xref: /linux/fs/ext4/namei.c (revision 2dbc0838bcf24ca59cabc3130cf3b1d6809cdcd4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/namei.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  from
11  *
12  *  linux/fs/minix/namei.c
13  *
14  *  Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  *  Big-endian to little-endian byte-swapping/bitmaps by
17  *        David S. Miller (davem@caip.rutgers.edu), 1995
18  *  Directory entry file type support and forward compatibility hooks
19  *	for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20  *  Hash Tree Directory indexing (c)
21  *	Daniel Phillips, 2001
22  *  Hash Tree Directory indexing porting
23  *	Christopher Li, 2002
24  *  Hash Tree Directory indexing cleanup
25  *	Theodore Ts'o, 2002
26  */
27 
28 #include <linux/fs.h>
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include <linux/iversion.h>
38 #include <linux/unicode.h>
39 #include "ext4.h"
40 #include "ext4_jbd2.h"
41 
42 #include "xattr.h"
43 #include "acl.h"
44 
45 #include <trace/events/ext4.h>
46 /*
47  * define how far ahead to read directories while searching them.
48  */
49 #define NAMEI_RA_CHUNKS  2
50 #define NAMEI_RA_BLOCKS  4
51 #define NAMEI_RA_SIZE	     (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
52 
53 static struct buffer_head *ext4_append(handle_t *handle,
54 					struct inode *inode,
55 					ext4_lblk_t *block)
56 {
57 	struct buffer_head *bh;
58 	int err;
59 
60 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
61 		     ((inode->i_size >> 10) >=
62 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
63 		return ERR_PTR(-ENOSPC);
64 
65 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
66 
67 	bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
68 	if (IS_ERR(bh))
69 		return bh;
70 	inode->i_size += inode->i_sb->s_blocksize;
71 	EXT4_I(inode)->i_disksize = inode->i_size;
72 	BUFFER_TRACE(bh, "get_write_access");
73 	err = ext4_journal_get_write_access(handle, bh);
74 	if (err) {
75 		brelse(bh);
76 		ext4_std_error(inode->i_sb, err);
77 		return ERR_PTR(err);
78 	}
79 	return bh;
80 }
81 
82 static int ext4_dx_csum_verify(struct inode *inode,
83 			       struct ext4_dir_entry *dirent);
84 
85 /*
86  * Hints to ext4_read_dirblock regarding whether we expect a directory
87  * block being read to be an index block, or a block containing
88  * directory entries (and if the latter, whether it was found via a
89  * logical block in an htree index block).  This is used to control
90  * what sort of sanity checkinig ext4_read_dirblock() will do on the
91  * directory block read from the storage device.  EITHER will means
92  * the caller doesn't know what kind of directory block will be read,
93  * so no specific verification will be done.
94  */
95 typedef enum {
96 	EITHER, INDEX, DIRENT, DIRENT_HTREE
97 } dirblock_type_t;
98 
99 #define ext4_read_dirblock(inode, block, type) \
100 	__ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
101 
102 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
103 						ext4_lblk_t block,
104 						dirblock_type_t type,
105 						const char *func,
106 						unsigned int line)
107 {
108 	struct buffer_head *bh;
109 	struct ext4_dir_entry *dirent;
110 	int is_dx_block = 0;
111 
112 	bh = ext4_bread(NULL, inode, block, 0);
113 	if (IS_ERR(bh)) {
114 		__ext4_warning(inode->i_sb, func, line,
115 			       "inode #%lu: lblock %lu: comm %s: "
116 			       "error %ld reading directory block",
117 			       inode->i_ino, (unsigned long)block,
118 			       current->comm, PTR_ERR(bh));
119 
120 		return bh;
121 	}
122 	if (!bh && (type == INDEX || type == DIRENT_HTREE)) {
123 		ext4_error_inode(inode, func, line, block,
124 				 "Directory hole found for htree %s block",
125 				 (type == INDEX) ? "index" : "leaf");
126 		return ERR_PTR(-EFSCORRUPTED);
127 	}
128 	if (!bh)
129 		return NULL;
130 	dirent = (struct ext4_dir_entry *) bh->b_data;
131 	/* Determine whether or not we have an index block */
132 	if (is_dx(inode)) {
133 		if (block == 0)
134 			is_dx_block = 1;
135 		else if (ext4_rec_len_from_disk(dirent->rec_len,
136 						inode->i_sb->s_blocksize) ==
137 			 inode->i_sb->s_blocksize)
138 			is_dx_block = 1;
139 	}
140 	if (!is_dx_block && type == INDEX) {
141 		ext4_error_inode(inode, func, line, block,
142 		       "directory leaf block found instead of index block");
143 		brelse(bh);
144 		return ERR_PTR(-EFSCORRUPTED);
145 	}
146 	if (!ext4_has_metadata_csum(inode->i_sb) ||
147 	    buffer_verified(bh))
148 		return bh;
149 
150 	/*
151 	 * An empty leaf block can get mistaken for a index block; for
152 	 * this reason, we can only check the index checksum when the
153 	 * caller is sure it should be an index block.
154 	 */
155 	if (is_dx_block && type == INDEX) {
156 		if (ext4_dx_csum_verify(inode, dirent))
157 			set_buffer_verified(bh);
158 		else {
159 			ext4_error_inode(inode, func, line, block,
160 					 "Directory index failed checksum");
161 			brelse(bh);
162 			return ERR_PTR(-EFSBADCRC);
163 		}
164 	}
165 	if (!is_dx_block) {
166 		if (ext4_dirblock_csum_verify(inode, bh))
167 			set_buffer_verified(bh);
168 		else {
169 			ext4_error_inode(inode, func, line, block,
170 					 "Directory block failed checksum");
171 			brelse(bh);
172 			return ERR_PTR(-EFSBADCRC);
173 		}
174 	}
175 	return bh;
176 }
177 
178 #ifndef assert
179 #define assert(test) J_ASSERT(test)
180 #endif
181 
182 #ifdef DX_DEBUG
183 #define dxtrace(command) command
184 #else
185 #define dxtrace(command)
186 #endif
187 
188 struct fake_dirent
189 {
190 	__le32 inode;
191 	__le16 rec_len;
192 	u8 name_len;
193 	u8 file_type;
194 };
195 
196 struct dx_countlimit
197 {
198 	__le16 limit;
199 	__le16 count;
200 };
201 
202 struct dx_entry
203 {
204 	__le32 hash;
205 	__le32 block;
206 };
207 
208 /*
209  * dx_root_info is laid out so that if it should somehow get overlaid by a
210  * dirent the two low bits of the hash version will be zero.  Therefore, the
211  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
212  */
213 
214 struct dx_root
215 {
216 	struct fake_dirent dot;
217 	char dot_name[4];
218 	struct fake_dirent dotdot;
219 	char dotdot_name[4];
220 	struct dx_root_info
221 	{
222 		__le32 reserved_zero;
223 		u8 hash_version;
224 		u8 info_length; /* 8 */
225 		u8 indirect_levels;
226 		u8 unused_flags;
227 	}
228 	info;
229 	struct dx_entry	entries[0];
230 };
231 
232 struct dx_node
233 {
234 	struct fake_dirent fake;
235 	struct dx_entry	entries[0];
236 };
237 
238 
239 struct dx_frame
240 {
241 	struct buffer_head *bh;
242 	struct dx_entry *entries;
243 	struct dx_entry *at;
244 };
245 
246 struct dx_map_entry
247 {
248 	u32 hash;
249 	u16 offs;
250 	u16 size;
251 };
252 
253 /*
254  * This goes at the end of each htree block.
255  */
256 struct dx_tail {
257 	u32 dt_reserved;
258 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
259 };
260 
261 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
262 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
263 static inline unsigned dx_get_hash(struct dx_entry *entry);
264 static void dx_set_hash(struct dx_entry *entry, unsigned value);
265 static unsigned dx_get_count(struct dx_entry *entries);
266 static unsigned dx_get_limit(struct dx_entry *entries);
267 static void dx_set_count(struct dx_entry *entries, unsigned value);
268 static void dx_set_limit(struct dx_entry *entries, unsigned value);
269 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
270 static unsigned dx_node_limit(struct inode *dir);
271 static struct dx_frame *dx_probe(struct ext4_filename *fname,
272 				 struct inode *dir,
273 				 struct dx_hash_info *hinfo,
274 				 struct dx_frame *frame);
275 static void dx_release(struct dx_frame *frames);
276 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
277 		       unsigned blocksize, struct dx_hash_info *hinfo,
278 		       struct dx_map_entry map[]);
279 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
280 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
281 		struct dx_map_entry *offsets, int count, unsigned blocksize);
282 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
283 static void dx_insert_block(struct dx_frame *frame,
284 					u32 hash, ext4_lblk_t block);
285 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
286 				 struct dx_frame *frame,
287 				 struct dx_frame *frames,
288 				 __u32 *start_hash);
289 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
290 		struct ext4_filename *fname,
291 		struct ext4_dir_entry_2 **res_dir);
292 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
293 			     struct inode *dir, struct inode *inode);
294 
295 /* checksumming functions */
296 void ext4_initialize_dirent_tail(struct buffer_head *bh,
297 				 unsigned int blocksize)
298 {
299 	struct ext4_dir_entry_tail *t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
300 
301 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
302 	t->det_rec_len = ext4_rec_len_to_disk(
303 			sizeof(struct ext4_dir_entry_tail), blocksize);
304 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
305 }
306 
307 /* Walk through a dirent block to find a checksum "dirent" at the tail */
308 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
309 						   struct buffer_head *bh)
310 {
311 	struct ext4_dir_entry_tail *t;
312 
313 #ifdef PARANOID
314 	struct ext4_dir_entry *d, *top;
315 
316 	d = (struct ext4_dir_entry *)bh->b_data;
317 	top = (struct ext4_dir_entry *)(bh->b_data +
318 		(EXT4_BLOCK_SIZE(inode->i_sb) -
319 		 sizeof(struct ext4_dir_entry_tail)));
320 	while (d < top && d->rec_len)
321 		d = (struct ext4_dir_entry *)(((void *)d) +
322 		    le16_to_cpu(d->rec_len));
323 
324 	if (d != top)
325 		return NULL;
326 
327 	t = (struct ext4_dir_entry_tail *)d;
328 #else
329 	t = EXT4_DIRENT_TAIL(bh->b_data, EXT4_BLOCK_SIZE(inode->i_sb));
330 #endif
331 
332 	if (t->det_reserved_zero1 ||
333 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
334 	    t->det_reserved_zero2 ||
335 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
336 		return NULL;
337 
338 	return t;
339 }
340 
341 static __le32 ext4_dirblock_csum(struct inode *inode, void *dirent, int size)
342 {
343 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
344 	struct ext4_inode_info *ei = EXT4_I(inode);
345 	__u32 csum;
346 
347 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
348 	return cpu_to_le32(csum);
349 }
350 
351 #define warn_no_space_for_csum(inode)					\
352 	__warn_no_space_for_csum((inode), __func__, __LINE__)
353 
354 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
355 				     unsigned int line)
356 {
357 	__ext4_warning_inode(inode, func, line,
358 		"No space for directory leaf checksum. Please run e2fsck -D.");
359 }
360 
361 int ext4_dirblock_csum_verify(struct inode *inode, struct buffer_head *bh)
362 {
363 	struct ext4_dir_entry_tail *t;
364 
365 	if (!ext4_has_metadata_csum(inode->i_sb))
366 		return 1;
367 
368 	t = get_dirent_tail(inode, bh);
369 	if (!t) {
370 		warn_no_space_for_csum(inode);
371 		return 0;
372 	}
373 
374 	if (t->det_checksum != ext4_dirblock_csum(inode, bh->b_data,
375 						  (char *)t - bh->b_data))
376 		return 0;
377 
378 	return 1;
379 }
380 
381 static void ext4_dirblock_csum_set(struct inode *inode,
382 				 struct buffer_head *bh)
383 {
384 	struct ext4_dir_entry_tail *t;
385 
386 	if (!ext4_has_metadata_csum(inode->i_sb))
387 		return;
388 
389 	t = get_dirent_tail(inode, bh);
390 	if (!t) {
391 		warn_no_space_for_csum(inode);
392 		return;
393 	}
394 
395 	t->det_checksum = ext4_dirblock_csum(inode, bh->b_data,
396 					     (char *)t - bh->b_data);
397 }
398 
399 int ext4_handle_dirty_dirblock(handle_t *handle,
400 			       struct inode *inode,
401 			       struct buffer_head *bh)
402 {
403 	ext4_dirblock_csum_set(inode, bh);
404 	return ext4_handle_dirty_metadata(handle, inode, bh);
405 }
406 
407 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
408 					       struct ext4_dir_entry *dirent,
409 					       int *offset)
410 {
411 	struct ext4_dir_entry *dp;
412 	struct dx_root_info *root;
413 	int count_offset;
414 
415 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
416 		count_offset = 8;
417 	else if (le16_to_cpu(dirent->rec_len) == 12) {
418 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
419 		if (le16_to_cpu(dp->rec_len) !=
420 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
421 			return NULL;
422 		root = (struct dx_root_info *)(((void *)dp + 12));
423 		if (root->reserved_zero ||
424 		    root->info_length != sizeof(struct dx_root_info))
425 			return NULL;
426 		count_offset = 32;
427 	} else
428 		return NULL;
429 
430 	if (offset)
431 		*offset = count_offset;
432 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
433 }
434 
435 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
436 			   int count_offset, int count, struct dx_tail *t)
437 {
438 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
439 	struct ext4_inode_info *ei = EXT4_I(inode);
440 	__u32 csum;
441 	int size;
442 	__u32 dummy_csum = 0;
443 	int offset = offsetof(struct dx_tail, dt_checksum);
444 
445 	size = count_offset + (count * sizeof(struct dx_entry));
446 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
447 	csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
448 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
449 
450 	return cpu_to_le32(csum);
451 }
452 
453 static int ext4_dx_csum_verify(struct inode *inode,
454 			       struct ext4_dir_entry *dirent)
455 {
456 	struct dx_countlimit *c;
457 	struct dx_tail *t;
458 	int count_offset, limit, count;
459 
460 	if (!ext4_has_metadata_csum(inode->i_sb))
461 		return 1;
462 
463 	c = get_dx_countlimit(inode, dirent, &count_offset);
464 	if (!c) {
465 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
466 		return 0;
467 	}
468 	limit = le16_to_cpu(c->limit);
469 	count = le16_to_cpu(c->count);
470 	if (count_offset + (limit * sizeof(struct dx_entry)) >
471 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
472 		warn_no_space_for_csum(inode);
473 		return 0;
474 	}
475 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
476 
477 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
478 					    count, t))
479 		return 0;
480 	return 1;
481 }
482 
483 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
484 {
485 	struct dx_countlimit *c;
486 	struct dx_tail *t;
487 	int count_offset, limit, count;
488 
489 	if (!ext4_has_metadata_csum(inode->i_sb))
490 		return;
491 
492 	c = get_dx_countlimit(inode, dirent, &count_offset);
493 	if (!c) {
494 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
495 		return;
496 	}
497 	limit = le16_to_cpu(c->limit);
498 	count = le16_to_cpu(c->count);
499 	if (count_offset + (limit * sizeof(struct dx_entry)) >
500 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
501 		warn_no_space_for_csum(inode);
502 		return;
503 	}
504 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
505 
506 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
507 }
508 
509 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
510 					    struct inode *inode,
511 					    struct buffer_head *bh)
512 {
513 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
514 	return ext4_handle_dirty_metadata(handle, inode, bh);
515 }
516 
517 /*
518  * p is at least 6 bytes before the end of page
519  */
520 static inline struct ext4_dir_entry_2 *
521 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
522 {
523 	return (struct ext4_dir_entry_2 *)((char *)p +
524 		ext4_rec_len_from_disk(p->rec_len, blocksize));
525 }
526 
527 /*
528  * Future: use high four bits of block for coalesce-on-delete flags
529  * Mask them off for now.
530  */
531 
532 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
533 {
534 	return le32_to_cpu(entry->block) & 0x0fffffff;
535 }
536 
537 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
538 {
539 	entry->block = cpu_to_le32(value);
540 }
541 
542 static inline unsigned dx_get_hash(struct dx_entry *entry)
543 {
544 	return le32_to_cpu(entry->hash);
545 }
546 
547 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
548 {
549 	entry->hash = cpu_to_le32(value);
550 }
551 
552 static inline unsigned dx_get_count(struct dx_entry *entries)
553 {
554 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
555 }
556 
557 static inline unsigned dx_get_limit(struct dx_entry *entries)
558 {
559 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
560 }
561 
562 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
563 {
564 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
565 }
566 
567 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
568 {
569 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
570 }
571 
572 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
573 {
574 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
575 		EXT4_DIR_REC_LEN(2) - infosize;
576 
577 	if (ext4_has_metadata_csum(dir->i_sb))
578 		entry_space -= sizeof(struct dx_tail);
579 	return entry_space / sizeof(struct dx_entry);
580 }
581 
582 static inline unsigned dx_node_limit(struct inode *dir)
583 {
584 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
585 
586 	if (ext4_has_metadata_csum(dir->i_sb))
587 		entry_space -= sizeof(struct dx_tail);
588 	return entry_space / sizeof(struct dx_entry);
589 }
590 
591 /*
592  * Debug
593  */
594 #ifdef DX_DEBUG
595 static void dx_show_index(char * label, struct dx_entry *entries)
596 {
597 	int i, n = dx_get_count (entries);
598 	printk(KERN_DEBUG "%s index", label);
599 	for (i = 0; i < n; i++) {
600 		printk(KERN_CONT " %x->%lu",
601 		       i ? dx_get_hash(entries + i) : 0,
602 		       (unsigned long)dx_get_block(entries + i));
603 	}
604 	printk(KERN_CONT "\n");
605 }
606 
607 struct stats
608 {
609 	unsigned names;
610 	unsigned space;
611 	unsigned bcount;
612 };
613 
614 static struct stats dx_show_leaf(struct inode *dir,
615 				struct dx_hash_info *hinfo,
616 				struct ext4_dir_entry_2 *de,
617 				int size, int show_names)
618 {
619 	unsigned names = 0, space = 0;
620 	char *base = (char *) de;
621 	struct dx_hash_info h = *hinfo;
622 
623 	printk("names: ");
624 	while ((char *) de < base + size)
625 	{
626 		if (de->inode)
627 		{
628 			if (show_names)
629 			{
630 #ifdef CONFIG_FS_ENCRYPTION
631 				int len;
632 				char *name;
633 				struct fscrypt_str fname_crypto_str =
634 					FSTR_INIT(NULL, 0);
635 				int res = 0;
636 
637 				name  = de->name;
638 				len = de->name_len;
639 				if (IS_ENCRYPTED(dir))
640 					res = fscrypt_get_encryption_info(dir);
641 				if (res) {
642 					printk(KERN_WARNING "Error setting up"
643 					       " fname crypto: %d\n", res);
644 				}
645 				if (!fscrypt_has_encryption_key(dir)) {
646 					/* Directory is not encrypted */
647 					ext4fs_dirhash(dir, de->name,
648 						de->name_len, &h);
649 					printk("%*.s:(U)%x.%u ", len,
650 					       name, h.hash,
651 					       (unsigned) ((char *) de
652 							   - base));
653 				} else {
654 					struct fscrypt_str de_name =
655 						FSTR_INIT(name, len);
656 
657 					/* Directory is encrypted */
658 					res = fscrypt_fname_alloc_buffer(
659 						dir, len,
660 						&fname_crypto_str);
661 					if (res)
662 						printk(KERN_WARNING "Error "
663 							"allocating crypto "
664 							"buffer--skipping "
665 							"crypto\n");
666 					res = fscrypt_fname_disk_to_usr(dir,
667 						0, 0, &de_name,
668 						&fname_crypto_str);
669 					if (res) {
670 						printk(KERN_WARNING "Error "
671 							"converting filename "
672 							"from disk to usr"
673 							"\n");
674 						name = "??";
675 						len = 2;
676 					} else {
677 						name = fname_crypto_str.name;
678 						len = fname_crypto_str.len;
679 					}
680 					ext4fs_dirhash(dir, de->name,
681 						       de->name_len, &h);
682 					printk("%*.s:(E)%x.%u ", len, name,
683 					       h.hash, (unsigned) ((char *) de
684 								   - base));
685 					fscrypt_fname_free_buffer(
686 							&fname_crypto_str);
687 				}
688 #else
689 				int len = de->name_len;
690 				char *name = de->name;
691 				ext4fs_dirhash(dir, de->name, de->name_len, &h);
692 				printk("%*.s:%x.%u ", len, name, h.hash,
693 				       (unsigned) ((char *) de - base));
694 #endif
695 			}
696 			space += EXT4_DIR_REC_LEN(de->name_len);
697 			names++;
698 		}
699 		de = ext4_next_entry(de, size);
700 	}
701 	printk(KERN_CONT "(%i)\n", names);
702 	return (struct stats) { names, space, 1 };
703 }
704 
705 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
706 			     struct dx_entry *entries, int levels)
707 {
708 	unsigned blocksize = dir->i_sb->s_blocksize;
709 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
710 	unsigned bcount = 0;
711 	struct buffer_head *bh;
712 	printk("%i indexed blocks...\n", count);
713 	for (i = 0; i < count; i++, entries++)
714 	{
715 		ext4_lblk_t block = dx_get_block(entries);
716 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
717 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
718 		struct stats stats;
719 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
720 		bh = ext4_bread(NULL,dir, block, 0);
721 		if (!bh || IS_ERR(bh))
722 			continue;
723 		stats = levels?
724 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
725 		   dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
726 			bh->b_data, blocksize, 0);
727 		names += stats.names;
728 		space += stats.space;
729 		bcount += stats.bcount;
730 		brelse(bh);
731 	}
732 	if (bcount)
733 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
734 		       levels ? "" : "   ", names, space/bcount,
735 		       (space/bcount)*100/blocksize);
736 	return (struct stats) { names, space, bcount};
737 }
738 #endif /* DX_DEBUG */
739 
740 /*
741  * Probe for a directory leaf block to search.
742  *
743  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
744  * error in the directory index, and the caller should fall back to
745  * searching the directory normally.  The callers of dx_probe **MUST**
746  * check for this error code, and make sure it never gets reflected
747  * back to userspace.
748  */
749 static struct dx_frame *
750 dx_probe(struct ext4_filename *fname, struct inode *dir,
751 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
752 {
753 	unsigned count, indirect;
754 	struct dx_entry *at, *entries, *p, *q, *m;
755 	struct dx_root *root;
756 	struct dx_frame *frame = frame_in;
757 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
758 	u32 hash;
759 
760 	memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
761 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
762 	if (IS_ERR(frame->bh))
763 		return (struct dx_frame *) frame->bh;
764 
765 	root = (struct dx_root *) frame->bh->b_data;
766 	if (root->info.hash_version != DX_HASH_TEA &&
767 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
768 	    root->info.hash_version != DX_HASH_LEGACY) {
769 		ext4_warning_inode(dir, "Unrecognised inode hash code %u",
770 				   root->info.hash_version);
771 		goto fail;
772 	}
773 	if (fname)
774 		hinfo = &fname->hinfo;
775 	hinfo->hash_version = root->info.hash_version;
776 	if (hinfo->hash_version <= DX_HASH_TEA)
777 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
778 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
779 	if (fname && fname_name(fname))
780 		ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), hinfo);
781 	hash = hinfo->hash;
782 
783 	if (root->info.unused_flags & 1) {
784 		ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
785 				   root->info.unused_flags);
786 		goto fail;
787 	}
788 
789 	indirect = root->info.indirect_levels;
790 	if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
791 		ext4_warning(dir->i_sb,
792 			     "Directory (ino: %lu) htree depth %#06x exceed"
793 			     "supported value", dir->i_ino,
794 			     ext4_dir_htree_level(dir->i_sb));
795 		if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
796 			ext4_warning(dir->i_sb, "Enable large directory "
797 						"feature to access it");
798 		}
799 		goto fail;
800 	}
801 
802 	entries = (struct dx_entry *)(((char *)&root->info) +
803 				      root->info.info_length);
804 
805 	if (dx_get_limit(entries) != dx_root_limit(dir,
806 						   root->info.info_length)) {
807 		ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
808 				   dx_get_limit(entries),
809 				   dx_root_limit(dir, root->info.info_length));
810 		goto fail;
811 	}
812 
813 	dxtrace(printk("Look up %x", hash));
814 	while (1) {
815 		count = dx_get_count(entries);
816 		if (!count || count > dx_get_limit(entries)) {
817 			ext4_warning_inode(dir,
818 					   "dx entry: count %u beyond limit %u",
819 					   count, dx_get_limit(entries));
820 			goto fail;
821 		}
822 
823 		p = entries + 1;
824 		q = entries + count - 1;
825 		while (p <= q) {
826 			m = p + (q - p) / 2;
827 			dxtrace(printk(KERN_CONT "."));
828 			if (dx_get_hash(m) > hash)
829 				q = m - 1;
830 			else
831 				p = m + 1;
832 		}
833 
834 		if (0) { // linear search cross check
835 			unsigned n = count - 1;
836 			at = entries;
837 			while (n--)
838 			{
839 				dxtrace(printk(KERN_CONT ","));
840 				if (dx_get_hash(++at) > hash)
841 				{
842 					at--;
843 					break;
844 				}
845 			}
846 			assert (at == p - 1);
847 		}
848 
849 		at = p - 1;
850 		dxtrace(printk(KERN_CONT " %x->%u\n",
851 			       at == entries ? 0 : dx_get_hash(at),
852 			       dx_get_block(at)));
853 		frame->entries = entries;
854 		frame->at = at;
855 		if (!indirect--)
856 			return frame;
857 		frame++;
858 		frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
859 		if (IS_ERR(frame->bh)) {
860 			ret_err = (struct dx_frame *) frame->bh;
861 			frame->bh = NULL;
862 			goto fail;
863 		}
864 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
865 
866 		if (dx_get_limit(entries) != dx_node_limit(dir)) {
867 			ext4_warning_inode(dir,
868 				"dx entry: limit %u != node limit %u",
869 				dx_get_limit(entries), dx_node_limit(dir));
870 			goto fail;
871 		}
872 	}
873 fail:
874 	while (frame >= frame_in) {
875 		brelse(frame->bh);
876 		frame--;
877 	}
878 
879 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
880 		ext4_warning_inode(dir,
881 			"Corrupt directory, running e2fsck is recommended");
882 	return ret_err;
883 }
884 
885 static void dx_release(struct dx_frame *frames)
886 {
887 	struct dx_root_info *info;
888 	int i;
889 	unsigned int indirect_levels;
890 
891 	if (frames[0].bh == NULL)
892 		return;
893 
894 	info = &((struct dx_root *)frames[0].bh->b_data)->info;
895 	/* save local copy, "info" may be freed after brelse() */
896 	indirect_levels = info->indirect_levels;
897 	for (i = 0; i <= indirect_levels; i++) {
898 		if (frames[i].bh == NULL)
899 			break;
900 		brelse(frames[i].bh);
901 		frames[i].bh = NULL;
902 	}
903 }
904 
905 /*
906  * This function increments the frame pointer to search the next leaf
907  * block, and reads in the necessary intervening nodes if the search
908  * should be necessary.  Whether or not the search is necessary is
909  * controlled by the hash parameter.  If the hash value is even, then
910  * the search is only continued if the next block starts with that
911  * hash value.  This is used if we are searching for a specific file.
912  *
913  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
914  *
915  * This function returns 1 if the caller should continue to search,
916  * or 0 if it should not.  If there is an error reading one of the
917  * index blocks, it will a negative error code.
918  *
919  * If start_hash is non-null, it will be filled in with the starting
920  * hash of the next page.
921  */
922 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
923 				 struct dx_frame *frame,
924 				 struct dx_frame *frames,
925 				 __u32 *start_hash)
926 {
927 	struct dx_frame *p;
928 	struct buffer_head *bh;
929 	int num_frames = 0;
930 	__u32 bhash;
931 
932 	p = frame;
933 	/*
934 	 * Find the next leaf page by incrementing the frame pointer.
935 	 * If we run out of entries in the interior node, loop around and
936 	 * increment pointer in the parent node.  When we break out of
937 	 * this loop, num_frames indicates the number of interior
938 	 * nodes need to be read.
939 	 */
940 	while (1) {
941 		if (++(p->at) < p->entries + dx_get_count(p->entries))
942 			break;
943 		if (p == frames)
944 			return 0;
945 		num_frames++;
946 		p--;
947 	}
948 
949 	/*
950 	 * If the hash is 1, then continue only if the next page has a
951 	 * continuation hash of any value.  This is used for readdir
952 	 * handling.  Otherwise, check to see if the hash matches the
953 	 * desired contiuation hash.  If it doesn't, return since
954 	 * there's no point to read in the successive index pages.
955 	 */
956 	bhash = dx_get_hash(p->at);
957 	if (start_hash)
958 		*start_hash = bhash;
959 	if ((hash & 1) == 0) {
960 		if ((bhash & ~1) != hash)
961 			return 0;
962 	}
963 	/*
964 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
965 	 * block so no check is necessary
966 	 */
967 	while (num_frames--) {
968 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
969 		if (IS_ERR(bh))
970 			return PTR_ERR(bh);
971 		p++;
972 		brelse(p->bh);
973 		p->bh = bh;
974 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
975 	}
976 	return 1;
977 }
978 
979 
980 /*
981  * This function fills a red-black tree with information from a
982  * directory block.  It returns the number directory entries loaded
983  * into the tree.  If there is an error it is returned in err.
984  */
985 static int htree_dirblock_to_tree(struct file *dir_file,
986 				  struct inode *dir, ext4_lblk_t block,
987 				  struct dx_hash_info *hinfo,
988 				  __u32 start_hash, __u32 start_minor_hash)
989 {
990 	struct buffer_head *bh;
991 	struct ext4_dir_entry_2 *de, *top;
992 	int err = 0, count = 0;
993 	struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
994 
995 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
996 							(unsigned long)block));
997 	bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
998 	if (IS_ERR(bh))
999 		return PTR_ERR(bh);
1000 
1001 	de = (struct ext4_dir_entry_2 *) bh->b_data;
1002 	top = (struct ext4_dir_entry_2 *) ((char *) de +
1003 					   dir->i_sb->s_blocksize -
1004 					   EXT4_DIR_REC_LEN(0));
1005 #ifdef CONFIG_FS_ENCRYPTION
1006 	/* Check if the directory is encrypted */
1007 	if (IS_ENCRYPTED(dir)) {
1008 		err = fscrypt_get_encryption_info(dir);
1009 		if (err < 0) {
1010 			brelse(bh);
1011 			return err;
1012 		}
1013 		err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
1014 						     &fname_crypto_str);
1015 		if (err < 0) {
1016 			brelse(bh);
1017 			return err;
1018 		}
1019 	}
1020 #endif
1021 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1022 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1023 				bh->b_data, bh->b_size,
1024 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1025 					 + ((char *)de - bh->b_data))) {
1026 			/* silently ignore the rest of the block */
1027 			break;
1028 		}
1029 		ext4fs_dirhash(dir, de->name, de->name_len, hinfo);
1030 		if ((hinfo->hash < start_hash) ||
1031 		    ((hinfo->hash == start_hash) &&
1032 		     (hinfo->minor_hash < start_minor_hash)))
1033 			continue;
1034 		if (de->inode == 0)
1035 			continue;
1036 		if (!IS_ENCRYPTED(dir)) {
1037 			tmp_str.name = de->name;
1038 			tmp_str.len = de->name_len;
1039 			err = ext4_htree_store_dirent(dir_file,
1040 				   hinfo->hash, hinfo->minor_hash, de,
1041 				   &tmp_str);
1042 		} else {
1043 			int save_len = fname_crypto_str.len;
1044 			struct fscrypt_str de_name = FSTR_INIT(de->name,
1045 								de->name_len);
1046 
1047 			/* Directory is encrypted */
1048 			err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1049 					hinfo->minor_hash, &de_name,
1050 					&fname_crypto_str);
1051 			if (err) {
1052 				count = err;
1053 				goto errout;
1054 			}
1055 			err = ext4_htree_store_dirent(dir_file,
1056 				   hinfo->hash, hinfo->minor_hash, de,
1057 					&fname_crypto_str);
1058 			fname_crypto_str.len = save_len;
1059 		}
1060 		if (err != 0) {
1061 			count = err;
1062 			goto errout;
1063 		}
1064 		count++;
1065 	}
1066 errout:
1067 	brelse(bh);
1068 #ifdef CONFIG_FS_ENCRYPTION
1069 	fscrypt_fname_free_buffer(&fname_crypto_str);
1070 #endif
1071 	return count;
1072 }
1073 
1074 
1075 /*
1076  * This function fills a red-black tree with information from a
1077  * directory.  We start scanning the directory in hash order, starting
1078  * at start_hash and start_minor_hash.
1079  *
1080  * This function returns the number of entries inserted into the tree,
1081  * or a negative error code.
1082  */
1083 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1084 			 __u32 start_minor_hash, __u32 *next_hash)
1085 {
1086 	struct dx_hash_info hinfo;
1087 	struct ext4_dir_entry_2 *de;
1088 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1089 	struct inode *dir;
1090 	ext4_lblk_t block;
1091 	int count = 0;
1092 	int ret, err;
1093 	__u32 hashval;
1094 	struct fscrypt_str tmp_str;
1095 
1096 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1097 		       start_hash, start_minor_hash));
1098 	dir = file_inode(dir_file);
1099 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1100 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1101 		if (hinfo.hash_version <= DX_HASH_TEA)
1102 			hinfo.hash_version +=
1103 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
1104 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1105 		if (ext4_has_inline_data(dir)) {
1106 			int has_inline_data = 1;
1107 			count = ext4_inlinedir_to_tree(dir_file, dir, 0,
1108 						       &hinfo, start_hash,
1109 						       start_minor_hash,
1110 						       &has_inline_data);
1111 			if (has_inline_data) {
1112 				*next_hash = ~0;
1113 				return count;
1114 			}
1115 		}
1116 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1117 					       start_hash, start_minor_hash);
1118 		*next_hash = ~0;
1119 		return count;
1120 	}
1121 	hinfo.hash = start_hash;
1122 	hinfo.minor_hash = 0;
1123 	frame = dx_probe(NULL, dir, &hinfo, frames);
1124 	if (IS_ERR(frame))
1125 		return PTR_ERR(frame);
1126 
1127 	/* Add '.' and '..' from the htree header */
1128 	if (!start_hash && !start_minor_hash) {
1129 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1130 		tmp_str.name = de->name;
1131 		tmp_str.len = de->name_len;
1132 		err = ext4_htree_store_dirent(dir_file, 0, 0,
1133 					      de, &tmp_str);
1134 		if (err != 0)
1135 			goto errout;
1136 		count++;
1137 	}
1138 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1139 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1140 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1141 		tmp_str.name = de->name;
1142 		tmp_str.len = de->name_len;
1143 		err = ext4_htree_store_dirent(dir_file, 2, 0,
1144 					      de, &tmp_str);
1145 		if (err != 0)
1146 			goto errout;
1147 		count++;
1148 	}
1149 
1150 	while (1) {
1151 		if (fatal_signal_pending(current)) {
1152 			err = -ERESTARTSYS;
1153 			goto errout;
1154 		}
1155 		cond_resched();
1156 		block = dx_get_block(frame->at);
1157 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1158 					     start_hash, start_minor_hash);
1159 		if (ret < 0) {
1160 			err = ret;
1161 			goto errout;
1162 		}
1163 		count += ret;
1164 		hashval = ~0;
1165 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1166 					    frame, frames, &hashval);
1167 		*next_hash = hashval;
1168 		if (ret < 0) {
1169 			err = ret;
1170 			goto errout;
1171 		}
1172 		/*
1173 		 * Stop if:  (a) there are no more entries, or
1174 		 * (b) we have inserted at least one entry and the
1175 		 * next hash value is not a continuation
1176 		 */
1177 		if ((ret == 0) ||
1178 		    (count && ((hashval & 1) == 0)))
1179 			break;
1180 	}
1181 	dx_release(frames);
1182 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1183 		       "next hash: %x\n", count, *next_hash));
1184 	return count;
1185 errout:
1186 	dx_release(frames);
1187 	return (err);
1188 }
1189 
1190 static inline int search_dirblock(struct buffer_head *bh,
1191 				  struct inode *dir,
1192 				  struct ext4_filename *fname,
1193 				  unsigned int offset,
1194 				  struct ext4_dir_entry_2 **res_dir)
1195 {
1196 	return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1197 			       fname, offset, res_dir);
1198 }
1199 
1200 /*
1201  * Directory block splitting, compacting
1202  */
1203 
1204 /*
1205  * Create map of hash values, offsets, and sizes, stored at end of block.
1206  * Returns number of entries mapped.
1207  */
1208 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1209 		       unsigned blocksize, struct dx_hash_info *hinfo,
1210 		       struct dx_map_entry *map_tail)
1211 {
1212 	int count = 0;
1213 	char *base = (char *) de;
1214 	struct dx_hash_info h = *hinfo;
1215 
1216 	while ((char *) de < base + blocksize) {
1217 		if (de->name_len && de->inode) {
1218 			ext4fs_dirhash(dir, de->name, de->name_len, &h);
1219 			map_tail--;
1220 			map_tail->hash = h.hash;
1221 			map_tail->offs = ((char *) de - base)>>2;
1222 			map_tail->size = le16_to_cpu(de->rec_len);
1223 			count++;
1224 			cond_resched();
1225 		}
1226 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1227 		de = ext4_next_entry(de, blocksize);
1228 	}
1229 	return count;
1230 }
1231 
1232 /* Sort map by hash value */
1233 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1234 {
1235 	struct dx_map_entry *p, *q, *top = map + count - 1;
1236 	int more;
1237 	/* Combsort until bubble sort doesn't suck */
1238 	while (count > 2) {
1239 		count = count*10/13;
1240 		if (count - 9 < 2) /* 9, 10 -> 11 */
1241 			count = 11;
1242 		for (p = top, q = p - count; q >= map; p--, q--)
1243 			if (p->hash < q->hash)
1244 				swap(*p, *q);
1245 	}
1246 	/* Garden variety bubble sort */
1247 	do {
1248 		more = 0;
1249 		q = top;
1250 		while (q-- > map) {
1251 			if (q[1].hash >= q[0].hash)
1252 				continue;
1253 			swap(*(q+1), *q);
1254 			more = 1;
1255 		}
1256 	} while(more);
1257 }
1258 
1259 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1260 {
1261 	struct dx_entry *entries = frame->entries;
1262 	struct dx_entry *old = frame->at, *new = old + 1;
1263 	int count = dx_get_count(entries);
1264 
1265 	assert(count < dx_get_limit(entries));
1266 	assert(old < entries + count);
1267 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1268 	dx_set_hash(new, hash);
1269 	dx_set_block(new, block);
1270 	dx_set_count(entries, count + 1);
1271 }
1272 
1273 #ifdef CONFIG_UNICODE
1274 /*
1275  * Test whether a case-insensitive directory entry matches the filename
1276  * being searched for.  If quick is set, assume the name being looked up
1277  * is already in the casefolded form.
1278  *
1279  * Returns: 0 if the directory entry matches, more than 0 if it
1280  * doesn't match or less than zero on error.
1281  */
1282 int ext4_ci_compare(const struct inode *parent, const struct qstr *name,
1283 		    const struct qstr *entry, bool quick)
1284 {
1285 	const struct ext4_sb_info *sbi = EXT4_SB(parent->i_sb);
1286 	const struct unicode_map *um = sbi->s_encoding;
1287 	int ret;
1288 
1289 	if (quick)
1290 		ret = utf8_strncasecmp_folded(um, name, entry);
1291 	else
1292 		ret = utf8_strncasecmp(um, name, entry);
1293 
1294 	if (ret < 0) {
1295 		/* Handle invalid character sequence as either an error
1296 		 * or as an opaque byte sequence.
1297 		 */
1298 		if (ext4_has_strict_mode(sbi))
1299 			return -EINVAL;
1300 
1301 		if (name->len != entry->len)
1302 			return 1;
1303 
1304 		return !!memcmp(name->name, entry->name, name->len);
1305 	}
1306 
1307 	return ret;
1308 }
1309 
1310 void ext4_fname_setup_ci_filename(struct inode *dir, const struct qstr *iname,
1311 				  struct fscrypt_str *cf_name)
1312 {
1313 	int len;
1314 
1315 	if (!IS_CASEFOLDED(dir)) {
1316 		cf_name->name = NULL;
1317 		return;
1318 	}
1319 
1320 	cf_name->name = kmalloc(EXT4_NAME_LEN, GFP_NOFS);
1321 	if (!cf_name->name)
1322 		return;
1323 
1324 	len = utf8_casefold(EXT4_SB(dir->i_sb)->s_encoding,
1325 			    iname, cf_name->name,
1326 			    EXT4_NAME_LEN);
1327 	if (len <= 0) {
1328 		kfree(cf_name->name);
1329 		cf_name->name = NULL;
1330 		return;
1331 	}
1332 	cf_name->len = (unsigned) len;
1333 
1334 }
1335 #endif
1336 
1337 /*
1338  * Test whether a directory entry matches the filename being searched for.
1339  *
1340  * Return: %true if the directory entry matches, otherwise %false.
1341  */
1342 static inline bool ext4_match(const struct inode *parent,
1343 			      const struct ext4_filename *fname,
1344 			      const struct ext4_dir_entry_2 *de)
1345 {
1346 	struct fscrypt_name f;
1347 #ifdef CONFIG_UNICODE
1348 	const struct qstr entry = {.name = de->name, .len = de->name_len};
1349 #endif
1350 
1351 	if (!de->inode)
1352 		return false;
1353 
1354 	f.usr_fname = fname->usr_fname;
1355 	f.disk_name = fname->disk_name;
1356 #ifdef CONFIG_FS_ENCRYPTION
1357 	f.crypto_buf = fname->crypto_buf;
1358 #endif
1359 
1360 #ifdef CONFIG_UNICODE
1361 	if (EXT4_SB(parent->i_sb)->s_encoding && IS_CASEFOLDED(parent)) {
1362 		if (fname->cf_name.name) {
1363 			struct qstr cf = {.name = fname->cf_name.name,
1364 					  .len = fname->cf_name.len};
1365 			return !ext4_ci_compare(parent, &cf, &entry, true);
1366 		}
1367 		return !ext4_ci_compare(parent, fname->usr_fname, &entry,
1368 					false);
1369 	}
1370 #endif
1371 
1372 	return fscrypt_match_name(&f, de->name, de->name_len);
1373 }
1374 
1375 /*
1376  * Returns 0 if not found, -1 on failure, and 1 on success
1377  */
1378 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1379 		    struct inode *dir, struct ext4_filename *fname,
1380 		    unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1381 {
1382 	struct ext4_dir_entry_2 * de;
1383 	char * dlimit;
1384 	int de_len;
1385 
1386 	de = (struct ext4_dir_entry_2 *)search_buf;
1387 	dlimit = search_buf + buf_size;
1388 	while ((char *) de < dlimit) {
1389 		/* this code is executed quadratically often */
1390 		/* do minimal checking `by hand' */
1391 		if ((char *) de + de->name_len <= dlimit &&
1392 		    ext4_match(dir, fname, de)) {
1393 			/* found a match - just to be sure, do
1394 			 * a full check */
1395 			if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1396 						 bh->b_size, offset))
1397 				return -1;
1398 			*res_dir = de;
1399 			return 1;
1400 		}
1401 		/* prevent looping on a bad block */
1402 		de_len = ext4_rec_len_from_disk(de->rec_len,
1403 						dir->i_sb->s_blocksize);
1404 		if (de_len <= 0)
1405 			return -1;
1406 		offset += de_len;
1407 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1408 	}
1409 	return 0;
1410 }
1411 
1412 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1413 			       struct ext4_dir_entry *de)
1414 {
1415 	struct super_block *sb = dir->i_sb;
1416 
1417 	if (!is_dx(dir))
1418 		return 0;
1419 	if (block == 0)
1420 		return 1;
1421 	if (de->inode == 0 &&
1422 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1423 			sb->s_blocksize)
1424 		return 1;
1425 	return 0;
1426 }
1427 
1428 /*
1429  *	__ext4_find_entry()
1430  *
1431  * finds an entry in the specified directory with the wanted name. It
1432  * returns the cache buffer in which the entry was found, and the entry
1433  * itself (as a parameter - res_dir). It does NOT read the inode of the
1434  * entry - you'll have to do that yourself if you want to.
1435  *
1436  * The returned buffer_head has ->b_count elevated.  The caller is expected
1437  * to brelse() it when appropriate.
1438  */
1439 static struct buffer_head *__ext4_find_entry(struct inode *dir,
1440 					     struct ext4_filename *fname,
1441 					     struct ext4_dir_entry_2 **res_dir,
1442 					     int *inlined)
1443 {
1444 	struct super_block *sb;
1445 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1446 	struct buffer_head *bh, *ret = NULL;
1447 	ext4_lblk_t start, block;
1448 	const u8 *name = fname->usr_fname->name;
1449 	size_t ra_max = 0;	/* Number of bh's in the readahead
1450 				   buffer, bh_use[] */
1451 	size_t ra_ptr = 0;	/* Current index into readahead
1452 				   buffer */
1453 	ext4_lblk_t  nblocks;
1454 	int i, namelen, retval;
1455 
1456 	*res_dir = NULL;
1457 	sb = dir->i_sb;
1458 	namelen = fname->usr_fname->len;
1459 	if (namelen > EXT4_NAME_LEN)
1460 		return NULL;
1461 
1462 	if (ext4_has_inline_data(dir)) {
1463 		int has_inline_data = 1;
1464 		ret = ext4_find_inline_entry(dir, fname, res_dir,
1465 					     &has_inline_data);
1466 		if (has_inline_data) {
1467 			if (inlined)
1468 				*inlined = 1;
1469 			goto cleanup_and_exit;
1470 		}
1471 	}
1472 
1473 	if ((namelen <= 2) && (name[0] == '.') &&
1474 	    (name[1] == '.' || name[1] == '\0')) {
1475 		/*
1476 		 * "." or ".." will only be in the first block
1477 		 * NFS may look up ".."; "." should be handled by the VFS
1478 		 */
1479 		block = start = 0;
1480 		nblocks = 1;
1481 		goto restart;
1482 	}
1483 	if (is_dx(dir)) {
1484 		ret = ext4_dx_find_entry(dir, fname, res_dir);
1485 		/*
1486 		 * On success, or if the error was file not found,
1487 		 * return.  Otherwise, fall back to doing a search the
1488 		 * old fashioned way.
1489 		 */
1490 		if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1491 			goto cleanup_and_exit;
1492 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1493 			       "falling back\n"));
1494 		ret = NULL;
1495 	}
1496 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1497 	if (!nblocks) {
1498 		ret = NULL;
1499 		goto cleanup_and_exit;
1500 	}
1501 	start = EXT4_I(dir)->i_dir_start_lookup;
1502 	if (start >= nblocks)
1503 		start = 0;
1504 	block = start;
1505 restart:
1506 	do {
1507 		/*
1508 		 * We deal with the read-ahead logic here.
1509 		 */
1510 		if (ra_ptr >= ra_max) {
1511 			/* Refill the readahead buffer */
1512 			ra_ptr = 0;
1513 			if (block < start)
1514 				ra_max = start - block;
1515 			else
1516 				ra_max = nblocks - block;
1517 			ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1518 			retval = ext4_bread_batch(dir, block, ra_max,
1519 						  false /* wait */, bh_use);
1520 			if (retval) {
1521 				ret = ERR_PTR(retval);
1522 				ra_max = 0;
1523 				goto cleanup_and_exit;
1524 			}
1525 		}
1526 		if ((bh = bh_use[ra_ptr++]) == NULL)
1527 			goto next;
1528 		wait_on_buffer(bh);
1529 		if (!buffer_uptodate(bh)) {
1530 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1531 					 (unsigned long) block);
1532 			brelse(bh);
1533 			ret = ERR_PTR(-EIO);
1534 			goto cleanup_and_exit;
1535 		}
1536 		if (!buffer_verified(bh) &&
1537 		    !is_dx_internal_node(dir, block,
1538 					 (struct ext4_dir_entry *)bh->b_data) &&
1539 		    !ext4_dirblock_csum_verify(dir, bh)) {
1540 			EXT4_ERROR_INODE(dir, "checksumming directory "
1541 					 "block %lu", (unsigned long)block);
1542 			brelse(bh);
1543 			ret = ERR_PTR(-EFSBADCRC);
1544 			goto cleanup_and_exit;
1545 		}
1546 		set_buffer_verified(bh);
1547 		i = search_dirblock(bh, dir, fname,
1548 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1549 		if (i == 1) {
1550 			EXT4_I(dir)->i_dir_start_lookup = block;
1551 			ret = bh;
1552 			goto cleanup_and_exit;
1553 		} else {
1554 			brelse(bh);
1555 			if (i < 0)
1556 				goto cleanup_and_exit;
1557 		}
1558 	next:
1559 		if (++block >= nblocks)
1560 			block = 0;
1561 	} while (block != start);
1562 
1563 	/*
1564 	 * If the directory has grown while we were searching, then
1565 	 * search the last part of the directory before giving up.
1566 	 */
1567 	block = nblocks;
1568 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1569 	if (block < nblocks) {
1570 		start = 0;
1571 		goto restart;
1572 	}
1573 
1574 cleanup_and_exit:
1575 	/* Clean up the read-ahead blocks */
1576 	for (; ra_ptr < ra_max; ra_ptr++)
1577 		brelse(bh_use[ra_ptr]);
1578 	return ret;
1579 }
1580 
1581 static struct buffer_head *ext4_find_entry(struct inode *dir,
1582 					   const struct qstr *d_name,
1583 					   struct ext4_dir_entry_2 **res_dir,
1584 					   int *inlined)
1585 {
1586 	int err;
1587 	struct ext4_filename fname;
1588 	struct buffer_head *bh;
1589 
1590 	err = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1591 	if (err == -ENOENT)
1592 		return NULL;
1593 	if (err)
1594 		return ERR_PTR(err);
1595 
1596 	bh = __ext4_find_entry(dir, &fname, res_dir, inlined);
1597 
1598 	ext4_fname_free_filename(&fname);
1599 	return bh;
1600 }
1601 
1602 static struct buffer_head *ext4_lookup_entry(struct inode *dir,
1603 					     struct dentry *dentry,
1604 					     struct ext4_dir_entry_2 **res_dir)
1605 {
1606 	int err;
1607 	struct ext4_filename fname;
1608 	struct buffer_head *bh;
1609 
1610 	err = ext4_fname_prepare_lookup(dir, dentry, &fname);
1611 	if (err == -ENOENT)
1612 		return NULL;
1613 	if (err)
1614 		return ERR_PTR(err);
1615 
1616 	bh = __ext4_find_entry(dir, &fname, res_dir, NULL);
1617 
1618 	ext4_fname_free_filename(&fname);
1619 	return bh;
1620 }
1621 
1622 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1623 			struct ext4_filename *fname,
1624 			struct ext4_dir_entry_2 **res_dir)
1625 {
1626 	struct super_block * sb = dir->i_sb;
1627 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1628 	struct buffer_head *bh;
1629 	ext4_lblk_t block;
1630 	int retval;
1631 
1632 #ifdef CONFIG_FS_ENCRYPTION
1633 	*res_dir = NULL;
1634 #endif
1635 	frame = dx_probe(fname, dir, NULL, frames);
1636 	if (IS_ERR(frame))
1637 		return (struct buffer_head *) frame;
1638 	do {
1639 		block = dx_get_block(frame->at);
1640 		bh = ext4_read_dirblock(dir, block, DIRENT_HTREE);
1641 		if (IS_ERR(bh))
1642 			goto errout;
1643 
1644 		retval = search_dirblock(bh, dir, fname,
1645 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1646 					 res_dir);
1647 		if (retval == 1)
1648 			goto success;
1649 		brelse(bh);
1650 		if (retval == -1) {
1651 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1652 			goto errout;
1653 		}
1654 
1655 		/* Check to see if we should continue to search */
1656 		retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1657 					       frames, NULL);
1658 		if (retval < 0) {
1659 			ext4_warning_inode(dir,
1660 				"error %d reading directory index block",
1661 				retval);
1662 			bh = ERR_PTR(retval);
1663 			goto errout;
1664 		}
1665 	} while (retval == 1);
1666 
1667 	bh = NULL;
1668 errout:
1669 	dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1670 success:
1671 	dx_release(frames);
1672 	return bh;
1673 }
1674 
1675 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1676 {
1677 	struct inode *inode;
1678 	struct ext4_dir_entry_2 *de;
1679 	struct buffer_head *bh;
1680 
1681 	if (dentry->d_name.len > EXT4_NAME_LEN)
1682 		return ERR_PTR(-ENAMETOOLONG);
1683 
1684 	bh = ext4_lookup_entry(dir, dentry, &de);
1685 	if (IS_ERR(bh))
1686 		return ERR_CAST(bh);
1687 	inode = NULL;
1688 	if (bh) {
1689 		__u32 ino = le32_to_cpu(de->inode);
1690 		brelse(bh);
1691 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1692 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1693 			return ERR_PTR(-EFSCORRUPTED);
1694 		}
1695 		if (unlikely(ino == dir->i_ino)) {
1696 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1697 					 dentry);
1698 			return ERR_PTR(-EFSCORRUPTED);
1699 		}
1700 		inode = ext4_iget(dir->i_sb, ino, EXT4_IGET_NORMAL);
1701 		if (inode == ERR_PTR(-ESTALE)) {
1702 			EXT4_ERROR_INODE(dir,
1703 					 "deleted inode referenced: %u",
1704 					 ino);
1705 			return ERR_PTR(-EFSCORRUPTED);
1706 		}
1707 		if (!IS_ERR(inode) && IS_ENCRYPTED(dir) &&
1708 		    (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1709 		    !fscrypt_has_permitted_context(dir, inode)) {
1710 			ext4_warning(inode->i_sb,
1711 				     "Inconsistent encryption contexts: %lu/%lu",
1712 				     dir->i_ino, inode->i_ino);
1713 			iput(inode);
1714 			return ERR_PTR(-EPERM);
1715 		}
1716 	}
1717 
1718 #ifdef CONFIG_UNICODE
1719 	if (!inode && IS_CASEFOLDED(dir)) {
1720 		/* Eventually we want to call d_add_ci(dentry, NULL)
1721 		 * for negative dentries in the encoding case as
1722 		 * well.  For now, prevent the negative dentry
1723 		 * from being cached.
1724 		 */
1725 		return NULL;
1726 	}
1727 #endif
1728 	return d_splice_alias(inode, dentry);
1729 }
1730 
1731 
1732 struct dentry *ext4_get_parent(struct dentry *child)
1733 {
1734 	__u32 ino;
1735 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1736 	struct ext4_dir_entry_2 * de;
1737 	struct buffer_head *bh;
1738 
1739 	bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1740 	if (IS_ERR(bh))
1741 		return ERR_CAST(bh);
1742 	if (!bh)
1743 		return ERR_PTR(-ENOENT);
1744 	ino = le32_to_cpu(de->inode);
1745 	brelse(bh);
1746 
1747 	if (!ext4_valid_inum(child->d_sb, ino)) {
1748 		EXT4_ERROR_INODE(d_inode(child),
1749 				 "bad parent inode number: %u", ino);
1750 		return ERR_PTR(-EFSCORRUPTED);
1751 	}
1752 
1753 	return d_obtain_alias(ext4_iget(child->d_sb, ino, EXT4_IGET_NORMAL));
1754 }
1755 
1756 /*
1757  * Move count entries from end of map between two memory locations.
1758  * Returns pointer to last entry moved.
1759  */
1760 static struct ext4_dir_entry_2 *
1761 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1762 		unsigned blocksize)
1763 {
1764 	unsigned rec_len = 0;
1765 
1766 	while (count--) {
1767 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1768 						(from + (map->offs<<2));
1769 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1770 		memcpy (to, de, rec_len);
1771 		((struct ext4_dir_entry_2 *) to)->rec_len =
1772 				ext4_rec_len_to_disk(rec_len, blocksize);
1773 		de->inode = 0;
1774 		map++;
1775 		to += rec_len;
1776 	}
1777 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1778 }
1779 
1780 /*
1781  * Compact each dir entry in the range to the minimal rec_len.
1782  * Returns pointer to last entry in range.
1783  */
1784 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1785 {
1786 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1787 	unsigned rec_len = 0;
1788 
1789 	prev = to = de;
1790 	while ((char*)de < base + blocksize) {
1791 		next = ext4_next_entry(de, blocksize);
1792 		if (de->inode && de->name_len) {
1793 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1794 			if (de > to)
1795 				memmove(to, de, rec_len);
1796 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1797 			prev = to;
1798 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1799 		}
1800 		de = next;
1801 	}
1802 	return prev;
1803 }
1804 
1805 /*
1806  * Split a full leaf block to make room for a new dir entry.
1807  * Allocate a new block, and move entries so that they are approx. equally full.
1808  * Returns pointer to de in block into which the new entry will be inserted.
1809  */
1810 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1811 			struct buffer_head **bh,struct dx_frame *frame,
1812 			struct dx_hash_info *hinfo)
1813 {
1814 	unsigned blocksize = dir->i_sb->s_blocksize;
1815 	unsigned count, continued;
1816 	struct buffer_head *bh2;
1817 	ext4_lblk_t newblock;
1818 	u32 hash2;
1819 	struct dx_map_entry *map;
1820 	char *data1 = (*bh)->b_data, *data2;
1821 	unsigned split, move, size;
1822 	struct ext4_dir_entry_2 *de = NULL, *de2;
1823 	int	csum_size = 0;
1824 	int	err = 0, i;
1825 
1826 	if (ext4_has_metadata_csum(dir->i_sb))
1827 		csum_size = sizeof(struct ext4_dir_entry_tail);
1828 
1829 	bh2 = ext4_append(handle, dir, &newblock);
1830 	if (IS_ERR(bh2)) {
1831 		brelse(*bh);
1832 		*bh = NULL;
1833 		return (struct ext4_dir_entry_2 *) bh2;
1834 	}
1835 
1836 	BUFFER_TRACE(*bh, "get_write_access");
1837 	err = ext4_journal_get_write_access(handle, *bh);
1838 	if (err)
1839 		goto journal_error;
1840 
1841 	BUFFER_TRACE(frame->bh, "get_write_access");
1842 	err = ext4_journal_get_write_access(handle, frame->bh);
1843 	if (err)
1844 		goto journal_error;
1845 
1846 	data2 = bh2->b_data;
1847 
1848 	/* create map in the end of data2 block */
1849 	map = (struct dx_map_entry *) (data2 + blocksize);
1850 	count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1851 			     blocksize, hinfo, map);
1852 	map -= count;
1853 	dx_sort_map(map, count);
1854 	/* Split the existing block in the middle, size-wise */
1855 	size = 0;
1856 	move = 0;
1857 	for (i = count-1; i >= 0; i--) {
1858 		/* is more than half of this entry in 2nd half of the block? */
1859 		if (size + map[i].size/2 > blocksize/2)
1860 			break;
1861 		size += map[i].size;
1862 		move++;
1863 	}
1864 	/* map index at which we will split */
1865 	split = count - move;
1866 	hash2 = map[split].hash;
1867 	continued = hash2 == map[split - 1].hash;
1868 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1869 			(unsigned long)dx_get_block(frame->at),
1870 					hash2, split, count-split));
1871 
1872 	/* Fancy dance to stay within two buffers */
1873 	de2 = dx_move_dirents(data1, data2, map + split, count - split,
1874 			      blocksize);
1875 	de = dx_pack_dirents(data1, blocksize);
1876 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1877 					   (char *) de,
1878 					   blocksize);
1879 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1880 					    (char *) de2,
1881 					    blocksize);
1882 	if (csum_size) {
1883 		ext4_initialize_dirent_tail(*bh, blocksize);
1884 		ext4_initialize_dirent_tail(bh2, blocksize);
1885 	}
1886 
1887 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1888 			blocksize, 1));
1889 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1890 			blocksize, 1));
1891 
1892 	/* Which block gets the new entry? */
1893 	if (hinfo->hash >= hash2) {
1894 		swap(*bh, bh2);
1895 		de = de2;
1896 	}
1897 	dx_insert_block(frame, hash2 + continued, newblock);
1898 	err = ext4_handle_dirty_dirblock(handle, dir, bh2);
1899 	if (err)
1900 		goto journal_error;
1901 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1902 	if (err)
1903 		goto journal_error;
1904 	brelse(bh2);
1905 	dxtrace(dx_show_index("frame", frame->entries));
1906 	return de;
1907 
1908 journal_error:
1909 	brelse(*bh);
1910 	brelse(bh2);
1911 	*bh = NULL;
1912 	ext4_std_error(dir->i_sb, err);
1913 	return ERR_PTR(err);
1914 }
1915 
1916 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1917 		      struct buffer_head *bh,
1918 		      void *buf, int buf_size,
1919 		      struct ext4_filename *fname,
1920 		      struct ext4_dir_entry_2 **dest_de)
1921 {
1922 	struct ext4_dir_entry_2 *de;
1923 	unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1924 	int nlen, rlen;
1925 	unsigned int offset = 0;
1926 	char *top;
1927 
1928 	de = (struct ext4_dir_entry_2 *)buf;
1929 	top = buf + buf_size - reclen;
1930 	while ((char *) de <= top) {
1931 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1932 					 buf, buf_size, offset))
1933 			return -EFSCORRUPTED;
1934 		if (ext4_match(dir, fname, de))
1935 			return -EEXIST;
1936 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1937 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1938 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1939 			break;
1940 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1941 		offset += rlen;
1942 	}
1943 	if ((char *) de > top)
1944 		return -ENOSPC;
1945 
1946 	*dest_de = de;
1947 	return 0;
1948 }
1949 
1950 void ext4_insert_dentry(struct inode *inode,
1951 			struct ext4_dir_entry_2 *de,
1952 			int buf_size,
1953 			struct ext4_filename *fname)
1954 {
1955 
1956 	int nlen, rlen;
1957 
1958 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1959 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1960 	if (de->inode) {
1961 		struct ext4_dir_entry_2 *de1 =
1962 			(struct ext4_dir_entry_2 *)((char *)de + nlen);
1963 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1964 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1965 		de = de1;
1966 	}
1967 	de->file_type = EXT4_FT_UNKNOWN;
1968 	de->inode = cpu_to_le32(inode->i_ino);
1969 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1970 	de->name_len = fname_len(fname);
1971 	memcpy(de->name, fname_name(fname), fname_len(fname));
1972 }
1973 
1974 /*
1975  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1976  * it points to a directory entry which is guaranteed to be large
1977  * enough for new directory entry.  If de is NULL, then
1978  * add_dirent_to_buf will attempt search the directory block for
1979  * space.  It will return -ENOSPC if no space is available, and -EIO
1980  * and -EEXIST if directory entry already exists.
1981  */
1982 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1983 			     struct inode *dir,
1984 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1985 			     struct buffer_head *bh)
1986 {
1987 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1988 	int		csum_size = 0;
1989 	int		err;
1990 
1991 	if (ext4_has_metadata_csum(inode->i_sb))
1992 		csum_size = sizeof(struct ext4_dir_entry_tail);
1993 
1994 	if (!de) {
1995 		err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1996 					blocksize - csum_size, fname, &de);
1997 		if (err)
1998 			return err;
1999 	}
2000 	BUFFER_TRACE(bh, "get_write_access");
2001 	err = ext4_journal_get_write_access(handle, bh);
2002 	if (err) {
2003 		ext4_std_error(dir->i_sb, err);
2004 		return err;
2005 	}
2006 
2007 	/* By now the buffer is marked for journaling */
2008 	ext4_insert_dentry(inode, de, blocksize, fname);
2009 
2010 	/*
2011 	 * XXX shouldn't update any times until successful
2012 	 * completion of syscall, but too many callers depend
2013 	 * on this.
2014 	 *
2015 	 * XXX similarly, too many callers depend on
2016 	 * ext4_new_inode() setting the times, but error
2017 	 * recovery deletes the inode, so the worst that can
2018 	 * happen is that the times are slightly out of date
2019 	 * and/or different from the directory change time.
2020 	 */
2021 	dir->i_mtime = dir->i_ctime = current_time(dir);
2022 	ext4_update_dx_flag(dir);
2023 	inode_inc_iversion(dir);
2024 	ext4_mark_inode_dirty(handle, dir);
2025 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2026 	err = ext4_handle_dirty_dirblock(handle, dir, bh);
2027 	if (err)
2028 		ext4_std_error(dir->i_sb, err);
2029 	return 0;
2030 }
2031 
2032 /*
2033  * This converts a one block unindexed directory to a 3 block indexed
2034  * directory, and adds the dentry to the indexed directory.
2035  */
2036 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
2037 			    struct inode *dir,
2038 			    struct inode *inode, struct buffer_head *bh)
2039 {
2040 	struct buffer_head *bh2;
2041 	struct dx_root	*root;
2042 	struct dx_frame	frames[EXT4_HTREE_LEVEL], *frame;
2043 	struct dx_entry *entries;
2044 	struct ext4_dir_entry_2	*de, *de2;
2045 	char		*data2, *top;
2046 	unsigned	len;
2047 	int		retval;
2048 	unsigned	blocksize;
2049 	ext4_lblk_t  block;
2050 	struct fake_dirent *fde;
2051 	int csum_size = 0;
2052 
2053 	if (ext4_has_metadata_csum(inode->i_sb))
2054 		csum_size = sizeof(struct ext4_dir_entry_tail);
2055 
2056 	blocksize =  dir->i_sb->s_blocksize;
2057 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
2058 	BUFFER_TRACE(bh, "get_write_access");
2059 	retval = ext4_journal_get_write_access(handle, bh);
2060 	if (retval) {
2061 		ext4_std_error(dir->i_sb, retval);
2062 		brelse(bh);
2063 		return retval;
2064 	}
2065 	root = (struct dx_root *) bh->b_data;
2066 
2067 	/* The 0th block becomes the root, move the dirents out */
2068 	fde = &root->dotdot;
2069 	de = (struct ext4_dir_entry_2 *)((char *)fde +
2070 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
2071 	if ((char *) de >= (((char *) root) + blocksize)) {
2072 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
2073 		brelse(bh);
2074 		return -EFSCORRUPTED;
2075 	}
2076 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
2077 
2078 	/* Allocate new block for the 0th block's dirents */
2079 	bh2 = ext4_append(handle, dir, &block);
2080 	if (IS_ERR(bh2)) {
2081 		brelse(bh);
2082 		return PTR_ERR(bh2);
2083 	}
2084 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2085 	data2 = bh2->b_data;
2086 
2087 	memcpy(data2, de, len);
2088 	de = (struct ext4_dir_entry_2 *) data2;
2089 	top = data2 + len;
2090 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2091 		de = de2;
2092 	de->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
2093 					   (char *) de, blocksize);
2094 
2095 	if (csum_size)
2096 		ext4_initialize_dirent_tail(bh2, blocksize);
2097 
2098 	/* Initialize the root; the dot dirents already exist */
2099 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2100 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2101 					   blocksize);
2102 	memset (&root->info, 0, sizeof(root->info));
2103 	root->info.info_length = sizeof(root->info);
2104 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2105 	entries = root->entries;
2106 	dx_set_block(entries, 1);
2107 	dx_set_count(entries, 1);
2108 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2109 
2110 	/* Initialize as for dx_probe */
2111 	fname->hinfo.hash_version = root->info.hash_version;
2112 	if (fname->hinfo.hash_version <= DX_HASH_TEA)
2113 		fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2114 	fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2115 	ext4fs_dirhash(dir, fname_name(fname), fname_len(fname), &fname->hinfo);
2116 
2117 	memset(frames, 0, sizeof(frames));
2118 	frame = frames;
2119 	frame->entries = entries;
2120 	frame->at = entries;
2121 	frame->bh = bh;
2122 
2123 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2124 	if (retval)
2125 		goto out_frames;
2126 	retval = ext4_handle_dirty_dirblock(handle, dir, bh2);
2127 	if (retval)
2128 		goto out_frames;
2129 
2130 	de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2131 	if (IS_ERR(de)) {
2132 		retval = PTR_ERR(de);
2133 		goto out_frames;
2134 	}
2135 
2136 	retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2137 out_frames:
2138 	/*
2139 	 * Even if the block split failed, we have to properly write
2140 	 * out all the changes we did so far. Otherwise we can end up
2141 	 * with corrupted filesystem.
2142 	 */
2143 	if (retval)
2144 		ext4_mark_inode_dirty(handle, dir);
2145 	dx_release(frames);
2146 	brelse(bh2);
2147 	return retval;
2148 }
2149 
2150 /*
2151  *	ext4_add_entry()
2152  *
2153  * adds a file entry to the specified directory, using the same
2154  * semantics as ext4_find_entry(). It returns NULL if it failed.
2155  *
2156  * NOTE!! The inode part of 'de' is left at 0 - which means you
2157  * may not sleep between calling this and putting something into
2158  * the entry, as someone else might have used it while you slept.
2159  */
2160 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2161 			  struct inode *inode)
2162 {
2163 	struct inode *dir = d_inode(dentry->d_parent);
2164 	struct buffer_head *bh = NULL;
2165 	struct ext4_dir_entry_2 *de;
2166 	struct super_block *sb;
2167 	struct ext4_sb_info *sbi;
2168 	struct ext4_filename fname;
2169 	int	retval;
2170 	int	dx_fallback=0;
2171 	unsigned blocksize;
2172 	ext4_lblk_t block, blocks;
2173 	int	csum_size = 0;
2174 
2175 	if (ext4_has_metadata_csum(inode->i_sb))
2176 		csum_size = sizeof(struct ext4_dir_entry_tail);
2177 
2178 	sb = dir->i_sb;
2179 	sbi = EXT4_SB(sb);
2180 	blocksize = sb->s_blocksize;
2181 	if (!dentry->d_name.len)
2182 		return -EINVAL;
2183 
2184 #ifdef CONFIG_UNICODE
2185 	if (ext4_has_strict_mode(sbi) && IS_CASEFOLDED(dir) &&
2186 	    utf8_validate(sbi->s_encoding, &dentry->d_name))
2187 		return -EINVAL;
2188 #endif
2189 
2190 	retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2191 	if (retval)
2192 		return retval;
2193 
2194 	if (ext4_has_inline_data(dir)) {
2195 		retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2196 		if (retval < 0)
2197 			goto out;
2198 		if (retval == 1) {
2199 			retval = 0;
2200 			goto out;
2201 		}
2202 	}
2203 
2204 	if (is_dx(dir)) {
2205 		retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2206 		if (!retval || (retval != ERR_BAD_DX_DIR))
2207 			goto out;
2208 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2209 		dx_fallback++;
2210 		ext4_mark_inode_dirty(handle, dir);
2211 	}
2212 	blocks = dir->i_size >> sb->s_blocksize_bits;
2213 	for (block = 0; block < blocks; block++) {
2214 		bh = ext4_read_dirblock(dir, block, DIRENT);
2215 		if (bh == NULL) {
2216 			bh = ext4_bread(handle, dir, block,
2217 					EXT4_GET_BLOCKS_CREATE);
2218 			goto add_to_new_block;
2219 		}
2220 		if (IS_ERR(bh)) {
2221 			retval = PTR_ERR(bh);
2222 			bh = NULL;
2223 			goto out;
2224 		}
2225 		retval = add_dirent_to_buf(handle, &fname, dir, inode,
2226 					   NULL, bh);
2227 		if (retval != -ENOSPC)
2228 			goto out;
2229 
2230 		if (blocks == 1 && !dx_fallback &&
2231 		    ext4_has_feature_dir_index(sb)) {
2232 			retval = make_indexed_dir(handle, &fname, dir,
2233 						  inode, bh);
2234 			bh = NULL; /* make_indexed_dir releases bh */
2235 			goto out;
2236 		}
2237 		brelse(bh);
2238 	}
2239 	bh = ext4_append(handle, dir, &block);
2240 add_to_new_block:
2241 	if (IS_ERR(bh)) {
2242 		retval = PTR_ERR(bh);
2243 		bh = NULL;
2244 		goto out;
2245 	}
2246 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2247 	de->inode = 0;
2248 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2249 
2250 	if (csum_size)
2251 		ext4_initialize_dirent_tail(bh, blocksize);
2252 
2253 	retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2254 out:
2255 	ext4_fname_free_filename(&fname);
2256 	brelse(bh);
2257 	if (retval == 0)
2258 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2259 	return retval;
2260 }
2261 
2262 /*
2263  * Returns 0 for success, or a negative error value
2264  */
2265 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2266 			     struct inode *dir, struct inode *inode)
2267 {
2268 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2269 	struct dx_entry *entries, *at;
2270 	struct buffer_head *bh;
2271 	struct super_block *sb = dir->i_sb;
2272 	struct ext4_dir_entry_2 *de;
2273 	int restart;
2274 	int err;
2275 
2276 again:
2277 	restart = 0;
2278 	frame = dx_probe(fname, dir, NULL, frames);
2279 	if (IS_ERR(frame))
2280 		return PTR_ERR(frame);
2281 	entries = frame->entries;
2282 	at = frame->at;
2283 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT_HTREE);
2284 	if (IS_ERR(bh)) {
2285 		err = PTR_ERR(bh);
2286 		bh = NULL;
2287 		goto cleanup;
2288 	}
2289 
2290 	BUFFER_TRACE(bh, "get_write_access");
2291 	err = ext4_journal_get_write_access(handle, bh);
2292 	if (err)
2293 		goto journal_error;
2294 
2295 	err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2296 	if (err != -ENOSPC)
2297 		goto cleanup;
2298 
2299 	err = 0;
2300 	/* Block full, should compress but for now just split */
2301 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2302 		       dx_get_count(entries), dx_get_limit(entries)));
2303 	/* Need to split index? */
2304 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2305 		ext4_lblk_t newblock;
2306 		int levels = frame - frames + 1;
2307 		unsigned int icount;
2308 		int add_level = 1;
2309 		struct dx_entry *entries2;
2310 		struct dx_node *node2;
2311 		struct buffer_head *bh2;
2312 
2313 		while (frame > frames) {
2314 			if (dx_get_count((frame - 1)->entries) <
2315 			    dx_get_limit((frame - 1)->entries)) {
2316 				add_level = 0;
2317 				break;
2318 			}
2319 			frame--; /* split higher index block */
2320 			at = frame->at;
2321 			entries = frame->entries;
2322 			restart = 1;
2323 		}
2324 		if (add_level && levels == ext4_dir_htree_level(sb)) {
2325 			ext4_warning(sb, "Directory (ino: %lu) index full, "
2326 					 "reach max htree level :%d",
2327 					 dir->i_ino, levels);
2328 			if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2329 				ext4_warning(sb, "Large directory feature is "
2330 						 "not enabled on this "
2331 						 "filesystem");
2332 			}
2333 			err = -ENOSPC;
2334 			goto cleanup;
2335 		}
2336 		icount = dx_get_count(entries);
2337 		bh2 = ext4_append(handle, dir, &newblock);
2338 		if (IS_ERR(bh2)) {
2339 			err = PTR_ERR(bh2);
2340 			goto cleanup;
2341 		}
2342 		node2 = (struct dx_node *)(bh2->b_data);
2343 		entries2 = node2->entries;
2344 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2345 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2346 							   sb->s_blocksize);
2347 		BUFFER_TRACE(frame->bh, "get_write_access");
2348 		err = ext4_journal_get_write_access(handle, frame->bh);
2349 		if (err)
2350 			goto journal_error;
2351 		if (!add_level) {
2352 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2353 			unsigned hash2 = dx_get_hash(entries + icount1);
2354 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2355 				       icount1, icount2));
2356 
2357 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2358 			err = ext4_journal_get_write_access(handle,
2359 							     (frame - 1)->bh);
2360 			if (err)
2361 				goto journal_error;
2362 
2363 			memcpy((char *) entries2, (char *) (entries + icount1),
2364 			       icount2 * sizeof(struct dx_entry));
2365 			dx_set_count(entries, icount1);
2366 			dx_set_count(entries2, icount2);
2367 			dx_set_limit(entries2, dx_node_limit(dir));
2368 
2369 			/* Which index block gets the new entry? */
2370 			if (at - entries >= icount1) {
2371 				frame->at = at = at - entries - icount1 + entries2;
2372 				frame->entries = entries = entries2;
2373 				swap(frame->bh, bh2);
2374 			}
2375 			dx_insert_block((frame - 1), hash2, newblock);
2376 			dxtrace(dx_show_index("node", frame->entries));
2377 			dxtrace(dx_show_index("node",
2378 			       ((struct dx_node *) bh2->b_data)->entries));
2379 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2380 			if (err)
2381 				goto journal_error;
2382 			brelse (bh2);
2383 			err = ext4_handle_dirty_dx_node(handle, dir,
2384 						   (frame - 1)->bh);
2385 			if (err)
2386 				goto journal_error;
2387 			if (restart) {
2388 				err = ext4_handle_dirty_dx_node(handle, dir,
2389 							   frame->bh);
2390 				goto journal_error;
2391 			}
2392 		} else {
2393 			struct dx_root *dxroot;
2394 			memcpy((char *) entries2, (char *) entries,
2395 			       icount * sizeof(struct dx_entry));
2396 			dx_set_limit(entries2, dx_node_limit(dir));
2397 
2398 			/* Set up root */
2399 			dx_set_count(entries, 1);
2400 			dx_set_block(entries + 0, newblock);
2401 			dxroot = (struct dx_root *)frames[0].bh->b_data;
2402 			dxroot->info.indirect_levels += 1;
2403 			dxtrace(printk(KERN_DEBUG
2404 				       "Creating %d level index...\n",
2405 				       dxroot->info.indirect_levels));
2406 			err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2407 			if (err)
2408 				goto journal_error;
2409 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2410 			brelse(bh2);
2411 			restart = 1;
2412 			goto journal_error;
2413 		}
2414 	}
2415 	de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2416 	if (IS_ERR(de)) {
2417 		err = PTR_ERR(de);
2418 		goto cleanup;
2419 	}
2420 	err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2421 	goto cleanup;
2422 
2423 journal_error:
2424 	ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2425 cleanup:
2426 	brelse(bh);
2427 	dx_release(frames);
2428 	/* @restart is true means htree-path has been changed, we need to
2429 	 * repeat dx_probe() to find out valid htree-path
2430 	 */
2431 	if (restart && err == 0)
2432 		goto again;
2433 	return err;
2434 }
2435 
2436 /*
2437  * ext4_generic_delete_entry deletes a directory entry by merging it
2438  * with the previous entry
2439  */
2440 int ext4_generic_delete_entry(handle_t *handle,
2441 			      struct inode *dir,
2442 			      struct ext4_dir_entry_2 *de_del,
2443 			      struct buffer_head *bh,
2444 			      void *entry_buf,
2445 			      int buf_size,
2446 			      int csum_size)
2447 {
2448 	struct ext4_dir_entry_2 *de, *pde;
2449 	unsigned int blocksize = dir->i_sb->s_blocksize;
2450 	int i;
2451 
2452 	i = 0;
2453 	pde = NULL;
2454 	de = (struct ext4_dir_entry_2 *)entry_buf;
2455 	while (i < buf_size - csum_size) {
2456 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2457 					 bh->b_data, bh->b_size, i))
2458 			return -EFSCORRUPTED;
2459 		if (de == de_del)  {
2460 			if (pde)
2461 				pde->rec_len = ext4_rec_len_to_disk(
2462 					ext4_rec_len_from_disk(pde->rec_len,
2463 							       blocksize) +
2464 					ext4_rec_len_from_disk(de->rec_len,
2465 							       blocksize),
2466 					blocksize);
2467 			else
2468 				de->inode = 0;
2469 			inode_inc_iversion(dir);
2470 			return 0;
2471 		}
2472 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2473 		pde = de;
2474 		de = ext4_next_entry(de, blocksize);
2475 	}
2476 	return -ENOENT;
2477 }
2478 
2479 static int ext4_delete_entry(handle_t *handle,
2480 			     struct inode *dir,
2481 			     struct ext4_dir_entry_2 *de_del,
2482 			     struct buffer_head *bh)
2483 {
2484 	int err, csum_size = 0;
2485 
2486 	if (ext4_has_inline_data(dir)) {
2487 		int has_inline_data = 1;
2488 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2489 					       &has_inline_data);
2490 		if (has_inline_data)
2491 			return err;
2492 	}
2493 
2494 	if (ext4_has_metadata_csum(dir->i_sb))
2495 		csum_size = sizeof(struct ext4_dir_entry_tail);
2496 
2497 	BUFFER_TRACE(bh, "get_write_access");
2498 	err = ext4_journal_get_write_access(handle, bh);
2499 	if (unlikely(err))
2500 		goto out;
2501 
2502 	err = ext4_generic_delete_entry(handle, dir, de_del,
2503 					bh, bh->b_data,
2504 					dir->i_sb->s_blocksize, csum_size);
2505 	if (err)
2506 		goto out;
2507 
2508 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2509 	err = ext4_handle_dirty_dirblock(handle, dir, bh);
2510 	if (unlikely(err))
2511 		goto out;
2512 
2513 	return 0;
2514 out:
2515 	if (err != -ENOENT)
2516 		ext4_std_error(dir->i_sb, err);
2517 	return err;
2518 }
2519 
2520 /*
2521  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2522  * since this indicates that nlinks count was previously 1 to avoid overflowing
2523  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2524  * that subdirectory link counts are not being maintained accurately.
2525  *
2526  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2527  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2528  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2529  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2530  */
2531 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2532 {
2533 	inc_nlink(inode);
2534 	if (is_dx(inode) &&
2535 	    (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2536 		set_nlink(inode, 1);
2537 }
2538 
2539 /*
2540  * If a directory had nlink == 1, then we should let it be 1. This indicates
2541  * directory has >EXT4_LINK_MAX subdirs.
2542  */
2543 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2544 {
2545 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2546 		drop_nlink(inode);
2547 }
2548 
2549 
2550 static int ext4_add_nondir(handle_t *handle,
2551 		struct dentry *dentry, struct inode *inode)
2552 {
2553 	int err = ext4_add_entry(handle, dentry, inode);
2554 	if (!err) {
2555 		ext4_mark_inode_dirty(handle, inode);
2556 		d_instantiate_new(dentry, inode);
2557 		return 0;
2558 	}
2559 	drop_nlink(inode);
2560 	unlock_new_inode(inode);
2561 	iput(inode);
2562 	return err;
2563 }
2564 
2565 /*
2566  * By the time this is called, we already have created
2567  * the directory cache entry for the new file, but it
2568  * is so far negative - it has no inode.
2569  *
2570  * If the create succeeds, we fill in the inode information
2571  * with d_instantiate().
2572  */
2573 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2574 		       bool excl)
2575 {
2576 	handle_t *handle;
2577 	struct inode *inode;
2578 	int err, credits, retries = 0;
2579 
2580 	err = dquot_initialize(dir);
2581 	if (err)
2582 		return err;
2583 
2584 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2585 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2586 retry:
2587 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2588 					    NULL, EXT4_HT_DIR, credits);
2589 	handle = ext4_journal_current_handle();
2590 	err = PTR_ERR(inode);
2591 	if (!IS_ERR(inode)) {
2592 		inode->i_op = &ext4_file_inode_operations;
2593 		inode->i_fop = &ext4_file_operations;
2594 		ext4_set_aops(inode);
2595 		err = ext4_add_nondir(handle, dentry, inode);
2596 		if (!err && IS_DIRSYNC(dir))
2597 			ext4_handle_sync(handle);
2598 	}
2599 	if (handle)
2600 		ext4_journal_stop(handle);
2601 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2602 		goto retry;
2603 	return err;
2604 }
2605 
2606 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2607 		      umode_t mode, dev_t rdev)
2608 {
2609 	handle_t *handle;
2610 	struct inode *inode;
2611 	int err, credits, retries = 0;
2612 
2613 	err = dquot_initialize(dir);
2614 	if (err)
2615 		return err;
2616 
2617 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2618 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2619 retry:
2620 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2621 					    NULL, EXT4_HT_DIR, credits);
2622 	handle = ext4_journal_current_handle();
2623 	err = PTR_ERR(inode);
2624 	if (!IS_ERR(inode)) {
2625 		init_special_inode(inode, inode->i_mode, rdev);
2626 		inode->i_op = &ext4_special_inode_operations;
2627 		err = ext4_add_nondir(handle, dentry, inode);
2628 		if (!err && IS_DIRSYNC(dir))
2629 			ext4_handle_sync(handle);
2630 	}
2631 	if (handle)
2632 		ext4_journal_stop(handle);
2633 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2634 		goto retry;
2635 	return err;
2636 }
2637 
2638 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2639 {
2640 	handle_t *handle;
2641 	struct inode *inode;
2642 	int err, retries = 0;
2643 
2644 	err = dquot_initialize(dir);
2645 	if (err)
2646 		return err;
2647 
2648 retry:
2649 	inode = ext4_new_inode_start_handle(dir, mode,
2650 					    NULL, 0, NULL,
2651 					    EXT4_HT_DIR,
2652 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2653 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2654 	handle = ext4_journal_current_handle();
2655 	err = PTR_ERR(inode);
2656 	if (!IS_ERR(inode)) {
2657 		inode->i_op = &ext4_file_inode_operations;
2658 		inode->i_fop = &ext4_file_operations;
2659 		ext4_set_aops(inode);
2660 		d_tmpfile(dentry, inode);
2661 		err = ext4_orphan_add(handle, inode);
2662 		if (err)
2663 			goto err_unlock_inode;
2664 		mark_inode_dirty(inode);
2665 		unlock_new_inode(inode);
2666 	}
2667 	if (handle)
2668 		ext4_journal_stop(handle);
2669 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2670 		goto retry;
2671 	return err;
2672 err_unlock_inode:
2673 	ext4_journal_stop(handle);
2674 	unlock_new_inode(inode);
2675 	return err;
2676 }
2677 
2678 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2679 			  struct ext4_dir_entry_2 *de,
2680 			  int blocksize, int csum_size,
2681 			  unsigned int parent_ino, int dotdot_real_len)
2682 {
2683 	de->inode = cpu_to_le32(inode->i_ino);
2684 	de->name_len = 1;
2685 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2686 					   blocksize);
2687 	strcpy(de->name, ".");
2688 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2689 
2690 	de = ext4_next_entry(de, blocksize);
2691 	de->inode = cpu_to_le32(parent_ino);
2692 	de->name_len = 2;
2693 	if (!dotdot_real_len)
2694 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2695 					(csum_size + EXT4_DIR_REC_LEN(1)),
2696 					blocksize);
2697 	else
2698 		de->rec_len = ext4_rec_len_to_disk(
2699 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2700 	strcpy(de->name, "..");
2701 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2702 
2703 	return ext4_next_entry(de, blocksize);
2704 }
2705 
2706 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2707 			     struct inode *inode)
2708 {
2709 	struct buffer_head *dir_block = NULL;
2710 	struct ext4_dir_entry_2 *de;
2711 	ext4_lblk_t block = 0;
2712 	unsigned int blocksize = dir->i_sb->s_blocksize;
2713 	int csum_size = 0;
2714 	int err;
2715 
2716 	if (ext4_has_metadata_csum(dir->i_sb))
2717 		csum_size = sizeof(struct ext4_dir_entry_tail);
2718 
2719 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2720 		err = ext4_try_create_inline_dir(handle, dir, inode);
2721 		if (err < 0 && err != -ENOSPC)
2722 			goto out;
2723 		if (!err)
2724 			goto out;
2725 	}
2726 
2727 	inode->i_size = 0;
2728 	dir_block = ext4_append(handle, inode, &block);
2729 	if (IS_ERR(dir_block))
2730 		return PTR_ERR(dir_block);
2731 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2732 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2733 	set_nlink(inode, 2);
2734 	if (csum_size)
2735 		ext4_initialize_dirent_tail(dir_block, blocksize);
2736 
2737 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2738 	err = ext4_handle_dirty_dirblock(handle, inode, dir_block);
2739 	if (err)
2740 		goto out;
2741 	set_buffer_verified(dir_block);
2742 out:
2743 	brelse(dir_block);
2744 	return err;
2745 }
2746 
2747 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2748 {
2749 	handle_t *handle;
2750 	struct inode *inode;
2751 	int err, credits, retries = 0;
2752 
2753 	if (EXT4_DIR_LINK_MAX(dir))
2754 		return -EMLINK;
2755 
2756 	err = dquot_initialize(dir);
2757 	if (err)
2758 		return err;
2759 
2760 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2761 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2762 retry:
2763 	inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2764 					    &dentry->d_name,
2765 					    0, NULL, EXT4_HT_DIR, credits);
2766 	handle = ext4_journal_current_handle();
2767 	err = PTR_ERR(inode);
2768 	if (IS_ERR(inode))
2769 		goto out_stop;
2770 
2771 	inode->i_op = &ext4_dir_inode_operations;
2772 	inode->i_fop = &ext4_dir_operations;
2773 	err = ext4_init_new_dir(handle, dir, inode);
2774 	if (err)
2775 		goto out_clear_inode;
2776 	err = ext4_mark_inode_dirty(handle, inode);
2777 	if (!err)
2778 		err = ext4_add_entry(handle, dentry, inode);
2779 	if (err) {
2780 out_clear_inode:
2781 		clear_nlink(inode);
2782 		unlock_new_inode(inode);
2783 		ext4_mark_inode_dirty(handle, inode);
2784 		iput(inode);
2785 		goto out_stop;
2786 	}
2787 	ext4_inc_count(handle, dir);
2788 	ext4_update_dx_flag(dir);
2789 	err = ext4_mark_inode_dirty(handle, dir);
2790 	if (err)
2791 		goto out_clear_inode;
2792 	d_instantiate_new(dentry, inode);
2793 	if (IS_DIRSYNC(dir))
2794 		ext4_handle_sync(handle);
2795 
2796 out_stop:
2797 	if (handle)
2798 		ext4_journal_stop(handle);
2799 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2800 		goto retry;
2801 	return err;
2802 }
2803 
2804 /*
2805  * routine to check that the specified directory is empty (for rmdir)
2806  */
2807 bool ext4_empty_dir(struct inode *inode)
2808 {
2809 	unsigned int offset;
2810 	struct buffer_head *bh;
2811 	struct ext4_dir_entry_2 *de, *de1;
2812 	struct super_block *sb;
2813 
2814 	if (ext4_has_inline_data(inode)) {
2815 		int has_inline_data = 1;
2816 		int ret;
2817 
2818 		ret = empty_inline_dir(inode, &has_inline_data);
2819 		if (has_inline_data)
2820 			return ret;
2821 	}
2822 
2823 	sb = inode->i_sb;
2824 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2825 		EXT4_ERROR_INODE(inode, "invalid size");
2826 		return true;
2827 	}
2828 	/* The first directory block must not be a hole,
2829 	 * so treat it as DIRENT_HTREE
2830 	 */
2831 	bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
2832 	if (IS_ERR(bh))
2833 		return true;
2834 
2835 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2836 	de1 = ext4_next_entry(de, sb->s_blocksize);
2837 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2838 			le32_to_cpu(de1->inode) == 0 ||
2839 			strcmp(".", de->name) || strcmp("..", de1->name)) {
2840 		ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2841 		brelse(bh);
2842 		return true;
2843 	}
2844 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2845 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2846 	de = ext4_next_entry(de1, sb->s_blocksize);
2847 	while (offset < inode->i_size) {
2848 		if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2849 			unsigned int lblock;
2850 			brelse(bh);
2851 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2852 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2853 			if (bh == NULL) {
2854 				offset += sb->s_blocksize;
2855 				continue;
2856 			}
2857 			if (IS_ERR(bh))
2858 				return true;
2859 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2860 		}
2861 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2862 					 bh->b_data, bh->b_size, offset)) {
2863 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2864 							 sb->s_blocksize);
2865 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2866 			continue;
2867 		}
2868 		if (le32_to_cpu(de->inode)) {
2869 			brelse(bh);
2870 			return false;
2871 		}
2872 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2873 		de = ext4_next_entry(de, sb->s_blocksize);
2874 	}
2875 	brelse(bh);
2876 	return true;
2877 }
2878 
2879 /*
2880  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2881  * such inodes, starting at the superblock, in case we crash before the
2882  * file is closed/deleted, or in case the inode truncate spans multiple
2883  * transactions and the last transaction is not recovered after a crash.
2884  *
2885  * At filesystem recovery time, we walk this list deleting unlinked
2886  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2887  *
2888  * Orphan list manipulation functions must be called under i_mutex unless
2889  * we are just creating the inode or deleting it.
2890  */
2891 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2892 {
2893 	struct super_block *sb = inode->i_sb;
2894 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2895 	struct ext4_iloc iloc;
2896 	int err = 0, rc;
2897 	bool dirty = false;
2898 
2899 	if (!sbi->s_journal || is_bad_inode(inode))
2900 		return 0;
2901 
2902 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2903 		     !inode_is_locked(inode));
2904 	/*
2905 	 * Exit early if inode already is on orphan list. This is a big speedup
2906 	 * since we don't have to contend on the global s_orphan_lock.
2907 	 */
2908 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2909 		return 0;
2910 
2911 	/*
2912 	 * Orphan handling is only valid for files with data blocks
2913 	 * being truncated, or files being unlinked. Note that we either
2914 	 * hold i_mutex, or the inode can not be referenced from outside,
2915 	 * so i_nlink should not be bumped due to race
2916 	 */
2917 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2918 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2919 
2920 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2921 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2922 	if (err)
2923 		goto out;
2924 
2925 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2926 	if (err)
2927 		goto out;
2928 
2929 	mutex_lock(&sbi->s_orphan_lock);
2930 	/*
2931 	 * Due to previous errors inode may be already a part of on-disk
2932 	 * orphan list. If so skip on-disk list modification.
2933 	 */
2934 	if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2935 	    (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2936 		/* Insert this inode at the head of the on-disk orphan list */
2937 		NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2938 		sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2939 		dirty = true;
2940 	}
2941 	list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2942 	mutex_unlock(&sbi->s_orphan_lock);
2943 
2944 	if (dirty) {
2945 		err = ext4_handle_dirty_super(handle, sb);
2946 		rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2947 		if (!err)
2948 			err = rc;
2949 		if (err) {
2950 			/*
2951 			 * We have to remove inode from in-memory list if
2952 			 * addition to on disk orphan list failed. Stray orphan
2953 			 * list entries can cause panics at unmount time.
2954 			 */
2955 			mutex_lock(&sbi->s_orphan_lock);
2956 			list_del_init(&EXT4_I(inode)->i_orphan);
2957 			mutex_unlock(&sbi->s_orphan_lock);
2958 		}
2959 	} else
2960 		brelse(iloc.bh);
2961 
2962 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2963 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2964 			inode->i_ino, NEXT_ORPHAN(inode));
2965 out:
2966 	ext4_std_error(sb, err);
2967 	return err;
2968 }
2969 
2970 /*
2971  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2972  * of such inodes stored on disk, because it is finally being cleaned up.
2973  */
2974 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2975 {
2976 	struct list_head *prev;
2977 	struct ext4_inode_info *ei = EXT4_I(inode);
2978 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2979 	__u32 ino_next;
2980 	struct ext4_iloc iloc;
2981 	int err = 0;
2982 
2983 	if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2984 		return 0;
2985 
2986 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2987 		     !inode_is_locked(inode));
2988 	/* Do this quick check before taking global s_orphan_lock. */
2989 	if (list_empty(&ei->i_orphan))
2990 		return 0;
2991 
2992 	if (handle) {
2993 		/* Grab inode buffer early before taking global s_orphan_lock */
2994 		err = ext4_reserve_inode_write(handle, inode, &iloc);
2995 	}
2996 
2997 	mutex_lock(&sbi->s_orphan_lock);
2998 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2999 
3000 	prev = ei->i_orphan.prev;
3001 	list_del_init(&ei->i_orphan);
3002 
3003 	/* If we're on an error path, we may not have a valid
3004 	 * transaction handle with which to update the orphan list on
3005 	 * disk, but we still need to remove the inode from the linked
3006 	 * list in memory. */
3007 	if (!handle || err) {
3008 		mutex_unlock(&sbi->s_orphan_lock);
3009 		goto out_err;
3010 	}
3011 
3012 	ino_next = NEXT_ORPHAN(inode);
3013 	if (prev == &sbi->s_orphan) {
3014 		jbd_debug(4, "superblock will point to %u\n", ino_next);
3015 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
3016 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
3017 		if (err) {
3018 			mutex_unlock(&sbi->s_orphan_lock);
3019 			goto out_brelse;
3020 		}
3021 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
3022 		mutex_unlock(&sbi->s_orphan_lock);
3023 		err = ext4_handle_dirty_super(handle, inode->i_sb);
3024 	} else {
3025 		struct ext4_iloc iloc2;
3026 		struct inode *i_prev =
3027 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
3028 
3029 		jbd_debug(4, "orphan inode %lu will point to %u\n",
3030 			  i_prev->i_ino, ino_next);
3031 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
3032 		if (err) {
3033 			mutex_unlock(&sbi->s_orphan_lock);
3034 			goto out_brelse;
3035 		}
3036 		NEXT_ORPHAN(i_prev) = ino_next;
3037 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
3038 		mutex_unlock(&sbi->s_orphan_lock);
3039 	}
3040 	if (err)
3041 		goto out_brelse;
3042 	NEXT_ORPHAN(inode) = 0;
3043 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
3044 out_err:
3045 	ext4_std_error(inode->i_sb, err);
3046 	return err;
3047 
3048 out_brelse:
3049 	brelse(iloc.bh);
3050 	goto out_err;
3051 }
3052 
3053 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
3054 {
3055 	int retval;
3056 	struct inode *inode;
3057 	struct buffer_head *bh;
3058 	struct ext4_dir_entry_2 *de;
3059 	handle_t *handle = NULL;
3060 
3061 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3062 		return -EIO;
3063 
3064 	/* Initialize quotas before so that eventual writes go in
3065 	 * separate transaction */
3066 	retval = dquot_initialize(dir);
3067 	if (retval)
3068 		return retval;
3069 	retval = dquot_initialize(d_inode(dentry));
3070 	if (retval)
3071 		return retval;
3072 
3073 	retval = -ENOENT;
3074 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3075 	if (IS_ERR(bh))
3076 		return PTR_ERR(bh);
3077 	if (!bh)
3078 		goto end_rmdir;
3079 
3080 	inode = d_inode(dentry);
3081 
3082 	retval = -EFSCORRUPTED;
3083 	if (le32_to_cpu(de->inode) != inode->i_ino)
3084 		goto end_rmdir;
3085 
3086 	retval = -ENOTEMPTY;
3087 	if (!ext4_empty_dir(inode))
3088 		goto end_rmdir;
3089 
3090 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3091 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3092 	if (IS_ERR(handle)) {
3093 		retval = PTR_ERR(handle);
3094 		handle = NULL;
3095 		goto end_rmdir;
3096 	}
3097 
3098 	if (IS_DIRSYNC(dir))
3099 		ext4_handle_sync(handle);
3100 
3101 	retval = ext4_delete_entry(handle, dir, de, bh);
3102 	if (retval)
3103 		goto end_rmdir;
3104 	if (!EXT4_DIR_LINK_EMPTY(inode))
3105 		ext4_warning_inode(inode,
3106 			     "empty directory '%.*s' has too many links (%u)",
3107 			     dentry->d_name.len, dentry->d_name.name,
3108 			     inode->i_nlink);
3109 	inode_inc_iversion(inode);
3110 	clear_nlink(inode);
3111 	/* There's no need to set i_disksize: the fact that i_nlink is
3112 	 * zero will ensure that the right thing happens during any
3113 	 * recovery. */
3114 	inode->i_size = 0;
3115 	ext4_orphan_add(handle, inode);
3116 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3117 	ext4_mark_inode_dirty(handle, inode);
3118 	ext4_dec_count(handle, dir);
3119 	ext4_update_dx_flag(dir);
3120 	ext4_mark_inode_dirty(handle, dir);
3121 
3122 #ifdef CONFIG_UNICODE
3123 	/* VFS negative dentries are incompatible with Encoding and
3124 	 * Case-insensitiveness. Eventually we'll want avoid
3125 	 * invalidating the dentries here, alongside with returning the
3126 	 * negative dentries at ext4_lookup(), when it is better
3127 	 * supported by the VFS for the CI case.
3128 	 */
3129 	if (IS_CASEFOLDED(dir))
3130 		d_invalidate(dentry);
3131 #endif
3132 
3133 end_rmdir:
3134 	brelse(bh);
3135 	if (handle)
3136 		ext4_journal_stop(handle);
3137 	return retval;
3138 }
3139 
3140 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3141 {
3142 	int retval;
3143 	struct inode *inode;
3144 	struct buffer_head *bh;
3145 	struct ext4_dir_entry_2 *de;
3146 	handle_t *handle = NULL;
3147 
3148 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3149 		return -EIO;
3150 
3151 	trace_ext4_unlink_enter(dir, dentry);
3152 	/* Initialize quotas before so that eventual writes go
3153 	 * in separate transaction */
3154 	retval = dquot_initialize(dir);
3155 	if (retval)
3156 		return retval;
3157 	retval = dquot_initialize(d_inode(dentry));
3158 	if (retval)
3159 		return retval;
3160 
3161 	retval = -ENOENT;
3162 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3163 	if (IS_ERR(bh))
3164 		return PTR_ERR(bh);
3165 	if (!bh)
3166 		goto end_unlink;
3167 
3168 	inode = d_inode(dentry);
3169 
3170 	retval = -EFSCORRUPTED;
3171 	if (le32_to_cpu(de->inode) != inode->i_ino)
3172 		goto end_unlink;
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_unlink;
3180 	}
3181 
3182 	if (IS_DIRSYNC(dir))
3183 		ext4_handle_sync(handle);
3184 
3185 	if (inode->i_nlink == 0) {
3186 		ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3187 				   dentry->d_name.len, dentry->d_name.name);
3188 		set_nlink(inode, 1);
3189 	}
3190 	retval = ext4_delete_entry(handle, dir, de, bh);
3191 	if (retval)
3192 		goto end_unlink;
3193 	dir->i_ctime = dir->i_mtime = current_time(dir);
3194 	ext4_update_dx_flag(dir);
3195 	ext4_mark_inode_dirty(handle, dir);
3196 	drop_nlink(inode);
3197 	if (!inode->i_nlink)
3198 		ext4_orphan_add(handle, inode);
3199 	inode->i_ctime = current_time(inode);
3200 	ext4_mark_inode_dirty(handle, inode);
3201 
3202 #ifdef CONFIG_UNICODE
3203 	/* VFS negative dentries are incompatible with Encoding and
3204 	 * Case-insensitiveness. Eventually we'll want avoid
3205 	 * invalidating the dentries here, alongside with returning the
3206 	 * negative dentries at ext4_lookup(), when it is  better
3207 	 * supported by the VFS for the CI case.
3208 	 */
3209 	if (IS_CASEFOLDED(dir))
3210 		d_invalidate(dentry);
3211 #endif
3212 
3213 end_unlink:
3214 	brelse(bh);
3215 	if (handle)
3216 		ext4_journal_stop(handle);
3217 	trace_ext4_unlink_exit(dentry, retval);
3218 	return retval;
3219 }
3220 
3221 static int ext4_symlink(struct inode *dir,
3222 			struct dentry *dentry, const char *symname)
3223 {
3224 	handle_t *handle;
3225 	struct inode *inode;
3226 	int err, len = strlen(symname);
3227 	int credits;
3228 	struct fscrypt_str disk_link;
3229 
3230 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3231 		return -EIO;
3232 
3233 	err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3234 				      &disk_link);
3235 	if (err)
3236 		return err;
3237 
3238 	err = dquot_initialize(dir);
3239 	if (err)
3240 		return err;
3241 
3242 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3243 		/*
3244 		 * For non-fast symlinks, we just allocate inode and put it on
3245 		 * orphan list in the first transaction => we need bitmap,
3246 		 * group descriptor, sb, inode block, quota blocks, and
3247 		 * possibly selinux xattr blocks.
3248 		 */
3249 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3250 			  EXT4_XATTR_TRANS_BLOCKS;
3251 	} else {
3252 		/*
3253 		 * Fast symlink. We have to add entry to directory
3254 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3255 		 * allocate new inode (bitmap, group descriptor, inode block,
3256 		 * quota blocks, sb is already counted in previous macros).
3257 		 */
3258 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3259 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3260 	}
3261 
3262 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3263 					    &dentry->d_name, 0, NULL,
3264 					    EXT4_HT_DIR, credits);
3265 	handle = ext4_journal_current_handle();
3266 	if (IS_ERR(inode)) {
3267 		if (handle)
3268 			ext4_journal_stop(handle);
3269 		return PTR_ERR(inode);
3270 	}
3271 
3272 	if (IS_ENCRYPTED(inode)) {
3273 		err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3274 		if (err)
3275 			goto err_drop_inode;
3276 		inode->i_op = &ext4_encrypted_symlink_inode_operations;
3277 	}
3278 
3279 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3280 		if (!IS_ENCRYPTED(inode))
3281 			inode->i_op = &ext4_symlink_inode_operations;
3282 		inode_nohighmem(inode);
3283 		ext4_set_aops(inode);
3284 		/*
3285 		 * We cannot call page_symlink() with transaction started
3286 		 * because it calls into ext4_write_begin() which can wait
3287 		 * for transaction commit if we are running out of space
3288 		 * and thus we deadlock. So we have to stop transaction now
3289 		 * and restart it when symlink contents is written.
3290 		 *
3291 		 * To keep fs consistent in case of crash, we have to put inode
3292 		 * to orphan list in the mean time.
3293 		 */
3294 		drop_nlink(inode);
3295 		err = ext4_orphan_add(handle, inode);
3296 		ext4_journal_stop(handle);
3297 		handle = NULL;
3298 		if (err)
3299 			goto err_drop_inode;
3300 		err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3301 		if (err)
3302 			goto err_drop_inode;
3303 		/*
3304 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3305 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3306 		 */
3307 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
3308 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3309 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3310 		if (IS_ERR(handle)) {
3311 			err = PTR_ERR(handle);
3312 			handle = NULL;
3313 			goto err_drop_inode;
3314 		}
3315 		set_nlink(inode, 1);
3316 		err = ext4_orphan_del(handle, inode);
3317 		if (err)
3318 			goto err_drop_inode;
3319 	} else {
3320 		/* clear the extent format for fast symlink */
3321 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3322 		if (!IS_ENCRYPTED(inode)) {
3323 			inode->i_op = &ext4_fast_symlink_inode_operations;
3324 			inode->i_link = (char *)&EXT4_I(inode)->i_data;
3325 		}
3326 		memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3327 		       disk_link.len);
3328 		inode->i_size = disk_link.len - 1;
3329 	}
3330 	EXT4_I(inode)->i_disksize = inode->i_size;
3331 	err = ext4_add_nondir(handle, dentry, inode);
3332 	if (!err && IS_DIRSYNC(dir))
3333 		ext4_handle_sync(handle);
3334 
3335 	if (handle)
3336 		ext4_journal_stop(handle);
3337 	goto out_free_encrypted_link;
3338 
3339 err_drop_inode:
3340 	if (handle)
3341 		ext4_journal_stop(handle);
3342 	clear_nlink(inode);
3343 	unlock_new_inode(inode);
3344 	iput(inode);
3345 out_free_encrypted_link:
3346 	if (disk_link.name != (unsigned char *)symname)
3347 		kfree(disk_link.name);
3348 	return err;
3349 }
3350 
3351 static int ext4_link(struct dentry *old_dentry,
3352 		     struct inode *dir, struct dentry *dentry)
3353 {
3354 	handle_t *handle;
3355 	struct inode *inode = d_inode(old_dentry);
3356 	int err, retries = 0;
3357 
3358 	if (inode->i_nlink >= EXT4_LINK_MAX)
3359 		return -EMLINK;
3360 
3361 	err = fscrypt_prepare_link(old_dentry, dir, dentry);
3362 	if (err)
3363 		return err;
3364 
3365 	if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3366 	    (!projid_eq(EXT4_I(dir)->i_projid,
3367 			EXT4_I(old_dentry->d_inode)->i_projid)))
3368 		return -EXDEV;
3369 
3370 	err = dquot_initialize(dir);
3371 	if (err)
3372 		return err;
3373 
3374 retry:
3375 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3376 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3377 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3378 	if (IS_ERR(handle))
3379 		return PTR_ERR(handle);
3380 
3381 	if (IS_DIRSYNC(dir))
3382 		ext4_handle_sync(handle);
3383 
3384 	inode->i_ctime = current_time(inode);
3385 	ext4_inc_count(handle, inode);
3386 	ihold(inode);
3387 
3388 	err = ext4_add_entry(handle, dentry, inode);
3389 	if (!err) {
3390 		ext4_mark_inode_dirty(handle, inode);
3391 		/* this can happen only for tmpfile being
3392 		 * linked the first time
3393 		 */
3394 		if (inode->i_nlink == 1)
3395 			ext4_orphan_del(handle, inode);
3396 		d_instantiate(dentry, inode);
3397 	} else {
3398 		drop_nlink(inode);
3399 		iput(inode);
3400 	}
3401 	ext4_journal_stop(handle);
3402 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3403 		goto retry;
3404 	return err;
3405 }
3406 
3407 
3408 /*
3409  * Try to find buffer head where contains the parent block.
3410  * It should be the inode block if it is inlined or the 1st block
3411  * if it is a normal dir.
3412  */
3413 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3414 					struct inode *inode,
3415 					int *retval,
3416 					struct ext4_dir_entry_2 **parent_de,
3417 					int *inlined)
3418 {
3419 	struct buffer_head *bh;
3420 
3421 	if (!ext4_has_inline_data(inode)) {
3422 		/* The first directory block must not be a hole, so
3423 		 * treat it as DIRENT_HTREE
3424 		 */
3425 		bh = ext4_read_dirblock(inode, 0, DIRENT_HTREE);
3426 		if (IS_ERR(bh)) {
3427 			*retval = PTR_ERR(bh);
3428 			return NULL;
3429 		}
3430 		*parent_de = ext4_next_entry(
3431 					(struct ext4_dir_entry_2 *)bh->b_data,
3432 					inode->i_sb->s_blocksize);
3433 		return bh;
3434 	}
3435 
3436 	*inlined = 1;
3437 	return ext4_get_first_inline_block(inode, parent_de, retval);
3438 }
3439 
3440 struct ext4_renament {
3441 	struct inode *dir;
3442 	struct dentry *dentry;
3443 	struct inode *inode;
3444 	bool is_dir;
3445 	int dir_nlink_delta;
3446 
3447 	/* entry for "dentry" */
3448 	struct buffer_head *bh;
3449 	struct ext4_dir_entry_2 *de;
3450 	int inlined;
3451 
3452 	/* entry for ".." in inode if it's a directory */
3453 	struct buffer_head *dir_bh;
3454 	struct ext4_dir_entry_2 *parent_de;
3455 	int dir_inlined;
3456 };
3457 
3458 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3459 {
3460 	int retval;
3461 
3462 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3463 					      &retval, &ent->parent_de,
3464 					      &ent->dir_inlined);
3465 	if (!ent->dir_bh)
3466 		return retval;
3467 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3468 		return -EFSCORRUPTED;
3469 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3470 	return ext4_journal_get_write_access(handle, ent->dir_bh);
3471 }
3472 
3473 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3474 				  unsigned dir_ino)
3475 {
3476 	int retval;
3477 
3478 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3479 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3480 	if (!ent->dir_inlined) {
3481 		if (is_dx(ent->inode)) {
3482 			retval = ext4_handle_dirty_dx_node(handle,
3483 							   ent->inode,
3484 							   ent->dir_bh);
3485 		} else {
3486 			retval = ext4_handle_dirty_dirblock(handle, ent->inode,
3487 							    ent->dir_bh);
3488 		}
3489 	} else {
3490 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3491 	}
3492 	if (retval) {
3493 		ext4_std_error(ent->dir->i_sb, retval);
3494 		return retval;
3495 	}
3496 	return 0;
3497 }
3498 
3499 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3500 		       unsigned ino, unsigned file_type)
3501 {
3502 	int retval;
3503 
3504 	BUFFER_TRACE(ent->bh, "get write access");
3505 	retval = ext4_journal_get_write_access(handle, ent->bh);
3506 	if (retval)
3507 		return retval;
3508 	ent->de->inode = cpu_to_le32(ino);
3509 	if (ext4_has_feature_filetype(ent->dir->i_sb))
3510 		ent->de->file_type = file_type;
3511 	inode_inc_iversion(ent->dir);
3512 	ent->dir->i_ctime = ent->dir->i_mtime =
3513 		current_time(ent->dir);
3514 	ext4_mark_inode_dirty(handle, ent->dir);
3515 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3516 	if (!ent->inlined) {
3517 		retval = ext4_handle_dirty_dirblock(handle, ent->dir, ent->bh);
3518 		if (unlikely(retval)) {
3519 			ext4_std_error(ent->dir->i_sb, retval);
3520 			return retval;
3521 		}
3522 	}
3523 	brelse(ent->bh);
3524 	ent->bh = NULL;
3525 
3526 	return 0;
3527 }
3528 
3529 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3530 				  const struct qstr *d_name)
3531 {
3532 	int retval = -ENOENT;
3533 	struct buffer_head *bh;
3534 	struct ext4_dir_entry_2 *de;
3535 
3536 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3537 	if (IS_ERR(bh))
3538 		return PTR_ERR(bh);
3539 	if (bh) {
3540 		retval = ext4_delete_entry(handle, dir, de, bh);
3541 		brelse(bh);
3542 	}
3543 	return retval;
3544 }
3545 
3546 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3547 			       int force_reread)
3548 {
3549 	int retval;
3550 	/*
3551 	 * ent->de could have moved from under us during htree split, so make
3552 	 * sure that we are deleting the right entry.  We might also be pointing
3553 	 * to a stale entry in the unused part of ent->bh so just checking inum
3554 	 * and the name isn't enough.
3555 	 */
3556 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3557 	    ent->de->name_len != ent->dentry->d_name.len ||
3558 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3559 		    ent->de->name_len) ||
3560 	    force_reread) {
3561 		retval = ext4_find_delete_entry(handle, ent->dir,
3562 						&ent->dentry->d_name);
3563 	} else {
3564 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3565 		if (retval == -ENOENT) {
3566 			retval = ext4_find_delete_entry(handle, ent->dir,
3567 							&ent->dentry->d_name);
3568 		}
3569 	}
3570 
3571 	if (retval) {
3572 		ext4_warning_inode(ent->dir,
3573 				   "Deleting old file: nlink %d, error=%d",
3574 				   ent->dir->i_nlink, retval);
3575 	}
3576 }
3577 
3578 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3579 {
3580 	if (ent->dir_nlink_delta) {
3581 		if (ent->dir_nlink_delta == -1)
3582 			ext4_dec_count(handle, ent->dir);
3583 		else
3584 			ext4_inc_count(handle, ent->dir);
3585 		ext4_mark_inode_dirty(handle, ent->dir);
3586 	}
3587 }
3588 
3589 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3590 					      int credits, handle_t **h)
3591 {
3592 	struct inode *wh;
3593 	handle_t *handle;
3594 	int retries = 0;
3595 
3596 	/*
3597 	 * for inode block, sb block, group summaries,
3598 	 * and inode bitmap
3599 	 */
3600 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3601 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3602 retry:
3603 	wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3604 					 &ent->dentry->d_name, 0, NULL,
3605 					 EXT4_HT_DIR, credits);
3606 
3607 	handle = ext4_journal_current_handle();
3608 	if (IS_ERR(wh)) {
3609 		if (handle)
3610 			ext4_journal_stop(handle);
3611 		if (PTR_ERR(wh) == -ENOSPC &&
3612 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3613 			goto retry;
3614 	} else {
3615 		*h = handle;
3616 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3617 		wh->i_op = &ext4_special_inode_operations;
3618 	}
3619 	return wh;
3620 }
3621 
3622 /*
3623  * Anybody can rename anything with this: the permission checks are left to the
3624  * higher-level routines.
3625  *
3626  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3627  * while new_{dentry,inode) refers to the destination dentry/inode
3628  * This comes from rename(const char *oldpath, const char *newpath)
3629  */
3630 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3631 		       struct inode *new_dir, struct dentry *new_dentry,
3632 		       unsigned int flags)
3633 {
3634 	handle_t *handle = NULL;
3635 	struct ext4_renament old = {
3636 		.dir = old_dir,
3637 		.dentry = old_dentry,
3638 		.inode = d_inode(old_dentry),
3639 	};
3640 	struct ext4_renament new = {
3641 		.dir = new_dir,
3642 		.dentry = new_dentry,
3643 		.inode = d_inode(new_dentry),
3644 	};
3645 	int force_reread;
3646 	int retval;
3647 	struct inode *whiteout = NULL;
3648 	int credits;
3649 	u8 old_file_type;
3650 
3651 	if (new.inode && new.inode->i_nlink == 0) {
3652 		EXT4_ERROR_INODE(new.inode,
3653 				 "target of rename is already freed");
3654 		return -EFSCORRUPTED;
3655 	}
3656 
3657 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3658 	    (!projid_eq(EXT4_I(new_dir)->i_projid,
3659 			EXT4_I(old_dentry->d_inode)->i_projid)))
3660 		return -EXDEV;
3661 
3662 	retval = dquot_initialize(old.dir);
3663 	if (retval)
3664 		return retval;
3665 	retval = dquot_initialize(new.dir);
3666 	if (retval)
3667 		return retval;
3668 
3669 	/* Initialize quotas before so that eventual writes go
3670 	 * in separate transaction */
3671 	if (new.inode) {
3672 		retval = dquot_initialize(new.inode);
3673 		if (retval)
3674 			return retval;
3675 	}
3676 
3677 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3678 	if (IS_ERR(old.bh))
3679 		return PTR_ERR(old.bh);
3680 	/*
3681 	 *  Check for inode number is _not_ due to possible IO errors.
3682 	 *  We might rmdir the source, keep it as pwd of some process
3683 	 *  and merrily kill the link to whatever was created under the
3684 	 *  same name. Goodbye sticky bit ;-<
3685 	 */
3686 	retval = -ENOENT;
3687 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3688 		goto end_rename;
3689 
3690 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3691 				 &new.de, &new.inlined);
3692 	if (IS_ERR(new.bh)) {
3693 		retval = PTR_ERR(new.bh);
3694 		new.bh = NULL;
3695 		goto end_rename;
3696 	}
3697 	if (new.bh) {
3698 		if (!new.inode) {
3699 			brelse(new.bh);
3700 			new.bh = NULL;
3701 		}
3702 	}
3703 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3704 		ext4_alloc_da_blocks(old.inode);
3705 
3706 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3707 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3708 	if (!(flags & RENAME_WHITEOUT)) {
3709 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3710 		if (IS_ERR(handle)) {
3711 			retval = PTR_ERR(handle);
3712 			handle = NULL;
3713 			goto end_rename;
3714 		}
3715 	} else {
3716 		whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3717 		if (IS_ERR(whiteout)) {
3718 			retval = PTR_ERR(whiteout);
3719 			whiteout = NULL;
3720 			goto end_rename;
3721 		}
3722 	}
3723 
3724 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3725 		ext4_handle_sync(handle);
3726 
3727 	if (S_ISDIR(old.inode->i_mode)) {
3728 		if (new.inode) {
3729 			retval = -ENOTEMPTY;
3730 			if (!ext4_empty_dir(new.inode))
3731 				goto end_rename;
3732 		} else {
3733 			retval = -EMLINK;
3734 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3735 				goto end_rename;
3736 		}
3737 		retval = ext4_rename_dir_prepare(handle, &old);
3738 		if (retval)
3739 			goto end_rename;
3740 	}
3741 	/*
3742 	 * If we're renaming a file within an inline_data dir and adding or
3743 	 * setting the new dirent causes a conversion from inline_data to
3744 	 * extents/blockmap, we need to force the dirent delete code to
3745 	 * re-read the directory, or else we end up trying to delete a dirent
3746 	 * from what is now the extent tree root (or a block map).
3747 	 */
3748 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3749 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3750 
3751 	old_file_type = old.de->file_type;
3752 	if (whiteout) {
3753 		/*
3754 		 * Do this before adding a new entry, so the old entry is sure
3755 		 * to be still pointing to the valid old entry.
3756 		 */
3757 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3758 				     EXT4_FT_CHRDEV);
3759 		if (retval)
3760 			goto end_rename;
3761 		ext4_mark_inode_dirty(handle, whiteout);
3762 	}
3763 	if (!new.bh) {
3764 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3765 		if (retval)
3766 			goto end_rename;
3767 	} else {
3768 		retval = ext4_setent(handle, &new,
3769 				     old.inode->i_ino, old_file_type);
3770 		if (retval)
3771 			goto end_rename;
3772 	}
3773 	if (force_reread)
3774 		force_reread = !ext4_test_inode_flag(new.dir,
3775 						     EXT4_INODE_INLINE_DATA);
3776 
3777 	/*
3778 	 * Like most other Unix systems, set the ctime for inodes on a
3779 	 * rename.
3780 	 */
3781 	old.inode->i_ctime = current_time(old.inode);
3782 	ext4_mark_inode_dirty(handle, old.inode);
3783 
3784 	if (!whiteout) {
3785 		/*
3786 		 * ok, that's it
3787 		 */
3788 		ext4_rename_delete(handle, &old, force_reread);
3789 	}
3790 
3791 	if (new.inode) {
3792 		ext4_dec_count(handle, new.inode);
3793 		new.inode->i_ctime = current_time(new.inode);
3794 	}
3795 	old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3796 	ext4_update_dx_flag(old.dir);
3797 	if (old.dir_bh) {
3798 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3799 		if (retval)
3800 			goto end_rename;
3801 
3802 		ext4_dec_count(handle, old.dir);
3803 		if (new.inode) {
3804 			/* checked ext4_empty_dir above, can't have another
3805 			 * parent, ext4_dec_count() won't work for many-linked
3806 			 * dirs */
3807 			clear_nlink(new.inode);
3808 		} else {
3809 			ext4_inc_count(handle, new.dir);
3810 			ext4_update_dx_flag(new.dir);
3811 			ext4_mark_inode_dirty(handle, new.dir);
3812 		}
3813 	}
3814 	ext4_mark_inode_dirty(handle, old.dir);
3815 	if (new.inode) {
3816 		ext4_mark_inode_dirty(handle, new.inode);
3817 		if (!new.inode->i_nlink)
3818 			ext4_orphan_add(handle, new.inode);
3819 	}
3820 	retval = 0;
3821 
3822 end_rename:
3823 	brelse(old.dir_bh);
3824 	brelse(old.bh);
3825 	brelse(new.bh);
3826 	if (whiteout) {
3827 		if (retval)
3828 			drop_nlink(whiteout);
3829 		unlock_new_inode(whiteout);
3830 		iput(whiteout);
3831 	}
3832 	if (handle)
3833 		ext4_journal_stop(handle);
3834 	return retval;
3835 }
3836 
3837 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3838 			     struct inode *new_dir, struct dentry *new_dentry)
3839 {
3840 	handle_t *handle = NULL;
3841 	struct ext4_renament old = {
3842 		.dir = old_dir,
3843 		.dentry = old_dentry,
3844 		.inode = d_inode(old_dentry),
3845 	};
3846 	struct ext4_renament new = {
3847 		.dir = new_dir,
3848 		.dentry = new_dentry,
3849 		.inode = d_inode(new_dentry),
3850 	};
3851 	u8 new_file_type;
3852 	int retval;
3853 	struct timespec64 ctime;
3854 
3855 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3856 	     !projid_eq(EXT4_I(new_dir)->i_projid,
3857 			EXT4_I(old_dentry->d_inode)->i_projid)) ||
3858 	    (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3859 	     !projid_eq(EXT4_I(old_dir)->i_projid,
3860 			EXT4_I(new_dentry->d_inode)->i_projid)))
3861 		return -EXDEV;
3862 
3863 	retval = dquot_initialize(old.dir);
3864 	if (retval)
3865 		return retval;
3866 	retval = dquot_initialize(new.dir);
3867 	if (retval)
3868 		return retval;
3869 
3870 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3871 				 &old.de, &old.inlined);
3872 	if (IS_ERR(old.bh))
3873 		return PTR_ERR(old.bh);
3874 	/*
3875 	 *  Check for inode number is _not_ due to possible IO errors.
3876 	 *  We might rmdir the source, keep it as pwd of some process
3877 	 *  and merrily kill the link to whatever was created under the
3878 	 *  same name. Goodbye sticky bit ;-<
3879 	 */
3880 	retval = -ENOENT;
3881 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3882 		goto end_rename;
3883 
3884 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3885 				 &new.de, &new.inlined);
3886 	if (IS_ERR(new.bh)) {
3887 		retval = PTR_ERR(new.bh);
3888 		new.bh = NULL;
3889 		goto end_rename;
3890 	}
3891 
3892 	/* RENAME_EXCHANGE case: old *and* new must both exist */
3893 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3894 		goto end_rename;
3895 
3896 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3897 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3898 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3899 	if (IS_ERR(handle)) {
3900 		retval = PTR_ERR(handle);
3901 		handle = NULL;
3902 		goto end_rename;
3903 	}
3904 
3905 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3906 		ext4_handle_sync(handle);
3907 
3908 	if (S_ISDIR(old.inode->i_mode)) {
3909 		old.is_dir = true;
3910 		retval = ext4_rename_dir_prepare(handle, &old);
3911 		if (retval)
3912 			goto end_rename;
3913 	}
3914 	if (S_ISDIR(new.inode->i_mode)) {
3915 		new.is_dir = true;
3916 		retval = ext4_rename_dir_prepare(handle, &new);
3917 		if (retval)
3918 			goto end_rename;
3919 	}
3920 
3921 	/*
3922 	 * Other than the special case of overwriting a directory, parents'
3923 	 * nlink only needs to be modified if this is a cross directory rename.
3924 	 */
3925 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
3926 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
3927 		new.dir_nlink_delta = -old.dir_nlink_delta;
3928 		retval = -EMLINK;
3929 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3930 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3931 			goto end_rename;
3932 	}
3933 
3934 	new_file_type = new.de->file_type;
3935 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3936 	if (retval)
3937 		goto end_rename;
3938 
3939 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3940 	if (retval)
3941 		goto end_rename;
3942 
3943 	/*
3944 	 * Like most other Unix systems, set the ctime for inodes on a
3945 	 * rename.
3946 	 */
3947 	ctime = current_time(old.inode);
3948 	old.inode->i_ctime = ctime;
3949 	new.inode->i_ctime = ctime;
3950 	ext4_mark_inode_dirty(handle, old.inode);
3951 	ext4_mark_inode_dirty(handle, new.inode);
3952 
3953 	if (old.dir_bh) {
3954 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3955 		if (retval)
3956 			goto end_rename;
3957 	}
3958 	if (new.dir_bh) {
3959 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3960 		if (retval)
3961 			goto end_rename;
3962 	}
3963 	ext4_update_dir_count(handle, &old);
3964 	ext4_update_dir_count(handle, &new);
3965 	retval = 0;
3966 
3967 end_rename:
3968 	brelse(old.dir_bh);
3969 	brelse(new.dir_bh);
3970 	brelse(old.bh);
3971 	brelse(new.bh);
3972 	if (handle)
3973 		ext4_journal_stop(handle);
3974 	return retval;
3975 }
3976 
3977 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3978 			struct inode *new_dir, struct dentry *new_dentry,
3979 			unsigned int flags)
3980 {
3981 	int err;
3982 
3983 	if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3984 		return -EIO;
3985 
3986 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3987 		return -EINVAL;
3988 
3989 	err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
3990 				     flags);
3991 	if (err)
3992 		return err;
3993 
3994 	if (flags & RENAME_EXCHANGE) {
3995 		return ext4_cross_rename(old_dir, old_dentry,
3996 					 new_dir, new_dentry);
3997 	}
3998 
3999 	return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
4000 }
4001 
4002 /*
4003  * directories can handle most operations...
4004  */
4005 const struct inode_operations ext4_dir_inode_operations = {
4006 	.create		= ext4_create,
4007 	.lookup		= ext4_lookup,
4008 	.link		= ext4_link,
4009 	.unlink		= ext4_unlink,
4010 	.symlink	= ext4_symlink,
4011 	.mkdir		= ext4_mkdir,
4012 	.rmdir		= ext4_rmdir,
4013 	.mknod		= ext4_mknod,
4014 	.tmpfile	= ext4_tmpfile,
4015 	.rename		= ext4_rename2,
4016 	.setattr	= ext4_setattr,
4017 	.getattr	= ext4_getattr,
4018 	.listxattr	= ext4_listxattr,
4019 	.get_acl	= ext4_get_acl,
4020 	.set_acl	= ext4_set_acl,
4021 	.fiemap         = ext4_fiemap,
4022 };
4023 
4024 const struct inode_operations ext4_special_inode_operations = {
4025 	.setattr	= ext4_setattr,
4026 	.getattr	= ext4_getattr,
4027 	.listxattr	= ext4_listxattr,
4028 	.get_acl	= ext4_get_acl,
4029 	.set_acl	= ext4_set_acl,
4030 };
4031