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 ext4_map_blocks map;
58 struct buffer_head *bh;
59 int err;
60
61 if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
62 ((inode->i_size >> 10) >=
63 EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
64 return ERR_PTR(-ENOSPC);
65
66 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
67 map.m_lblk = *block;
68 map.m_len = 1;
69
70 /*
71 * We're appending new directory block. Make sure the block is not
72 * allocated yet, otherwise we will end up corrupting the
73 * directory.
74 */
75 err = ext4_map_blocks(NULL, inode, &map, 0);
76 if (err < 0)
77 return ERR_PTR(err);
78 if (err) {
79 EXT4_ERROR_INODE(inode, "Logical block already allocated");
80 return ERR_PTR(-EFSCORRUPTED);
81 }
82
83 bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
84 if (IS_ERR(bh))
85 return bh;
86 inode->i_size += inode->i_sb->s_blocksize;
87 EXT4_I(inode)->i_disksize = inode->i_size;
88 err = ext4_mark_inode_dirty(handle, inode);
89 if (err)
90 goto out;
91 BUFFER_TRACE(bh, "get_write_access");
92 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
93 EXT4_JTR_NONE);
94 if (err)
95 goto out;
96 return bh;
97
98 out:
99 brelse(bh);
100 ext4_std_error(inode->i_sb, err);
101 return ERR_PTR(err);
102 }
103
104 static int ext4_dx_csum_verify(struct inode *inode,
105 struct ext4_dir_entry *dirent);
106
107 /*
108 * Hints to ext4_read_dirblock regarding whether we expect a directory
109 * block being read to be an index block, or a block containing
110 * directory entries (and if the latter, whether it was found via a
111 * logical block in an htree index block). This is used to control
112 * what sort of sanity checkinig ext4_read_dirblock() will do on the
113 * directory block read from the storage device. EITHER will means
114 * the caller doesn't know what kind of directory block will be read,
115 * so no specific verification will be done.
116 */
117 typedef enum {
118 EITHER, INDEX, DIRENT, DIRENT_HTREE
119 } dirblock_type_t;
120
121 #define ext4_read_dirblock(inode, block, type) \
122 __ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
123
124 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
125 ext4_lblk_t block,
126 dirblock_type_t type,
127 const char *func,
128 unsigned int line)
129 {
130 struct buffer_head *bh;
131 struct ext4_dir_entry *dirent;
132 int is_dx_block = 0;
133
134 if (block >= inode->i_size >> inode->i_blkbits) {
135 ext4_error_inode(inode, func, line, block,
136 "Attempting to read directory block (%u) that is past i_size (%llu)",
137 block, inode->i_size);
138 return ERR_PTR(-EFSCORRUPTED);
139 }
140
141 if (ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_EIO))
142 bh = ERR_PTR(-EIO);
143 else
144 bh = ext4_bread(NULL, inode, block, 0);
145 if (IS_ERR(bh)) {
146 __ext4_warning(inode->i_sb, func, line,
147 "inode #%lu: lblock %lu: comm %s: "
148 "error %ld reading directory block",
149 inode->i_ino, (unsigned long)block,
150 current->comm, PTR_ERR(bh));
151
152 return bh;
153 }
154 /* The first directory block must not be a hole. */
155 if (!bh && (type == INDEX || type == DIRENT_HTREE || block == 0)) {
156 ext4_error_inode(inode, func, line, block,
157 "Directory hole found for htree %s block %u",
158 (type == INDEX) ? "index" : "leaf", block);
159 return ERR_PTR(-EFSCORRUPTED);
160 }
161 if (!bh)
162 return NULL;
163 dirent = (struct ext4_dir_entry *) bh->b_data;
164 /* Determine whether or not we have an index block */
165 if (is_dx(inode)) {
166 if (block == 0)
167 is_dx_block = 1;
168 else if (ext4_rec_len_from_disk(dirent->rec_len,
169 inode->i_sb->s_blocksize) ==
170 inode->i_sb->s_blocksize)
171 is_dx_block = 1;
172 }
173 if (!is_dx_block && type == INDEX) {
174 ext4_error_inode(inode, func, line, block,
175 "directory leaf block found instead of index block");
176 brelse(bh);
177 return ERR_PTR(-EFSCORRUPTED);
178 }
179 if (!ext4_has_metadata_csum(inode->i_sb) ||
180 buffer_verified(bh))
181 return bh;
182
183 /*
184 * An empty leaf block can get mistaken for a index block; for
185 * this reason, we can only check the index checksum when the
186 * caller is sure it should be an index block.
187 */
188 if (is_dx_block && type == INDEX) {
189 if (ext4_dx_csum_verify(inode, dirent) &&
190 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
191 set_buffer_verified(bh);
192 else {
193 ext4_error_inode_err(inode, func, line, block,
194 EFSBADCRC,
195 "Directory index failed checksum");
196 brelse(bh);
197 return ERR_PTR(-EFSBADCRC);
198 }
199 }
200 if (!is_dx_block) {
201 if (ext4_dirblock_csum_verify(inode, bh) &&
202 !ext4_simulate_fail(inode->i_sb, EXT4_SIM_DIRBLOCK_CRC))
203 set_buffer_verified(bh);
204 else {
205 ext4_error_inode_err(inode, func, line, block,
206 EFSBADCRC,
207 "Directory block failed checksum");
208 brelse(bh);
209 return ERR_PTR(-EFSBADCRC);
210 }
211 }
212 return bh;
213 }
214
215 #ifdef DX_DEBUG
216 #define dxtrace(command) command
217 #else
218 #define dxtrace(command)
219 #endif
220
221 struct fake_dirent
222 {
223 __le32 inode;
224 __le16 rec_len;
225 u8 name_len;
226 u8 file_type;
227 };
228
229 struct dx_countlimit
230 {
231 __le16 limit;
232 __le16 count;
233 };
234
235 struct dx_entry
236 {
237 __le32 hash;
238 __le32 block;
239 };
240
241 /*
242 * dx_root_info is laid out so that if it should somehow get overlaid by a
243 * dirent the two low bits of the hash version will be zero. Therefore, the
244 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
245 */
246
247 struct dx_root
248 {
249 struct fake_dirent dot;
250 char dot_name[4];
251 struct fake_dirent dotdot;
252 char dotdot_name[4];
253 struct dx_root_info
254 {
255 __le32 reserved_zero;
256 u8 hash_version;
257 u8 info_length; /* 8 */
258 u8 indirect_levels;
259 u8 unused_flags;
260 }
261 info;
262 struct dx_entry entries[];
263 };
264
265 struct dx_node
266 {
267 struct fake_dirent fake;
268 struct dx_entry entries[];
269 };
270
271
272 struct dx_frame
273 {
274 struct buffer_head *bh;
275 struct dx_entry *entries;
276 struct dx_entry *at;
277 };
278
279 struct dx_map_entry
280 {
281 u32 hash;
282 u16 offs;
283 u16 size;
284 };
285
286 /*
287 * This goes at the end of each htree block.
288 */
289 struct dx_tail {
290 u32 dt_reserved;
291 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
292 };
293
294 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
295 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
296 static inline unsigned dx_get_hash(struct dx_entry *entry);
297 static void dx_set_hash(struct dx_entry *entry, unsigned value);
298 static unsigned dx_get_count(struct dx_entry *entries);
299 static unsigned dx_get_limit(struct dx_entry *entries);
300 static void dx_set_count(struct dx_entry *entries, unsigned value);
301 static void dx_set_limit(struct dx_entry *entries, unsigned value);
302 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
303 static unsigned dx_node_limit(struct inode *dir);
304 static struct dx_frame *dx_probe(struct ext4_filename *fname,
305 struct inode *dir,
306 struct dx_hash_info *hinfo,
307 struct dx_frame *frame);
308 static void dx_release(struct dx_frame *frames);
309 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
310 struct dx_hash_info *hinfo,
311 struct dx_map_entry *map_tail);
312 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
313 static struct ext4_dir_entry_2 *dx_move_dirents(struct inode *dir, char *from,
314 char *to, struct dx_map_entry *offsets,
315 int count, unsigned int blocksize);
316 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
317 unsigned int blocksize);
318 static void dx_insert_block(struct dx_frame *frame,
319 u32 hash, ext4_lblk_t block);
320 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
321 struct dx_frame *frame,
322 struct dx_frame *frames,
323 __u32 *start_hash);
324 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
325 struct ext4_filename *fname,
326 struct ext4_dir_entry_2 **res_dir);
327 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
328 struct inode *dir, struct inode *inode);
329
330 /* checksumming functions */
331 void ext4_initialize_dirent_tail(struct buffer_head *bh,
332 unsigned int blocksize)
333 {
334 struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
335
336 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
337 t->det_rec_len = ext4_rec_len_to_disk(
338 sizeof(struct ext4_dir_entry_tail), blocksize);
339 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
340 }
341
342 /* Walk through a dirent block to find a checksum "dirent" at the tail */
343 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
344 struct buffer_head *bh)
345 {
346 struct ext4_dir_entry_tail *t;
347 int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
348
349 #ifdef PARANOID
350 struct ext4_dir_entry *d, *top;
351
352 d = (struct ext4_dir_entry *)bh->b_data;
353 top = (struct ext4_dir_entry *)(bh->b_data +
354 (blocksize - sizeof(struct ext4_dir_entry_tail)));
355 while (d < top && ext4_rec_len_from_disk(d->rec_len, blocksize))
356 d = (struct ext4_dir_entry *)(((void *)d) +
357 ext4_rec_len_from_disk(d->rec_len, blocksize));
358
359 if (d != top)
360 return NULL;
361
362 t = (struct ext4_dir_entry_tail *)d;
363 #else
364 t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
365 #endif
366
367 if (t->det_reserved_zero1 ||
368 (ext4_rec_len_from_disk(t->det_rec_len, blocksize) !=
369 sizeof(struct ext4_dir_entry_tail)) ||
370 t->det_reserved_zero2 ||
371 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
372 return NULL;
373
374 return t;
375 }
376
377 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
378 {
379 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
380 struct ext4_inode_info *ei = EXT4_I(inode);
381 __u32 csum;
382
383 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
384 return cpu_to_le32(csum);
385 }
386
387 #define warn_no_space_for_csum(inode) \
388 __warn_no_space_for_csum((inode), __func__, __LINE__)
389
390 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
391 unsigned int line)
392 {
393 __ext4_warning_inode(inode, func, line,
394 "No space for directory leaf checksum. Please run e2fsck -D.");
395 }
396
397 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
398 {
399 struct ext4_dir_entry_tail *t;
400
401 if (!ext4_has_metadata_csum(inode->i_sb))
402 return 1;
403
404 t = get_dirent_tail(inode, bh);
405 if (!t) {
406 warn_no_space_for_csum(inode);
407 return 0;
408 }
409
410 if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
411 (char *)t - bh->b_data))
412 return 0;
413
414 return 1;
415 }
416
417 static void ext4_dirblock_csum_set(struct inode *inode,
418 struct buffer_head *bh)
419 {
420 struct ext4_dir_entry_tail *t;
421
422 if (!ext4_has_metadata_csum(inode->i_sb))
423 return;
424
425 t = get_dirent_tail(inode, bh);
426 if (!t) {
427 warn_no_space_for_csum(inode);
428 return;
429 }
430
431 t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
432 (char *)t - bh->b_data);
433 }
434
435 int ext4_handle_dirty_dirblock(handle_t *handle,
436 struct inode *inode,
437 struct buffer_head *bh)
438 {
439 ext4_dirblock_csum_set(inode, bh);
440 return ext4_handle_dirty_metadata(handle, inode, bh);
441 }
442
443 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
444 struct ext4_dir_entry *dirent,
445 int *offset)
446 {
447 struct ext4_dir_entry *dp;
448 struct dx_root_info *root;
449 int count_offset;
450 int blocksize = EXT4_BLOCK_SIZE(inode->i_sb);
451 unsigned int rlen = ext4_rec_len_from_disk(dirent->rec_len, blocksize);
452
453 if (rlen == blocksize)
454 count_offset = 8;
455 else if (rlen == 12) {
456 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
457 if (ext4_rec_len_from_disk(dp->rec_len, blocksize) != blocksize - 12)
458 return NULL;
459 root = (struct dx_root_info *)(((void *)dp + 12));
460 if (root->reserved_zero ||
461 root->info_length != sizeof(struct dx_root_info))
462 return NULL;
463 count_offset = 32;
464 } else
465 return NULL;
466
467 if (offset)
468 *offset = count_offset;
469 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
470 }
471
472 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
473 int count_offset, int count, struct dx_tail *t)
474 {
475 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
476 struct ext4_inode_info *ei = EXT4_I(inode);
477 __u32 csum;
478 int size;
479 __u32 dummy_csum = 0;
480 int offset = offsetof(struct dx_tail, dt_checksum);
481
482 size = count_offset + (count * sizeof(struct dx_entry));
483 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
484 csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
485 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
486
487 return cpu_to_le32(csum);
488 }
489
490 static int ext4_dx_csum_verify(struct inode *inode,
491 struct ext4_dir_entry *dirent)
492 {
493 struct dx_countlimit *c;
494 struct dx_tail *t;
495 int count_offset, limit, count;
496
497 if (!ext4_has_metadata_csum(inode->i_sb))
498 return 1;
499
500 c = get_dx_countlimit(inode, dirent, &count_offset);
501 if (!c) {
502 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
503 return 0;
504 }
505 limit = le16_to_cpu(c->limit);
506 count = le16_to_cpu(c->count);
507 if (count_offset + (limit * sizeof(struct dx_entry)) >
508 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
509 warn_no_space_for_csum(inode);
510 return 0;
511 }
512 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
513
514 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
515 count, t))
516 return 0;
517 return 1;
518 }
519
520 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
521 {
522 struct dx_countlimit *c;
523 struct dx_tail *t;
524 int count_offset, limit, count;
525
526 if (!ext4_has_metadata_csum(inode->i_sb))
527 return;
528
529 c = get_dx_countlimit(inode, dirent, &count_offset);
530 if (!c) {
531 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
532 return;
533 }
534 limit = le16_to_cpu(c->limit);
535 count = le16_to_cpu(c->count);
536 if (count_offset + (limit * sizeof(struct dx_entry)) >
537 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
538 warn_no_space_for_csum(inode);
539 return;
540 }
541 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
542
543 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
544 }
545
546 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
547 struct inode *inode,
548 struct buffer_head *bh)
549 {
550 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
551 return ext4_handle_dirty_metadata(handle, inode, bh);
552 }
553
554 /*
555 * p is at least 6 bytes before the end of page
556 */
557 static inline struct ext4_dir_entry_2 *
558 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
559 {
560 return (struct ext4_dir_entry_2 *)((char *)p +
561 ext4_rec_len_from_disk(p->rec_len, blocksize));
562 }
563
564 /*
565 * Future: use high four bits of block for coalesce-on-delete flags
566 * Mask them off for now.
567 */
568
569 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
570 {
571 return le32_to_cpu(entry->block) & 0x0fffffff;
572 }
573
574 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
575 {
576 entry->block = cpu_to_le32(value);
577 }
578
579 static inline unsigned dx_get_hash(struct dx_entry *entry)
580 {
581 return le32_to_cpu(entry->hash);
582 }
583
584 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
585 {
586 entry->hash = cpu_to_le32(value);
587 }
588
589 static inline unsigned dx_get_count(struct dx_entry *entries)
590 {
591 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
592 }
593
594 static inline unsigned dx_get_limit(struct dx_entry *entries)
595 {
596 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
597 }
598
599 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
600 {
601 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
602 }
603
604 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
605 {
606 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
607 }
608
609 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
610 {
611 unsigned int entry_space = dir->i_sb->s_blocksize -
612 ext4_dir_rec_len(1, NULL) -
613 ext4_dir_rec_len(2, NULL) - infosize;
614
615 if (ext4_has_metadata_csum(dir->i_sb))
616 entry_space -= sizeof(struct dx_tail);
617 return entry_space / sizeof(struct dx_entry);
618 }
619
620 static inline unsigned dx_node_limit(struct inode *dir)
621 {
622 unsigned int entry_space = dir->i_sb->s_blocksize -
623 ext4_dir_rec_len(0, dir);
624
625 if (ext4_has_metadata_csum(dir->i_sb))
626 entry_space -= sizeof(struct dx_tail);
627 return entry_space / sizeof(struct dx_entry);
628 }
629
630 /*
631 * Debug
632 */
633 #ifdef DX_DEBUG
634 static void dx_show_index(char * label, struct dx_entry *entries)
635 {
636 int i, n = dx_get_count (entries);
637 printk(KERN_DEBUG "%s index", label);
638 for (i = 0; i < n; i++) {
639 printk(KERN_CONT " %x->%lu",
640 i ? dx_get_hash(entries + i) : 0,
641 (unsigned long)dx_get_block(entries + i));
642 }
643 printk(KERN_CONT "\n");
644 }
645
646 struct stats
647 {
648 unsigned names;
649 unsigned space;
650 unsigned bcount;
651 };
652
653 static struct stats dx_show_leaf(struct inode *dir,
654 struct dx_hash_info *hinfo,
655 struct ext4_dir_entry_2 *de,
656 int size, int show_names)
657 {
658 unsigned names = 0, space = 0;
659 char *base = (char *) de;
660 struct dx_hash_info h = *hinfo;
661
662 printk("names: ");
663 while ((char *) de < base + size)
664 {
665 if (de->inode)
666 {
667 if (show_names)
668 {
669 #ifdef CONFIG_FS_ENCRYPTION
670 int len;
671 char *name;
672 struct fscrypt_str fname_crypto_str =
673 FSTR_INIT(NULL, 0);
674 int res = 0;
675
676 name = de->name;
677 len = de->name_len;
678 if (!IS_ENCRYPTED(dir)) {
679 /* Directory is not encrypted */
680 (void) ext4fs_dirhash(dir, de->name,
681 de->name_len, &h);
682 printk("%*.s:(U)%x.%u ", len,
683 name, h.hash,
684 (unsigned) ((char *) de
685 - base));
686 } else {
687 struct fscrypt_str de_name =
688 FSTR_INIT(name, len);
689
690 /* Directory is encrypted */
691 res = fscrypt_fname_alloc_buffer(
692 len, &fname_crypto_str);
693 if (res)
694 printk(KERN_WARNING "Error "
695 "allocating crypto "
696 "buffer--skipping "
697 "crypto\n");
698 res = fscrypt_fname_disk_to_usr(dir,
699 0, 0, &de_name,
700 &fname_crypto_str);
701 if (res) {
702 printk(KERN_WARNING "Error "
703 "converting filename "
704 "from disk to usr"
705 "\n");
706 name = "??";
707 len = 2;
708 } else {
709 name = fname_crypto_str.name;
710 len = fname_crypto_str.len;
711 }
712 if (IS_CASEFOLDED(dir))
713 h.hash = EXT4_DIRENT_HASH(de);
714 else
715 (void) ext4fs_dirhash(dir,
716 de->name,
717 de->name_len, &h);
718 printk("%*.s:(E)%x.%u ", len, name,
719 h.hash, (unsigned) ((char *) de
720 - base));
721 fscrypt_fname_free_buffer(
722 &fname_crypto_str);
723 }
724 #else
725 int len = de->name_len;
726 char *name = de->name;
727 (void) ext4fs_dirhash(dir, de->name,
728 de->name_len, &h);
729 printk("%*.s:%x.%u ", len, name, h.hash,
730 (unsigned) ((char *) de - base));
731 #endif
732 }
733 space += ext4_dir_rec_len(de->name_len, dir);
734 names++;
735 }
736 de = ext4_next_entry(de, size);
737 }
738 printk(KERN_CONT "(%i)\n", names);
739 return (struct stats) { names, space, 1 };
740 }
741
742 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
743 struct dx_entry *entries, int levels)
744 {
745 unsigned blocksize = dir->i_sb->s_blocksize;
746 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
747 unsigned bcount = 0;
748 struct buffer_head *bh;
749 printk("%i indexed blocks...\n", count);
750 for (i = 0; i < count; i++, entries++)
751 {
752 ext4_lblk_t block = dx_get_block(entries);
753 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
754 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
755 struct stats stats;
756 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
757 bh = ext4_bread(NULL,dir, block, 0);
758 if (!bh || IS_ERR(bh))
759 continue;
760 stats = levels?
761 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
762 dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
763 bh->b_data, blocksize, 0);
764 names += stats.names;
765 space += stats.space;
766 bcount += stats.bcount;
767 brelse(bh);
768 }
769 if (bcount)
770 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
771 levels ? "" : " ", names, space/bcount,
772 (space/bcount)*100/blocksize);
773 return (struct stats) { names, space, bcount};
774 }
775
776 /*
777 * Linear search cross check
778 */
779 static inline void htree_rep_invariant_check(struct dx_entry *at,
780 struct dx_entry *target,
781 u32 hash, unsigned int n)
782 {
783 while (n--) {
784 dxtrace(printk(KERN_CONT ","));
785 if (dx_get_hash(++at) > hash) {
786 at--;
787 break;
788 }
789 }
790 ASSERT(at == target - 1);
791 }
792 #else /* DX_DEBUG */
793 static inline void htree_rep_invariant_check(struct dx_entry *at,
794 struct dx_entry *target,
795 u32 hash, unsigned int n)
796 {
797 }
798 #endif /* DX_DEBUG */
799
800 /*
801 * Probe for a directory leaf block to search.
802 *
803 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
804 * error in the directory index, and the caller should fall back to
805 * searching the directory normally. The callers of dx_probe **MUST**
806 * check for this error code, and make sure it never gets reflected
807 * back to userspace.
808 */
809 static struct dx_frame *
810 dx_probe(struct ext4_filename *fname, struct inode *dir,
811 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
812 {
813 unsigned count, indirect, level, i;
814 struct dx_entry *at, *entries, *p, *q, *m;
815 struct dx_root *root;
816 struct dx_frame *frame = frame_in;
817 struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
818 u32 hash;
819 ext4_lblk_t block;
820 ext4_lblk_t blocks[EXT4_HTREE_LEVEL];
821
822 memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
823 frame->bh = ext4_read_dirblock(dir, 0, INDEX);
824 if (IS_ERR(frame->bh))
825 return (struct dx_frame *) frame->bh;
826
827 root = (struct dx_root *) frame->bh->b_data;
828 if (root->info.hash_version != DX_HASH_TEA &&
829 root->info.hash_version != DX_HASH_HALF_MD4 &&
830 root->info.hash_version != DX_HASH_LEGACY &&
831 root->info.hash_version != DX_HASH_SIPHASH) {
832 ext4_warning_inode(dir, "Unrecognised inode hash code %u",
833 root->info.hash_version);
834 goto fail;
835 }
836 if (ext4_hash_in_dirent(dir)) {
837 if (root->info.hash_version != DX_HASH_SIPHASH) {
838 ext4_warning_inode(dir,
839 "Hash in dirent, but hash is not SIPHASH");
840 goto fail;
841 }
842 } else {
843 if (root->info.hash_version == DX_HASH_SIPHASH) {
844 ext4_warning_inode(dir,
845 "Hash code is SIPHASH, but hash not in dirent");
846 goto fail;
847 }
848 }
849 if (fname)
850 hinfo = &fname->hinfo;
851 hinfo->hash_version = root->info.hash_version;
852 if (hinfo->hash_version <= DX_HASH_TEA)
853 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
854 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
855 /* hash is already computed for encrypted casefolded directory */
856 if (fname && fname_name(fname) &&
857 !(IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir))) {
858 int ret = ext4fs_dirhash(dir, fname_name(fname),
859 fname_len(fname), hinfo);
860 if (ret < 0) {
861 ret_err = ERR_PTR(ret);
862 goto fail;
863 }
864 }
865 hash = hinfo->hash;
866
867 if (root->info.unused_flags & 1) {
868 ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
869 root->info.unused_flags);
870 goto fail;
871 }
872
873 indirect = root->info.indirect_levels;
874 if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
875 ext4_warning(dir->i_sb,
876 "Directory (ino: %lu) htree depth %#06x exceed"
877 "supported value", dir->i_ino,
878 ext4_dir_htree_level(dir->i_sb));
879 if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
880 ext4_warning(dir->i_sb, "Enable large directory "
881 "feature to access it");
882 }
883 goto fail;
884 }
885
886 entries = (struct dx_entry *)(((char *)&root->info) +
887 root->info.info_length);
888
889 if (dx_get_limit(entries) != dx_root_limit(dir,
890 root->info.info_length)) {
891 ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
892 dx_get_limit(entries),
893 dx_root_limit(dir, root->info.info_length));
894 goto fail;
895 }
896
897 dxtrace(printk("Look up %x", hash));
898 level = 0;
899 blocks[0] = 0;
900 while (1) {
901 count = dx_get_count(entries);
902 if (!count || count > dx_get_limit(entries)) {
903 ext4_warning_inode(dir,
904 "dx entry: count %u beyond limit %u",
905 count, dx_get_limit(entries));
906 goto fail;
907 }
908
909 p = entries + 1;
910 q = entries + count - 1;
911 while (p <= q) {
912 m = p + (q - p) / 2;
913 dxtrace(printk(KERN_CONT "."));
914 if (dx_get_hash(m) > hash)
915 q = m - 1;
916 else
917 p = m + 1;
918 }
919
920 htree_rep_invariant_check(entries, p, hash, count - 1);
921
922 at = p - 1;
923 dxtrace(printk(KERN_CONT " %x->%u\n",
924 at == entries ? 0 : dx_get_hash(at),
925 dx_get_block(at)));
926 frame->entries = entries;
927 frame->at = at;
928
929 block = dx_get_block(at);
930 for (i = 0; i <= level; i++) {
931 if (blocks[i] == block) {
932 ext4_warning_inode(dir,
933 "dx entry: tree cycle block %u points back to block %u",
934 blocks[level], block);
935 goto fail;
936 }
937 }
938 if (++level > indirect)
939 return frame;
940 blocks[level] = block;
941 frame++;
942 frame->bh = ext4_read_dirblock(dir, block, INDEX);
943 if (IS_ERR(frame->bh)) {
944 ret_err = (struct dx_frame *) frame->bh;
945 frame->bh = NULL;
946 goto fail;
947 }
948
949 entries = ((struct dx_node *) frame->bh->b_data)->entries;
950
951 if (dx_get_limit(entries) != dx_node_limit(dir)) {
952 ext4_warning_inode(dir,
953 "dx entry: limit %u != node limit %u",
954 dx_get_limit(entries), dx_node_limit(dir));
955 goto fail;
956 }
957 }
958 fail:
959 while (frame >= frame_in) {
960 brelse(frame->bh);
961 frame--;
962 }
963
964 if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
965 ext4_warning_inode(dir,
966 "Corrupt directory, running e2fsck is recommended");
967 return ret_err;
968 }
969
970 static void dx_release(struct dx_frame *frames)
971 {
972 struct dx_root_info *info;
973 int i;
974 unsigned int indirect_levels;
975
976 if (frames[0].bh == NULL)
977 return;
978
979 info = &((struct dx_root *)frames[0].bh->b_data)->info;
980 /* save local copy, "info" may be freed after brelse() */
981 indirect_levels = info->indirect_levels;
982 for (i = 0; i <= indirect_levels; i++) {
983 if (frames[i].bh == NULL)
984 break;
985 brelse(frames[i].bh);
986 frames[i].bh = NULL;
987 }
988 }
989
990 /*
991 * This function increments the frame pointer to search the next leaf
992 * block, and reads in the necessary intervening nodes if the search
993 * should be necessary. Whether or not the search is necessary is
994 * controlled by the hash parameter. If the hash value is even, then
995 * the search is only continued if the next block starts with that
996 * hash value. This is used if we are searching for a specific file.
997 *
998 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
999 *
1000 * This function returns 1 if the caller should continue to search,
1001 * or 0 if it should not. If there is an error reading one of the
1002 * index blocks, it will a negative error code.
1003 *
1004 * If start_hash is non-null, it will be filled in with the starting
1005 * hash of the next page.
1006 */
1007 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
1008 struct dx_frame *frame,
1009 struct dx_frame *frames,
1010 __u32 *start_hash)
1011 {
1012 struct dx_frame *p;
1013 struct buffer_head *bh;
1014 int num_frames = 0;
1015 __u32 bhash;
1016
1017 p = frame;
1018 /*
1019 * Find the next leaf page by incrementing the frame pointer.
1020 * If we run out of entries in the interior node, loop around and
1021 * increment pointer in the parent node. When we break out of
1022 * this loop, num_frames indicates the number of interior
1023 * nodes need to be read.
1024 */
1025 while (1) {
1026 if (++(p->at) < p->entries + dx_get_count(p->entries))
1027 break;
1028 if (p == frames)
1029 return 0;
1030 num_frames++;
1031 p--;
1032 }
1033
1034 /*
1035 * If the hash is 1, then continue only if the next page has a
1036 * continuation hash of any value. This is used for readdir
1037 * handling. Otherwise, check to see if the hash matches the
1038 * desired continuation hash. If it doesn't, return since
1039 * there's no point to read in the successive index pages.
1040 */
1041 bhash = dx_get_hash(p->at);
1042 if (start_hash)
1043 *start_hash = bhash;
1044 if ((hash & 1) == 0) {
1045 if ((bhash & ~1) != hash)
1046 return 0;
1047 }
1048 /*
1049 * If the hash is HASH_NB_ALWAYS, we always go to the next
1050 * block so no check is necessary
1051 */
1052 while (num_frames--) {
1053 bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
1054 if (IS_ERR(bh))
1055 return PTR_ERR(bh);
1056 p++;
1057 brelse(p->bh);
1058 p->bh = bh;
1059 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
1060 }
1061 return 1;
1062 }
1063
1064
1065 /*
1066 * This function fills a red-black tree with information from a
1067 * directory block. It returns the number directory entries loaded
1068 * into the tree. If there is an error it is returned in err.
1069 */
1070 static int htree_dirblock_to_tree(struct file *dir_file,
1071 struct inode *dir, ext4_lblk_t block,
1072 struct dx_hash_info *hinfo,
1073 __u32 start_hash, __u32 start_minor_hash)
1074 {
1075 struct buffer_head *bh;
1076 struct ext4_dir_entry_2 *de, *top;
1077 int err = 0, count = 0;
1078 struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
1079 int csum = ext4_has_metadata_csum(dir->i_sb);
1080
1081 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
1082 (unsigned long)block));
1083 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1084 if (IS_ERR(bh))
1085 return PTR_ERR(bh);
1086
1087 de = (struct ext4_dir_entry_2 *) bh->b_data;
1088 /* csum entries are not larger in the casefolded encrypted case */
1089 top = (struct ext4_dir_entry_2 *) ((char *) de +
1090 dir->i_sb->s_blocksize -
1091 ext4_dir_rec_len(0,
1092 csum ? NULL : dir));
1093 /* Check if the directory is encrypted */
1094 if (IS_ENCRYPTED(dir)) {
1095 err = fscrypt_prepare_readdir(dir);
1096 if (err < 0) {
1097 brelse(bh);
1098 return err;
1099 }
1100 err = fscrypt_fname_alloc_buffer(EXT4_NAME_LEN,
1101 &fname_crypto_str);
1102 if (err < 0) {
1103 brelse(bh);
1104 return err;
1105 }
1106 }
1107
1108 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1109 if (ext4_check_dir_entry(dir, NULL, de, bh,
1110 bh->b_data, bh->b_size,
1111 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1112 + ((char *)de - bh->b_data))) {
1113 /* silently ignore the rest of the block */
1114 break;
1115 }
1116 if (ext4_hash_in_dirent(dir)) {
1117 if (de->name_len && de->inode) {
1118 hinfo->hash = EXT4_DIRENT_HASH(de);
1119 hinfo->minor_hash = EXT4_DIRENT_MINOR_HASH(de);
1120 } else {
1121 hinfo->hash = 0;
1122 hinfo->minor_hash = 0;
1123 }
1124 } else {
1125 err = ext4fs_dirhash(dir, de->name,
1126 de->name_len, hinfo);
1127 if (err < 0) {
1128 count = err;
1129 goto errout;
1130 }
1131 }
1132 if ((hinfo->hash < start_hash) ||
1133 ((hinfo->hash == start_hash) &&
1134 (hinfo->minor_hash < start_minor_hash)))
1135 continue;
1136 if (de->inode == 0)
1137 continue;
1138 if (!IS_ENCRYPTED(dir)) {
1139 tmp_str.name = de->name;
1140 tmp_str.len = de->name_len;
1141 err = ext4_htree_store_dirent(dir_file,
1142 hinfo->hash, hinfo->minor_hash, de,
1143 &tmp_str);
1144 } else {
1145 int save_len = fname_crypto_str.len;
1146 struct fscrypt_str de_name = FSTR_INIT(de->name,
1147 de->name_len);
1148
1149 /* Directory is encrypted */
1150 err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1151 hinfo->minor_hash, &de_name,
1152 &fname_crypto_str);
1153 if (err) {
1154 count = err;
1155 goto errout;
1156 }
1157 err = ext4_htree_store_dirent(dir_file,
1158 hinfo->hash, hinfo->minor_hash, de,
1159 &fname_crypto_str);
1160 fname_crypto_str.len = save_len;
1161 }
1162 if (err != 0) {
1163 count = err;
1164 goto errout;
1165 }
1166 count++;
1167 }
1168 errout:
1169 brelse(bh);
1170 fscrypt_fname_free_buffer(&fname_crypto_str);
1171 return count;
1172 }
1173
1174
1175 /*
1176 * This function fills a red-black tree with information from a
1177 * directory. We start scanning the directory in hash order, starting
1178 * at start_hash and start_minor_hash.
1179 *
1180 * This function returns the number of entries inserted into the tree,
1181 * or a negative error code.
1182 */
1183 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1184 __u32 start_minor_hash, __u32 *next_hash)
1185 {
1186 struct dx_hash_info hinfo;
1187 struct ext4_dir_entry_2 *de;
1188 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1189 struct inode *dir;
1190 ext4_lblk_t block;
1191 int count = 0;
1192 int ret, err;
1193 __u32 hashval;
1194 struct fscrypt_str tmp_str;
1195
1196 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1197 start_hash, start_minor_hash));
1198 dir = file_inode(dir_file);
1199 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1200 if (ext4_hash_in_dirent(dir))
1201 hinfo.hash_version = DX_HASH_SIPHASH;
1202 else
1203 hinfo.hash_version =
1204 EXT4_SB(dir->i_sb)->s_def_hash_version;
1205 if (hinfo.hash_version <= DX_HASH_TEA)
1206 hinfo.hash_version +=
1207 EXT4_SB(dir->i_sb)->s_hash_unsigned;
1208 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1209 if (ext4_has_inline_data(dir)) {
1210 int has_inline_data = 1;
1211 count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1212 &hinfo, start_hash,
1213 start_minor_hash,
1214 &has_inline_data);
1215 if (has_inline_data) {
1216 *next_hash = ~0;
1217 return count;
1218 }
1219 }
1220 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1221 start_hash, start_minor_hash);
1222 *next_hash = ~0;
1223 return count;
1224 }
1225 hinfo.hash = start_hash;
1226 hinfo.minor_hash = 0;
1227 frame = dx_probe(NULL, dir, &hinfo, frames);
1228 if (IS_ERR(frame))
1229 return PTR_ERR(frame);
1230
1231 /* Add '.' and '..' from the htree header */
1232 if (!start_hash && !start_minor_hash) {
1233 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1234 tmp_str.name = de->name;
1235 tmp_str.len = de->name_len;
1236 err = ext4_htree_store_dirent(dir_file, 0, 0,
1237 de, &tmp_str);
1238 if (err != 0)
1239 goto errout;
1240 count++;
1241 }
1242 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1243 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1244 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1245 tmp_str.name = de->name;
1246 tmp_str.len = de->name_len;
1247 err = ext4_htree_store_dirent(dir_file, 2, 0,
1248 de, &tmp_str);
1249 if (err != 0)
1250 goto errout;
1251 count++;
1252 }
1253
1254 while (1) {
1255 if (fatal_signal_pending(current)) {
1256 err = -ERESTARTSYS;
1257 goto errout;
1258 }
1259 cond_resched();
1260 block = dx_get_block(frame->at);
1261 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1262 start_hash, start_minor_hash);
1263 if (ret < 0) {
1264 err = ret;
1265 goto errout;
1266 }
1267 count += ret;
1268 hashval = ~0;
1269 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1270 frame, frames, &hashval);
1271 *next_hash = hashval;
1272 if (ret < 0) {
1273 err = ret;
1274 goto errout;
1275 }
1276 /*
1277 * Stop if: (a) there are no more entries, or
1278 * (b) we have inserted at least one entry and the
1279 * next hash value is not a continuation
1280 */
1281 if ((ret == 0) ||
1282 (count && ((hashval & 1) == 0)))
1283 break;
1284 }
1285 dx_release(frames);
1286 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1287 "next hash: %x\n", count, *next_hash));
1288 return count;
1289 errout:
1290 dx_release(frames);
1291 return (err);
1292 }
1293
1294 static inline int search_dirblock(struct buffer_head *bh,
1295 struct inode *dir,
1296 struct ext4_filename *fname,
1297 unsigned int offset,
1298 struct ext4_dir_entry_2 **res_dir)
1299 {
1300 return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1301 fname, offset, res_dir);
1302 }
1303
1304 /*
1305 * Directory block splitting, compacting
1306 */
1307
1308 /*
1309 * Create map of hash values, offsets, and sizes, stored at end of block.
1310 * Returns number of entries mapped.
1311 */
1312 static int dx_make_map(struct inode *dir, struct buffer_head *bh,
1313 struct dx_hash_info *hinfo,
1314 struct dx_map_entry *map_tail)
1315 {
1316 int count = 0;
1317 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)bh->b_data;
1318 unsigned int buflen = bh->b_size;
1319 char *base = bh->b_data;
1320 struct dx_hash_info h = *hinfo;
1321 int blocksize = EXT4_BLOCK_SIZE(dir->i_sb);
1322
1323 if (ext4_has_metadata_csum(dir->i_sb))
1324 buflen -= sizeof(struct ext4_dir_entry_tail);
1325
1326 while ((char *) de < base + buflen) {
1327 if (ext4_check_dir_entry(dir, NULL, de, bh, base, buflen,
1328 ((char *)de) - base))
1329 return -EFSCORRUPTED;
1330 if (de->name_len && de->inode) {
1331 if (ext4_hash_in_dirent(dir))
1332 h.hash = EXT4_DIRENT_HASH(de);
1333 else {
1334 int err = ext4fs_dirhash(dir, de->name,
1335 de->name_len, &h);
1336 if (err < 0)
1337 return err;
1338 }
1339 map_tail--;
1340 map_tail->hash = h.hash;
1341 map_tail->offs = ((char *) de - base)>>2;
1342 map_tail->size = ext4_rec_len_from_disk(de->rec_len,
1343 blocksize);
1344 count++;
1345 cond_resched();
1346 }
1347 de = ext4_next_entry(de, blocksize);
1348 }
1349 return count;
1350 }
1351
1352 /* Sort map by hash value */
1353 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1354 {
1355 struct dx_map_entry *p, *q, *top = map + count - 1;
1356 int more;
1357 /* Combsort until bubble sort doesn't suck */
1358 while (count > 2) {
1359 count = count*10/13;
1360 if (count - 9 < 2) /* 9, 10 -> 11 */
1361 count = 11;
1362 for (p = top, q = p - count; q >= map; p--, q--)
1363 if (p->hash < q->hash)
1364 swap(*p, *q);
1365 }
1366 /* Garden variety bubble sort */
1367 do {
1368 more = 0;
1369 q = top;
1370 while (q-- > map) {
1371 if (q[1].hash >= q[0].hash)
1372 continue;
1373 swap(*(q+1), *q);
1374 more = 1;
1375 }
1376 } while(more);
1377 }
1378
1379 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1380 {
1381 struct dx_entry *entries = frame->entries;
1382 struct dx_entry *old = frame->at, *new = old + 1;
1383 int count = dx_get_count(entries);
1384
1385 ASSERT(count < dx_get_limit(entries));
1386 ASSERT(old < entries + count);
1387 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1388 dx_set_hash(new, hash);
1389 dx_set_block(new, block);
1390 dx_set_count(entries, count + 1);
1391 }
1392
1393 #if IS_ENABLED(CONFIG_UNICODE)
1394 int ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1395 struct ext4_filename *name)
1396 {
1397 struct qstr *cf_name = &name->cf_name;
1398 unsigned char *buf;
1399 struct dx_hash_info *hinfo = &name->hinfo;
1400 int len;
1401
1402 if (!IS_CASEFOLDED(dir) ||
1403 (IS_ENCRYPTED(dir) && !fscrypt_has_encryption_key(dir))) {
1404 cf_name->name = NULL;
1405 return 0;
1406 }
1407
1408 buf = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1409 if (!buf)
1410 return -ENOMEM;
1411
1412 len = utf8_casefold(dir->i_sb->s_encoding, iname, buf, EXT4_NAME_LEN);
1413 if (len <= 0) {
1414 kfree(buf);
1415 buf = NULL;
1416 }
1417 cf_name->name = buf;
1418 cf_name->len = (unsigned) len;
1419
1420 if (!IS_ENCRYPTED(dir))
1421 return 0;
1422
1423 hinfo->hash_version = DX_HASH_SIPHASH;
1424 hinfo->seed = NULL;
1425 if (cf_name->name)
1426 return ext4fs_dirhash(dir, cf_name->name, cf_name->len, hinfo);
1427 else
1428 return ext4fs_dirhash(dir, iname->name, iname->len, hinfo);
1429 }
1430 #endif
1431
1432 /*
1433 * Test whether a directory entry matches the filename being searched for.
1434 *
1435 * Return: %true if the directory entry matches, otherwise %false.
1436 */
1437 static bool ext4_match(struct inode *parent,
1438 const struct ext4_filename *fname,
1439 struct ext4_dir_entry_2 *de)
1440 {
1441 struct fscrypt_name f;
1442
1443 if (!de->inode)
1444 return false;
1445
1446 f.usr_fname = fname->usr_fname;
1447 f.disk_name = fname->disk_name;
1448 #ifdef CONFIG_FS_ENCRYPTION
1449 f.crypto_buf = fname->crypto_buf;
1450 #endif
1451
1452 #if IS_ENABLED(CONFIG_UNICODE)
1453 if (IS_CASEFOLDED(parent) &&
1454 (!IS_ENCRYPTED(parent) || fscrypt_has_encryption_key(parent))) {
1455 /*
1456 * Just checking IS_ENCRYPTED(parent) below is not
1457 * sufficient to decide whether one can use the hash for
1458 * skipping the string comparison, because the key might
1459 * have been added right after
1460 * ext4_fname_setup_ci_filename(). In this case, a hash
1461 * mismatch will be a false negative. Therefore, make
1462 * sure cf_name was properly initialized before
1463 * considering the calculated hash.
1464 */
1465 if (IS_ENCRYPTED(parent) && fname->cf_name.name &&
1466 (fname->hinfo.hash != EXT4_DIRENT_HASH(de) ||
1467 fname->hinfo.minor_hash != EXT4_DIRENT_MINOR_HASH(de)))
1468 return false;
1469 /*
1470 * Treat comparison errors as not a match. The
1471 * only case where it happens is on a disk
1472 * corruption or ENOMEM.
1473 */
1474
1475 return generic_ci_match(parent, fname->usr_fname,
1476 &fname->cf_name, de->name,
1477 de->name_len) > 0;
1478 }
1479 #endif
1480
1481 return fscrypt_match_name(&f, de->name, de->name_len);
1482 }
1483
1484 /*
1485 * Returns 0 if not found, -1 on failure, and 1 on success
1486 */
1487 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1488 struct inode *dir, struct ext4_filename *fname,
1489 unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1490 {
1491 struct ext4_dir_entry_2 * de;
1492 char * dlimit;
1493 int de_len;
1494
1495 de = (struct ext4_dir_entry_2 *)search_buf;
1496 dlimit = search_buf + buf_size;
1497 while ((char *) de < dlimit - EXT4_BASE_DIR_LEN) {
1498 /* this code is executed quadratically often */
1499 /* do minimal checking `by hand' */
1500 if (de->name + de->name_len <= dlimit &&
1501 ext4_match(dir, fname, de)) {
1502 /* found a match - just to be sure, do
1503 * a full check */
1504 if (ext4_check_dir_entry(dir, NULL, de, bh, search_buf,
1505 buf_size, offset))
1506 return -1;
1507 *res_dir = de;
1508 return 1;
1509 }
1510 /* prevent looping on a bad block */
1511 de_len = ext4_rec_len_from_disk(de->rec_len,
1512 dir->i_sb->s_blocksize);
1513 if (de_len <= 0)
1514 return -1;
1515 offset += de_len;
1516 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1517 }
1518 return 0;
1519 }
1520
1521 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1522 struct ext4_dir_entry *de)
1523 {
1524 struct super_block *sb = dir->i_sb;
1525
1526 if (!is_dx(dir))
1527 return 0;
1528 if (block == 0)
1529 return 1;
1530 if (de->inode == 0 &&
1531 ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1532 sb->s_blocksize)
1533 return 1;
1534 return 0;
1535 }
1536
1537 /*
1538 * __ext4_find_entry()
1539 *
1540 * finds an entry in the specified directory with the wanted name. It
1541 * returns the cache buffer in which the entry was found, and the entry
1542 * itself (as a parameter - res_dir). It does NOT read the inode of the
1543 * entry - you'll have to do that yourself if you want to.
1544 *
1545 * The returned buffer_head has ->b_count elevated. The caller is expected
1546 * to brelse() it when appropriate.
1547 */
1548 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1549 struct ext4_filename *fname,
1550 struct ext4_dir_entry_2 **res_dir,
1551 int *inlined)
1552 {
1553 struct super_block *sb;
1554 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1555 struct buffer_head *bh, *ret = NULL;
1556 ext4_lblk_t start, block;
1557 const u8 *name = fname->usr_fname->name;
1558 size_t ra_max = 0; /* Number of bh's in the readahead
1559 buffer, bh_use[] */
1560 size_t ra_ptr = 0; /* Current index into readahead
1561 buffer */
1562 ext4_lblk_t nblocks;
1563 int i, namelen, retval;
1564
1565 *res_dir = NULL;
1566 sb = dir->i_sb;
1567 namelen = fname->usr_fname->len;
1568 if (namelen > EXT4_NAME_LEN)
1569 return NULL;
1570
1571 if (ext4_has_inline_data(dir)) {
1572 int has_inline_data = 1;
1573 ret = ext4_find_inline_entry(dir, fname, res_dir,
1574 &has_inline_data);
1575 if (inlined)
1576 *inlined = has_inline_data;
1577 if (has_inline_data)
1578 goto cleanup_and_exit;
1579 }
1580
1581 if ((namelen <= 2) && (name[0] == '.') &&
1582 (name[1] == '.' || name[1] == '\0')) {
1583 /*
1584 * "." or ".." will only be in the first block
1585 * NFS may look up ".."; "." should be handled by the VFS
1586 */
1587 block = start = 0;
1588 nblocks = 1;
1589 goto restart;
1590 }
1591 if (is_dx(dir)) {
1592 ret = ext4_dx_find_entry(dir, fname, res_dir);
1593 /*
1594 * On success, or if the error was file not found,
1595 * return. Otherwise, fall back to doing a search the
1596 * old fashioned way.
1597 */
1598 if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1599 goto cleanup_and_exit;
1600 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1601 "falling back\n"));
1602 ret = NULL;
1603 }
1604 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1605 if (!nblocks) {
1606 ret = NULL;
1607 goto cleanup_and_exit;
1608 }
1609 start = EXT4_I(dir)->i_dir_start_lookup;
1610 if (start >= nblocks)
1611 start = 0;
1612 block = start;
1613 restart:
1614 do {
1615 /*
1616 * We deal with the read-ahead logic here.
1617 */
1618 cond_resched();
1619 if (ra_ptr >= ra_max) {
1620 /* Refill the readahead buffer */
1621 ra_ptr = 0;
1622 if (block < start)
1623 ra_max = start - block;
1624 else
1625 ra_max = nblocks - block;
1626 ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1627 retval = ext4_bread_batch(dir, block, ra_max,
1628 false /* wait */, bh_use);
1629 if (retval) {
1630 ret = ERR_PTR(retval);
1631 ra_max = 0;
1632 goto cleanup_and_exit;
1633 }
1634 }
1635 if ((bh = bh_use[ra_ptr++]) == NULL)
1636 goto next;
1637 wait_on_buffer(bh);
1638 if (!buffer_uptodate(bh)) {
1639 EXT4_ERROR_INODE_ERR(dir, EIO,
1640 "reading directory lblock %lu",
1641 (unsigned long) block);
1642 brelse(bh);
1643 ret = ERR_PTR(-EIO);
1644 goto cleanup_and_exit;
1645 }
1646 if (!buffer_verified(bh) &&
1647 !is_dx_internal_node(dir, block,
1648 (struct ext4_dir_entry *)bh->b_data) &&
1649 !ext4_dirblock_csum_verify(dir, bh)) {
1650 EXT4_ERROR_INODE_ERR(dir, EFSBADCRC,
1651 "checksumming directory "
1652 "block %lu", (unsigned long)block);
1653 brelse(bh);
1654 ret = ERR_PTR(-EFSBADCRC);
1655 goto cleanup_and_exit;
1656 }
1657 set_buffer_verified(bh);
1658 i = search_dirblock(bh, dir, fname,
1659 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1660 if (i == 1) {
1661 EXT4_I(dir)->i_dir_start_lookup = block;
1662 ret = bh;
1663 goto cleanup_and_exit;
1664 } else {
1665 brelse(bh);
1666 if (i < 0)
1667 goto cleanup_and_exit;
1668 }
1669 next:
1670 if (++block >= nblocks)
1671 block = 0;
1672 } while (block != start);
1673
1674 /*
1675 * If the directory has grown while we were searching, then
1676 * search the last part of the directory before giving up.
1677 */
1678 block = nblocks;
1679 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1680 if (block < nblocks) {
1681 start = 0;
1682 goto restart;
1683 }
1684
1685 cleanup_and_exit:
1686 /* Clean up the read-ahead blocks */
1687 for (; ra_ptr < ra_max; ra_ptr++)
1688 brelse(bh_use[ra_ptr]);
1689 return ret;
1690 }
1691
1692 static struct buffer_head *ext4_find_entry(struct inode *dir,
1693 const struct qstr *d_name,
1694 struct ext4_dir_entry_2 **res_dir,
1695 int *inlined)
1696 {
1697 int err;
1698 struct ext4_filename fname;
1699 struct buffer_head *bh;
1700
1701 err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1702 if (err == -ENOENT)
1703 return NULL;
1704 if (err)
1705 return ERR_PTR(err);
1706
1707 bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1708
1709 ext4_fname_free_filename(&fname);
1710 return bh;
1711 }
1712
1713 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1714 struct dentry *dentry,
1715 struct ext4_dir_entry_2 **res_dir)
1716 {
1717 int err;
1718 struct ext4_filename fname;
1719 struct buffer_head *bh;
1720
1721 err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1722 if (err == -ENOENT)
1723 return NULL;
1724 if (err)
1725 return ERR_PTR(err);
1726
1727 bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1728
1729 ext4_fname_free_filename(&fname);
1730 return bh;
1731 }
1732
1733 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1734 struct ext4_filename *fname,
1735 struct ext4_dir_entry_2 **res_dir)
1736 {
1737 struct super_block * sb = dir->i_sb;
1738 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1739 struct buffer_head *bh;
1740 ext4_lblk_t block;
1741 int retval;
1742
1743 #ifdef CONFIG_FS_ENCRYPTION
1744 *res_dir = NULL;
1745 #endif
1746 frame = dx_probe(fname, dir, NULL, frames);
1747 if (IS_ERR(frame))
1748 return (struct buffer_head *) frame;
1749 do {
1750 block = dx_get_block(frame->at);
1751 bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1752 if (IS_ERR(bh))
1753 goto errout;
1754
1755 retval = search_dirblock(bh, dir, fname,
1756 block << EXT4_BLOCK_SIZE_BITS(sb),
1757 res_dir);
1758 if (retval == 1)
1759 goto success;
1760 brelse(bh);
1761 if (retval == -1) {
1762 bh = ERR_PTR(ERR_BAD_DX_DIR);
1763 goto errout;
1764 }
1765
1766 /* Check to see if we should continue to search */
1767 retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1768 frames, NULL);
1769 if (retval < 0) {
1770 ext4_warning_inode(dir,
1771 "error %d reading directory index block",
1772 retval);
1773 bh = ERR_PTR(retval);
1774 goto errout;
1775 }
1776 } while (retval == 1);
1777
1778 bh = NULL;
1779 errout:
1780 dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1781 success:
1782 dx_release(frames);
1783 return bh;
1784 }
1785
1786 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1787 {
1788 struct inode *inode;
1789 struct ext4_dir_entry_2 *de;
1790 struct buffer_head *bh;
1791
1792 if (dentry->d_name.len > EXT4_NAME_LEN)
1793 return ERR_PTR(-ENAMETOOLONG);
1794
1795 bh = ext4_lookup_entry(dir, dentry, &de);
1796 if (IS_ERR(bh))
1797 return ERR_CAST(bh);
1798 inode = NULL;
1799 if (bh) {
1800 __u32 ino = le32_to_cpu(de->inode);
1801 brelse(bh);
1802 if (!ext4_valid_inum(dir->i_sb, ino)) {
1803 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1804 return ERR_PTR(-EFSCORRUPTED);
1805 }
1806 if (unlikely(ino == dir->i_ino)) {
1807 EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1808 dentry);
1809 return ERR_PTR(-EFSCORRUPTED);
1810 }
1811 inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1812 if (inode == ERR_PTR(-ESTALE)) {
1813 EXT4_ERROR_INODE(dir,
1814 "deleted inode referenced: %u",
1815 ino);
1816 return ERR_PTR(-EFSCORRUPTED);
1817 }
1818 if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1819 (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1820 !fscrypt_has_permitted_context(dir, inode)) {
1821 ext4_warning(inode->i_sb,
1822 "Inconsistent encryption contexts: %lu/%lu",
1823 dir->i_ino, inode->i_ino);
1824 iput(inode);
1825 return ERR_PTR(-EPERM);
1826 }
1827 }
1828
1829 if (IS_ENABLED(CONFIG_UNICODE) && !inode && IS_CASEFOLDED(dir)) {
1830 /* Eventually we want to call d_add_ci(dentry, NULL)
1831 * for negative dentries in the encoding case as
1832 * well. For now, prevent the negative dentry
1833 * from being cached.
1834 */
1835 return NULL;
1836 }
1837
1838 return d_splice_alias(inode, dentry);
1839 }
1840
1841
1842 struct dentry *ext4_get_parent(struct dentry *child)
1843 {
1844 __u32 ino;
1845 struct ext4_dir_entry_2 * de;
1846 struct buffer_head *bh;
1847
1848 bh = ext4_find_entry(d_inode(child), &dotdot_name, &de, NULL);
1849 if (IS_ERR(bh))
1850 return ERR_CAST(bh);
1851 if (!bh)
1852 return ERR_PTR(-ENOENT);
1853 ino = le32_to_cpu(de->inode);
1854 brelse(bh);
1855
1856 if (!ext4_valid_inum(child->d_sb, ino)) {
1857 EXT4_ERROR_INODE(d_inode(child),
1858 "bad parent inode number: %u", ino);
1859 return ERR_PTR(-EFSCORRUPTED);
1860 }
1861
1862 return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1863 }
1864
1865 /*
1866 * Move count entries from end of map between two memory locations.
1867 * Returns pointer to last entry moved.
1868 */
1869 static struct ext4_dir_entry_2 *
1870 dx_move_dirents(struct inode *dir, char *from, char *to,
1871 struct dx_map_entry *map, int count,
1872 unsigned blocksize)
1873 {
1874 unsigned rec_len = 0;
1875
1876 while (count--) {
1877 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1878 (from + (map->offs<<2));
1879 rec_len = ext4_dir_rec_len(de->name_len, dir);
1880
1881 memcpy (to, de, rec_len);
1882 ((struct ext4_dir_entry_2 *) to)->rec_len =
1883 ext4_rec_len_to_disk(rec_len, blocksize);
1884
1885 /* wipe dir_entry excluding the rec_len field */
1886 de->inode = 0;
1887 memset(&de->name_len, 0, ext4_rec_len_from_disk(de->rec_len,
1888 blocksize) -
1889 offsetof(struct ext4_dir_entry_2,
1890 name_len));
1891
1892 map++;
1893 to += rec_len;
1894 }
1895 return (struct ext4_dir_entry_2 *) (to - rec_len);
1896 }
1897
1898 /*
1899 * Compact each dir entry in the range to the minimal rec_len.
1900 * Returns pointer to last entry in range.
1901 */
1902 static struct ext4_dir_entry_2 *dx_pack_dirents(struct inode *dir, char *base,
1903 unsigned int blocksize)
1904 {
1905 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1906 unsigned rec_len = 0;
1907
1908 prev = to = de;
1909 while ((char*)de < base + blocksize) {
1910 next = ext4_next_entry(de, blocksize);
1911 if (de->inode && de->name_len) {
1912 rec_len = ext4_dir_rec_len(de->name_len, dir);
1913 if (de > to)
1914 memmove(to, de, rec_len);
1915 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1916 prev = to;
1917 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1918 }
1919 de = next;
1920 }
1921 return prev;
1922 }
1923
1924 /*
1925 * Split a full leaf block to make room for a new dir entry.
1926 * Allocate a new block, and move entries so that they are approx. equally full.
1927 * Returns pointer to de in block into which the new entry will be inserted.
1928 */
1929 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1930 struct buffer_head **bh,struct dx_frame *frame,
1931 struct dx_hash_info *hinfo)
1932 {
1933 unsigned blocksize = dir->i_sb->s_blocksize;
1934 unsigned continued;
1935 int count;
1936 struct buffer_head *bh2;
1937 ext4_lblk_t newblock;
1938 u32 hash2;
1939 struct dx_map_entry *map;
1940 char *data1 = (*bh)->b_data, *data2;
1941 unsigned split, move, size;
1942 struct ext4_dir_entry_2 *de = NULL, *de2;
1943 int csum_size = 0;
1944 int err = 0, i;
1945
1946 if (ext4_has_metadata_csum(dir->i_sb))
1947 csum_size = sizeof(struct ext4_dir_entry_tail);
1948
1949 bh2 = ext4_append(handle, dir, &newblock);
1950 if (IS_ERR(bh2)) {
1951 brelse(*bh);
1952 *bh = NULL;
1953 return (struct ext4_dir_entry_2 *) bh2;
1954 }
1955
1956 BUFFER_TRACE(*bh, "get_write_access");
1957 err = ext4_journal_get_write_access(handle, dir->i_sb, *bh,
1958 EXT4_JTR_NONE);
1959 if (err)
1960 goto journal_error;
1961
1962 BUFFER_TRACE(frame->bh, "get_write_access");
1963 err = ext4_journal_get_write_access(handle, dir->i_sb, frame->bh,
1964 EXT4_JTR_NONE);
1965 if (err)
1966 goto journal_error;
1967
1968 data2 = bh2->b_data;
1969
1970 /* create map in the end of data2 block */
1971 map = (struct dx_map_entry *) (data2 + blocksize);
1972 count = dx_make_map(dir, *bh, hinfo, map);
1973 if (count < 0) {
1974 err = count;
1975 goto journal_error;
1976 }
1977 map -= count;
1978 dx_sort_map(map, count);
1979 /* Ensure that neither split block is over half full */
1980 size = 0;
1981 move = 0;
1982 for (i = count-1; i >= 0; i--) {
1983 /* is more than half of this entry in 2nd half of the block? */
1984 if (size + map[i].size/2 > blocksize/2)
1985 break;
1986 size += map[i].size;
1987 move++;
1988 }
1989 /*
1990 * map index at which we will split
1991 *
1992 * If the sum of active entries didn't exceed half the block size, just
1993 * split it in half by count; each resulting block will have at least
1994 * half the space free.
1995 */
1996 if (i > 0)
1997 split = count - move;
1998 else
1999 split = count/2;
2000
2001 hash2 = map[split].hash;
2002 continued = hash2 == map[split - 1].hash;
2003 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
2004 (unsigned long)dx_get_block(frame->at),
2005 hash2, split, count-split));
2006
2007 /* Fancy dance to stay within two buffers */
2008 de2 = dx_move_dirents(dir, data1, data2, map + split, count - split,
2009 blocksize);
2010 de = dx_pack_dirents(dir, data1, blocksize);
2011 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2012 (char *) de,
2013 blocksize);
2014 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2015 (char *) de2,
2016 blocksize);
2017 if (csum_size) {
2018 ext4_initialize_dirent_tail(*bh, blocksize);
2019 ext4_initialize_dirent_tail(bh2, blocksize);
2020 }
2021
2022 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
2023 blocksize, 1));
2024 dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
2025 blocksize, 1));
2026
2027 /* Which block gets the new entry? */
2028 if (hinfo->hash >= hash2) {
2029 swap(*bh, bh2);
2030 de = de2;
2031 }
2032 dx_insert_block(frame, hash2 + continued, newblock);
2033 err = ext4_handle_dirty_dirblock(handle, dir, bh2);
2034 if (err)
2035 goto journal_error;
2036 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2037 if (err)
2038 goto journal_error;
2039 brelse(bh2);
2040 dxtrace(dx_show_index("frame", frame->entries));
2041 return de;
2042
2043 journal_error:
2044 brelse(*bh);
2045 brelse(bh2);
2046 *bh = NULL;
2047 ext4_std_error(dir->i_sb, err);
2048 return ERR_PTR(err);
2049 }
2050
2051 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
2052 struct buffer_head *bh,
2053 void *buf, int buf_size,
2054 struct ext4_filename *fname,
2055 struct ext4_dir_entry_2 **dest_de)
2056 {
2057 struct ext4_dir_entry_2 *de;
2058 unsigned short reclen = ext4_dir_rec_len(fname_len(fname), dir);
2059 int nlen, rlen;
2060 unsigned int offset = 0;
2061 char *top;
2062
2063 de = buf;
2064 top = buf + buf_size - reclen;
2065 while ((char *) de <= top) {
2066 if (ext4_check_dir_entry(dir, NULL, de, bh,
2067 buf, buf_size, offset))
2068 return -EFSCORRUPTED;
2069 if (ext4_match(dir, fname, de))
2070 return -EEXIST;
2071 nlen = ext4_dir_rec_len(de->name_len, dir);
2072 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2073 if ((de->inode ? rlen - nlen : rlen) >= reclen)
2074 break;
2075 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
2076 offset += rlen;
2077 }
2078 if ((char *) de > top)
2079 return -ENOSPC;
2080
2081 *dest_de = de;
2082 return 0;
2083 }
2084
2085 void ext4_insert_dentry(struct inode *dir,
2086 struct inode *inode,
2087 struct ext4_dir_entry_2 *de,
2088 int buf_size,
2089 struct ext4_filename *fname)
2090 {
2091
2092 int nlen, rlen;
2093
2094 nlen = ext4_dir_rec_len(de->name_len, dir);
2095 rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
2096 if (de->inode) {
2097 struct ext4_dir_entry_2 *de1 =
2098 (struct ext4_dir_entry_2 *)((char *)de + nlen);
2099 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
2100 de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
2101 de = de1;
2102 }
2103 de->file_type = EXT4_FT_UNKNOWN;
2104 de->inode = cpu_to_le32(inode->i_ino);
2105 ext4_set_de_type(inode->i_sb, de, inode->i_mode);
2106 de->name_len = fname_len(fname);
2107 memcpy(de->name, fname_name(fname), fname_len(fname));
2108 if (ext4_hash_in_dirent(dir)) {
2109 struct dx_hash_info *hinfo = &fname->hinfo;
2110
2111 EXT4_DIRENT_HASHES(de)->hash = cpu_to_le32(hinfo->hash);
2112 EXT4_DIRENT_HASHES(de)->minor_hash =
2113 cpu_to_le32(hinfo->minor_hash);
2114 }
2115 }
2116
2117 /*
2118 * Add a new entry into a directory (leaf) block. If de is non-NULL,
2119 * it points to a directory entry which is guaranteed to be large
2120 * enough for new directory entry. If de is NULL, then
2121 * add_dirent_to_buf will attempt search the directory block for
2122 * space. It will return -ENOSPC if no space is available, and -EIO
2123 * and -EEXIST if directory entry already exists.
2124 */
2125 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
2126 struct inode *dir,
2127 struct inode *inode, struct ext4_dir_entry_2 *de,
2128 struct buffer_head *bh)
2129 {
2130 unsigned int blocksize = dir->i_sb->s_blocksize;
2131 int csum_size = 0;
2132 int err, err2;
2133
2134 if (ext4_has_metadata_csum(inode->i_sb))
2135 csum_size = sizeof(struct ext4_dir_entry_tail);
2136
2137 if (!de) {
2138 err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
2139 blocksize - csum_size, fname, &de);
2140 if (err)
2141 return err;
2142 }
2143 BUFFER_TRACE(bh, "get_write_access");
2144 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2145 EXT4_JTR_NONE);
2146 if (err) {
2147 ext4_std_error(dir->i_sb, err);
2148 return err;
2149 }
2150
2151 /* By now the buffer is marked for journaling */
2152 ext4_insert_dentry(dir, inode, de, blocksize, fname);
2153
2154 /*
2155 * XXX shouldn't update any times until successful
2156 * completion of syscall, but too many callers depend
2157 * on this.
2158 *
2159 * XXX similarly, too many callers depend on
2160 * ext4_new_inode() setting the times, but error
2161 * recovery deletes the inode, so the worst that can
2162 * happen is that the times are slightly out of date
2163 * and/or different from the directory change time.
2164 */
2165 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
2166 ext4_update_dx_flag(dir);
2167 inode_inc_iversion(dir);
2168 err2 = ext4_mark_inode_dirty(handle, dir);
2169 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2170 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2171 if (err)
2172 ext4_std_error(dir->i_sb, err);
2173 return err ? err : err2;
2174 }
2175
2176 static bool ext4_check_dx_root(struct inode *dir, struct dx_root *root)
2177 {
2178 struct fake_dirent *fde;
2179 const char *error_msg;
2180 unsigned int rlen;
2181 unsigned int blocksize = dir->i_sb->s_blocksize;
2182 char *blockend = (char *)root + dir->i_sb->s_blocksize;
2183
2184 fde = &root->dot;
2185 if (unlikely(fde->name_len != 1)) {
2186 error_msg = "invalid name_len for '.'";
2187 goto corrupted;
2188 }
2189 if (unlikely(strncmp(root->dot_name, ".", fde->name_len))) {
2190 error_msg = "invalid name for '.'";
2191 goto corrupted;
2192 }
2193 rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
2194 if (unlikely((char *)fde + rlen >= blockend)) {
2195 error_msg = "invalid rec_len for '.'";
2196 goto corrupted;
2197 }
2198
2199 fde = &root->dotdot;
2200 if (unlikely(fde->name_len != 2)) {
2201 error_msg = "invalid name_len for '..'";
2202 goto corrupted;
2203 }
2204 if (unlikely(strncmp(root->dotdot_name, "..", fde->name_len))) {
2205 error_msg = "invalid name for '..'";
2206 goto corrupted;
2207 }
2208 rlen = ext4_rec_len_from_disk(fde->rec_len, blocksize);
2209 if (unlikely((char *)fde + rlen >= blockend)) {
2210 error_msg = "invalid rec_len for '..'";
2211 goto corrupted;
2212 }
2213
2214 return true;
2215
2216 corrupted:
2217 EXT4_ERROR_INODE(dir, "Corrupt dir, %s, running e2fsck is recommended",
2218 error_msg);
2219 return false;
2220 }
2221
2222 /*
2223 * This converts a one block unindexed directory to a 3 block indexed
2224 * directory, and adds the dentry to the indexed directory.
2225 */
2226 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2227 struct inode *dir,
2228 struct inode *inode, struct buffer_head *bh)
2229 {
2230 struct buffer_head *bh2;
2231 struct dx_root *root;
2232 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2233 struct dx_entry *entries;
2234 struct ext4_dir_entry_2 *de, *de2;
2235 char *data2, *top;
2236 unsigned len;
2237 int retval;
2238 unsigned blocksize;
2239 ext4_lblk_t block;
2240 struct fake_dirent *fde;
2241 int csum_size = 0;
2242
2243 if (ext4_has_metadata_csum(inode->i_sb))
2244 csum_size = sizeof(struct ext4_dir_entry_tail);
2245
2246 blocksize = dir->i_sb->s_blocksize;
2247 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2248 BUFFER_TRACE(bh, "get_write_access");
2249 retval = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2250 EXT4_JTR_NONE);
2251 if (retval) {
2252 ext4_std_error(dir->i_sb, retval);
2253 brelse(bh);
2254 return retval;
2255 }
2256
2257 root = (struct dx_root *) bh->b_data;
2258 if (!ext4_check_dx_root(dir, root)) {
2259 brelse(bh);
2260 return -EFSCORRUPTED;
2261 }
2262
2263 /* The 0th block becomes the root, move the dirents out */
2264 fde = &root->dotdot;
2265 de = (struct ext4_dir_entry_2 *)((char *)fde +
2266 ext4_rec_len_from_disk(fde->rec_len, blocksize));
2267 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2268
2269 /* Allocate new block for the 0th block's dirents */
2270 bh2 = ext4_append(handle, dir, &block);
2271 if (IS_ERR(bh2)) {
2272 brelse(bh);
2273 return PTR_ERR(bh2);
2274 }
2275 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2276 data2 = bh2->b_data;
2277
2278 memcpy(data2, de, len);
2279 memset(de, 0, len); /* wipe old data */
2280 de = (struct ext4_dir_entry_2 *) data2;
2281 top = data2 + len;
2282 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
2283 if (ext4_check_dir_entry(dir, NULL, de, bh2, data2, len,
2284 (char *)de - data2)) {
2285 brelse(bh2);
2286 brelse(bh);
2287 return -EFSCORRUPTED;
2288 }
2289 de = de2;
2290 }
2291 de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2292 (char *) de, blocksize);
2293
2294 if (csum_size)
2295 ext4_initialize_dirent_tail(bh2, blocksize);
2296
2297 /* Initialize the root; the dot dirents already exist */
2298 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2299 de->rec_len = ext4_rec_len_to_disk(
2300 blocksize - ext4_dir_rec_len(2, NULL), blocksize);
2301 memset (&root->info, 0, sizeof(root->info));
2302 root->info.info_length = sizeof(root->info);
2303 if (ext4_hash_in_dirent(dir))
2304 root->info.hash_version = DX_HASH_SIPHASH;
2305 else
2306 root->info.hash_version =
2307 EXT4_SB(dir->i_sb)->s_def_hash_version;
2308
2309 entries = root->entries;
2310 dx_set_block(entries, 1);
2311 dx_set_count(entries, 1);
2312 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2313
2314 /* Initialize as for dx_probe */
2315 fname->hinfo.hash_version = root->info.hash_version;
2316 if (fname->hinfo.hash_version <= DX_HASH_TEA)
2317 fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2318 fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2319
2320 /* casefolded encrypted hashes are computed on fname setup */
2321 if (!ext4_hash_in_dirent(dir)) {
2322 int err = ext4fs_dirhash(dir, fname_name(fname),
2323 fname_len(fname), &fname->hinfo);
2324 if (err < 0) {
2325 brelse(bh2);
2326 brelse(bh);
2327 return err;
2328 }
2329 }
2330 memset(frames, 0, sizeof(frames));
2331 frame = frames;
2332 frame->entries = entries;
2333 frame->at = entries;
2334 frame->bh = bh;
2335
2336 retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2337 if (retval)
2338 goto out_frames;
2339 retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2340 if (retval)
2341 goto out_frames;
2342
2343 de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2344 if (IS_ERR(de)) {
2345 retval = PTR_ERR(de);
2346 goto out_frames;
2347 }
2348
2349 retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2350 out_frames:
2351 /*
2352 * Even if the block split failed, we have to properly write
2353 * out all the changes we did so far. Otherwise we can end up
2354 * with corrupted filesystem.
2355 */
2356 if (retval)
2357 ext4_mark_inode_dirty(handle, dir);
2358 dx_release(frames);
2359 brelse(bh2);
2360 return retval;
2361 }
2362
2363 /*
2364 * ext4_add_entry()
2365 *
2366 * adds a file entry to the specified directory, using the same
2367 * semantics as ext4_find_entry(). It returns NULL if it failed.
2368 *
2369 * NOTE!! The inode part of 'de' is left at 0 - which means you
2370 * may not sleep between calling this and putting something into
2371 * the entry, as someone else might have used it while you slept.
2372 */
2373 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2374 struct inode *inode)
2375 {
2376 struct inode *dir = d_inode(dentry->d_parent);
2377 struct buffer_head *bh = NULL;
2378 struct ext4_dir_entry_2 *de;
2379 struct super_block *sb;
2380 struct ext4_filename fname;
2381 int retval;
2382 int dx_fallback=0;
2383 unsigned blocksize;
2384 ext4_lblk_t block, blocks;
2385 int csum_size = 0;
2386
2387 if (ext4_has_metadata_csum(inode->i_sb))
2388 csum_size = sizeof(struct ext4_dir_entry_tail);
2389
2390 sb = dir->i_sb;
2391 blocksize = sb->s_blocksize;
2392
2393 if (fscrypt_is_nokey_name(dentry))
2394 return -ENOKEY;
2395
2396 #if IS_ENABLED(CONFIG_UNICODE)
2397 if (sb_has_strict_encoding(sb) && IS_CASEFOLDED(dir) &&
2398 utf8_validate(sb->s_encoding, &dentry->d_name))
2399 return -EINVAL;
2400 #endif
2401
2402 retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2403 if (retval)
2404 return retval;
2405
2406 if (ext4_has_inline_data(dir)) {
2407 retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2408 if (retval < 0)
2409 goto out;
2410 if (retval == 1) {
2411 retval = 0;
2412 goto out;
2413 }
2414 }
2415
2416 if (is_dx(dir)) {
2417 retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2418 if (!retval || (retval != ERR_BAD_DX_DIR))
2419 goto out;
2420 /* Can we just ignore htree data? */
2421 if (ext4_has_metadata_csum(sb)) {
2422 EXT4_ERROR_INODE(dir,
2423 "Directory has corrupted htree index.");
2424 retval = -EFSCORRUPTED;
2425 goto out;
2426 }
2427 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2428 dx_fallback++;
2429 retval = ext4_mark_inode_dirty(handle, dir);
2430 if (unlikely(retval))
2431 goto out;
2432 }
2433 blocks = dir->i_size >> sb->s_blocksize_bits;
2434 for (block = 0; block < blocks; block++) {
2435 bh = ext4_read_dirblock(dir, block, DIRENT);
2436 if (bh == NULL) {
2437 bh = ext4_bread(handle, dir, block,
2438 EXT4_GET_BLOCKS_CREATE);
2439 goto add_to_new_block;
2440 }
2441 if (IS_ERR(bh)) {
2442 retval = PTR_ERR(bh);
2443 bh = NULL;
2444 goto out;
2445 }
2446 retval = add_dirent_to_buf(handle, &fname, dir, inode,
2447 NULL, bh);
2448 if (retval != -ENOSPC)
2449 goto out;
2450
2451 if (blocks == 1 && !dx_fallback &&
2452 ext4_has_feature_dir_index(sb)) {
2453 retval = make_indexed_dir(handle, &fname, dir,
2454 inode, bh);
2455 bh = NULL; /* make_indexed_dir releases bh */
2456 goto out;
2457 }
2458 brelse(bh);
2459 }
2460 bh = ext4_append(handle, dir, &block);
2461 add_to_new_block:
2462 if (IS_ERR(bh)) {
2463 retval = PTR_ERR(bh);
2464 bh = NULL;
2465 goto out;
2466 }
2467 de = (struct ext4_dir_entry_2 *) bh->b_data;
2468 de->inode = 0;
2469 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2470
2471 if (csum_size)
2472 ext4_initialize_dirent_tail(bh, blocksize);
2473
2474 retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2475 out:
2476 ext4_fname_free_filename(&fname);
2477 brelse(bh);
2478 if (retval == 0)
2479 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2480 return retval;
2481 }
2482
2483 /*
2484 * Returns 0 for success, or a negative error value
2485 */
2486 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2487 struct inode *dir, struct inode *inode)
2488 {
2489 struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2490 struct dx_entry *entries, *at;
2491 struct buffer_head *bh;
2492 struct super_block *sb = dir->i_sb;
2493 struct ext4_dir_entry_2 *de;
2494 int restart;
2495 int err;
2496
2497 again:
2498 restart = 0;
2499 frame = dx_probe(fname, dir, NULL, frames);
2500 if (IS_ERR(frame))
2501 return PTR_ERR(frame);
2502 entries = frame->entries;
2503 at = frame->at;
2504 bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2505 if (IS_ERR(bh)) {
2506 err = PTR_ERR(bh);
2507 bh = NULL;
2508 goto cleanup;
2509 }
2510
2511 BUFFER_TRACE(bh, "get_write_access");
2512 err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
2513 if (err)
2514 goto journal_error;
2515
2516 err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2517 if (err != -ENOSPC)
2518 goto cleanup;
2519
2520 err = 0;
2521 /* Block full, should compress but for now just split */
2522 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2523 dx_get_count(entries), dx_get_limit(entries)));
2524 /* Need to split index? */
2525 if (dx_get_count(entries) == dx_get_limit(entries)) {
2526 ext4_lblk_t newblock;
2527 int levels = frame - frames + 1;
2528 unsigned int icount;
2529 int add_level = 1;
2530 struct dx_entry *entries2;
2531 struct dx_node *node2;
2532 struct buffer_head *bh2;
2533
2534 while (frame > frames) {
2535 if (dx_get_count((frame - 1)->entries) <
2536 dx_get_limit((frame - 1)->entries)) {
2537 add_level = 0;
2538 break;
2539 }
2540 frame--; /* split higher index block */
2541 at = frame->at;
2542 entries = frame->entries;
2543 restart = 1;
2544 }
2545 if (add_level && levels == ext4_dir_htree_level(sb)) {
2546 ext4_warning(sb, "Directory (ino: %lu) index full, "
2547 "reach max htree level :%d",
2548 dir->i_ino, levels);
2549 if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2550 ext4_warning(sb, "Large directory feature is "
2551 "not enabled on this "
2552 "filesystem");
2553 }
2554 err = -ENOSPC;
2555 goto cleanup;
2556 }
2557 icount = dx_get_count(entries);
2558 bh2 = ext4_append(handle, dir, &newblock);
2559 if (IS_ERR(bh2)) {
2560 err = PTR_ERR(bh2);
2561 goto cleanup;
2562 }
2563 node2 = (struct dx_node *)(bh2->b_data);
2564 entries2 = node2->entries;
2565 memset(&node2->fake, 0, sizeof(struct fake_dirent));
2566 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2567 sb->s_blocksize);
2568 BUFFER_TRACE(frame->bh, "get_write_access");
2569 err = ext4_journal_get_write_access(handle, sb, frame->bh,
2570 EXT4_JTR_NONE);
2571 if (err)
2572 goto journal_error;
2573 if (!add_level) {
2574 unsigned icount1 = icount/2, icount2 = icount - icount1;
2575 unsigned hash2 = dx_get_hash(entries + icount1);
2576 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2577 icount1, icount2));
2578
2579 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2580 err = ext4_journal_get_write_access(handle, sb,
2581 (frame - 1)->bh,
2582 EXT4_JTR_NONE);
2583 if (err)
2584 goto journal_error;
2585
2586 memcpy((char *) entries2, (char *) (entries + icount1),
2587 icount2 * sizeof(struct dx_entry));
2588 dx_set_count(entries, icount1);
2589 dx_set_count(entries2, icount2);
2590 dx_set_limit(entries2, dx_node_limit(dir));
2591
2592 /* Which index block gets the new entry? */
2593 if (at - entries >= icount1) {
2594 frame->at = at - entries - icount1 + entries2;
2595 frame->entries = entries = entries2;
2596 swap(frame->bh, bh2);
2597 }
2598 dx_insert_block((frame - 1), hash2, newblock);
2599 dxtrace(dx_show_index("node", frame->entries));
2600 dxtrace(dx_show_index("node",
2601 ((struct dx_node *) bh2->b_data)->entries));
2602 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2603 if (err)
2604 goto journal_error;
2605 brelse (bh2);
2606 err = ext4_handle_dirty_dx_node(handle, dir,
2607 (frame - 1)->bh);
2608 if (err)
2609 goto journal_error;
2610 err = ext4_handle_dirty_dx_node(handle, dir,
2611 frame->bh);
2612 if (restart || err)
2613 goto journal_error;
2614 } else {
2615 struct dx_root *dxroot;
2616 memcpy((char *) entries2, (char *) entries,
2617 icount * sizeof(struct dx_entry));
2618 dx_set_limit(entries2, dx_node_limit(dir));
2619
2620 /* Set up root */
2621 dx_set_count(entries, 1);
2622 dx_set_block(entries + 0, newblock);
2623 dxroot = (struct dx_root *)frames[0].bh->b_data;
2624 dxroot->info.indirect_levels += 1;
2625 dxtrace(printk(KERN_DEBUG
2626 "Creating %d level index...\n",
2627 dxroot->info.indirect_levels));
2628 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2629 if (err)
2630 goto journal_error;
2631 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2632 brelse(bh2);
2633 restart = 1;
2634 goto journal_error;
2635 }
2636 }
2637 de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2638 if (IS_ERR(de)) {
2639 err = PTR_ERR(de);
2640 goto cleanup;
2641 }
2642 err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2643 goto cleanup;
2644
2645 journal_error:
2646 ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2647 cleanup:
2648 brelse(bh);
2649 dx_release(frames);
2650 /* @restart is true means htree-path has been changed, we need to
2651 * repeat dx_probe() to find out valid htree-path
2652 */
2653 if (restart && err == 0)
2654 goto again;
2655 return err;
2656 }
2657
2658 /*
2659 * ext4_generic_delete_entry deletes a directory entry by merging it
2660 * with the previous entry
2661 */
2662 int ext4_generic_delete_entry(struct inode *dir,
2663 struct ext4_dir_entry_2 *de_del,
2664 struct buffer_head *bh,
2665 void *entry_buf,
2666 int buf_size,
2667 int csum_size)
2668 {
2669 struct ext4_dir_entry_2 *de, *pde;
2670 unsigned int blocksize = dir->i_sb->s_blocksize;
2671 int i;
2672
2673 i = 0;
2674 pde = NULL;
2675 de = entry_buf;
2676 while (i < buf_size - csum_size) {
2677 if (ext4_check_dir_entry(dir, NULL, de, bh,
2678 entry_buf, buf_size, i))
2679 return -EFSCORRUPTED;
2680 if (de == de_del) {
2681 if (pde) {
2682 pde->rec_len = ext4_rec_len_to_disk(
2683 ext4_rec_len_from_disk(pde->rec_len,
2684 blocksize) +
2685 ext4_rec_len_from_disk(de->rec_len,
2686 blocksize),
2687 blocksize);
2688
2689 /* wipe entire dir_entry */
2690 memset(de, 0, ext4_rec_len_from_disk(de->rec_len,
2691 blocksize));
2692 } else {
2693 /* wipe dir_entry excluding the rec_len field */
2694 de->inode = 0;
2695 memset(&de->name_len, 0,
2696 ext4_rec_len_from_disk(de->rec_len,
2697 blocksize) -
2698 offsetof(struct ext4_dir_entry_2,
2699 name_len));
2700 }
2701
2702 inode_inc_iversion(dir);
2703 return 0;
2704 }
2705 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2706 pde = de;
2707 de = ext4_next_entry(de, blocksize);
2708 }
2709 return -ENOENT;
2710 }
2711
2712 static int ext4_delete_entry(handle_t *handle,
2713 struct inode *dir,
2714 struct ext4_dir_entry_2 *de_del,
2715 struct buffer_head *bh)
2716 {
2717 int err, csum_size = 0;
2718
2719 if (ext4_has_inline_data(dir)) {
2720 int has_inline_data = 1;
2721 err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2722 &has_inline_data);
2723 if (has_inline_data)
2724 return err;
2725 }
2726
2727 if (ext4_has_metadata_csum(dir->i_sb))
2728 csum_size = sizeof(struct ext4_dir_entry_tail);
2729
2730 BUFFER_TRACE(bh, "get_write_access");
2731 err = ext4_journal_get_write_access(handle, dir->i_sb, bh,
2732 EXT4_JTR_NONE);
2733 if (unlikely(err))
2734 goto out;
2735
2736 err = ext4_generic_delete_entry(dir, de_del, bh, bh->b_data,
2737 dir->i_sb->s_blocksize, csum_size);
2738 if (err)
2739 goto out;
2740
2741 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2742 err = ext4_handle_dirty_dirblock(handle, dir, bh);
2743 if (unlikely(err))
2744 goto out;
2745
2746 return 0;
2747 out:
2748 if (err != -ENOENT)
2749 ext4_std_error(dir->i_sb, err);
2750 return err;
2751 }
2752
2753 /*
2754 * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2755 * since this indicates that nlinks count was previously 1 to avoid overflowing
2756 * the 16-bit i_links_count field on disk. Directories with i_nlink == 1 mean
2757 * that subdirectory link counts are not being maintained accurately.
2758 *
2759 * The caller has already checked for i_nlink overflow in case the DIR_LINK
2760 * feature is not enabled and returned -EMLINK. The is_dx() check is a proxy
2761 * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2762 * on regular files) and to avoid creating huge/slow non-HTREE directories.
2763 */
2764 static void ext4_inc_count(struct inode *inode)
2765 {
2766 inc_nlink(inode);
2767 if (is_dx(inode) &&
2768 (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2769 set_nlink(inode, 1);
2770 }
2771
2772 /*
2773 * If a directory had nlink == 1, then we should let it be 1. This indicates
2774 * directory has >EXT4_LINK_MAX subdirs.
2775 */
2776 static void ext4_dec_count(struct inode *inode)
2777 {
2778 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2779 drop_nlink(inode);
2780 }
2781
2782
2783 /*
2784 * Add non-directory inode to a directory. On success, the inode reference is
2785 * consumed by dentry is instantiation. This is also indicated by clearing of
2786 * *inodep pointer. On failure, the caller is responsible for dropping the
2787 * inode reference in the safe context.
2788 */
2789 static int ext4_add_nondir(handle_t *handle,
2790 struct dentry *dentry, struct inode **inodep)
2791 {
2792 struct inode *dir = d_inode(dentry->d_parent);
2793 struct inode *inode = *inodep;
2794 int err = ext4_add_entry(handle, dentry, inode);
2795 if (!err) {
2796 err = ext4_mark_inode_dirty(handle, inode);
2797 if (IS_DIRSYNC(dir))
2798 ext4_handle_sync(handle);
2799 d_instantiate_new(dentry, inode);
2800 *inodep = NULL;
2801 return err;
2802 }
2803 drop_nlink(inode);
2804 ext4_mark_inode_dirty(handle, inode);
2805 ext4_orphan_add(handle, inode);
2806 unlock_new_inode(inode);
2807 return err;
2808 }
2809
2810 /*
2811 * By the time this is called, we already have created
2812 * the directory cache entry for the new file, but it
2813 * is so far negative - it has no inode.
2814 *
2815 * If the create succeeds, we fill in the inode information
2816 * with d_instantiate().
2817 */
2818 static int ext4_create(struct mnt_idmap *idmap, struct inode *dir,
2819 struct dentry *dentry, umode_t mode, bool excl)
2820 {
2821 handle_t *handle;
2822 struct inode *inode;
2823 int err, credits, retries = 0;
2824
2825 err = dquot_initialize(dir);
2826 if (err)
2827 return err;
2828
2829 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2830 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2831 retry:
2832 inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2833 0, NULL, EXT4_HT_DIR, credits);
2834 handle = ext4_journal_current_handle();
2835 err = PTR_ERR(inode);
2836 if (!IS_ERR(inode)) {
2837 inode->i_op = &ext4_file_inode_operations;
2838 inode->i_fop = &ext4_file_operations;
2839 ext4_set_aops(inode);
2840 err = ext4_add_nondir(handle, dentry, &inode);
2841 if (!err)
2842 ext4_fc_track_create(handle, dentry);
2843 }
2844 if (handle)
2845 ext4_journal_stop(handle);
2846 if (!IS_ERR_OR_NULL(inode))
2847 iput(inode);
2848 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2849 goto retry;
2850 return err;
2851 }
2852
2853 static int ext4_mknod(struct mnt_idmap *idmap, struct inode *dir,
2854 struct dentry *dentry, umode_t mode, dev_t rdev)
2855 {
2856 handle_t *handle;
2857 struct inode *inode;
2858 int err, credits, retries = 0;
2859
2860 err = dquot_initialize(dir);
2861 if (err)
2862 return err;
2863
2864 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2865 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2866 retry:
2867 inode = ext4_new_inode_start_handle(idmap, dir, mode, &dentry->d_name,
2868 0, NULL, EXT4_HT_DIR, credits);
2869 handle = ext4_journal_current_handle();
2870 err = PTR_ERR(inode);
2871 if (!IS_ERR(inode)) {
2872 init_special_inode(inode, inode->i_mode, rdev);
2873 inode->i_op = &ext4_special_inode_operations;
2874 err = ext4_add_nondir(handle, dentry, &inode);
2875 if (!err)
2876 ext4_fc_track_create(handle, dentry);
2877 }
2878 if (handle)
2879 ext4_journal_stop(handle);
2880 if (!IS_ERR_OR_NULL(inode))
2881 iput(inode);
2882 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2883 goto retry;
2884 return err;
2885 }
2886
2887 static int ext4_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
2888 struct file *file, umode_t mode)
2889 {
2890 handle_t *handle;
2891 struct inode *inode;
2892 int err, retries = 0;
2893
2894 err = dquot_initialize(dir);
2895 if (err)
2896 return err;
2897
2898 retry:
2899 inode = ext4_new_inode_start_handle(idmap, dir, mode,
2900 NULL, 0, NULL,
2901 EXT4_HT_DIR,
2902 EXT4_MAXQUOTAS_TRANS_BLOCKS(dir->i_sb) +
2903 4 + EXT4_XATTR_TRANS_BLOCKS);
2904 handle = ext4_journal_current_handle();
2905 err = PTR_ERR(inode);
2906 if (!IS_ERR(inode)) {
2907 inode->i_op = &ext4_file_inode_operations;
2908 inode->i_fop = &ext4_file_operations;
2909 ext4_set_aops(inode);
2910 d_tmpfile(file, inode);
2911 err = ext4_orphan_add(handle, inode);
2912 if (err)
2913 goto err_unlock_inode;
2914 mark_inode_dirty(inode);
2915 unlock_new_inode(inode);
2916 }
2917 if (handle)
2918 ext4_journal_stop(handle);
2919 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2920 goto retry;
2921 return finish_open_simple(file, err);
2922 err_unlock_inode:
2923 ext4_journal_stop(handle);
2924 unlock_new_inode(inode);
2925 return err;
2926 }
2927
2928 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2929 struct ext4_dir_entry_2 *de,
2930 int blocksize, int csum_size,
2931 unsigned int parent_ino, int dotdot_real_len)
2932 {
2933 de->inode = cpu_to_le32(inode->i_ino);
2934 de->name_len = 1;
2935 de->rec_len = ext4_rec_len_to_disk(ext4_dir_rec_len(de->name_len, NULL),
2936 blocksize);
2937 strcpy(de->name, ".");
2938 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2939
2940 de = ext4_next_entry(de, blocksize);
2941 de->inode = cpu_to_le32(parent_ino);
2942 de->name_len = 2;
2943 if (!dotdot_real_len)
2944 de->rec_len = ext4_rec_len_to_disk(blocksize -
2945 (csum_size + ext4_dir_rec_len(1, NULL)),
2946 blocksize);
2947 else
2948 de->rec_len = ext4_rec_len_to_disk(
2949 ext4_dir_rec_len(de->name_len, NULL),
2950 blocksize);
2951 strcpy(de->name, "..");
2952 ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2953
2954 return ext4_next_entry(de, blocksize);
2955 }
2956
2957 int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2958 struct inode *inode)
2959 {
2960 struct buffer_head *dir_block = NULL;
2961 struct ext4_dir_entry_2 *de;
2962 ext4_lblk_t block = 0;
2963 unsigned int blocksize = dir->i_sb->s_blocksize;
2964 int csum_size = 0;
2965 int err;
2966
2967 if (ext4_has_metadata_csum(dir->i_sb))
2968 csum_size = sizeof(struct ext4_dir_entry_tail);
2969
2970 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2971 err = ext4_try_create_inline_dir(handle, dir, inode);
2972 if (err < 0 && err != -ENOSPC)
2973 goto out;
2974 if (!err)
2975 goto out;
2976 }
2977
2978 inode->i_size = 0;
2979 dir_block = ext4_append(handle, inode, &block);
2980 if (IS_ERR(dir_block))
2981 return PTR_ERR(dir_block);
2982 de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2983 ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2984 set_nlink(inode, 2);
2985 if (csum_size)
2986 ext4_initialize_dirent_tail(dir_block, blocksize);
2987
2988 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2989 err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2990 if (err)
2991 goto out;
2992 set_buffer_verified(dir_block);
2993 out:
2994 brelse(dir_block);
2995 return err;
2996 }
2997
2998 static int ext4_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2999 struct dentry *dentry, umode_t mode)
3000 {
3001 handle_t *handle;
3002 struct inode *inode;
3003 int err, err2 = 0, credits, retries = 0;
3004
3005 if (EXT4_DIR_LINK_MAX(dir))
3006 return -EMLINK;
3007
3008 err = dquot_initialize(dir);
3009 if (err)
3010 return err;
3011
3012 credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3013 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
3014 retry:
3015 inode = ext4_new_inode_start_handle(idmap, dir, S_IFDIR | mode,
3016 &dentry->d_name,
3017 0, NULL, EXT4_HT_DIR, credits);
3018 handle = ext4_journal_current_handle();
3019 err = PTR_ERR(inode);
3020 if (IS_ERR(inode))
3021 goto out_stop;
3022
3023 inode->i_op = &ext4_dir_inode_operations;
3024 inode->i_fop = &ext4_dir_operations;
3025 err = ext4_init_new_dir(handle, dir, inode);
3026 if (err)
3027 goto out_clear_inode;
3028 err = ext4_mark_inode_dirty(handle, inode);
3029 if (!err)
3030 err = ext4_add_entry(handle, dentry, inode);
3031 if (err) {
3032 out_clear_inode:
3033 clear_nlink(inode);
3034 ext4_orphan_add(handle, inode);
3035 unlock_new_inode(inode);
3036 err2 = ext4_mark_inode_dirty(handle, inode);
3037 if (unlikely(err2))
3038 err = err2;
3039 ext4_journal_stop(handle);
3040 iput(inode);
3041 goto out_retry;
3042 }
3043 ext4_inc_count(dir);
3044
3045 ext4_update_dx_flag(dir);
3046 err = ext4_mark_inode_dirty(handle, dir);
3047 if (err)
3048 goto out_clear_inode;
3049 d_instantiate_new(dentry, inode);
3050 ext4_fc_track_create(handle, dentry);
3051 if (IS_DIRSYNC(dir))
3052 ext4_handle_sync(handle);
3053
3054 out_stop:
3055 if (handle)
3056 ext4_journal_stop(handle);
3057 out_retry:
3058 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3059 goto retry;
3060 return err;
3061 }
3062
3063 /*
3064 * routine to check that the specified directory is empty (for rmdir)
3065 */
3066 bool ext4_empty_dir(struct inode *inode)
3067 {
3068 unsigned int offset;
3069 struct buffer_head *bh;
3070 struct ext4_dir_entry_2 *de;
3071 struct super_block *sb;
3072
3073 if (ext4_has_inline_data(inode)) {
3074 int has_inline_data = 1;
3075 int ret;
3076
3077 ret = empty_inline_dir(inode, &has_inline_data);
3078 if (has_inline_data)
3079 return ret;
3080 }
3081
3082 sb = inode->i_sb;
3083 if (inode->i_size < ext4_dir_rec_len(1, NULL) +
3084 ext4_dir_rec_len(2, NULL)) {
3085 EXT4_ERROR_INODE(inode, "invalid size");
3086 return false;
3087 }
3088 bh = ext4_read_dirblock(inode, 0, EITHER);
3089 if (IS_ERR(bh))
3090 return false;
3091
3092 de = (struct ext4_dir_entry_2 *) bh->b_data;
3093 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3094 0) ||
3095 le32_to_cpu(de->inode) != inode->i_ino || strcmp(".", de->name)) {
3096 ext4_warning_inode(inode, "directory missing '.'");
3097 brelse(bh);
3098 return false;
3099 }
3100 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3101 de = ext4_next_entry(de, sb->s_blocksize);
3102 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data, bh->b_size,
3103 offset) ||
3104 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3105 ext4_warning_inode(inode, "directory missing '..'");
3106 brelse(bh);
3107 return false;
3108 }
3109 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3110 while (offset < inode->i_size) {
3111 if (!(offset & (sb->s_blocksize - 1))) {
3112 unsigned int lblock;
3113 brelse(bh);
3114 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
3115 bh = ext4_read_dirblock(inode, lblock, EITHER);
3116 if (bh == NULL) {
3117 offset += sb->s_blocksize;
3118 continue;
3119 }
3120 if (IS_ERR(bh))
3121 return false;
3122 }
3123 de = (struct ext4_dir_entry_2 *) (bh->b_data +
3124 (offset & (sb->s_blocksize - 1)));
3125 if (ext4_check_dir_entry(inode, NULL, de, bh,
3126 bh->b_data, bh->b_size, offset) ||
3127 le32_to_cpu(de->inode)) {
3128 brelse(bh);
3129 return false;
3130 }
3131 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
3132 }
3133 brelse(bh);
3134 return true;
3135 }
3136
3137 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3138 {
3139 int retval;
3140 struct inode *inode;
3141 struct buffer_head *bh;
3142 struct ext4_dir_entry_2 *de;
3143 handle_t *handle = NULL;
3144
3145 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3146 return -EIO;
3147
3148 /* Initialize quotas before so that eventual writes go in
3149 * separate transaction */
3150 retval = dquot_initialize(dir);
3151 if (retval)
3152 return retval;
3153 retval = dquot_initialize(d_inode(dentry));
3154 if (retval)
3155 return retval;
3156
3157 retval = -ENOENT;
3158 bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3159 if (IS_ERR(bh))
3160 return PTR_ERR(bh);
3161 if (!bh)
3162 goto end_rmdir;
3163
3164 inode = d_inode(dentry);
3165
3166 retval = -EFSCORRUPTED;
3167 if (le32_to_cpu(de->inode) != inode->i_ino)
3168 goto end_rmdir;
3169
3170 retval = -ENOTEMPTY;
3171 if (!ext4_empty_dir(inode))
3172 goto end_rmdir;
3173
3174 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3175 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3176 if (IS_ERR(handle)) {
3177 retval = PTR_ERR(handle);
3178 handle = NULL;
3179 goto end_rmdir;
3180 }
3181
3182 if (IS_DIRSYNC(dir))
3183 ext4_handle_sync(handle);
3184
3185 retval = ext4_delete_entry(handle, dir, de, bh);
3186 if (retval)
3187 goto end_rmdir;
3188 if (!EXT4_DIR_LINK_EMPTY(inode))
3189 ext4_warning_inode(inode,
3190 "empty directory '%.*s' has too many links (%u)",
3191 dentry->d_name.len, dentry->d_name.name,
3192 inode->i_nlink);
3193 inode_inc_iversion(inode);
3194 clear_nlink(inode);
3195 /* There's no need to set i_disksize: the fact that i_nlink is
3196 * zero will ensure that the right thing happens during any
3197 * recovery. */
3198 inode->i_size = 0;
3199 ext4_orphan_add(handle, inode);
3200 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3201 inode_set_ctime_current(inode);
3202 retval = ext4_mark_inode_dirty(handle, inode);
3203 if (retval)
3204 goto end_rmdir;
3205 ext4_dec_count(dir);
3206 ext4_update_dx_flag(dir);
3207 ext4_fc_track_unlink(handle, dentry);
3208 retval = ext4_mark_inode_dirty(handle, dir);
3209
3210 /* VFS negative dentries are incompatible with Encoding and
3211 * Case-insensitiveness. Eventually we'll want avoid
3212 * invalidating the dentries here, alongside with returning the
3213 * negative dentries at ext4_lookup(), when it is better
3214 * supported by the VFS for the CI case.
3215 */
3216 if (IS_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
3217 d_invalidate(dentry);
3218
3219 end_rmdir:
3220 brelse(bh);
3221 if (handle)
3222 ext4_journal_stop(handle);
3223 return retval;
3224 }
3225
3226 int __ext4_unlink(struct inode *dir, const struct qstr *d_name,
3227 struct inode *inode,
3228 struct dentry *dentry /* NULL during fast_commit recovery */)
3229 {
3230 int retval = -ENOENT;
3231 struct buffer_head *bh;
3232 struct ext4_dir_entry_2 *de;
3233 handle_t *handle;
3234 int skip_remove_dentry = 0;
3235
3236 /*
3237 * Keep this outside the transaction; it may have to set up the
3238 * directory's encryption key, which isn't GFP_NOFS-safe.
3239 */
3240 bh = ext4_find_entry(dir, d_name, &de, NULL);
3241 if (IS_ERR(bh))
3242 return PTR_ERR(bh);
3243
3244 if (!bh)
3245 return -ENOENT;
3246
3247 if (le32_to_cpu(de->inode) != inode->i_ino) {
3248 /*
3249 * It's okay if we find dont find dentry which matches
3250 * the inode. That's because it might have gotten
3251 * renamed to a different inode number
3252 */
3253 if (EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
3254 skip_remove_dentry = 1;
3255 else
3256 goto out_bh;
3257 }
3258
3259 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3260 EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3261 if (IS_ERR(handle)) {
3262 retval = PTR_ERR(handle);
3263 goto out_bh;
3264 }
3265
3266 if (IS_DIRSYNC(dir))
3267 ext4_handle_sync(handle);
3268
3269 if (!skip_remove_dentry) {
3270 retval = ext4_delete_entry(handle, dir, de, bh);
3271 if (retval)
3272 goto out_handle;
3273 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
3274 ext4_update_dx_flag(dir);
3275 retval = ext4_mark_inode_dirty(handle, dir);
3276 if (retval)
3277 goto out_handle;
3278 } else {
3279 retval = 0;
3280 }
3281 if (inode->i_nlink == 0)
3282 ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3283 d_name->len, d_name->name);
3284 else
3285 drop_nlink(inode);
3286 if (!inode->i_nlink)
3287 ext4_orphan_add(handle, inode);
3288 inode_set_ctime_current(inode);
3289 retval = ext4_mark_inode_dirty(handle, inode);
3290 if (dentry && !retval)
3291 ext4_fc_track_unlink(handle, dentry);
3292 out_handle:
3293 ext4_journal_stop(handle);
3294 out_bh:
3295 brelse(bh);
3296 return retval;
3297 }
3298
3299 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3300 {
3301 int retval;
3302
3303 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3304 return -EIO;
3305
3306 trace_ext4_unlink_enter(dir, dentry);
3307 /*
3308 * Initialize quotas before so that eventual writes go
3309 * in separate transaction
3310 */
3311 retval = dquot_initialize(dir);
3312 if (retval)
3313 goto out_trace;
3314 retval = dquot_initialize(d_inode(dentry));
3315 if (retval)
3316 goto out_trace;
3317
3318 retval = __ext4_unlink(dir, &dentry->d_name, d_inode(dentry), dentry);
3319
3320 /* VFS negative dentries are incompatible with Encoding and
3321 * Case-insensitiveness. Eventually we'll want avoid
3322 * invalidating the dentries here, alongside with returning the
3323 * negative dentries at ext4_lookup(), when it is better
3324 * supported by the VFS for the CI case.
3325 */
3326 if (IS_ENABLED(CONFIG_UNICODE) && IS_CASEFOLDED(dir))
3327 d_invalidate(dentry);
3328
3329 out_trace:
3330 trace_ext4_unlink_exit(dentry, retval);
3331 return retval;
3332 }
3333
3334 static int ext4_init_symlink_block(handle_t *handle, struct inode *inode,
3335 struct fscrypt_str *disk_link)
3336 {
3337 struct buffer_head *bh;
3338 char *kaddr;
3339 int err = 0;
3340
3341 bh = ext4_bread(handle, inode, 0, EXT4_GET_BLOCKS_CREATE);
3342 if (IS_ERR(bh))
3343 return PTR_ERR(bh);
3344
3345 BUFFER_TRACE(bh, "get_write_access");
3346 err = ext4_journal_get_write_access(handle, inode->i_sb, bh, EXT4_JTR_NONE);
3347 if (err)
3348 goto out;
3349
3350 kaddr = (char *)bh->b_data;
3351 memcpy(kaddr, disk_link->name, disk_link->len);
3352 inode->i_size = disk_link->len - 1;
3353 EXT4_I(inode)->i_disksize = inode->i_size;
3354 err = ext4_handle_dirty_metadata(handle, inode, bh);
3355 out:
3356 brelse(bh);
3357 return err;
3358 }
3359
3360 static int ext4_symlink(struct mnt_idmap *idmap, struct inode *dir,
3361 struct dentry *dentry, const char *symname)
3362 {
3363 handle_t *handle;
3364 struct inode *inode;
3365 int err, len = strlen(symname);
3366 int credits;
3367 struct fscrypt_str disk_link;
3368 int retries = 0;
3369
3370 if (unlikely(ext4_forced_shutdown(dir->i_sb)))
3371 return -EIO;
3372
3373 err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3374 &disk_link);
3375 if (err)
3376 return err;
3377
3378 err = dquot_initialize(dir);
3379 if (err)
3380 return err;
3381
3382 /*
3383 * EXT4_INDEX_EXTRA_TRANS_BLOCKS for addition of entry into the
3384 * directory. +3 for inode, inode bitmap, group descriptor allocation.
3385 * EXT4_DATA_TRANS_BLOCKS for the data block allocation and
3386 * modification.
3387 */
3388 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3389 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3390 retry:
3391 inode = ext4_new_inode_start_handle(idmap, dir, S_IFLNK|S_IRWXUGO,
3392 &dentry->d_name, 0, NULL,
3393 EXT4_HT_DIR, credits);
3394 handle = ext4_journal_current_handle();
3395 if (IS_ERR(inode)) {
3396 if (handle)
3397 ext4_journal_stop(handle);
3398 err = PTR_ERR(inode);
3399 goto out_retry;
3400 }
3401
3402 if (IS_ENCRYPTED(inode)) {
3403 err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3404 if (err)
3405 goto err_drop_inode;
3406 inode->i_op = &ext4_encrypted_symlink_inode_operations;
3407 } else {
3408 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3409 inode->i_op = &ext4_symlink_inode_operations;
3410 } else {
3411 inode->i_op = &ext4_fast_symlink_inode_operations;
3412 inode->i_link = (char *)&EXT4_I(inode)->i_data;
3413 }
3414 }
3415
3416 if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3417 /* alloc symlink block and fill it */
3418 err = ext4_init_symlink_block(handle, inode, &disk_link);
3419 if (err)
3420 goto err_drop_inode;
3421 } else {
3422 /* clear the extent format for fast symlink */
3423 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3424 memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3425 disk_link.len);
3426 inode->i_size = disk_link.len - 1;
3427 EXT4_I(inode)->i_disksize = inode->i_size;
3428 }
3429 err = ext4_add_nondir(handle, dentry, &inode);
3430 if (handle)
3431 ext4_journal_stop(handle);
3432 iput(inode);
3433 goto out_retry;
3434
3435 err_drop_inode:
3436 clear_nlink(inode);
3437 ext4_mark_inode_dirty(handle, inode);
3438 ext4_orphan_add(handle, inode);
3439 unlock_new_inode(inode);
3440 if (handle)
3441 ext4_journal_stop(handle);
3442 iput(inode);
3443 out_retry:
3444 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3445 goto retry;
3446 if (disk_link.name != (unsigned char *)symname)
3447 kfree(disk_link.name);
3448 return err;
3449 }
3450
3451 int __ext4_link(struct inode *dir, struct inode *inode, struct dentry *dentry)
3452 {
3453 handle_t *handle;
3454 int err, retries = 0;
3455 retry:
3456 handle = ext4_journal_start(dir, EXT4_HT_DIR,
3457 (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3458 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3459 if (IS_ERR(handle))
3460 return PTR_ERR(handle);
3461
3462 if (IS_DIRSYNC(dir))
3463 ext4_handle_sync(handle);
3464
3465 inode_set_ctime_current(inode);
3466 ext4_inc_count(inode);
3467 ihold(inode);
3468
3469 err = ext4_add_entry(handle, dentry, inode);
3470 if (!err) {
3471 err = ext4_mark_inode_dirty(handle, inode);
3472 /* this can happen only for tmpfile being
3473 * linked the first time
3474 */
3475 if (inode->i_nlink == 1)
3476 ext4_orphan_del(handle, inode);
3477 d_instantiate(dentry, inode);
3478 ext4_fc_track_link(handle, dentry);
3479 } else {
3480 drop_nlink(inode);
3481 iput(inode);
3482 }
3483 ext4_journal_stop(handle);
3484 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3485 goto retry;
3486 return err;
3487 }
3488
3489 static int ext4_link(struct dentry *old_dentry,
3490 struct inode *dir, struct dentry *dentry)
3491 {
3492 struct inode *inode = d_inode(old_dentry);
3493 int err;
3494
3495 if (inode->i_nlink >= EXT4_LINK_MAX)
3496 return -EMLINK;
3497
3498 err = fscrypt_prepare_link(old_dentry, dir, dentry);
3499 if (err)
3500 return err;
3501
3502 if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3503 (!projid_eq(EXT4_I(dir)->i_projid,
3504 EXT4_I(old_dentry->d_inode)->i_projid)))
3505 return -EXDEV;
3506
3507 err = dquot_initialize(dir);
3508 if (err)
3509 return err;
3510 return __ext4_link(dir, inode, dentry);
3511 }
3512
3513 /*
3514 * Try to find buffer head where contains the parent block.
3515 * It should be the inode block if it is inlined or the 1st block
3516 * if it is a normal dir.
3517 */
3518 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3519 struct inode *inode,
3520 int *retval,
3521 struct ext4_dir_entry_2 **parent_de,
3522 int *inlined)
3523 {
3524 struct buffer_head *bh;
3525
3526 if (!ext4_has_inline_data(inode)) {
3527 struct ext4_dir_entry_2 *de;
3528 unsigned int offset;
3529
3530 bh = ext4_read_dirblock(inode, 0, EITHER);
3531 if (IS_ERR(bh)) {
3532 *retval = PTR_ERR(bh);
3533 return NULL;
3534 }
3535
3536 de = (struct ext4_dir_entry_2 *) bh->b_data;
3537 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3538 bh->b_size, 0) ||
3539 le32_to_cpu(de->inode) != inode->i_ino ||
3540 strcmp(".", de->name)) {
3541 EXT4_ERROR_INODE(inode, "directory missing '.'");
3542 brelse(bh);
3543 *retval = -EFSCORRUPTED;
3544 return NULL;
3545 }
3546 offset = ext4_rec_len_from_disk(de->rec_len,
3547 inode->i_sb->s_blocksize);
3548 de = ext4_next_entry(de, inode->i_sb->s_blocksize);
3549 if (ext4_check_dir_entry(inode, NULL, de, bh, bh->b_data,
3550 bh->b_size, offset) ||
3551 le32_to_cpu(de->inode) == 0 || strcmp("..", de->name)) {
3552 EXT4_ERROR_INODE(inode, "directory missing '..'");
3553 brelse(bh);
3554 *retval = -EFSCORRUPTED;
3555 return NULL;
3556 }
3557 *parent_de = de;
3558
3559 return bh;
3560 }
3561
3562 *inlined = 1;
3563 return ext4_get_first_inline_block(inode, parent_de, retval);
3564 }
3565
3566 struct ext4_renament {
3567 struct inode *dir;
3568 struct dentry *dentry;
3569 struct inode *inode;
3570 bool is_dir;
3571 int dir_nlink_delta;
3572
3573 /* entry for "dentry" */
3574 struct buffer_head *bh;
3575 struct ext4_dir_entry_2 *de;
3576 int inlined;
3577
3578 /* entry for ".." in inode if it's a directory */
3579 struct buffer_head *dir_bh;
3580 struct ext4_dir_entry_2 *parent_de;
3581 int dir_inlined;
3582 };
3583
3584 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent, bool is_cross)
3585 {
3586 int retval;
3587
3588 ent->is_dir = true;
3589 if (!is_cross)
3590 return 0;
3591
3592 ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3593 &retval, &ent->parent_de,
3594 &ent->dir_inlined);
3595 if (!ent->dir_bh)
3596 return retval;
3597 if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3598 return -EFSCORRUPTED;
3599 BUFFER_TRACE(ent->dir_bh, "get_write_access");
3600 return ext4_journal_get_write_access(handle, ent->dir->i_sb,
3601 ent->dir_bh, EXT4_JTR_NONE);
3602 }
3603
3604 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3605 unsigned dir_ino)
3606 {
3607 int retval;
3608
3609 if (!ent->dir_bh)
3610 return 0;
3611
3612 ent->parent_de->inode = cpu_to_le32(dir_ino);
3613 BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3614 if (!ent->dir_inlined) {
3615 if (is_dx(ent->inode)) {
3616 retval = ext4_handle_dirty_dx_node(handle,
3617 ent->inode,
3618 ent->dir_bh);
3619 } else {
3620 retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3621 ent->dir_bh);
3622 }
3623 } else {
3624 retval = ext4_mark_inode_dirty(handle, ent->inode);
3625 }
3626 if (retval) {
3627 ext4_std_error(ent->dir->i_sb, retval);
3628 return retval;
3629 }
3630 return 0;
3631 }
3632
3633 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3634 unsigned ino, unsigned file_type)
3635 {
3636 int retval, retval2;
3637
3638 BUFFER_TRACE(ent->bh, "get write access");
3639 retval = ext4_journal_get_write_access(handle, ent->dir->i_sb, ent->bh,
3640 EXT4_JTR_NONE);
3641 if (retval)
3642 return retval;
3643 ent->de->inode = cpu_to_le32(ino);
3644 if (ext4_has_feature_filetype(ent->dir->i_sb))
3645 ent->de->file_type = file_type;
3646 inode_inc_iversion(ent->dir);
3647 inode_set_mtime_to_ts(ent->dir, inode_set_ctime_current(ent->dir));
3648 retval = ext4_mark_inode_dirty(handle, ent->dir);
3649 BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3650 if (!ent->inlined) {
3651 retval2 = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3652 if (unlikely(retval2)) {
3653 ext4_std_error(ent->dir->i_sb, retval2);
3654 return retval2;
3655 }
3656 }
3657 return retval;
3658 }
3659
3660 static void ext4_resetent(handle_t *handle, struct ext4_renament *ent,
3661 unsigned ino, unsigned file_type)
3662 {
3663 struct ext4_renament old = *ent;
3664 int retval = 0;
3665
3666 /*
3667 * old->de could have moved from under us during make indexed dir,
3668 * so the old->de may no longer valid and need to find it again
3669 * before reset old inode info.
3670 */
3671 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3672 &old.inlined);
3673 if (IS_ERR(old.bh))
3674 retval = PTR_ERR(old.bh);
3675 if (!old.bh)
3676 retval = -ENOENT;
3677 if (retval) {
3678 ext4_std_error(old.dir->i_sb, retval);
3679 return;
3680 }
3681
3682 ext4_setent(handle, &old, ino, file_type);
3683 brelse(old.bh);
3684 }
3685
3686 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3687 const struct qstr *d_name)
3688 {
3689 int retval = -ENOENT;
3690 struct buffer_head *bh;
3691 struct ext4_dir_entry_2 *de;
3692
3693 bh = ext4_find_entry(dir, d_name, &de, NULL);
3694 if (IS_ERR(bh))
3695 return PTR_ERR(bh);
3696 if (bh) {
3697 retval = ext4_delete_entry(handle, dir, de, bh);
3698 brelse(bh);
3699 }
3700 return retval;
3701 }
3702
3703 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3704 int force_reread)
3705 {
3706 int retval;
3707 /*
3708 * ent->de could have moved from under us during htree split, so make
3709 * sure that we are deleting the right entry. We might also be pointing
3710 * to a stale entry in the unused part of ent->bh so just checking inum
3711 * and the name isn't enough.
3712 */
3713 if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3714 ent->de->name_len != ent->dentry->d_name.len ||
3715 strncmp(ent->de->name, ent->dentry->d_name.name,
3716 ent->de->name_len) ||
3717 force_reread) {
3718 retval = ext4_find_delete_entry(handle, ent->dir,
3719 &ent->dentry->d_name);
3720 } else {
3721 retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3722 if (retval == -ENOENT) {
3723 retval = ext4_find_delete_entry(handle, ent->dir,
3724 &ent->dentry->d_name);
3725 }
3726 }
3727
3728 if (retval) {
3729 ext4_warning_inode(ent->dir,
3730 "Deleting old file: nlink %d, error=%d",
3731 ent->dir->i_nlink, retval);
3732 }
3733 }
3734
3735 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3736 {
3737 if (ent->dir_nlink_delta) {
3738 if (ent->dir_nlink_delta == -1)
3739 ext4_dec_count(ent->dir);
3740 else
3741 ext4_inc_count(ent->dir);
3742 ext4_mark_inode_dirty(handle, ent->dir);
3743 }
3744 }
3745
3746 static struct inode *ext4_whiteout_for_rename(struct mnt_idmap *idmap,
3747 struct ext4_renament *ent,
3748 int credits, handle_t **h)
3749 {
3750 struct inode *wh;
3751 handle_t *handle;
3752 int retries = 0;
3753
3754 /*
3755 * for inode block, sb block, group summaries,
3756 * and inode bitmap
3757 */
3758 credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3759 EXT4_XATTR_TRANS_BLOCKS + 4);
3760 retry:
3761 wh = ext4_new_inode_start_handle(idmap, ent->dir,
3762 S_IFCHR | WHITEOUT_MODE,
3763 &ent->dentry->d_name, 0, NULL,
3764 EXT4_HT_DIR, credits);
3765
3766 handle = ext4_journal_current_handle();
3767 if (IS_ERR(wh)) {
3768 if (handle)
3769 ext4_journal_stop(handle);
3770 if (PTR_ERR(wh) == -ENOSPC &&
3771 ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3772 goto retry;
3773 } else {
3774 *h = handle;
3775 init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3776 wh->i_op = &ext4_special_inode_operations;
3777 }
3778 return wh;
3779 }
3780
3781 /*
3782 * Anybody can rename anything with this: the permission checks are left to the
3783 * higher-level routines.
3784 *
3785 * n.b. old_{dentry,inode) refers to the source dentry/inode
3786 * while new_{dentry,inode) refers to the destination dentry/inode
3787 * This comes from rename(const char *oldpath, const char *newpath)
3788 */
3789 static int ext4_rename(struct mnt_idmap *idmap, struct inode *old_dir,
3790 struct dentry *old_dentry, struct inode *new_dir,
3791 struct dentry *new_dentry, unsigned int flags)
3792 {
3793 handle_t *handle = NULL;
3794 struct ext4_renament old = {
3795 .dir = old_dir,
3796 .dentry = old_dentry,
3797 .inode = d_inode(old_dentry),
3798 };
3799 struct ext4_renament new = {
3800 .dir = new_dir,
3801 .dentry = new_dentry,
3802 .inode = d_inode(new_dentry),
3803 };
3804 int force_reread;
3805 int retval;
3806 struct inode *whiteout = NULL;
3807 int credits;
3808 u8 old_file_type;
3809
3810 if (new.inode && new.inode->i_nlink == 0) {
3811 EXT4_ERROR_INODE(new.inode,
3812 "target of rename is already freed");
3813 return -EFSCORRUPTED;
3814 }
3815
3816 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3817 (!projid_eq(EXT4_I(new_dir)->i_projid,
3818 EXT4_I(old_dentry->d_inode)->i_projid)))
3819 return -EXDEV;
3820
3821 retval = dquot_initialize(old.dir);
3822 if (retval)
3823 return retval;
3824 retval = dquot_initialize(old.inode);
3825 if (retval)
3826 return retval;
3827 retval = dquot_initialize(new.dir);
3828 if (retval)
3829 return retval;
3830
3831 /* Initialize quotas before so that eventual writes go
3832 * in separate transaction */
3833 if (new.inode) {
3834 retval = dquot_initialize(new.inode);
3835 if (retval)
3836 return retval;
3837 }
3838
3839 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de,
3840 &old.inlined);
3841 if (IS_ERR(old.bh))
3842 return PTR_ERR(old.bh);
3843
3844 /*
3845 * Check for inode number is _not_ due to possible IO errors.
3846 * We might rmdir the source, keep it as pwd of some process
3847 * and merrily kill the link to whatever was created under the
3848 * same name. Goodbye sticky bit ;-<
3849 */
3850 retval = -ENOENT;
3851 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3852 goto release_bh;
3853
3854 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3855 &new.de, &new.inlined);
3856 if (IS_ERR(new.bh)) {
3857 retval = PTR_ERR(new.bh);
3858 new.bh = NULL;
3859 goto release_bh;
3860 }
3861 if (new.bh) {
3862 if (!new.inode) {
3863 brelse(new.bh);
3864 new.bh = NULL;
3865 }
3866 }
3867 if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3868 ext4_alloc_da_blocks(old.inode);
3869
3870 credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3871 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3872 if (!(flags & RENAME_WHITEOUT)) {
3873 handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3874 if (IS_ERR(handle)) {
3875 retval = PTR_ERR(handle);
3876 goto release_bh;
3877 }
3878 } else {
3879 whiteout = ext4_whiteout_for_rename(idmap, &old, credits, &handle);
3880 if (IS_ERR(whiteout)) {
3881 retval = PTR_ERR(whiteout);
3882 goto release_bh;
3883 }
3884 }
3885
3886 old_file_type = old.de->file_type;
3887 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3888 ext4_handle_sync(handle);
3889
3890 if (S_ISDIR(old.inode->i_mode)) {
3891 if (new.inode) {
3892 retval = -ENOTEMPTY;
3893 if (!ext4_empty_dir(new.inode))
3894 goto end_rename;
3895 } else {
3896 retval = -EMLINK;
3897 if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3898 goto end_rename;
3899 }
3900 retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
3901 if (retval)
3902 goto end_rename;
3903 }
3904 /*
3905 * If we're renaming a file within an inline_data dir and adding or
3906 * setting the new dirent causes a conversion from inline_data to
3907 * extents/blockmap, we need to force the dirent delete code to
3908 * re-read the directory, or else we end up trying to delete a dirent
3909 * from what is now the extent tree root (or a block map).
3910 */
3911 force_reread = (new.dir->i_ino == old.dir->i_ino &&
3912 ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3913
3914 if (whiteout) {
3915 /*
3916 * Do this before adding a new entry, so the old entry is sure
3917 * to be still pointing to the valid old entry.
3918 */
3919 retval = ext4_setent(handle, &old, whiteout->i_ino,
3920 EXT4_FT_CHRDEV);
3921 if (retval)
3922 goto end_rename;
3923 retval = ext4_mark_inode_dirty(handle, whiteout);
3924 if (unlikely(retval))
3925 goto end_rename;
3926
3927 }
3928 if (!new.bh) {
3929 retval = ext4_add_entry(handle, new.dentry, old.inode);
3930 if (retval)
3931 goto end_rename;
3932 } else {
3933 retval = ext4_setent(handle, &new,
3934 old.inode->i_ino, old_file_type);
3935 if (retval)
3936 goto end_rename;
3937 }
3938 if (force_reread)
3939 force_reread = !ext4_test_inode_flag(new.dir,
3940 EXT4_INODE_INLINE_DATA);
3941
3942 /*
3943 * Like most other Unix systems, set the ctime for inodes on a
3944 * rename.
3945 */
3946 inode_set_ctime_current(old.inode);
3947 retval = ext4_mark_inode_dirty(handle, old.inode);
3948 if (unlikely(retval))
3949 goto end_rename;
3950
3951 if (!whiteout) {
3952 /*
3953 * ok, that's it
3954 */
3955 ext4_rename_delete(handle, &old, force_reread);
3956 }
3957
3958 if (new.inode) {
3959 ext4_dec_count(new.inode);
3960 inode_set_ctime_current(new.inode);
3961 }
3962 inode_set_mtime_to_ts(old.dir, inode_set_ctime_current(old.dir));
3963 ext4_update_dx_flag(old.dir);
3964 if (old.is_dir) {
3965 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3966 if (retval)
3967 goto end_rename;
3968
3969 ext4_dec_count(old.dir);
3970 if (new.inode) {
3971 /* checked ext4_empty_dir above, can't have another
3972 * parent, ext4_dec_count() won't work for many-linked
3973 * dirs */
3974 clear_nlink(new.inode);
3975 } else {
3976 ext4_inc_count(new.dir);
3977 ext4_update_dx_flag(new.dir);
3978 retval = ext4_mark_inode_dirty(handle, new.dir);
3979 if (unlikely(retval))
3980 goto end_rename;
3981 }
3982 }
3983 retval = ext4_mark_inode_dirty(handle, old.dir);
3984 if (unlikely(retval))
3985 goto end_rename;
3986
3987 if (old.is_dir) {
3988 /*
3989 * We disable fast commits here that's because the
3990 * replay code is not yet capable of changing dot dot
3991 * dirents in directories.
3992 */
3993 ext4_fc_mark_ineligible(old.inode->i_sb,
3994 EXT4_FC_REASON_RENAME_DIR, handle);
3995 } else {
3996 struct super_block *sb = old.inode->i_sb;
3997
3998 if (new.inode)
3999 ext4_fc_track_unlink(handle, new.dentry);
4000 if (test_opt2(sb, JOURNAL_FAST_COMMIT) &&
4001 !(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4002 !(ext4_test_mount_flag(sb, EXT4_MF_FC_INELIGIBLE))) {
4003 __ext4_fc_track_link(handle, old.inode, new.dentry);
4004 __ext4_fc_track_unlink(handle, old.inode, old.dentry);
4005 if (whiteout)
4006 __ext4_fc_track_create(handle, whiteout,
4007 old.dentry);
4008 }
4009 }
4010
4011 if (new.inode) {
4012 retval = ext4_mark_inode_dirty(handle, new.inode);
4013 if (unlikely(retval))
4014 goto end_rename;
4015 if (!new.inode->i_nlink)
4016 ext4_orphan_add(handle, new.inode);
4017 }
4018 retval = 0;
4019
4020 end_rename:
4021 if (whiteout) {
4022 if (retval) {
4023 ext4_resetent(handle, &old,
4024 old.inode->i_ino, old_file_type);
4025 drop_nlink(whiteout);
4026 ext4_mark_inode_dirty(handle, whiteout);
4027 ext4_orphan_add(handle, whiteout);
4028 }
4029 unlock_new_inode(whiteout);
4030 ext4_journal_stop(handle);
4031 iput(whiteout);
4032 } else {
4033 ext4_journal_stop(handle);
4034 }
4035 release_bh:
4036 brelse(old.dir_bh);
4037 brelse(old.bh);
4038 brelse(new.bh);
4039
4040 return retval;
4041 }
4042
4043 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
4044 struct inode *new_dir, struct dentry *new_dentry)
4045 {
4046 handle_t *handle = NULL;
4047 struct ext4_renament old = {
4048 .dir = old_dir,
4049 .dentry = old_dentry,
4050 .inode = d_inode(old_dentry),
4051 };
4052 struct ext4_renament new = {
4053 .dir = new_dir,
4054 .dentry = new_dentry,
4055 .inode = d_inode(new_dentry),
4056 };
4057 u8 new_file_type;
4058 int retval;
4059
4060 if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
4061 !projid_eq(EXT4_I(new_dir)->i_projid,
4062 EXT4_I(old_dentry->d_inode)->i_projid)) ||
4063 (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
4064 !projid_eq(EXT4_I(old_dir)->i_projid,
4065 EXT4_I(new_dentry->d_inode)->i_projid)))
4066 return -EXDEV;
4067
4068 retval = dquot_initialize(old.dir);
4069 if (retval)
4070 return retval;
4071 retval = dquot_initialize(new.dir);
4072 if (retval)
4073 return retval;
4074
4075 old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
4076 &old.de, &old.inlined);
4077 if (IS_ERR(old.bh))
4078 return PTR_ERR(old.bh);
4079 /*
4080 * Check for inode number is _not_ due to possible IO errors.
4081 * We might rmdir the source, keep it as pwd of some process
4082 * and merrily kill the link to whatever was created under the
4083 * same name. Goodbye sticky bit ;-<
4084 */
4085 retval = -ENOENT;
4086 if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
4087 goto end_rename;
4088
4089 new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
4090 &new.de, &new.inlined);
4091 if (IS_ERR(new.bh)) {
4092 retval = PTR_ERR(new.bh);
4093 new.bh = NULL;
4094 goto end_rename;
4095 }
4096
4097 /* RENAME_EXCHANGE case: old *and* new must both exist */
4098 if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
4099 goto end_rename;
4100
4101 handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
4102 (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
4103 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
4104 if (IS_ERR(handle)) {
4105 retval = PTR_ERR(handle);
4106 handle = NULL;
4107 goto end_rename;
4108 }
4109
4110 if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
4111 ext4_handle_sync(handle);
4112
4113 if (S_ISDIR(old.inode->i_mode)) {
4114 retval = ext4_rename_dir_prepare(handle, &old, new.dir != old.dir);
4115 if (retval)
4116 goto end_rename;
4117 }
4118 if (S_ISDIR(new.inode->i_mode)) {
4119 retval = ext4_rename_dir_prepare(handle, &new, new.dir != old.dir);
4120 if (retval)
4121 goto end_rename;
4122 }
4123
4124 /*
4125 * Other than the special case of overwriting a directory, parents'
4126 * nlink only needs to be modified if this is a cross directory rename.
4127 */
4128 if (old.dir != new.dir && old.is_dir != new.is_dir) {
4129 old.dir_nlink_delta = old.is_dir ? -1 : 1;
4130 new.dir_nlink_delta = -old.dir_nlink_delta;
4131 retval = -EMLINK;
4132 if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
4133 (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
4134 goto end_rename;
4135 }
4136
4137 new_file_type = new.de->file_type;
4138 retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
4139 if (retval)
4140 goto end_rename;
4141
4142 retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
4143 if (retval)
4144 goto end_rename;
4145
4146 /*
4147 * Like most other Unix systems, set the ctime for inodes on a
4148 * rename.
4149 */
4150 inode_set_ctime_current(old.inode);
4151 inode_set_ctime_current(new.inode);
4152 retval = ext4_mark_inode_dirty(handle, old.inode);
4153 if (unlikely(retval))
4154 goto end_rename;
4155 retval = ext4_mark_inode_dirty(handle, new.inode);
4156 if (unlikely(retval))
4157 goto end_rename;
4158 ext4_fc_mark_ineligible(new.inode->i_sb,
4159 EXT4_FC_REASON_CROSS_RENAME, handle);
4160 if (old.dir_bh) {
4161 retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
4162 if (retval)
4163 goto end_rename;
4164 }
4165 if (new.dir_bh) {
4166 retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
4167 if (retval)
4168 goto end_rename;
4169 }
4170 ext4_update_dir_count(handle, &old);
4171 ext4_update_dir_count(handle, &new);
4172 retval = 0;
4173
4174 end_rename:
4175 brelse(old.dir_bh);
4176 brelse(new.dir_bh);
4177 brelse(old.bh);
4178 brelse(new.bh);
4179 if (handle)
4180 ext4_journal_stop(handle);
4181 return retval;
4182 }
4183
4184 static int ext4_rename2(struct mnt_idmap *idmap,
4185 struct inode *old_dir, struct dentry *old_dentry,
4186 struct inode *new_dir, struct dentry *new_dentry,
4187 unsigned int flags)
4188 {
4189 int err;
4190
4191 if (unlikely(ext4_forced_shutdown(old_dir->i_sb)))
4192 return -EIO;
4193
4194 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4195 return -EINVAL;
4196
4197 err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
4198 flags);
4199 if (err)
4200 return err;
4201
4202 if (flags & RENAME_EXCHANGE) {
4203 return ext4_cross_rename(old_dir, old_dentry,
4204 new_dir, new_dentry);
4205 }
4206
4207 return ext4_rename(idmap, old_dir, old_dentry, new_dir, new_dentry, flags);
4208 }
4209
4210 /*
4211 * directories can handle most operations...
4212 */
4213 const struct inode_operations ext4_dir_inode_operations = {
4214 .create = ext4_create,
4215 .lookup = ext4_lookup,
4216 .link = ext4_link,
4217 .unlink = ext4_unlink,
4218 .symlink = ext4_symlink,
4219 .mkdir = ext4_mkdir,
4220 .rmdir = ext4_rmdir,
4221 .mknod = ext4_mknod,
4222 .tmpfile = ext4_tmpfile,
4223 .rename = ext4_rename2,
4224 .setattr = ext4_setattr,
4225 .getattr = ext4_getattr,
4226 .listxattr = ext4_listxattr,
4227 .get_inode_acl = ext4_get_acl,
4228 .set_acl = ext4_set_acl,
4229 .fiemap = ext4_fiemap,
4230 .fileattr_get = ext4_fileattr_get,
4231 .fileattr_set = ext4_fileattr_set,
4232 };
4233
4234 const struct inode_operations ext4_special_inode_operations = {
4235 .setattr = ext4_setattr,
4236 .getattr = ext4_getattr,
4237 .listxattr = ext4_listxattr,
4238 .get_inode_acl = ext4_get_acl,
4239 .set_acl = ext4_set_acl,
4240 };
4241
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.