991 struct inode *inode, struct ext4_dir_entry_2 *de,
1992 struct buffer_head *bh)
1993 {
1994 unsigned int blocksize = dir->i_sb->s_blocksize;
1995 int csum_size = 0;
1996 int err, err2;
1997
1998 if (ext4_has_metadata_csum(inode->i_sb))
1999 csum_size = sizeof(struct ext4_dir_entry_tail);
2000
2001 if (!de) {
2002 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2003 blocksize - csum_size, fname, &de);
2004 if (err)
2005 return err;
2006 }
2007 BUFFER_TRACE(bh, "get_write_access");
2008 err = ext4_journal_get_write_access(handle, bh);
2009 if (err) {
2010 ext4_std_error(dir->i_sb, err);
2011 return err;
2012 }
2013
2014 /* By now the buffer is marked for journaling */
2015 ext4_insert_dentry(inode, de, blocksize, fname);
2016
2017 /*
2018 * XXX shouldn't update any times until successful
2019 * completion of syscall, but too many callers depend
2020 * on this.
2021 *
2022 * XXX similarly, too many callers depend on
2023 * ext4_new_inode() setting the times, but error
2024 * recovery deletes the inode, so the worst that can
2025 * happen is that the times are slightly out of date
2026 * and/or different from the directory change time.
2027 */
2028 dir->i_mtime = dir->i_ctime = current_time(dir);
2029 ext4_update_dx_flag(dir);
2030 inode_inc_iversion(dir);
2031 err2 = ext4_mark_inode_dirty(handle, dir);
2032 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2033 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2034 if (err)
2035 ext4_std_error(dir->i_sb, err);
2036 return err ? err : err2;
2037 }
2038
2039 /*
2040 * This converts a one block unindexed directory to a 3 block indexed
2041 * directory, and adds the dentry to the indexed directory.
2042 */
2043 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2044 struct inode *dir,
2045 struct inode *inode, struct buffer_head *bh)
2046 {
2047 struct buffer_head *bh2;
2048 struct dx_root *root;
2049 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2050 struct dx_entry *entries;
2051 struct ext4_dir_entry_2 *de, *de2;
2052 char *data2, *top;
2053 unsigned len;
2054 int retval;
2055 unsigned blocksize;
2056 ext4_lblk_t block;
2057 struct fake_dirent *fde;
2058 int csum_size = 0;
2059
2060 if (ext4_has_metadata_csum(inode->i_sb))
2061 csum_size = sizeof(struct ext4_dir_entry_tail);
2062
2063 blocksize = dir->i_sb->s_blocksize;
2064 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2065 BUFFER_TRACE(bh, "get_write_access");
2066 retval = ext4_journal_get_write_access(handle, bh);
2067 if (retval) {
2068 ext4_std_error(dir->i_sb, retval);
2069 brelse(bh);
2070 return retval;
2071 }
2072 root = (struct dx_root *) bh->b_data;
2073
2074 /* The 0th block becomes the root, move the dirents out */
2075 fde = &root->dotdot;
2076 de = (struct ext4_dir_entry_2 *)((char *)fde +
2077 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2078 if ((char *) de >= (((char *) root) + blocksize)) {
2079 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2080 brelse(bh);
2081 return -EFSCORRUPTED;
2082 }
2083 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2084
2085 /* Allocate new block for the 0th block's dirents */
2086 bh2 = ext4_append(handle, dir, &block);
2087 if (IS_ERR(bh2)) {
2088 brelse(bh);
2089 return PTR_ERR(bh2);
2090 }
2091 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2092 data2 = bh2->b_data;
2093
2094 memcpy(data2, de, len);
2095 de = (struct ext4_dir_entry_2 *) data2;
2096 top = data2 + len;
2097 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2098 de = de2;
2099 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2100 (char *) de, blocksize);
2101
2102 if (csum_size)
2103 ext4_initialize_dirent_tail(bh2, blocksize);
2104
2105 /* Initialize the root; the dot dirents already exist */
2106 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2107 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2108 blocksize);
2109 memset (&root->info, 0, sizeof(root->info));
2110 root->info.info_length = sizeof(root->info);
2111 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2112 entries = root->entries;
2113 dx_set_block(entries, 1);
2114 dx_set_count(entries, 1);
2115 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2116
2117 /* Initialize as for dx_probe */
2118 fname->hinfo.hash_version = root->info.hash_version;
2119 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2120 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2121 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2122 ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2123
2124 memset(frames, 0, sizeof(frames));
2125 frame = frames;
2126 frame->entries = entries;
2127 frame->at = entries;
2128 frame->bh = bh;
2129
2130 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2131 if (retval)
2132 goto out_frames;
2133 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2134 if (retval)
2135 goto out_frames;
2136
2137 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2138 if (IS_ERR(de)) {
2139 retval = PTR_ERR(de);
2140 goto out_frames;
2141 }
2142
2143 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2144 out_frames:
2145 /*
2146 * Even if the block split failed, we have to properly write
2147 * out all the changes we did so far. Otherwise we can end up
2148 * with corrupted filesystem.
2149 */
2150 if (retval)
2151 ext4_mark_inode_dirty(handle, dir);
2152 dx_release(frames);
2153 brelse(bh2);
2154 return retval;
2155 }
2156
2157 /*
2158 * ext4_add_entry()
2159 *
2160 * adds a file entry to the specified directory, using the same
2161 * semantics as ext4_find_entry(). It returns NULL if it failed.
2162 *
2163 * NOTE!! The inode part of 'de' is left at 0 - which means you
2164 * may not sleep between calling this and putting something into
2165 * the entry, as someone else might have used it while you slept.
2166 */
2167 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2168 struct inode *inode)
2169 {
2170 struct inode *dir = d_inode(dentry->d_parent);
2171 struct buffer_head *bh = NULL;
2172 struct ext4_dir_entry_2 *de;
2173 struct super_block *sb;
2174 #ifdef CONFIG_UNICODE
2175 struct ext4_sb_info *sbi;
2176 #endif
2177 struct ext4_filename fname;
2178 int retval;
2179 int dx_fallback=0;
2180 unsigned blocksize;
2181 ext4_lblk_t block, blocks;
2182 int csum_size = 0;
2183
2184 if (ext4_has_metadata_csum(inode->i_sb))
2185 csum_size = sizeof(struct ext4_dir_entry_tail);
2186
2187 sb = dir->i_sb;
2188 blocksize = sb->s_blocksize;
2189 if (!dentry->d_name.len)
2190 return -EINVAL;
2191
2192 #ifdef CONFIG_UNICODE
2193 sbi = EXT4_SB(sb);
2194 if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2195 sbi->s_encoding && utf8_validate(sbi->s_encoding, &dentry->d_name))
2196 return -EINVAL;
2197 #endif
2198
2199 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2200 if (retval)
2201 return retval;
2202
2203 if (ext4_has_inline_data(dir)) {
2204 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2205 if (retval < 0)
2206 goto out;
2207 if (retval == 1) {
2208 retval = 0;
2209 goto out;
2210 }
2211 }
2212
2213 if (is_dx(dir)) {
2214 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2215 if (!retval || (retval != ERR_BAD_DX_DIR))
2216 goto out;
2217 /* Can we just ignore htree data? */
2218 if (ext4_has_metadata_csum(sb)) {
2219 EXT4_ERROR_INODE(dir,
2220 "Directory has corrupted htree index.");
2221 retval = -EFSCORRUPTED;
2222 goto out;
2223 }
2224 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2225 dx_fallback++;
2226 retval = ext4_mark_inode_dirty(handle, dir);
2227 if (unlikely(retval))
2228 goto out;
2229 }
2230 blocks = dir->i_size >> sb->s_blocksize_bits;
2231 for (block = 0; block < blocks; block++) {
2232 bh = ext4_read_dirblock(dir, block, DIRENT);
2233 if (bh == NULL) {
2234 bh = ext4_bread(handle, dir, block,
2235 EXT4_GET_BLOCKS_CREATE);
2236 goto add_to_new_block;
2237 }
2238 if (IS_ERR(bh)) {
2239 retval = PTR_ERR(bh);
2240 bh = NULL;
2241 goto out;
2242 }
2243 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2244 NULL, bh);
2245 if (retval != -ENOSPC)
2246 goto out;
2247
2248 if (blocks == 1 && !dx_fallback &&
2249 ext4_has_feature_dir_index(sb)) {
2250 retval = make_indexed_dir(handle, &fname, dir,
2251 inode, bh);
2252 bh = NULL; /* make_indexed_dir releases bh */
2253 goto out;
2254 }
2255 brelse(bh);
2256 }
2257 bh = ext4_append(handle, dir, &block);
2258 add_to_new_block:
2259 if (IS_ERR(bh)) {
2260 retval = PTR_ERR(bh);
2261 bh = NULL;
2262 goto out;
2263 }
2264 de = (struct ext4_dir_entry_2 *) bh->b_data;
2265 de->inode = 0;
2266 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2267
2268 if (csum_size)
2269 ext4_initialize_dirent_tail(bh, blocksize);
2270
2271 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2272 out:
2273 ext4_fname_free_filename(&fname);
2274 brelse(bh);
2275 if (retval == 0)
2276 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2277 return retval;
2278 }
2279
2280 /*
2281 * Returns 0 for success, or a negative error value
2282 */
2283 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2284 struct inode *dir, struct inode *inode)
2285 {
2286 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2287 struct dx_entry *entries, *at;
2288 struct buffer_head *bh;
2289 struct super_block *sb = dir->i_sb;
2290 struct ext4_dir_entry_2 *de;
2291 int restart;
2292 int err;
2293
2294 again:
2295 restart = 0;
2296 frame = dx_probe(fname, dir, NULL, frames);
2297 if (IS_ERR(frame))
2298 return PTR_ERR(frame);
2299 entries = frame->entries;
2300 at = frame->at;
2301 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2302 if (IS_ERR(bh)) {
2303 err = PTR_ERR(bh);
2304 bh = NULL;
2305 goto cleanup;
2306 }
2307
2308 BUFFER_TRACE(bh, "get_write_access");
2309 err = ext4_journal_get_write_access(handle, bh);
2310 if (err)
2311 goto journal_error;
2312
2313 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2314 if (err != -ENOSPC)
2315 goto cleanup;
2316
2317 err = 0;
2318 /* Block full, should compress but for now just split */
2319 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2320 dx_get_count(entries), dx_get_limit(entries)));
2321 /* Need to split index? */
2322 if (dx_get_count(entries) == dx_get_limit(entries)) {
2323 ext4_lblk_t newblock;
2324 int levels = frame - frames + 1;
2325 unsigned int icount;
2326 int add_level = 1;
2327 struct dx_entry *entries2;
2328 struct dx_node *node2;
2329 struct buffer_head *bh2;
2330
2331 while (frame > frames) {
2332 if (dx_get_count((frame - 1)->entries) <
2333 dx_get_limit((frame - 1)->entries)) {
2334 add_level = 0;
2335 break;
2336 }
2337 frame--; /* split higher index block */
2338 at = frame->at;
2339 entries = frame->entries;
2340 restart = 1;
2341 }
2342 if (add_level && levels == ext4_dir_htree_level(sb)) {
2343 ext4_warning(sb, "Directory (ino: %lu) index full, "
2344 "reach max htree level :%d",
2345 dir->i_ino, levels);
2346 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2347 ext4_warning(sb, "Large directory feature is "
2348 "not enabled on this "
2349 "filesystem");
2350 }
2351 err = -ENOSPC;
2352 goto cleanup;
2353 }
2354 icount = dx_get_count(entries);
2355 bh2 = ext4_append(handle, dir, &newblock);
2356 if (IS_ERR(bh2)) {
2357 err = PTR_ERR(bh2);
2358 goto cleanup;
2359 }
2360 node2 = (struct dx_node *)(bh2->b_data);
2361 entries2 = node2->entries;
2362 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2363 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2364 sb->s_blocksize);
2365 BUFFER_TRACE(frame->bh, "get_write_access");
2366 err = ext4_journal_get_write_access(handle, frame->bh);
2367 if (err)
2368 goto journal_error;
2369 if (!add_level) {
2370 unsigned icount1 = icount/2, icount2 = icount - icount1;
2371 unsigned hash2 = dx_get_hash(entries + icount1);
2372 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2373 icount1, icount2));
2374
2375 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2376 err = ext4_journal_get_write_access(handle,
2377 (frame - 1)->bh);
2378 if (err)
2379 goto journal_error;
2380
2381 memcpy((char *) entries2, (char *) (entries + icount1),
2382 icount2 * sizeof(struct dx_entry));
2383 dx_set_count(entries, icount1);
2384 dx_set_count(entries2, icount2);
2385 dx_set_limit(entries2, dx_node_limit(dir));
2386
2387 /* Which index block gets the new entry? */
2388 if (at - entries >= icount1) {
2389 frame->at = at = at - entries - icount1 + entries2;
2390 frame->entries = entries = entries2;
2391 swap(frame->bh, bh2);
2392 }
2393 dx_insert_block((frame - 1), hash2, newblock);
2394 dxtrace(dx_show_index("node", frame->entries));
2395 dxtrace(dx_show_index("node",
2396 ((struct dx_node *) bh2->b_data)->entries));
2397 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2398 if (err)
2399 goto journal_error;
2400 brelse (bh2);
2401 err = ext4_handle_dirty_dx_node(handle, dir,
2402 (frame - 1)->bh);
2403 if (err)
2404 goto journal_error;
2405 if (restart) {
2406 err = ext4_handle_dirty_dx_node(handle, dir,
2407 frame->bh);
2408 goto journal_error;
2409 }
2410 } else {
2411 struct dx_root *dxroot;
2412 memcpy((char *) entries2, (char *) entries,
2413 icount * sizeof(struct dx_entry));
2414 dx_set_limit(entries2, dx_node_limit(dir));
2415
2416 /* Set up root */
2417 dx_set_count(entries, 1);
2418 dx_set_block(entries + 0, newblock);
2419 dxroot = (struct dx_root *)frames[0].bh->b_data;
2420 dxroot->info.indirect_levels += 1;
2421 dxtrace(printk(KERN_DEBUG
2422 "Creating %d level index...\n",
2423 dxroot->info.indirect_levels));
2424 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2425 if (err)
2426 goto journal_error;
2427 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2428 brelse(bh2);
2429 restart = 1;
2430 goto journal_error;
2431 }
2432 }
2433 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2434 if (IS_ERR(de)) {
2435 err = PTR_ERR(de);
2436 goto cleanup;
2437 }
2438 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2439 goto cleanup;
2440
2441 journal_error:
2442 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2443 cleanup:
2444 brelse(bh);
2445 dx_release(frames);
2446 /* @restart is true means htree-path has been changed, we need to
2447 * repeat dx_probe() to find out valid htree-path
2448 */
2449 if (restart && err == 0)
2450 goto again;
2451 return err;
2452 }
2453
2454 /*
2455 * ext4_generic_delete_entry deletes a directory entry by merging it
2456 * with the previous entry
2457 */
2458 int ext4_generic_delete_entry(handle_t *handle,
2459 struct inode *dir,
2460 struct ext4_dir_entry_2 *de_del,
2461 struct buffer_head *bh,
2462 void *entry_buf,
2463 int buf_size,
2464 int csum_size)
2465 {
2466 struct ext4_dir_entry_2 *de, *pde;
2467 unsigned int blocksize = dir->i_sb->s_blocksize;
2468 int i;
2469
2470 i = 0;
2471 pde = NULL;
2472 de = (struct ext4_dir_entry_2 *)entry_buf;
2473 while (i < buf_size - csum_size) {
2474 if (ext4_check_dir_entry(dir, NULL, de, bh,
2475 bh->b_data, bh->b_size, i))
2476 return -EFSCORRUPTED;
2477 if (de == de_del) {
2478 if (pde)
2479 pde->rec_len = ext4_rec_len_to_disk(
2480 ext4_rec_len_from_disk(pde->rec_len,
2481 blocksize) +
2482 ext4_rec_len_from_disk(de->rec_len,
2483 blocksize),
2484 blocksize);
2485 else
2486 de->inode = 0;
2487 inode_inc_iversion(dir);
2488 return 0;
2489 }
2490 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2491 pde = de;
2492 de = ext4_next_entry(de, blocksize);
2493 }
2494 return -ENOENT;
2495 }
2496
2497 static int ext4_delete_entry(handle_t *handle,
2498 struct inode *dir,
2499 struct ext4_dir_entry_2 *de_del,
2500 struct buffer_head *bh)
2501 {
2502 int err, csum_size = 0;
2503
2504 if (ext4_has_inline_data(dir)) {
2505 int has_inline_data = 1;
2506 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2507 &has_inline_data);
2508 if (has_inline_data)
2509 return err;
2510 }
2511
2512 if (ext4_has_metadata_csum(dir->i_sb))
2513 csum_size = sizeof(struct ext4_dir_entry_tail);
2514
2515 BUFFER_TRACE(bh, "get_write_access");
2516 err = ext4_journal_get_write_access(handle, bh);
2517 if (unlikely(err))
2518 goto out;
2519
2520 err = ext4_generic_delete_entry(handle, dir, de_del,
2521 bh, bh->b_data,
2522 dir->i_sb->s_blocksize, csum_size);
2523 if (err)
2524 goto out;
2525
2526 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2527 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2528 if (unlikely(err))
2529 goto out;
2530
2531 return 0;
2532 out:
2533 if (err != -ENOENT)
2534 ext4_std_error(dir->i_sb, err);
2535 return err;
2536 }
2537
2538 /*
2539 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2540 * since this indicates that nlinks count was previously 1 to avoid overflowing
2541 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2542 * that subdirectory link counts are not being maintained accurately.
2543 *
2544 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2545 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2546 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2547 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2548 */
2549 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2550 {
2551 inc_nlink(inode);
2552 if (is_dx(inode) &&
2553 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2554 set_nlink(inode, 1);
2555 }
2556
2557 /*
2558 * If a directory had nlink == 1, then we should let it be 1. This indicates
2559 * directory has >EXT4_LINK_MAX subdirs.
2560 */
2561 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2562 {
2563 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2564 drop_nlink(inode);
2565 }
2566
2567
2568 /*
2569 * Add non-directory inode to a directory. On success, the inode reference is
2570 * consumed by dentry is instantiation. This is also indicated by clearing of
2571 * *inodep pointer. On failure, the caller is responsible for dropping the
2572 * inode reference in the safe context.
2573 */
2574 static int ext4_add_nondir(handle_t *handle,
2575 struct dentry *dentry, struct inode **inodep)
2576 {
2577 struct inode *dir = d_inode(dentry->d_parent);
2578 struct inode *inode = *inodep;
2579 int err = ext4_add_entry(handle, dentry, inode);
2580 if (!err) {
2581 err = ext4_mark_inode_dirty(handle, inode);
2582 if (IS_DIRSYNC(dir))
2583 ext4_handle_sync(handle);
2584 d_instantiate_new(dentry, inode);
2585 *inodep = NULL;
2586 return err;
2587 }
2588 drop_nlink(inode);
2589 ext4_orphan_add(handle, inode);
2590 unlock_new_inode(inode);
2591 return err;
2592 }
2593
2594 /*
2595 * By the time this is called, we already have created
2596 * the directory cache entry for the new file, but it
2597 * is so far negative - it has no inode.
2598 *
2599 * If the create succeeds, we fill in the inode information
2600 * with d_instantiate().
2601 */
2602 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2603 bool excl)
2604 {
2605 handle_t *handle;
2606 struct inode *inode;
2607 int err, credits, retries = 0;
2608
2609 err = dquot_initialize(dir);
2610 if (err)
2611 return err;
2612
2613 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2614 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2615 retry:
2616 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2617 NULL, EXT4_HT_DIR, credits);
2618 handle = ext4_journal_current_handle();
2619 err = PTR_ERR(inode);
2620 if (!IS_ERR(inode)) {
2621 inode->i_op = &ext4_file_inode_operations;
2622 inode->i_fop = &ext4_file_operations;
2623 ext4_set_aops(inode);
2624 err = ext4_add_nondir(handle, dentry, &inode);
2625 }
2626 if (handle)
2627 ext4_journal_stop(handle);
2628 if (!IS_ERR_OR_NULL(inode))
2629 iput(inode);
2630 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2631 goto retry;
2632 return err;
2633 }
2634
2635 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2636 umode_t mode, dev_t rdev)
2637 {
2638 handle_t *handle;
2639 struct inode *inode;
2640 int err, credits, retries = 0;
2641
2642 err = dquot_initialize(dir);
2643 if (err)
2644 return err;
2645
2646 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2647 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2648 retry:
2649 inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2650 NULL, EXT4_HT_DIR, credits);
2651 handle = ext4_journal_current_handle();
2652 err = PTR_ERR(inode);
2653 if (!IS_ERR(inode)) {
2654 init_special_inode(inode, inode->i_mode, rdev);
2655 inode->i_op = &ext4_special_inode_operations;
2656 err = ext4_add_nondir(handle, dentry, &inode);
2657 }
2658 if (handle)
2659 ext4_journal_stop(handle);
2660 if (!IS_ERR_OR_NULL(inode))
2661 iput(inode);
2662 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2663 goto retry;
2664 return err;
2665 }
2666
2667 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2668 {
2669 handle_t *handle;
2670 struct inode *inode;
2671 int err, retries = 0;
2672
2673 err = dquot_initialize(dir);
2674 if (err)
2675 return err;
2676
2677 retry:
2678 inode = ext4_new_inode_start_handle(dir, mode,
2679 NULL, 0, NULL,
2680 EXT4_HT_DIR,
2681 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2682 4 + EXT4_XATTR_TRANS_BLOCKS);
2683 handle = ext4_journal_current_handle();
2684 err = PTR_ERR(inode);
2685 if (!IS_ERR(inode)) {
2686 inode->i_op = &ext4_file_inode_operations;
2687 inode->i_fop = &ext4_file_operations;
2688 ext4_set_aops(inode);
2689 d_tmpfile(dentry, inode);
2690 err = ext4_orphan_add(handle, inode);
2691 if (err)
2692 goto err_unlock_inode;
2693 mark_inode_dirty(inode);
2694 unlock_new_inode(inode);
2695 }
2696 if (handle)
2697 ext4_journal_stop(handle);
2698 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2699 goto retry;
2700 return err;
2701 err_unlock_inode:
2702 ext4_journal_stop(handle);
2703 unlock_new_inode(inode);
2704 return err;
2705 }
2706
2707 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2708 struct ext4_dir_entry_2 *de,
2709 int blocksize, int csum_size,
2710 unsigned int parent_ino, int dotdot_real_len)
2711 {
2712 de->inode = cpu_to_le32(inode->i_ino);
2713 de->name_len = 1;
2714 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2715 blocksize);
2716 strcpy(de->name, ".");
2717 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2718
2719 de = ext4_next_entry(de, blocksize);
2720 de->inode = cpu_to_le32(parent_ino);
2721 de->name_len = 2;
2722 if (!dotdot_real_len)
2723 de->rec_len = ext4_rec_len_to_disk(blocksize -
2724 (csum_size + EXT4_DIR_REC_LEN(1)),
2725 blocksize);
2726 else
2727 de->rec_len = ext4_rec_len_to_disk(
2728 EXT4_DIR_REC_LEN(de->name_len), blocksize);
2729 strcpy(de->name, "..");
2730 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2731
2732 return ext4_next_entry(de, blocksize);
2733 }
2734
2735 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2736 struct inode *inode)
2737 {
2738 struct buffer_head *dir_block = NULL;
2739 struct ext4_dir_entry_2 *de;
2740 ext4_lblk_t block = 0;
2741 unsigned int blocksize = dir->i_sb->s_blocksize;
2742 int csum_size = 0;
2743 int err;
2744
2745 if (ext4_has_metadata_csum(dir->i_sb))
2746 csum_size = sizeof(struct ext4_dir_entry_tail);
2747
2748 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2749 err = ext4_try_create_inline_dir(handle, dir, inode);
2750 if (err < 0 && err != -ENOSPC)
2751 goto out;
2752 if (!err)
2753 goto out;
2754 }
2755
2756 inode->i_size = 0;
2757 dir_block = ext4_append(handle, inode, &block);
2758 if (IS_ERR(dir_block))
2759 return PTR_ERR(dir_block);
2760 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2761 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2762 set_nlink(inode, 2);
2763 if (csum_size)
2764 ext4_initialize_dirent_tail(dir_block, blocksize);
2765
2766 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2767 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2768 if (err)
2769 goto out;
2770 set_buffer_verified(dir_block);
2771 out:
2772 brelse(dir_block);
2773 return err;
2774 }
2775
2776 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2777 {
2778 handle_t *handle;
2779 struct inode *inode;
2780 int err, err2 = 0, credits, retries = 0;
2781
2782 if (EXT4_DIR_LINK_MAX(dir))
2783 return -EMLINK;
2784
2785 err = dquot_initialize(dir);
2786 if (err)
2787 return err;
2788
2789 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2790 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2791 retry:
2792 inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2793 &dentry->d_name,
2794 0, NULL, EXT4_HT_DIR, credits);
2795 handle = ext4_journal_current_handle();
2796 err = PTR_ERR(inode);
2797 if (IS_ERR(inode))
2798 goto out_stop;
2799
2800 inode->i_op = &ext4_dir_inode_operations;
2801 inode->i_fop = &ext4_dir_operations;
2802 err = ext4_init_new_dir(handle, dir, inode);
2803 if (err)
2804 goto out_clear_inode;
2805 err = ext4_mark_inode_dirty(handle, inode);
2806 if (!err)
2807 err = ext4_add_entry(handle, dentry, inode);
2808 if (err) {
2809 out_clear_inode:
2810 clear_nlink(inode);
2811 ext4_orphan_add(handle, inode);
2812 unlock_new_inode(inode);
2813 err2 = ext4_mark_inode_dirty(handle, inode);
2814 if (unlikely(err2))
2815 err = err2;
2816 ext4_journal_stop(handle);
2817 iput(inode);
2818 goto out_retry;
2819 }
2820 ext4_inc_count(handle, dir);
2821 ext4_update_dx_flag(dir);
2822 err = ext4_mark_inode_dirty(handle, dir);
2823 if (err)
2824 goto out_clear_inode;
2825 d_instantiate_new(dentry, inode);
2826 if (IS_DIRSYNC(dir))
2827 ext4_handle_sync(handle);
2828
2829 out_stop:
2830 if (handle)
2831 ext4_journal_stop(handle);
2832 out_retry:
2833 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2834 goto retry;
2835 return err;
2836 }
2837
2838 /*
2839 * routine to check that the specified directory is empty (for rmdir)
2840 */
2841 bool ext4_empty_dir(struct inode *inode)
2842 {
2843 unsigned int offset;
2844 struct buffer_head *bh;
2845 struct ext4_dir_entry_2 *de;
2846 struct super_block *sb;
2847
2848 if (ext4_has_inline_data(inode)) {
2849 int has_inline_data = 1;
2850 int ret;
2851
2852 ret = empty_inline_dir(inode, &has_inline_data);
2853 if (has_inline_data)
2854 return ret;
2855 }
2856
2857 sb = inode->i_sb;
2858 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2859 EXT4_ERROR_INODE(inode, "invalid size");
2860 return true;
2861 }
2862 /* The first directory block must not be a hole,
2863 * so treat it as DIRENT_HTREE
2864 */
2865 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2866 if (IS_ERR(bh))
2867 return true;
2868
2869 de = (struct ext4_dir_entry_2 *) bh->b_data;
2870 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2871 0) ||
2872 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
2873 ext4_warning_inode(inode, "directory missing '.'");
2874 brelse(bh);
2875 return true;
2876 }
2877 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2878 de = ext4_next_entry(de, sb->s_blocksize);
2879 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
2880 offset) ||
2881 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
2882 ext4_warning_inode(inode, "directory missing '..'");
2883 brelse(bh);
2884 return true;
2885 }
2886 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2887 while (offset < inode->i_size) {
2888 if (!(offset & (sb->s_blocksize - 1))) {
2889 unsigned int lblock;
2890 brelse(bh);
2891 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2892 bh = ext4_read_dirblock(inode, lblock, EITHER);
2893 if (bh == NULL) {
2894 offset += sb->s_blocksize;
2895 continue;
2896 }
2897 if (IS_ERR(bh))
2898 return true;
2899 }
2900 de = (struct ext4_dir_entry_2 *) (bh->b_data +
2901 (offset & (sb->s_blocksize - 1)));
2902 if (ext4_check_dir_entry(inode, NULL, de, bh,
2903 bh->b_data, bh->b_size, offset)) {
2904 offset = (offset | (sb->s_blocksize - 1)) + 1;
2905 continue;
2906 }
2907 if (le32_to_cpu(de->inode)) {
2908 brelse(bh);
2909 return false;
2910 }
2911 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2912 }
2913 brelse(bh);
2914 return true;
2915 }
2916
2917 /*
2918 * ext4_orphan_add() links an unlinked or truncated inode into a list of
2919 * such inodes, starting at the superblock, in case we crash before the
2920 * file is closed/deleted, or in case the inode truncate spans multiple
2921 * transactions and the last transaction is not recovered after a crash.
2922 *
2923 * At filesystem recovery time, we walk this list deleting unlinked
2924 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2925 *
2926 * Orphan list manipulation functions must be called under i_mutex unless
2927 * we are just creating the inode or deleting it.
2928 */
2929 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2930 {
2931 struct super_block *sb = inode->i_sb;
2932 struct ext4_sb_info *sbi = EXT4_SB(sb);
2933 struct ext4_iloc iloc;
2934 int err = 0, rc;
2935 bool dirty = false;
2936
2937 if (!sbi->s_journal || is_bad_inode(inode))
2938 return 0;
2939
2940 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2941 !inode_is_locked(inode));
2942 /*
2943 * Exit early if inode already is on orphan list. This is a big speedup
2944 * since we don't have to contend on the global s_orphan_lock.
2945 */
2946 if (!list_empty(&EXT4_I(inode)->i_orphan))
2947 return 0;
2948
2949 /*
2950 * Orphan handling is only valid for files with data blocks
2951 * being truncated, or files being unlinked. Note that we either
2952 * hold i_mutex, or the inode can not be referenced from outside,
2953 * so i_nlink should not be bumped due to race
2954 */
2955 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2956 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2957
2958 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2959 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2960 if (err)
2961 goto out;
2962
2963 err = ext4_reserve_inode_write(handle, inode, &iloc);
2964 if (err)
2965 goto out;
2966
2967 mutex_lock(&sbi->s_orphan_lock);
2968 /*
2969 * Due to previous errors inode may be already a part of on-disk
2970 * orphan list. If so skip on-disk list modification.
2971 */
2972 if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2973 (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2974 /* Insert this inode at the head of the on-disk orphan list */
2975 NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2976 sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2977 dirty = true;
2978 }
2979 list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2980 mutex_unlock(&sbi->s_orphan_lock);
2981
2982 if (dirty) {
2983 err = ext4_handle_dirty_super(handle, sb);
2984 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2985 if (!err)
2986 err = rc;
2987 if (err) {
2988 /*
2989 * We have to remove inode from in-memory list if
2990 * addition to on disk orphan list failed. Stray orphan
2991 * list entries can cause panics at unmount time.
2992 */
2993 mutex_lock(&sbi->s_orphan_lock);
2994 list_del_init(&EXT4_I(inode)->i_orphan);
2995 mutex_unlock(&sbi->s_orphan_lock);
2996 }
2997 } else
2998 brelse(iloc.bh);
2999
3000 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
3001 jbd_debug(4, "orphan inode %lu will point to %d\n",
3002 inode->i_ino, NEXT_ORPHAN(inode));
3003 out:
3004 ext4_std_error(sb, err);
3005 return err;
3006 }
3007
3008 /*
3009 * ext4_orphan_del() removes an unlinked or truncated inode from the list
3010 * of such inodes stored on disk, because it is finally being cleaned up.
3011 */
3012 int ext4_orphan_del(handle_t *handle, struct inode *inode)
3013 {
3014 struct list_head *prev;
3015 struct ext4_inode_info *ei = EXT4_I(inode);
3016 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3017 __u32 ino_next;
3018 struct ext4_iloc iloc;
3019 int err = 0;
3020
3021 if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
3022 return 0;
3023
3024 WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
3025 !inode_is_locked(inode));
3026 /* Do this quick check before taking global s_orphan_lock. */
3027 if (list_empty(&ei->i_orphan))
3028 return 0;
3029
3030 if (handle) {
3031 /* Grab inode buffer early before taking global s_orphan_lock */
3032 err = ext4_reserve_inode_write(handle, inode, &iloc);
3033 }
3034
3035 mutex_lock(&sbi->s_orphan_lock);
3036 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
3037
3038 prev = ei->i_orphan.prev;
3039 list_del_init(&ei->i_orphan);
3040
3041 /* If we're on an error path, we may not have a valid
3042 * transaction handle with which to update the orphan list on
3043 * disk, but we still need to remove the inode from the linked
3044 * list in memory. */
3045 if (!handle || err) {
3046 mutex_unlock(&sbi->s_orphan_lock);
3047 goto out_err;
3048 }
3049
3050 ino_next = NEXT_ORPHAN(inode);
3051 if (prev == &sbi->s_orphan) {
3052 jbd_debug(4, "superblock will point to %u\n", ino_next);
3053 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3054 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3055 if (err) {
3056 mutex_unlock(&sbi->s_orphan_lock);
3057 goto out_brelse;
3058 }
3059 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3060 mutex_unlock(&sbi->s_orphan_lock);
3061 err = ext4_handle_dirty_super(handle, inode->i_sb);
3062 } else {
3063 struct ext4_iloc iloc2;
3064 struct inode *i_prev =
3065 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3066
3067 jbd_debug(4, "orphan inode %lu will point to %u\n",
3068 i_prev->i_ino, ino_next);
3069 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3070 if (err) {
3071 mutex_unlock(&sbi->s_orphan_lock);
3072 goto out_brelse;
3073 }
3074 NEXT_ORPHAN(i_prev) = ino_next;
3075 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3076 mutex_unlock(&sbi->s_orphan_lock);
3077 }
3078 if (err)
3079 goto out_brelse;
3080 NEXT_ORPHAN(inode) = 0;
3081 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3082 out_err:
3083 ext4_std_error(inode->i_sb, err);
3084 return err;
3085
3086 out_brelse:
3087 brelse(iloc.bh);
3088 goto out_err;
3089 }
3090
3091 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3092 {
3093 int retval;
3094 struct inode *inode;
3095 struct buffer_head *bh;
3096 struct ext4_dir_entry_2 *de;
3097 handle_t *handle = NULL;
3098
3099 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3100 return -EIO;
3101
3102 /* Initialize quotas before so that eventual writes go in
3103 * separate transaction */
3104 retval = dquot_initialize(dir);
3105 if (retval)
3106 return retval;
3107 retval = dquot_initialize(d_inode(dentry));
3108 if (retval)
3109 return retval;
3110
3111 retval = -ENOENT;
3112 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3113 if (IS_ERR(bh))
3114 return PTR_ERR(bh);
3115 if (!bh)
3116 goto end_rmdir;
3117
3118 inode = d_inode(dentry);
3119
3120 retval = -EFSCORRUPTED;
3121 if (le32_to_cpu(de->inode) != inode->i_ino)
3122 goto end_rmdir;
3123
3124 retval = -ENOTEMPTY;
3125 if (!ext4_empty_dir(inode))
3126 goto end_rmdir;
3127
3128 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3129 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3130 if (IS_ERR(handle)) {
3131 retval = PTR_ERR(handle);
3132 handle = NULL;
3133 goto end_rmdir;
3134 }
3135
3136 if (IS_DIRSYNC(dir))
3137 ext4_handle_sync(handle);
3138
3139 retval = ext4_delete_entry(handle, dir, de, bh);
3140 if (retval)
3141 goto end_rmdir;
3142 if (!EXT4_DIR_LINK_EMPTY(inode))
3143 ext4_warning_inode(inode,
3144 "empty directory '%.*s' has too many links (%u)",
3145 dentry->d_name.len, dentry->d_name.name,
3146 inode->i_nlink);
3147 inode_inc_iversion(inode);
3148 clear_nlink(inode);
3149 /* There's no need to set i_disksize: the fact that i_nlink is
3150 * zero will ensure that the right thing happens during any
3151 * recovery. */
3152 inode->i_size = 0;
3153 ext4_orphan_add(handle, inode);
3154 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3155 retval = ext4_mark_inode_dirty(handle, inode);
3156 if (retval)
3157 goto end_rmdir;
3158 ext4_dec_count(handle, dir);
3159 ext4_update_dx_flag(dir);
3160 retval = ext4_mark_inode_dirty(handle, dir);
3161
3162 #ifdef CONFIG_UNICODE
3163 /* VFS negative dentries are incompatible with Encoding and
3164 * Case-insensitiveness. Eventually we'll want avoid
3165 * invalidating the dentries here, alongside with returning the
3166 * negative dentries at ext4_lookup(), when it is better
3167 * supported by the VFS for the CI case.
3168 */
3169 if (IS_CASEFOLDED(dir))
3170 d_invalidate(dentry);
3171 #endif
3172
3173 end_rmdir:
3174 brelse(bh);
3175 if (handle)
3176 ext4_journal_stop(handle);
3177 return retval;
3178 }
3179
3180 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3181 {
3182 int retval;
3183 struct inode *inode;
3184 struct buffer_head *bh;
3185 struct ext4_dir_entry_2 *de;
3186 handle_t *handle = NULL;
3187
3188 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3189 return -EIO;
3190
3191 trace_ext4_unlink_enter(dir, dentry);
3192 /* Initialize quotas before so that eventual writes go
3193 * in separate transaction */
3194 retval = dquot_initialize(dir);
3195 if (retval)
3196 return retval;
3197 retval = dquot_initialize(d_inode(dentry));
3198 if (retval)
3199 return retval;
3200
3201 retval = -ENOENT;
3202 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3203 if (IS_ERR(bh))
3204 return PTR_ERR(bh);
3205 if (!bh)
3206 goto end_unlink;
3207
3208 inode = d_inode(dentry);
3209
3210 retval = -EFSCORRUPTED;
3211 if (le32_to_cpu(de->inode) != inode->i_ino)
3212 goto end_unlink;
3213
3214 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3215 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3216 if (IS_ERR(handle)) {
3217 retval = PTR_ERR(handle);
3218 handle = NULL;
3219 goto end_unlink;
3220 }
3221
3222 if (IS_DIRSYNC(dir))
3223 ext4_handle_sync(handle);
3224
3225 retval = ext4_delete_entry(handle, dir, de, bh);
3226 if (retval)
3227 goto end_unlink;
3228 dir->i_ctime = dir->i_mtime = current_time(dir);
3229 ext4_update_dx_flag(dir);
3230 retval = ext4_mark_inode_dirty(handle, dir);
3231 if (retval)
3232 goto end_unlink;
3233 if (inode->i_nlink == 0)
3234 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3235 dentry->d_name.len, dentry->d_name.name);
3236 else
3237 drop_nlink(inode);
3238 if (!inode->i_nlink)
3239 ext4_orphan_add(handle, inode);
3240 inode->i_ctime = current_time(inode);
3241 retval = ext4_mark_inode_dirty(handle, inode);
3242
3243 #ifdef CONFIG_UNICODE
3244 /* VFS negative dentries are incompatible with Encoding and
3245 * Case-insensitiveness. Eventually we'll want avoid
3246 * invalidating the dentries here, alongside with returning the
3247 * negative dentries at ext4_lookup(), when it is better
3248 * supported by the VFS for the CI case.
3249 */
3250 if (IS_CASEFOLDED(dir))
3251 d_invalidate(dentry);
3252 #endif
3253
3254 end_unlink:
3255 brelse(bh);
3256 if (handle)
3257 ext4_journal_stop(handle);
3258 trace_ext4_unlink_exit(dentry, retval);
3259 return retval;
3260 }
3261
3262 static int ext4_symlink(struct inode *dir,
3263 struct dentry *dentry, const char *symname)
3264 {
3265 handle_t *handle;
3266 struct inode *inode;
3267 int err, len = strlen(symname);
3268 int credits;
3269 struct fscrypt_str disk_link;
3270
3271 if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3272 return -EIO;
3273
3274 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3275 &disk_link);
3276 if (err)
3277 return err;
3278
3279 err = dquot_initialize(dir);
3280 if (err)
3281 return err;
3282
3283 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3284 /*
3285 * For non-fast symlinks, we just allocate inode and put it on
3286 * orphan list in the first transaction => we need bitmap,
3287 * group descriptor, sb, inode block, quota blocks, and
3288 * possibly selinux xattr blocks.
3289 */
3290 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3291 EXT4_XATTR_TRANS_BLOCKS;
3292 } else {
3293 /*
3294 * Fast symlink. We have to add entry to directory
3295 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3296 * allocate new inode (bitmap, group descriptor, inode block,
3297 * quota blocks, sb is already counted in previous macros).
3298 */
3299 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3300 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3301 }
3302
3303 inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3304 &dentry->d_name, 0, NULL,
3305 EXT4_HT_DIR, credits);
3306 handle = ext4_journal_current_handle();
3307 if (IS_ERR(inode)) {
3308 if (handle)
3309 ext4_journal_stop(handle);
3310 return PTR_ERR(inode);
3311 }
3312
3313 if (IS_ENCRYPTED(inode)) {
3314 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3315 if (err)
3316 goto err_drop_inode;
3317 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3318 }
3319
3320 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3321 if (!IS_ENCRYPTED(inode))
3322 inode->i_op = &ext4_symlink_inode_operations;
3323 inode_nohighmem(inode);
3324 ext4_set_aops(inode);
3325 /*
3326 * We cannot call page_symlink() with transaction started
3327 * because it calls into ext4_write_begin() which can wait
3328 * for transaction commit if we are running out of space
3329 * and thus we deadlock. So we have to stop transaction now
3330 * and restart it when symlink contents is written.
3331 *
3332 * To keep fs consistent in case of crash, we have to put inode
3333 * to orphan list in the mean time.
3334 */
3335 drop_nlink(inode);
3336 err = ext4_orphan_add(handle, inode);
3337 ext4_journal_stop(handle);
3338 handle = NULL;
3339 if (err)
3340 goto err_drop_inode;
3341 err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3342 if (err)
3343 goto err_drop_inode;
3344 /*
3345 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3346 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3347 */
3348 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3349 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3350 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3351 if (IS_ERR(handle)) {
3352 err = PTR_ERR(handle);
3353 handle = NULL;
3354 goto err_drop_inode;
3355 }
3356 set_nlink(inode, 1);
3357 err = ext4_orphan_del(handle, inode);
3358 if (err)
3359 goto err_drop_inode;
3360 } else {
3361 /* clear the extent format for fast symlink */
3362 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3363 if (!IS_ENCRYPTED(inode)) {
3364 inode->i_op = &ext4_fast_symlink_inode_operations;
3365 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3366 }
3367 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3368 disk_link.len);
3369 inode->i_size = disk_link.len - 1;
3370 }
3371 EXT4_I(inode)->i_disksize = inode->i_size;
3372 err = ext4_add_nondir(handle, dentry, &inode);
3373 if (handle)
3374 ext4_journal_stop(handle);
3375 if (inode)
3376 iput(inode);
3377 goto out_free_encrypted_link;
3378
3379 err_drop_inode:
3380 if (handle)
3381 ext4_journal_stop(handle);
3382 clear_nlink(inode);
3383 unlock_new_inode(inode);
3384 iput(inode);
3385 out_free_encrypted_link:
3386 if (disk_link.name != (unsigned char *)symname)
3387 kfree(disk_link.name);
3388 return err;
3389 }
3390
3391 static int ext4_link(struct dentry *old_dentry,
3392 struct inode *dir, struct dentry *dentry)
3393 {
3394 handle_t *handle;
3395 struct inode *inode = d_inode(old_dentry);
3396 int err, retries = 0;
3397
3398 if (inode->i_nlink >= EXT4_LINK_MAX)
3399 return -EMLINK;
3400
3401 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3402 if (err)
3403 return err;
3404
3405 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3406 (!projid_eq(EXT4_I(dir)->i_projid,
3407 EXT4_I(old_dentry->d_inode)->i_projid)))
3408 return -EXDEV;
3409
3410 err = dquot_initialize(dir);
3411 if (err)
3412 return err;
3413
3414 retry:
3415 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3416 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3417 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3418 if (IS_ERR(handle))
3419 return PTR_ERR(handle);
3420
3421 if (IS_DIRSYNC(dir))
3422 ext4_handle_sync(handle);
3423
3424 inode->i_ctime = current_time(inode);
3425 ext4_inc_count(handle, inode);
3426 ihold(inode);
3427
3428 err = ext4_add_entry(handle, dentry, inode);
3429 if (!err) {
3430 err = ext4_mark_inode_dirty(handle, inode);
3431 /* this can happen only for tmpfile being
3432 * linked the first time
3433 */
3434 if (inode->i_nlink == 1)
3435 ext4_orphan_del(handle, inode);
3436 d_instantiate(dentry, inode);
3437 } else {
3438 drop_nlink(inode);
3439 iput(inode);
3440 }
3441 ext4_journal_stop(handle);
3442 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3443 goto retry;
3444 return err;
3445 }
3446
3447
3448 /*
3449 * Try to find buffer head where contains the parent block.
3450 * It should be the inode block if it is inlined or the 1st block
3451 * if it is a normal dir.
3452 */
3453 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3454 struct inode *inode,
3455 int *retval,
3456 struct ext4_dir_entry_2 **parent_de,
3457 int *inlined)
3458 {
3459 struct buffer_head *bh;
3460
3461 if (!ext4_has_inline_data(inode)) {
3462 /* The first directory block must not be a hole, so
3463 * treat it as DIRENT_HTREE
3464 */
3465 bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3466 if (IS_ERR(bh)) {
3467 *retval = PTR_ERR(bh);
3468 return NULL;
3469 }
3470 *parent_de = ext4_next_entry(
3471 (struct ext4_dir_entry_2 *)bh->b_data,
3472 inode->i_sb->s_blocksize);
3473 return bh;
3474 }
3475
3476 *inlined = 1;
3477 return ext4_get_first_inline_block(inode, parent_de, retval);
3478 }
3479
3480 struct ext4_renament {
3481 struct inode *dir;
3482 struct dentry *dentry;
3483 struct inode *inode;
3484 bool is_dir;
3485 int dir_nlink_delta;
3486
3487 /* entry for "dentry" */
3488 struct buffer_head *bh;
3489 struct ext4_dir_entry_2 *de;
3490 int inlined;
3491
3492 /* entry for ".." in inode if it's a directory */
3493 struct buffer_head *dir_bh;
3494 struct ext4_dir_entry_2 *parent_de;
3495 int dir_inlined;
3496 };
3497
3498 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3499 {
3500 int retval;
3501
3502 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3503 &retval, &ent->parent_de,
3504 &ent->dir_inlined);
3505 if (!ent->dir_bh)
3506 return retval;
3507 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3508 return -EFSCORRUPTED;
3509 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3510 return ext4_journal_get_write_access(handle, ent->dir_bh);
3511 }
3512
3513 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3514 unsigned dir_ino)
3515 {
3516 int retval;
3517
3518 ent->parent_de->inode = cpu_to_le32(dir_ino);
3519 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3520 if (!ent->dir_inlined) {
3521 if (is_dx(ent->inode)) {
3522 retval = ext4_handle_dirty_dx_node(handle,
3523 ent->inode,
3524 ent->dir_bh);
3525 } else {
3526 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3527 ent->dir_bh);
3528 }
3529 } else {
3530 retval = ext4_mark_inode_dirty(handle, ent->inode);
3531 }
3532 if (retval) {
3533 ext4_std_error(ent->dir->i_sb, retval);
3534 return retval;
3535 }
3536 return 0;
3537 }
3538
3539 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3540 unsigned ino, unsigned file_type)
3541 {
3542 int retval, retval2;
3543
3544 BUFFER_TRACE(ent->bh, "get write access");
3545 retval = ext4_journal_get_write_access(handle, ent->bh);
3546 if (retval)
3547 return retval;
3548 ent->de->inode = cpu_to_le32(ino);
3549 if (ext4_has_feature_filetype(ent->dir->i_sb))
3550 ent->de->file_type = file_type;
3551 inode_inc_iversion(ent->dir);
3552 ent->dir->i_ctime = ent->dir->i_mtime =
3553 current_time(ent->dir);
3554 retval = ext4_mark_inode_dirty(handle, ent->dir);
3555 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3556 if (!ent->inlined) {
3557 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3558 if (unlikely(retval2)) {
3559 ext4_std_error(ent->dir->i_sb, retval2);
3560 return retval2;
3561 }
3562 }
3563 brelse(ent->bh);
3564 ent->bh = NULL;
3565
3566 return retval;
3567 }
3568
3569 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3570 const struct qstr *d_name)
3571 {
3572 int retval = -ENOENT;
3573 struct buffer_head *bh;
3574 struct ext4_dir_entry_2 *de;
3575
3576 bh = ext4_find_entry(dir, d_name, &de, NULL);
3577 if (IS_ERR(bh))
3578 return PTR_ERR(bh);
3579 if (bh) {
3580 retval = ext4_delete_entry(handle, dir, de, bh);
3581 brelse(bh);
3582 }
3583 return retval;
3584 }
3585
3586 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3587 int force_reread)
3588 {
3589 int retval;
3590 /*
3591 * ent->de could have moved from under us during htree split, so make
3592 * sure that we are deleting the right entry. We might also be pointing
3593 * to a stale entry in the unused part of ent->bh so just checking inum
3594 * and the name isn't enough.
3595 */
3596 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3597 ent->de->name_len != ent->dentry->d_name.len ||
3598 strncmp(ent->de->name, ent->dentry->d_name.name,
3599 ent->de->name_len) ||
3600 force_reread) {
3601 retval = ext4_find_delete_entry(handle, ent->dir,
3602 &ent->dentry->d_name);
3603 } else {
3604 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3605 if (retval == -ENOENT) {
3606 retval = ext4_find_delete_entry(handle, ent->dir,
3607 &ent->dentry->d_name);
3608 }
3609 }
3610
3611 if (retval) {
3612 ext4_warning_inode(ent->dir,
3613 "Deleting old file: nlink %d, error=%d",
3614 ent->dir->i_nlink, retval);
3615 }
3616 }
3617
3618 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3619 {
3620 if (ent->dir_nlink_delta) {
3621 if (ent->dir_nlink_delta == -1)
3622 ext4_dec_count(handle, ent->dir);
3623 else
3624 ext4_inc_count(handle, ent->dir);
3625 ext4_mark_inode_dirty(handle, ent->dir);
3626 }
3627 }
3628
3629 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3630 int credits, handle_t **h)
3631 {
3632 struct inode *wh;
3633 handle_t *handle;
3634 int retries = 0;
3635
3636 /*
3637 * for inode block, sb block, group summaries,
3638 * and inode bitmap
3639 */
3640 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3641 EXT4_XATTR_TRANS_BLOCKS + 4);
3642 retry:
3643 wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3644 &ent->dentry->d_name, 0, NULL,
3645 EXT4_HT_DIR, credits);
3646
3647 handle = ext4_journal_current_handle();
3648 if (IS_ERR(wh)) {
3649 if (handle)
3650 ext4_journal_stop(handle);
3651 if (PTR_ERR(wh) == -ENOSPC &&
3652 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3653 goto retry;
3654 } else {
3655 *h = handle;
3656 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3657 wh->i_op = &ext4_special_inode_operations;
3658 }
3659 return wh;
3660 }
3661
3662 /*
3663 * Anybody can rename anything with this: the permission checks are left to the
3664 * higher-level routines.
3665 *
3666 * n.b. old_{dentry,inode) refers to the source dentry/inode
3667 * while new_{dentry,inode) refers to the destination dentry/inode
3668 * This comes from rename(const char *oldpath, const char *newpath)
3669 */
3670 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3671 struct inode *new_dir, struct dentry *new_dentry,
3672 unsigned int flags)
3673 {
3674 handle_t *handle = NULL;
3675 struct ext4_renament old = {
3676 .dir = old_dir,
3677 .dentry = old_dentry,
3678 .inode = d_inode(old_dentry),
3679 };
3680 struct ext4_renament new = {
3681 .dir = new_dir,
3682 .dentry = new_dentry,
3683 .inode = d_inode(new_dentry),
3684 };
3685 int force_reread;
3686 int retval;
3687 struct inode *whiteout = NULL;
3688 int credits;
3689 u8 old_file_type;
3690
3691 if (new.inode && new.inode->i_nlink == 0) {
3692 EXT4_ERROR_INODE(new.inode,
3693 "target of rename is already freed");
3694 return -EFSCORRUPTED;
3695 }
3696
3697 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3698 (!projid_eq(EXT4_I(new_dir)->i_projid,
3699 EXT4_I(old_dentry->d_inode)->i_projid)))
3700 return -EXDEV;
3701
3702 retval = dquot_initialize(old.dir);
3703 if (retval)
3704 return retval;
3705 retval = dquot_initialize(new.dir);
3706 if (retval)
3707 return retval;
3708
3709 /* Initialize quotas before so that eventual writes go
3710 * in separate transaction */
3711 if (new.inode) {
3712 retval = dquot_initialize(new.inode);
3713 if (retval)
3714 return retval;
3715 }
3716
3717 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3718 if (IS_ERR(old.bh))
3719 return PTR_ERR(old.bh);
3720 /*
3721 * Check for inode number is _not_ due to possible IO errors.
3722 * We might rmdir the source, keep it as pwd of some process
3723 * and merrily kill the link to whatever was created under the
3724 * same name. Goodbye sticky bit ;-<
3725 */
3726 retval = -ENOENT;
3727 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3728 goto end_rename;
3729
3730 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3731 &new.de, &new.inlined);
3732 if (IS_ERR(new.bh)) {
3733 retval = PTR_ERR(new.bh);
3734 new.bh = NULL;
3735 goto end_rename;
3736 }
3737 if (new.bh) {
3738 if (!new.inode) {
3739 brelse(new.bh);
3740 new.bh = NULL;
3741 }
3742 }
3743 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3744 ext4_alloc_da_blocks(old.inode);
3745
3746 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3747 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3748 if (!(flags & RENAME_WHITEOUT)) {
3749 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3750 if (IS_ERR(handle)) {
3751 retval = PTR_ERR(handle);
3752 handle = NULL;
3753 goto end_rename;
3754 }
3755 } else {
3756 whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3757 if (IS_ERR(whiteout)) {
3758 retval = PTR_ERR(whiteout);
3759 whiteout = NULL;
3760 goto end_rename;
3761 }
3762 }
3763
3764 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3765 ext4_handle_sync(handle);
3766
3767 if (S_ISDIR(old.inode->i_mode)) {
3768 if (new.inode) {
3769 retval = -ENOTEMPTY;
3770 if (!ext4_empty_dir(new.inode))
3771 goto end_rename;
3772 } else {
3773 retval = -EMLINK;
3774 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3775 goto end_rename;
3776 }
3777 retval = ext4_rename_dir_prepare(handle, &old);
3778 if (retval)
3779 goto end_rename;
3780 }
3781 /*
3782 * If we're renaming a file within an inline_data dir and adding or
3783 * setting the new dirent causes a conversion from inline_data to
3784 * extents/blockmap, we need to force the dirent delete code to
3785 * re-read the directory, or else we end up trying to delete a dirent
3786 * from what is now the extent tree root (or a block map).
3787 */
3788 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3789 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3790
3791 old_file_type = old.de->file_type;
3792 if (whiteout) {
3793 /*
3794 * Do this before adding a new entry, so the old entry is sure
3795 * to be still pointing to the valid old entry.
3796 */
3797 retval = ext4_setent(handle, &old, whiteout->i_ino,
3798 EXT4_FT_CHRDEV);
3799 if (retval)
3800 goto end_rename;
3801 retval = ext4_mark_inode_dirty(handle, whiteout);
3802 if (unlikely(retval))
3803 goto end_rename;
3804 }
3805 if (!new.bh) {
3806 retval = ext4_add_entry(handle, new.dentry, old.inode);
3807 if (retval)
3808 goto end_rename;
3809 } else {
3810 retval = ext4_setent(handle, &new,
3811 old.inode->i_ino, old_file_type);
3812 if (retval)
3813 goto end_rename;
3814 }
3815 if (force_reread)
3816 force_reread = !ext4_test_inode_flag(new.dir,
3817 EXT4_INODE_INLINE_DATA);
3818
3819 /*
3820 * Like most other Unix systems, set the ctime for inodes on a
3821 * rename.
3822 */
3823 old.inode->i_ctime = current_time(old.inode);
3824 retval = ext4_mark_inode_dirty(handle, old.inode);
3825 if (unlikely(retval))
3826 goto end_rename;
3827
3828 if (!whiteout) {
3829 /*
3830 * ok, that's it
3831 */
3832 ext4_rename_delete(handle, &old, force_reread);
3833 }
3834
3835 if (new.inode) {
3836 ext4_dec_count(handle, new.inode);
3837 new.inode->i_ctime = current_time(new.inode);
3838 }
3839 old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3840 ext4_update_dx_flag(old.dir);
3841 if (old.dir_bh) {
3842 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3843 if (retval)
3844 goto end_rename;
3845
3846 ext4_dec_count(handle, old.dir);
3847 if (new.inode) {
3848 /* checked ext4_empty_dir above, can't have another
3849 * parent, ext4_dec_count() won't work for many-linked
3850 * dirs */
3851 clear_nlink(new.inode);
3852 } else {
3853 ext4_inc_count(handle, new.dir);
3854 ext4_update_dx_flag(new.dir);
3855 retval = ext4_mark_inode_dirty(handle, new.dir);
3856 if (unlikely(retval))
3857 goto end_rename;
3858 }
3859 }
3860 retval = ext4_mark_inode_dirty(handle, old.dir);
3861 if (unlikely(retval))
3862 goto end_rename;
3863 if (new.inode) {
3864 retval = ext4_mark_inode_dirty(handle, new.inode);
3865 if (unlikely(retval))
3866 goto end_rename;
3867 if (!new.inode->i_nlink)
3868 ext4_orphan_add(handle, new.inode);
3869 }
3870 retval = 0;
3871
3872 end_rename:
3873 brelse(old.dir_bh);
3874 brelse(old.bh);
3875 brelse(new.bh);
3876 if (whiteout) {
3877 if (retval)
3878 drop_nlink(whiteout);
3879 unlock_new_inode(whiteout);
3880 iput(whiteout);
3881 }
3882 if (handle)
3883 ext4_journal_stop(handle);
3884 return retval;
3885 }
3886
3887 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3888 struct inode *new_dir, struct dentry *new_dentry)
3889 {
3890 handle_t *handle = NULL;
3891 struct ext4_renament old = {
3892 .dir = old_dir,
3893 .dentry = old_dentry,
3894 .inode = d_inode(old_dentry),
3895 };
3896 struct ext4_renament new = {
3897 .dir = new_dir,
3898 .dentry = new_dentry,
3899 .inode = d_inode(new_dentry),
3900 };
3901 u8 new_file_type;
3902 int retval;
3903 struct timespec64 ctime;
3904
3905 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3906 !projid_eq(EXT4_I(new_dir)->i_projid,
3907 EXT4_I(old_dentry->d_inode)->i_projid)) ||
3908 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3909 !projid_eq(EXT4_I(old_dir)->i_projid,
3910 EXT4_I(new_dentry->d_inode)->i_projid)))
3911 return -EXDEV;
3912
3913 retval = dquot_initialize(old.dir);
3914 if (retval)
3915 return retval;
3916 retval = dquot_initialize(new.dir);
3917 if (retval)
3918 return retval;
3919
3920 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3921 &old.de, &old.inlined);
3922 if (IS_ERR(old.bh))
3923 return PTR_ERR(old.bh);
3924 /*
3925 * Check for inode number is _not_ due to possible IO errors.
3926 * We might rmdir the source, keep it as pwd of some process
3927 * and merrily kill the link to whatever was created under the
3928 * same name. Goodbye sticky bit ;-<
3929 */
3930 retval = -ENOENT;
3931 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3932 goto end_rename;
3933
3934 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3935 &new.de, &new.inlined);
3936 if (IS_ERR(new.bh)) {
3937 retval = PTR_ERR(new.bh);
3938 new.bh = NULL;
3939 goto end_rename;
3940 }
3941
3942 /* RENAME_EXCHANGE case: old *and* new must both exist */
3943 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3944 goto end_rename;
3945
3946 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3947 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3948 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3949 if (IS_ERR(handle)) {
3950 retval = PTR_ERR(handle);
3951 handle = NULL;
3952 goto end_rename;
3953 }
3954
3955 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3956 ext4_handle_sync(handle);
3957
3958 if (S_ISDIR(old.inode->i_mode)) {
3959 old.is_dir = true;
3960 retval = ext4_rename_dir_prepare(handle, &old);
3961 if (retval)
3962 goto end_rename;
3963 }
3964 if (S_ISDIR(new.inode->i_mode)) {
3965 new.is_dir = true;
3966 retval = ext4_rename_dir_prepare(handle, &new);
3967 if (retval)
3968 goto end_rename;
3969 }
3970
3971 /*
3972 * Other than the special case of overwriting a directory, parents'
3973 * nlink only needs to be modified if this is a cross directory rename.
3974 */
3975 if (old.dir != new.dir && old.is_dir != new.is_dir) {
3976 old.dir_nlink_delta = old.is_dir ? -1 : 1;
3977 new.dir_nlink_delta = -old.dir_nlink_delta;
3978 retval = -EMLINK;
3979 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3980 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3981 goto end_rename;
3982 }
3983
3984 new_file_type = new.de->file_type;
3985 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3986 if (retval)
3987 goto end_rename;
3988
3989 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3990 if (retval)
3991 goto end_rename;
3992
3993 /*
3994 * Like most other Unix systems, set the ctime for inodes on a
3995 * rename.
3996 */
3997 ctime = current_time(old.inode);
3998 old.inode->i_ctime = ctime;
3999 new.inode->i_ctime = ctime;
4000 retval = ext4_mark_inode_dirty(handle, old.inode);
4001 if (unlikely(retval))
4002 goto end_rename;
4003 retval = ext4_mark_inode_dirty(handle, new.inode);
4004 if (unlikely(retval))
4005 goto end_rename;
4006
4007 if (old.dir_bh) {
4008 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4009 if (retval)
4010 goto end_rename;
4011 }
4012 if (new.dir_bh) {
4013 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4014 if (retval)
4015 goto end_rename;
4016 }
4017 ext4_update_dir_count(handle, &old);
4018 ext4_update_dir_count(handle, &new);
4019 retval = 0;
4020
4021 end_rename:
4022 brelse(old.dir_bh);
4023 brelse(new.dir_bh);
4024 brelse(old.bh);
4025 brelse(new.bh);
4026 if (handle)
4027 ext4_journal_stop(handle);
4028 return retval;
4029 }
4030
4031 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
4032 struct inode *new_dir, struct dentry *new_dentry,
4033 unsigned int flags)
4034 {
4035 int err;
4036
4037 if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
4038 return -EIO;
4039
4040 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4041 return -EINVAL;
4042
4043 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4044 flags);
4045 if (err)
4046 return err;
4047
4048 if (flags & RENAME_EXCHANGE) {
4049 return ext4_cross_rename(old_dir, old_dentry,
4050 new_dir, new_dentry);
4051 }
4052
4053 return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4054 }
4055
4056 /*
4057 * directories can handle most operations...
4058 */
4059 const struct inode_operations ext4_dir_inode_operations = {
4060 .create = ext4_create,
4061 .lookup = ext4_lookup,
4062 .link = ext4_link,
4063 .unlink = ext4_unlink,
4064 .symlink = ext4_symlink,
4065 .mkdir = ext4_mkdir,
4066 .rmdir = ext4_rmdir,
4067 .mknod = ext4_mknod,
4068 .tmpfile = ext4_tmpfile,
4069 .rename = ext4_rename2,
4070 .setattr = ext4_setattr,
4071 .getattr = ext4_getattr,
4072 .listxattr = ext4_listxattr,
4073 .get_acl = ext4_get_acl,
4074 .set_acl = ext4_set_acl,
4075 .fiemap = ext4_fiemap,
4076 };
4077
4078 const struct inode_operations ext4_special_inode_operations = {
4079 .setattr = ext4_setattr,
4080 .getattr = ext4_getattr,
4081 .listxattr = ext4_listxattr,
4082 .get_acl = ext4_get_acl,
4083 .set_acl = ext4_set_acl,
4084 };
Linux with added FPC-III support