xref: /linux/fs/ext4/namei.c (revision 37cb8e1f8e10c6e9bd2a1b95cdda0620a21b0551)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/namei.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  from
11  *
12  *  linux/fs/minix/namei.c
13  *
14  *  Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  *  Big-endian to little-endian byte-swapping/bitmaps by
17  *        David S. Miller (davem@caip.rutgers.edu), 1995
18  *  Directory entry file type support and forward compatibility hooks
19  *	for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20  *  Hash Tree Directory indexing (c)
21  *	Daniel Phillips, 2001
22  *  Hash Tree Directory indexing porting
23  *	Christopher Li, 2002
24  *  Hash Tree Directory indexing cleanup
25  *	Theodore Ts'o, 2002
26  */
27 
28 #include <linux/fs.h>
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include "ext4.h"
38 #include "ext4_jbd2.h"
39 
40 #include "xattr.h"
41 #include "acl.h"
42 
43 #include <trace/events/ext4.h>
44 /*
45  * define how far ahead to read directories while searching them.
46  */
47 #define NAMEI_RA_CHUNKS  2
48 #define NAMEI_RA_BLOCKS  4
49 #define NAMEI_RA_SIZE	     (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50 
51 static struct buffer_head *ext4_append(handle_t *handle,
52 					struct inode *inode,
53 					ext4_lblk_t *block)
54 {
55 	struct buffer_head *bh;
56 	int err;
57 
58 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
59 		     ((inode->i_size >> 10) >=
60 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
61 		return ERR_PTR(-ENOSPC);
62 
63 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
64 
65 	bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
66 	if (IS_ERR(bh))
67 		return bh;
68 	inode->i_size += inode->i_sb->s_blocksize;
69 	EXT4_I(inode)->i_disksize = inode->i_size;
70 	BUFFER_TRACE(bh, "get_write_access");
71 	err = ext4_journal_get_write_access(handle, bh);
72 	if (err) {
73 		brelse(bh);
74 		ext4_std_error(inode->i_sb, err);
75 		return ERR_PTR(err);
76 	}
77 	return bh;
78 }
79 
80 static int ext4_dx_csum_verify(struct inode *inode,
81 			       struct ext4_dir_entry *dirent);
82 
83 typedef enum {
84 	EITHER, INDEX, DIRENT
85 } dirblock_type_t;
86 
87 #define ext4_read_dirblock(inode, block, type) \
88 	__ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
89 
90 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
91 						ext4_lblk_t block,
92 						dirblock_type_t type,
93 						const char *func,
94 						unsigned int line)
95 {
96 	struct buffer_head *bh;
97 	struct ext4_dir_entry *dirent;
98 	int is_dx_block = 0;
99 
100 	bh = ext4_bread(NULL, inode, block, 0);
101 	if (IS_ERR(bh)) {
102 		__ext4_warning(inode->i_sb, func, line,
103 			       "inode #%lu: lblock %lu: comm %s: "
104 			       "error %ld reading directory block",
105 			       inode->i_ino, (unsigned long)block,
106 			       current->comm, PTR_ERR(bh));
107 
108 		return bh;
109 	}
110 	if (!bh) {
111 		ext4_error_inode(inode, func, line, block,
112 				 "Directory hole found");
113 		return ERR_PTR(-EFSCORRUPTED);
114 	}
115 	dirent = (struct ext4_dir_entry *) bh->b_data;
116 	/* Determine whether or not we have an index block */
117 	if (is_dx(inode)) {
118 		if (block == 0)
119 			is_dx_block = 1;
120 		else if (ext4_rec_len_from_disk(dirent->rec_len,
121 						inode->i_sb->s_blocksize) ==
122 			 inode->i_sb->s_blocksize)
123 			is_dx_block = 1;
124 	}
125 	if (!is_dx_block && type == INDEX) {
126 		ext4_error_inode(inode, func, line, block,
127 		       "directory leaf block found instead of index block");
128 		return ERR_PTR(-EFSCORRUPTED);
129 	}
130 	if (!ext4_has_metadata_csum(inode->i_sb) ||
131 	    buffer_verified(bh))
132 		return bh;
133 
134 	/*
135 	 * An empty leaf block can get mistaken for a index block; for
136 	 * this reason, we can only check the index checksum when the
137 	 * caller is sure it should be an index block.
138 	 */
139 	if (is_dx_block && type == INDEX) {
140 		if (ext4_dx_csum_verify(inode, dirent))
141 			set_buffer_verified(bh);
142 		else {
143 			ext4_error_inode(inode, func, line, block,
144 					 "Directory index failed checksum");
145 			brelse(bh);
146 			return ERR_PTR(-EFSBADCRC);
147 		}
148 	}
149 	if (!is_dx_block) {
150 		if (ext4_dirent_csum_verify(inode, dirent))
151 			set_buffer_verified(bh);
152 		else {
153 			ext4_error_inode(inode, func, line, block,
154 					 "Directory block failed checksum");
155 			brelse(bh);
156 			return ERR_PTR(-EFSBADCRC);
157 		}
158 	}
159 	return bh;
160 }
161 
162 #ifndef assert
163 #define assert(test) J_ASSERT(test)
164 #endif
165 
166 #ifdef DX_DEBUG
167 #define dxtrace(command) command
168 #else
169 #define dxtrace(command)
170 #endif
171 
172 struct fake_dirent
173 {
174 	__le32 inode;
175 	__le16 rec_len;
176 	u8 name_len;
177 	u8 file_type;
178 };
179 
180 struct dx_countlimit
181 {
182 	__le16 limit;
183 	__le16 count;
184 };
185 
186 struct dx_entry
187 {
188 	__le32 hash;
189 	__le32 block;
190 };
191 
192 /*
193  * dx_root_info is laid out so that if it should somehow get overlaid by a
194  * dirent the two low bits of the hash version will be zero.  Therefore, the
195  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
196  */
197 
198 struct dx_root
199 {
200 	struct fake_dirent dot;
201 	char dot_name[4];
202 	struct fake_dirent dotdot;
203 	char dotdot_name[4];
204 	struct dx_root_info
205 	{
206 		__le32 reserved_zero;
207 		u8 hash_version;
208 		u8 info_length; /* 8 */
209 		u8 indirect_levels;
210 		u8 unused_flags;
211 	}
212 	info;
213 	struct dx_entry	entries[0];
214 };
215 
216 struct dx_node
217 {
218 	struct fake_dirent fake;
219 	struct dx_entry	entries[0];
220 };
221 
222 
223 struct dx_frame
224 {
225 	struct buffer_head *bh;
226 	struct dx_entry *entries;
227 	struct dx_entry *at;
228 };
229 
230 struct dx_map_entry
231 {
232 	u32 hash;
233 	u16 offs;
234 	u16 size;
235 };
236 
237 /*
238  * This goes at the end of each htree block.
239  */
240 struct dx_tail {
241 	u32 dt_reserved;
242 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
243 };
244 
245 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
246 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
247 static inline unsigned dx_get_hash(struct dx_entry *entry);
248 static void dx_set_hash(struct dx_entry *entry, unsigned value);
249 static unsigned dx_get_count(struct dx_entry *entries);
250 static unsigned dx_get_limit(struct dx_entry *entries);
251 static void dx_set_count(struct dx_entry *entries, unsigned value);
252 static void dx_set_limit(struct dx_entry *entries, unsigned value);
253 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
254 static unsigned dx_node_limit(struct inode *dir);
255 static struct dx_frame *dx_probe(struct ext4_filename *fname,
256 				 struct inode *dir,
257 				 struct dx_hash_info *hinfo,
258 				 struct dx_frame *frame);
259 static void dx_release(struct dx_frame *frames);
260 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
261 		       unsigned blocksize, struct dx_hash_info *hinfo,
262 		       struct dx_map_entry map[]);
263 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
264 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
265 		struct dx_map_entry *offsets, int count, unsigned blocksize);
266 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
267 static void dx_insert_block(struct dx_frame *frame,
268 					u32 hash, ext4_lblk_t block);
269 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
270 				 struct dx_frame *frame,
271 				 struct dx_frame *frames,
272 				 __u32 *start_hash);
273 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
274 		struct ext4_filename *fname,
275 		struct ext4_dir_entry_2 **res_dir);
276 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
277 			     struct inode *dir, struct inode *inode);
278 
279 /* checksumming functions */
280 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
281 			    unsigned int blocksize)
282 {
283 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
284 	t->det_rec_len = ext4_rec_len_to_disk(
285 			sizeof(struct ext4_dir_entry_tail), blocksize);
286 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
287 }
288 
289 /* Walk through a dirent block to find a checksum "dirent" at the tail */
290 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
291 						   struct ext4_dir_entry *de)
292 {
293 	struct ext4_dir_entry_tail *t;
294 
295 #ifdef PARANOID
296 	struct ext4_dir_entry *d, *top;
297 
298 	d = de;
299 	top = (struct ext4_dir_entry *)(((void *)de) +
300 		(EXT4_BLOCK_SIZE(inode->i_sb) -
301 		sizeof(struct ext4_dir_entry_tail)));
302 	while (d < top && d->rec_len)
303 		d = (struct ext4_dir_entry *)(((void *)d) +
304 		    le16_to_cpu(d->rec_len));
305 
306 	if (d != top)
307 		return NULL;
308 
309 	t = (struct ext4_dir_entry_tail *)d;
310 #else
311 	t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
312 #endif
313 
314 	if (t->det_reserved_zero1 ||
315 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
316 	    t->det_reserved_zero2 ||
317 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
318 		return NULL;
319 
320 	return t;
321 }
322 
323 static __le32 ext4_dirent_csum(struct inode *inode,
324 			       struct ext4_dir_entry *dirent, int size)
325 {
326 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
327 	struct ext4_inode_info *ei = EXT4_I(inode);
328 	__u32 csum;
329 
330 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
331 	return cpu_to_le32(csum);
332 }
333 
334 #define warn_no_space_for_csum(inode)					\
335 	__warn_no_space_for_csum((inode), __func__, __LINE__)
336 
337 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
338 				     unsigned int line)
339 {
340 	__ext4_warning_inode(inode, func, line,
341 		"No space for directory leaf checksum. Please run e2fsck -D.");
342 }
343 
344 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
345 {
346 	struct ext4_dir_entry_tail *t;
347 
348 	if (!ext4_has_metadata_csum(inode->i_sb))
349 		return 1;
350 
351 	t = get_dirent_tail(inode, dirent);
352 	if (!t) {
353 		warn_no_space_for_csum(inode);
354 		return 0;
355 	}
356 
357 	if (t->det_checksum != ext4_dirent_csum(inode, dirent,
358 						(void *)t - (void *)dirent))
359 		return 0;
360 
361 	return 1;
362 }
363 
364 static void ext4_dirent_csum_set(struct inode *inode,
365 				 struct ext4_dir_entry *dirent)
366 {
367 	struct ext4_dir_entry_tail *t;
368 
369 	if (!ext4_has_metadata_csum(inode->i_sb))
370 		return;
371 
372 	t = get_dirent_tail(inode, dirent);
373 	if (!t) {
374 		warn_no_space_for_csum(inode);
375 		return;
376 	}
377 
378 	t->det_checksum = ext4_dirent_csum(inode, dirent,
379 					   (void *)t - (void *)dirent);
380 }
381 
382 int ext4_handle_dirty_dirent_node(handle_t *handle,
383 				  struct inode *inode,
384 				  struct buffer_head *bh)
385 {
386 	ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
387 	return ext4_handle_dirty_metadata(handle, inode, bh);
388 }
389 
390 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
391 					       struct ext4_dir_entry *dirent,
392 					       int *offset)
393 {
394 	struct ext4_dir_entry *dp;
395 	struct dx_root_info *root;
396 	int count_offset;
397 
398 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
399 		count_offset = 8;
400 	else if (le16_to_cpu(dirent->rec_len) == 12) {
401 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
402 		if (le16_to_cpu(dp->rec_len) !=
403 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
404 			return NULL;
405 		root = (struct dx_root_info *)(((void *)dp + 12));
406 		if (root->reserved_zero ||
407 		    root->info_length != sizeof(struct dx_root_info))
408 			return NULL;
409 		count_offset = 32;
410 	} else
411 		return NULL;
412 
413 	if (offset)
414 		*offset = count_offset;
415 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
416 }
417 
418 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
419 			   int count_offset, int count, struct dx_tail *t)
420 {
421 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
422 	struct ext4_inode_info *ei = EXT4_I(inode);
423 	__u32 csum;
424 	int size;
425 	__u32 dummy_csum = 0;
426 	int offset = offsetof(struct dx_tail, dt_checksum);
427 
428 	size = count_offset + (count * sizeof(struct dx_entry));
429 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
430 	csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
431 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
432 
433 	return cpu_to_le32(csum);
434 }
435 
436 static int ext4_dx_csum_verify(struct inode *inode,
437 			       struct ext4_dir_entry *dirent)
438 {
439 	struct dx_countlimit *c;
440 	struct dx_tail *t;
441 	int count_offset, limit, count;
442 
443 	if (!ext4_has_metadata_csum(inode->i_sb))
444 		return 1;
445 
446 	c = get_dx_countlimit(inode, dirent, &count_offset);
447 	if (!c) {
448 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
449 		return 0;
450 	}
451 	limit = le16_to_cpu(c->limit);
452 	count = le16_to_cpu(c->count);
453 	if (count_offset + (limit * sizeof(struct dx_entry)) >
454 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
455 		warn_no_space_for_csum(inode);
456 		return 0;
457 	}
458 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
459 
460 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
461 					    count, t))
462 		return 0;
463 	return 1;
464 }
465 
466 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
467 {
468 	struct dx_countlimit *c;
469 	struct dx_tail *t;
470 	int count_offset, limit, count;
471 
472 	if (!ext4_has_metadata_csum(inode->i_sb))
473 		return;
474 
475 	c = get_dx_countlimit(inode, dirent, &count_offset);
476 	if (!c) {
477 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
478 		return;
479 	}
480 	limit = le16_to_cpu(c->limit);
481 	count = le16_to_cpu(c->count);
482 	if (count_offset + (limit * sizeof(struct dx_entry)) >
483 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
484 		warn_no_space_for_csum(inode);
485 		return;
486 	}
487 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
488 
489 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
490 }
491 
492 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
493 					    struct inode *inode,
494 					    struct buffer_head *bh)
495 {
496 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
497 	return ext4_handle_dirty_metadata(handle, inode, bh);
498 }
499 
500 /*
501  * p is at least 6 bytes before the end of page
502  */
503 static inline struct ext4_dir_entry_2 *
504 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
505 {
506 	return (struct ext4_dir_entry_2 *)((char *)p +
507 		ext4_rec_len_from_disk(p->rec_len, blocksize));
508 }
509 
510 /*
511  * Future: use high four bits of block for coalesce-on-delete flags
512  * Mask them off for now.
513  */
514 
515 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
516 {
517 	return le32_to_cpu(entry->block) & 0x0fffffff;
518 }
519 
520 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
521 {
522 	entry->block = cpu_to_le32(value);
523 }
524 
525 static inline unsigned dx_get_hash(struct dx_entry *entry)
526 {
527 	return le32_to_cpu(entry->hash);
528 }
529 
530 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
531 {
532 	entry->hash = cpu_to_le32(value);
533 }
534 
535 static inline unsigned dx_get_count(struct dx_entry *entries)
536 {
537 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
538 }
539 
540 static inline unsigned dx_get_limit(struct dx_entry *entries)
541 {
542 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
543 }
544 
545 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
546 {
547 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
548 }
549 
550 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
551 {
552 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
553 }
554 
555 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
556 {
557 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
558 		EXT4_DIR_REC_LEN(2) - infosize;
559 
560 	if (ext4_has_metadata_csum(dir->i_sb))
561 		entry_space -= sizeof(struct dx_tail);
562 	return entry_space / sizeof(struct dx_entry);
563 }
564 
565 static inline unsigned dx_node_limit(struct inode *dir)
566 {
567 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
568 
569 	if (ext4_has_metadata_csum(dir->i_sb))
570 		entry_space -= sizeof(struct dx_tail);
571 	return entry_space / sizeof(struct dx_entry);
572 }
573 
574 /*
575  * Debug
576  */
577 #ifdef DX_DEBUG
578 static void dx_show_index(char * label, struct dx_entry *entries)
579 {
580 	int i, n = dx_get_count (entries);
581 	printk(KERN_DEBUG "%s index", label);
582 	for (i = 0; i < n; i++) {
583 		printk(KERN_CONT " %x->%lu",
584 		       i ? dx_get_hash(entries + i) : 0,
585 		       (unsigned long)dx_get_block(entries + i));
586 	}
587 	printk(KERN_CONT "\n");
588 }
589 
590 struct stats
591 {
592 	unsigned names;
593 	unsigned space;
594 	unsigned bcount;
595 };
596 
597 static struct stats dx_show_leaf(struct inode *dir,
598 				struct dx_hash_info *hinfo,
599 				struct ext4_dir_entry_2 *de,
600 				int size, int show_names)
601 {
602 	unsigned names = 0, space = 0;
603 	char *base = (char *) de;
604 	struct dx_hash_info h = *hinfo;
605 
606 	printk("names: ");
607 	while ((char *) de < base + size)
608 	{
609 		if (de->inode)
610 		{
611 			if (show_names)
612 			{
613 #ifdef CONFIG_EXT4_FS_ENCRYPTION
614 				int len;
615 				char *name;
616 				struct fscrypt_str fname_crypto_str =
617 					FSTR_INIT(NULL, 0);
618 				int res = 0;
619 
620 				name  = de->name;
621 				len = de->name_len;
622 				if (ext4_encrypted_inode(dir))
623 					res = fscrypt_get_encryption_info(dir);
624 				if (res) {
625 					printk(KERN_WARNING "Error setting up"
626 					       " fname crypto: %d\n", res);
627 				}
628 				if (!fscrypt_has_encryption_key(dir)) {
629 					/* Directory is not encrypted */
630 					ext4fs_dirhash(de->name,
631 						de->name_len, &h);
632 					printk("%*.s:(U)%x.%u ", len,
633 					       name, h.hash,
634 					       (unsigned) ((char *) de
635 							   - base));
636 				} else {
637 					struct fscrypt_str de_name =
638 						FSTR_INIT(name, len);
639 
640 					/* Directory is encrypted */
641 					res = fscrypt_fname_alloc_buffer(
642 						dir, len,
643 						&fname_crypto_str);
644 					if (res)
645 						printk(KERN_WARNING "Error "
646 							"allocating crypto "
647 							"buffer--skipping "
648 							"crypto\n");
649 					res = fscrypt_fname_disk_to_usr(dir,
650 						0, 0, &de_name,
651 						&fname_crypto_str);
652 					if (res) {
653 						printk(KERN_WARNING "Error "
654 							"converting filename "
655 							"from disk to usr"
656 							"\n");
657 						name = "??";
658 						len = 2;
659 					} else {
660 						name = fname_crypto_str.name;
661 						len = fname_crypto_str.len;
662 					}
663 					ext4fs_dirhash(de->name, de->name_len,
664 						       &h);
665 					printk("%*.s:(E)%x.%u ", len, name,
666 					       h.hash, (unsigned) ((char *) de
667 								   - base));
668 					fscrypt_fname_free_buffer(
669 							&fname_crypto_str);
670 				}
671 #else
672 				int len = de->name_len;
673 				char *name = de->name;
674 				ext4fs_dirhash(de->name, de->name_len, &h);
675 				printk("%*.s:%x.%u ", len, name, h.hash,
676 				       (unsigned) ((char *) de - base));
677 #endif
678 			}
679 			space += EXT4_DIR_REC_LEN(de->name_len);
680 			names++;
681 		}
682 		de = ext4_next_entry(de, size);
683 	}
684 	printk(KERN_CONT "(%i)\n", names);
685 	return (struct stats) { names, space, 1 };
686 }
687 
688 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
689 			     struct dx_entry *entries, int levels)
690 {
691 	unsigned blocksize = dir->i_sb->s_blocksize;
692 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
693 	unsigned bcount = 0;
694 	struct buffer_head *bh;
695 	printk("%i indexed blocks...\n", count);
696 	for (i = 0; i < count; i++, entries++)
697 	{
698 		ext4_lblk_t block = dx_get_block(entries);
699 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
700 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
701 		struct stats stats;
702 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
703 		bh = ext4_bread(NULL,dir, block, 0);
704 		if (!bh || IS_ERR(bh))
705 			continue;
706 		stats = levels?
707 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
708 		   dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
709 			bh->b_data, blocksize, 0);
710 		names += stats.names;
711 		space += stats.space;
712 		bcount += stats.bcount;
713 		brelse(bh);
714 	}
715 	if (bcount)
716 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
717 		       levels ? "" : "   ", names, space/bcount,
718 		       (space/bcount)*100/blocksize);
719 	return (struct stats) { names, space, bcount};
720 }
721 #endif /* DX_DEBUG */
722 
723 /*
724  * Probe for a directory leaf block to search.
725  *
726  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
727  * error in the directory index, and the caller should fall back to
728  * searching the directory normally.  The callers of dx_probe **MUST**
729  * check for this error code, and make sure it never gets reflected
730  * back to userspace.
731  */
732 static struct dx_frame *
733 dx_probe(struct ext4_filename *fname, struct inode *dir,
734 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
735 {
736 	unsigned count, indirect;
737 	struct dx_entry *at, *entries, *p, *q, *m;
738 	struct dx_root *root;
739 	struct dx_frame *frame = frame_in;
740 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
741 	u32 hash;
742 
743 	memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
744 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
745 	if (IS_ERR(frame->bh))
746 		return (struct dx_frame *) frame->bh;
747 
748 	root = (struct dx_root *) frame->bh->b_data;
749 	if (root->info.hash_version != DX_HASH_TEA &&
750 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
751 	    root->info.hash_version != DX_HASH_LEGACY) {
752 		ext4_warning_inode(dir, "Unrecognised inode hash code %u",
753 				   root->info.hash_version);
754 		goto fail;
755 	}
756 	if (fname)
757 		hinfo = &fname->hinfo;
758 	hinfo->hash_version = root->info.hash_version;
759 	if (hinfo->hash_version <= DX_HASH_TEA)
760 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
761 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
762 	if (fname && fname_name(fname))
763 		ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
764 	hash = hinfo->hash;
765 
766 	if (root->info.unused_flags & 1) {
767 		ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
768 				   root->info.unused_flags);
769 		goto fail;
770 	}
771 
772 	indirect = root->info.indirect_levels;
773 	if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
774 		ext4_warning(dir->i_sb,
775 			     "Directory (ino: %lu) htree depth %#06x exceed"
776 			     "supported value", dir->i_ino,
777 			     ext4_dir_htree_level(dir->i_sb));
778 		if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
779 			ext4_warning(dir->i_sb, "Enable large directory "
780 						"feature to access it");
781 		}
782 		goto fail;
783 	}
784 
785 	entries = (struct dx_entry *)(((char *)&root->info) +
786 				      root->info.info_length);
787 
788 	if (dx_get_limit(entries) != dx_root_limit(dir,
789 						   root->info.info_length)) {
790 		ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
791 				   dx_get_limit(entries),
792 				   dx_root_limit(dir, root->info.info_length));
793 		goto fail;
794 	}
795 
796 	dxtrace(printk("Look up %x", hash));
797 	while (1) {
798 		count = dx_get_count(entries);
799 		if (!count || count > dx_get_limit(entries)) {
800 			ext4_warning_inode(dir,
801 					   "dx entry: count %u beyond limit %u",
802 					   count, dx_get_limit(entries));
803 			goto fail;
804 		}
805 
806 		p = entries + 1;
807 		q = entries + count - 1;
808 		while (p <= q) {
809 			m = p + (q - p) / 2;
810 			dxtrace(printk(KERN_CONT "."));
811 			if (dx_get_hash(m) > hash)
812 				q = m - 1;
813 			else
814 				p = m + 1;
815 		}
816 
817 		if (0) { // linear search cross check
818 			unsigned n = count - 1;
819 			at = entries;
820 			while (n--)
821 			{
822 				dxtrace(printk(KERN_CONT ","));
823 				if (dx_get_hash(++at) > hash)
824 				{
825 					at--;
826 					break;
827 				}
828 			}
829 			assert (at == p - 1);
830 		}
831 
832 		at = p - 1;
833 		dxtrace(printk(KERN_CONT " %x->%u\n",
834 			       at == entries ? 0 : dx_get_hash(at),
835 			       dx_get_block(at)));
836 		frame->entries = entries;
837 		frame->at = at;
838 		if (!indirect--)
839 			return frame;
840 		frame++;
841 		frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
842 		if (IS_ERR(frame->bh)) {
843 			ret_err = (struct dx_frame *) frame->bh;
844 			frame->bh = NULL;
845 			goto fail;
846 		}
847 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
848 
849 		if (dx_get_limit(entries) != dx_node_limit(dir)) {
850 			ext4_warning_inode(dir,
851 				"dx entry: limit %u != node limit %u",
852 				dx_get_limit(entries), dx_node_limit(dir));
853 			goto fail;
854 		}
855 	}
856 fail:
857 	while (frame >= frame_in) {
858 		brelse(frame->bh);
859 		frame--;
860 	}
861 
862 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
863 		ext4_warning_inode(dir,
864 			"Corrupt directory, running e2fsck is recommended");
865 	return ret_err;
866 }
867 
868 static void dx_release(struct dx_frame *frames)
869 {
870 	struct dx_root_info *info;
871 	int i;
872 
873 	if (frames[0].bh == NULL)
874 		return;
875 
876 	info = &((struct dx_root *)frames[0].bh->b_data)->info;
877 	for (i = 0; i <= info->indirect_levels; i++) {
878 		if (frames[i].bh == NULL)
879 			break;
880 		brelse(frames[i].bh);
881 		frames[i].bh = NULL;
882 	}
883 }
884 
885 /*
886  * This function increments the frame pointer to search the next leaf
887  * block, and reads in the necessary intervening nodes if the search
888  * should be necessary.  Whether or not the search is necessary is
889  * controlled by the hash parameter.  If the hash value is even, then
890  * the search is only continued if the next block starts with that
891  * hash value.  This is used if we are searching for a specific file.
892  *
893  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
894  *
895  * This function returns 1 if the caller should continue to search,
896  * or 0 if it should not.  If there is an error reading one of the
897  * index blocks, it will a negative error code.
898  *
899  * If start_hash is non-null, it will be filled in with the starting
900  * hash of the next page.
901  */
902 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
903 				 struct dx_frame *frame,
904 				 struct dx_frame *frames,
905 				 __u32 *start_hash)
906 {
907 	struct dx_frame *p;
908 	struct buffer_head *bh;
909 	int num_frames = 0;
910 	__u32 bhash;
911 
912 	p = frame;
913 	/*
914 	 * Find the next leaf page by incrementing the frame pointer.
915 	 * If we run out of entries in the interior node, loop around and
916 	 * increment pointer in the parent node.  When we break out of
917 	 * this loop, num_frames indicates the number of interior
918 	 * nodes need to be read.
919 	 */
920 	while (1) {
921 		if (++(p->at) < p->entries + dx_get_count(p->entries))
922 			break;
923 		if (p == frames)
924 			return 0;
925 		num_frames++;
926 		p--;
927 	}
928 
929 	/*
930 	 * If the hash is 1, then continue only if the next page has a
931 	 * continuation hash of any value.  This is used for readdir
932 	 * handling.  Otherwise, check to see if the hash matches the
933 	 * desired contiuation hash.  If it doesn't, return since
934 	 * there's no point to read in the successive index pages.
935 	 */
936 	bhash = dx_get_hash(p->at);
937 	if (start_hash)
938 		*start_hash = bhash;
939 	if ((hash & 1) == 0) {
940 		if ((bhash & ~1) != hash)
941 			return 0;
942 	}
943 	/*
944 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
945 	 * block so no check is necessary
946 	 */
947 	while (num_frames--) {
948 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
949 		if (IS_ERR(bh))
950 			return PTR_ERR(bh);
951 		p++;
952 		brelse(p->bh);
953 		p->bh = bh;
954 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
955 	}
956 	return 1;
957 }
958 
959 
960 /*
961  * This function fills a red-black tree with information from a
962  * directory block.  It returns the number directory entries loaded
963  * into the tree.  If there is an error it is returned in err.
964  */
965 static int htree_dirblock_to_tree(struct file *dir_file,
966 				  struct inode *dir, ext4_lblk_t block,
967 				  struct dx_hash_info *hinfo,
968 				  __u32 start_hash, __u32 start_minor_hash)
969 {
970 	struct buffer_head *bh;
971 	struct ext4_dir_entry_2 *de, *top;
972 	int err = 0, count = 0;
973 	struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
974 
975 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
976 							(unsigned long)block));
977 	bh = ext4_read_dirblock(dir, block, DIRENT);
978 	if (IS_ERR(bh))
979 		return PTR_ERR(bh);
980 
981 	de = (struct ext4_dir_entry_2 *) bh->b_data;
982 	top = (struct ext4_dir_entry_2 *) ((char *) de +
983 					   dir->i_sb->s_blocksize -
984 					   EXT4_DIR_REC_LEN(0));
985 #ifdef CONFIG_EXT4_FS_ENCRYPTION
986 	/* Check if the directory is encrypted */
987 	if (ext4_encrypted_inode(dir)) {
988 		err = fscrypt_get_encryption_info(dir);
989 		if (err < 0) {
990 			brelse(bh);
991 			return err;
992 		}
993 		err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
994 						     &fname_crypto_str);
995 		if (err < 0) {
996 			brelse(bh);
997 			return err;
998 		}
999 	}
1000 #endif
1001 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1002 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1003 				bh->b_data, bh->b_size,
1004 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1005 					 + ((char *)de - bh->b_data))) {
1006 			/* silently ignore the rest of the block */
1007 			break;
1008 		}
1009 		ext4fs_dirhash(de->name, de->name_len, hinfo);
1010 		if ((hinfo->hash < start_hash) ||
1011 		    ((hinfo->hash == start_hash) &&
1012 		     (hinfo->minor_hash < start_minor_hash)))
1013 			continue;
1014 		if (de->inode == 0)
1015 			continue;
1016 		if (!ext4_encrypted_inode(dir)) {
1017 			tmp_str.name = de->name;
1018 			tmp_str.len = de->name_len;
1019 			err = ext4_htree_store_dirent(dir_file,
1020 				   hinfo->hash, hinfo->minor_hash, de,
1021 				   &tmp_str);
1022 		} else {
1023 			int save_len = fname_crypto_str.len;
1024 			struct fscrypt_str de_name = FSTR_INIT(de->name,
1025 								de->name_len);
1026 
1027 			/* Directory is encrypted */
1028 			err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1029 					hinfo->minor_hash, &de_name,
1030 					&fname_crypto_str);
1031 			if (err) {
1032 				count = err;
1033 				goto errout;
1034 			}
1035 			err = ext4_htree_store_dirent(dir_file,
1036 				   hinfo->hash, hinfo->minor_hash, de,
1037 					&fname_crypto_str);
1038 			fname_crypto_str.len = save_len;
1039 		}
1040 		if (err != 0) {
1041 			count = err;
1042 			goto errout;
1043 		}
1044 		count++;
1045 	}
1046 errout:
1047 	brelse(bh);
1048 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1049 	fscrypt_fname_free_buffer(&fname_crypto_str);
1050 #endif
1051 	return count;
1052 }
1053 
1054 
1055 /*
1056  * This function fills a red-black tree with information from a
1057  * directory.  We start scanning the directory in hash order, starting
1058  * at start_hash and start_minor_hash.
1059  *
1060  * This function returns the number of entries inserted into the tree,
1061  * or a negative error code.
1062  */
1063 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1064 			 __u32 start_minor_hash, __u32 *next_hash)
1065 {
1066 	struct dx_hash_info hinfo;
1067 	struct ext4_dir_entry_2 *de;
1068 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1069 	struct inode *dir;
1070 	ext4_lblk_t block;
1071 	int count = 0;
1072 	int ret, err;
1073 	__u32 hashval;
1074 	struct fscrypt_str tmp_str;
1075 
1076 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1077 		       start_hash, start_minor_hash));
1078 	dir = file_inode(dir_file);
1079 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1080 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1081 		if (hinfo.hash_version <= DX_HASH_TEA)
1082 			hinfo.hash_version +=
1083 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
1084 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1085 		if (ext4_has_inline_data(dir)) {
1086 			int has_inline_data = 1;
1087 			count = htree_inlinedir_to_tree(dir_file, dir, 0,
1088 							&hinfo, start_hash,
1089 							start_minor_hash,
1090 							&has_inline_data);
1091 			if (has_inline_data) {
1092 				*next_hash = ~0;
1093 				return count;
1094 			}
1095 		}
1096 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1097 					       start_hash, start_minor_hash);
1098 		*next_hash = ~0;
1099 		return count;
1100 	}
1101 	hinfo.hash = start_hash;
1102 	hinfo.minor_hash = 0;
1103 	frame = dx_probe(NULL, dir, &hinfo, frames);
1104 	if (IS_ERR(frame))
1105 		return PTR_ERR(frame);
1106 
1107 	/* Add '.' and '..' from the htree header */
1108 	if (!start_hash && !start_minor_hash) {
1109 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1110 		tmp_str.name = de->name;
1111 		tmp_str.len = de->name_len;
1112 		err = ext4_htree_store_dirent(dir_file, 0, 0,
1113 					      de, &tmp_str);
1114 		if (err != 0)
1115 			goto errout;
1116 		count++;
1117 	}
1118 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1119 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1120 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1121 		tmp_str.name = de->name;
1122 		tmp_str.len = de->name_len;
1123 		err = ext4_htree_store_dirent(dir_file, 2, 0,
1124 					      de, &tmp_str);
1125 		if (err != 0)
1126 			goto errout;
1127 		count++;
1128 	}
1129 
1130 	while (1) {
1131 		if (fatal_signal_pending(current)) {
1132 			err = -ERESTARTSYS;
1133 			goto errout;
1134 		}
1135 		cond_resched();
1136 		block = dx_get_block(frame->at);
1137 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1138 					     start_hash, start_minor_hash);
1139 		if (ret < 0) {
1140 			err = ret;
1141 			goto errout;
1142 		}
1143 		count += ret;
1144 		hashval = ~0;
1145 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1146 					    frame, frames, &hashval);
1147 		*next_hash = hashval;
1148 		if (ret < 0) {
1149 			err = ret;
1150 			goto errout;
1151 		}
1152 		/*
1153 		 * Stop if:  (a) there are no more entries, or
1154 		 * (b) we have inserted at least one entry and the
1155 		 * next hash value is not a continuation
1156 		 */
1157 		if ((ret == 0) ||
1158 		    (count && ((hashval & 1) == 0)))
1159 			break;
1160 	}
1161 	dx_release(frames);
1162 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1163 		       "next hash: %x\n", count, *next_hash));
1164 	return count;
1165 errout:
1166 	dx_release(frames);
1167 	return (err);
1168 }
1169 
1170 static inline int search_dirblock(struct buffer_head *bh,
1171 				  struct inode *dir,
1172 				  struct ext4_filename *fname,
1173 				  unsigned int offset,
1174 				  struct ext4_dir_entry_2 **res_dir)
1175 {
1176 	return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1177 			       fname, offset, res_dir);
1178 }
1179 
1180 /*
1181  * Directory block splitting, compacting
1182  */
1183 
1184 /*
1185  * Create map of hash values, offsets, and sizes, stored at end of block.
1186  * Returns number of entries mapped.
1187  */
1188 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1189 		       unsigned blocksize, struct dx_hash_info *hinfo,
1190 		       struct dx_map_entry *map_tail)
1191 {
1192 	int count = 0;
1193 	char *base = (char *) de;
1194 	struct dx_hash_info h = *hinfo;
1195 
1196 	while ((char *) de < base + blocksize) {
1197 		if (de->name_len && de->inode) {
1198 			ext4fs_dirhash(de->name, de->name_len, &h);
1199 			map_tail--;
1200 			map_tail->hash = h.hash;
1201 			map_tail->offs = ((char *) de - base)>>2;
1202 			map_tail->size = le16_to_cpu(de->rec_len);
1203 			count++;
1204 			cond_resched();
1205 		}
1206 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1207 		de = ext4_next_entry(de, blocksize);
1208 	}
1209 	return count;
1210 }
1211 
1212 /* Sort map by hash value */
1213 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1214 {
1215 	struct dx_map_entry *p, *q, *top = map + count - 1;
1216 	int more;
1217 	/* Combsort until bubble sort doesn't suck */
1218 	while (count > 2) {
1219 		count = count*10/13;
1220 		if (count - 9 < 2) /* 9, 10 -> 11 */
1221 			count = 11;
1222 		for (p = top, q = p - count; q >= map; p--, q--)
1223 			if (p->hash < q->hash)
1224 				swap(*p, *q);
1225 	}
1226 	/* Garden variety bubble sort */
1227 	do {
1228 		more = 0;
1229 		q = top;
1230 		while (q-- > map) {
1231 			if (q[1].hash >= q[0].hash)
1232 				continue;
1233 			swap(*(q+1), *q);
1234 			more = 1;
1235 		}
1236 	} while(more);
1237 }
1238 
1239 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1240 {
1241 	struct dx_entry *entries = frame->entries;
1242 	struct dx_entry *old = frame->at, *new = old + 1;
1243 	int count = dx_get_count(entries);
1244 
1245 	assert(count < dx_get_limit(entries));
1246 	assert(old < entries + count);
1247 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1248 	dx_set_hash(new, hash);
1249 	dx_set_block(new, block);
1250 	dx_set_count(entries, count + 1);
1251 }
1252 
1253 /*
1254  * Test whether a directory entry matches the filename being searched for.
1255  *
1256  * Return: %true if the directory entry matches, otherwise %false.
1257  */
1258 static inline bool ext4_match(const struct ext4_filename *fname,
1259 			      const struct ext4_dir_entry_2 *de)
1260 {
1261 	struct fscrypt_name f;
1262 
1263 	if (!de->inode)
1264 		return false;
1265 
1266 	f.usr_fname = fname->usr_fname;
1267 	f.disk_name = fname->disk_name;
1268 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1269 	f.crypto_buf = fname->crypto_buf;
1270 #endif
1271 	return fscrypt_match_name(&f, de->name, de->name_len);
1272 }
1273 
1274 /*
1275  * Returns 0 if not found, -1 on failure, and 1 on success
1276  */
1277 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1278 		    struct inode *dir, struct ext4_filename *fname,
1279 		    unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1280 {
1281 	struct ext4_dir_entry_2 * de;
1282 	char * dlimit;
1283 	int de_len;
1284 
1285 	de = (struct ext4_dir_entry_2 *)search_buf;
1286 	dlimit = search_buf + buf_size;
1287 	while ((char *) de < dlimit) {
1288 		/* this code is executed quadratically often */
1289 		/* do minimal checking `by hand' */
1290 		if ((char *) de + de->name_len <= dlimit &&
1291 		    ext4_match(fname, de)) {
1292 			/* found a match - just to be sure, do
1293 			 * a full check */
1294 			if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1295 						 bh->b_size, offset))
1296 				return -1;
1297 			*res_dir = de;
1298 			return 1;
1299 		}
1300 		/* prevent looping on a bad block */
1301 		de_len = ext4_rec_len_from_disk(de->rec_len,
1302 						dir->i_sb->s_blocksize);
1303 		if (de_len <= 0)
1304 			return -1;
1305 		offset += de_len;
1306 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1307 	}
1308 	return 0;
1309 }
1310 
1311 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1312 			       struct ext4_dir_entry *de)
1313 {
1314 	struct super_block *sb = dir->i_sb;
1315 
1316 	if (!is_dx(dir))
1317 		return 0;
1318 	if (block == 0)
1319 		return 1;
1320 	if (de->inode == 0 &&
1321 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1322 			sb->s_blocksize)
1323 		return 1;
1324 	return 0;
1325 }
1326 
1327 /*
1328  *	ext4_find_entry()
1329  *
1330  * finds an entry in the specified directory with the wanted name. It
1331  * returns the cache buffer in which the entry was found, and the entry
1332  * itself (as a parameter - res_dir). It does NOT read the inode of the
1333  * entry - you'll have to do that yourself if you want to.
1334  *
1335  * The returned buffer_head has ->b_count elevated.  The caller is expected
1336  * to brelse() it when appropriate.
1337  */
1338 static struct buffer_head * ext4_find_entry (struct inode *dir,
1339 					const struct qstr *d_name,
1340 					struct ext4_dir_entry_2 **res_dir,
1341 					int *inlined)
1342 {
1343 	struct super_block *sb;
1344 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1345 	struct buffer_head *bh, *ret = NULL;
1346 	ext4_lblk_t start, block;
1347 	const u8 *name = d_name->name;
1348 	size_t ra_max = 0;	/* Number of bh's in the readahead
1349 				   buffer, bh_use[] */
1350 	size_t ra_ptr = 0;	/* Current index into readahead
1351 				   buffer */
1352 	ext4_lblk_t  nblocks;
1353 	int i, namelen, retval;
1354 	struct ext4_filename fname;
1355 
1356 	*res_dir = NULL;
1357 	sb = dir->i_sb;
1358 	namelen = d_name->len;
1359 	if (namelen > EXT4_NAME_LEN)
1360 		return NULL;
1361 
1362 	retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1363 	if (retval == -ENOENT)
1364 		return NULL;
1365 	if (retval)
1366 		return ERR_PTR(retval);
1367 
1368 	if (ext4_has_inline_data(dir)) {
1369 		int has_inline_data = 1;
1370 		ret = ext4_find_inline_entry(dir, &fname, res_dir,
1371 					     &has_inline_data);
1372 		if (has_inline_data) {
1373 			if (inlined)
1374 				*inlined = 1;
1375 			goto cleanup_and_exit;
1376 		}
1377 	}
1378 
1379 	if ((namelen <= 2) && (name[0] == '.') &&
1380 	    (name[1] == '.' || name[1] == '\0')) {
1381 		/*
1382 		 * "." or ".." will only be in the first block
1383 		 * NFS may look up ".."; "." should be handled by the VFS
1384 		 */
1385 		block = start = 0;
1386 		nblocks = 1;
1387 		goto restart;
1388 	}
1389 	if (is_dx(dir)) {
1390 		ret = ext4_dx_find_entry(dir, &fname, res_dir);
1391 		/*
1392 		 * On success, or if the error was file not found,
1393 		 * return.  Otherwise, fall back to doing a search the
1394 		 * old fashioned way.
1395 		 */
1396 		if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1397 			goto cleanup_and_exit;
1398 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1399 			       "falling back\n"));
1400 	}
1401 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1402 	start = EXT4_I(dir)->i_dir_start_lookup;
1403 	if (start >= nblocks)
1404 		start = 0;
1405 	block = start;
1406 restart:
1407 	do {
1408 		/*
1409 		 * We deal with the read-ahead logic here.
1410 		 */
1411 		if (ra_ptr >= ra_max) {
1412 			/* Refill the readahead buffer */
1413 			ra_ptr = 0;
1414 			if (block < start)
1415 				ra_max = start - block;
1416 			else
1417 				ra_max = nblocks - block;
1418 			ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1419 			retval = ext4_bread_batch(dir, block, ra_max,
1420 						  false /* wait */, bh_use);
1421 			if (retval) {
1422 				ret = ERR_PTR(retval);
1423 				ra_max = 0;
1424 				goto cleanup_and_exit;
1425 			}
1426 		}
1427 		if ((bh = bh_use[ra_ptr++]) == NULL)
1428 			goto next;
1429 		wait_on_buffer(bh);
1430 		if (!buffer_uptodate(bh)) {
1431 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1432 					 (unsigned long) block);
1433 			brelse(bh);
1434 			ret = ERR_PTR(-EIO);
1435 			goto cleanup_and_exit;
1436 		}
1437 		if (!buffer_verified(bh) &&
1438 		    !is_dx_internal_node(dir, block,
1439 					 (struct ext4_dir_entry *)bh->b_data) &&
1440 		    !ext4_dirent_csum_verify(dir,
1441 				(struct ext4_dir_entry *)bh->b_data)) {
1442 			EXT4_ERROR_INODE(dir, "checksumming directory "
1443 					 "block %lu", (unsigned long)block);
1444 			brelse(bh);
1445 			ret = ERR_PTR(-EFSBADCRC);
1446 			goto cleanup_and_exit;
1447 		}
1448 		set_buffer_verified(bh);
1449 		i = search_dirblock(bh, dir, &fname,
1450 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1451 		if (i == 1) {
1452 			EXT4_I(dir)->i_dir_start_lookup = block;
1453 			ret = bh;
1454 			goto cleanup_and_exit;
1455 		} else {
1456 			brelse(bh);
1457 			if (i < 0)
1458 				goto cleanup_and_exit;
1459 		}
1460 	next:
1461 		if (++block >= nblocks)
1462 			block = 0;
1463 	} while (block != start);
1464 
1465 	/*
1466 	 * If the directory has grown while we were searching, then
1467 	 * search the last part of the directory before giving up.
1468 	 */
1469 	block = nblocks;
1470 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1471 	if (block < nblocks) {
1472 		start = 0;
1473 		goto restart;
1474 	}
1475 
1476 cleanup_and_exit:
1477 	/* Clean up the read-ahead blocks */
1478 	for (; ra_ptr < ra_max; ra_ptr++)
1479 		brelse(bh_use[ra_ptr]);
1480 	ext4_fname_free_filename(&fname);
1481 	return ret;
1482 }
1483 
1484 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1485 			struct ext4_filename *fname,
1486 			struct ext4_dir_entry_2 **res_dir)
1487 {
1488 	struct super_block * sb = dir->i_sb;
1489 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1490 	struct buffer_head *bh;
1491 	ext4_lblk_t block;
1492 	int retval;
1493 
1494 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1495 	*res_dir = NULL;
1496 #endif
1497 	frame = dx_probe(fname, dir, NULL, frames);
1498 	if (IS_ERR(frame))
1499 		return (struct buffer_head *) frame;
1500 	do {
1501 		block = dx_get_block(frame->at);
1502 		bh = ext4_read_dirblock(dir, block, DIRENT);
1503 		if (IS_ERR(bh))
1504 			goto errout;
1505 
1506 		retval = search_dirblock(bh, dir, fname,
1507 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1508 					 res_dir);
1509 		if (retval == 1)
1510 			goto success;
1511 		brelse(bh);
1512 		if (retval == -1) {
1513 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1514 			goto errout;
1515 		}
1516 
1517 		/* Check to see if we should continue to search */
1518 		retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1519 					       frames, NULL);
1520 		if (retval < 0) {
1521 			ext4_warning_inode(dir,
1522 				"error %d reading directory index block",
1523 				retval);
1524 			bh = ERR_PTR(retval);
1525 			goto errout;
1526 		}
1527 	} while (retval == 1);
1528 
1529 	bh = NULL;
1530 errout:
1531 	dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1532 success:
1533 	dx_release(frames);
1534 	return bh;
1535 }
1536 
1537 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1538 {
1539 	struct inode *inode;
1540 	struct ext4_dir_entry_2 *de;
1541 	struct buffer_head *bh;
1542 	int err;
1543 
1544 	err = fscrypt_prepare_lookup(dir, dentry, flags);
1545 	if (err)
1546 		return ERR_PTR(err);
1547 
1548 	if (dentry->d_name.len > EXT4_NAME_LEN)
1549 		return ERR_PTR(-ENAMETOOLONG);
1550 
1551 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1552 	if (IS_ERR(bh))
1553 		return (struct dentry *) bh;
1554 	inode = NULL;
1555 	if (bh) {
1556 		__u32 ino = le32_to_cpu(de->inode);
1557 		brelse(bh);
1558 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1559 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1560 			return ERR_PTR(-EFSCORRUPTED);
1561 		}
1562 		if (unlikely(ino == dir->i_ino)) {
1563 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1564 					 dentry);
1565 			return ERR_PTR(-EFSCORRUPTED);
1566 		}
1567 		inode = ext4_iget_normal(dir->i_sb, ino);
1568 		if (inode == ERR_PTR(-ESTALE)) {
1569 			EXT4_ERROR_INODE(dir,
1570 					 "deleted inode referenced: %u",
1571 					 ino);
1572 			return ERR_PTR(-EFSCORRUPTED);
1573 		}
1574 		if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1575 		    (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1576 		    !fscrypt_has_permitted_context(dir, inode)) {
1577 			ext4_warning(inode->i_sb,
1578 				     "Inconsistent encryption contexts: %lu/%lu",
1579 				     dir->i_ino, inode->i_ino);
1580 			iput(inode);
1581 			return ERR_PTR(-EPERM);
1582 		}
1583 	}
1584 	return d_splice_alias(inode, dentry);
1585 }
1586 
1587 
1588 struct dentry *ext4_get_parent(struct dentry *child)
1589 {
1590 	__u32 ino;
1591 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1592 	struct ext4_dir_entry_2 * de;
1593 	struct buffer_head *bh;
1594 
1595 	bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1596 	if (IS_ERR(bh))
1597 		return (struct dentry *) bh;
1598 	if (!bh)
1599 		return ERR_PTR(-ENOENT);
1600 	ino = le32_to_cpu(de->inode);
1601 	brelse(bh);
1602 
1603 	if (!ext4_valid_inum(child->d_sb, ino)) {
1604 		EXT4_ERROR_INODE(d_inode(child),
1605 				 "bad parent inode number: %u", ino);
1606 		return ERR_PTR(-EFSCORRUPTED);
1607 	}
1608 
1609 	return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1610 }
1611 
1612 /*
1613  * Move count entries from end of map between two memory locations.
1614  * Returns pointer to last entry moved.
1615  */
1616 static struct ext4_dir_entry_2 *
1617 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1618 		unsigned blocksize)
1619 {
1620 	unsigned rec_len = 0;
1621 
1622 	while (count--) {
1623 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1624 						(from + (map->offs<<2));
1625 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1626 		memcpy (to, de, rec_len);
1627 		((struct ext4_dir_entry_2 *) to)->rec_len =
1628 				ext4_rec_len_to_disk(rec_len, blocksize);
1629 		de->inode = 0;
1630 		map++;
1631 		to += rec_len;
1632 	}
1633 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1634 }
1635 
1636 /*
1637  * Compact each dir entry in the range to the minimal rec_len.
1638  * Returns pointer to last entry in range.
1639  */
1640 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1641 {
1642 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1643 	unsigned rec_len = 0;
1644 
1645 	prev = to = de;
1646 	while ((char*)de < base + blocksize) {
1647 		next = ext4_next_entry(de, blocksize);
1648 		if (de->inode && de->name_len) {
1649 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1650 			if (de > to)
1651 				memmove(to, de, rec_len);
1652 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1653 			prev = to;
1654 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1655 		}
1656 		de = next;
1657 	}
1658 	return prev;
1659 }
1660 
1661 /*
1662  * Split a full leaf block to make room for a new dir entry.
1663  * Allocate a new block, and move entries so that they are approx. equally full.
1664  * Returns pointer to de in block into which the new entry will be inserted.
1665  */
1666 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1667 			struct buffer_head **bh,struct dx_frame *frame,
1668 			struct dx_hash_info *hinfo)
1669 {
1670 	unsigned blocksize = dir->i_sb->s_blocksize;
1671 	unsigned count, continued;
1672 	struct buffer_head *bh2;
1673 	ext4_lblk_t newblock;
1674 	u32 hash2;
1675 	struct dx_map_entry *map;
1676 	char *data1 = (*bh)->b_data, *data2;
1677 	unsigned split, move, size;
1678 	struct ext4_dir_entry_2 *de = NULL, *de2;
1679 	struct ext4_dir_entry_tail *t;
1680 	int	csum_size = 0;
1681 	int	err = 0, i;
1682 
1683 	if (ext4_has_metadata_csum(dir->i_sb))
1684 		csum_size = sizeof(struct ext4_dir_entry_tail);
1685 
1686 	bh2 = ext4_append(handle, dir, &newblock);
1687 	if (IS_ERR(bh2)) {
1688 		brelse(*bh);
1689 		*bh = NULL;
1690 		return (struct ext4_dir_entry_2 *) bh2;
1691 	}
1692 
1693 	BUFFER_TRACE(*bh, "get_write_access");
1694 	err = ext4_journal_get_write_access(handle, *bh);
1695 	if (err)
1696 		goto journal_error;
1697 
1698 	BUFFER_TRACE(frame->bh, "get_write_access");
1699 	err = ext4_journal_get_write_access(handle, frame->bh);
1700 	if (err)
1701 		goto journal_error;
1702 
1703 	data2 = bh2->b_data;
1704 
1705 	/* create map in the end of data2 block */
1706 	map = (struct dx_map_entry *) (data2 + blocksize);
1707 	count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1708 			     blocksize, hinfo, map);
1709 	map -= count;
1710 	dx_sort_map(map, count);
1711 	/* Split the existing block in the middle, size-wise */
1712 	size = 0;
1713 	move = 0;
1714 	for (i = count-1; i >= 0; i--) {
1715 		/* is more than half of this entry in 2nd half of the block? */
1716 		if (size + map[i].size/2 > blocksize/2)
1717 			break;
1718 		size += map[i].size;
1719 		move++;
1720 	}
1721 	/* map index at which we will split */
1722 	split = count - move;
1723 	hash2 = map[split].hash;
1724 	continued = hash2 == map[split - 1].hash;
1725 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1726 			(unsigned long)dx_get_block(frame->at),
1727 					hash2, split, count-split));
1728 
1729 	/* Fancy dance to stay within two buffers */
1730 	de2 = dx_move_dirents(data1, data2, map + split, count - split,
1731 			      blocksize);
1732 	de = dx_pack_dirents(data1, blocksize);
1733 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1734 					   (char *) de,
1735 					   blocksize);
1736 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1737 					    (char *) de2,
1738 					    blocksize);
1739 	if (csum_size) {
1740 		t = EXT4_DIRENT_TAIL(data2, blocksize);
1741 		initialize_dirent_tail(t, blocksize);
1742 
1743 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1744 		initialize_dirent_tail(t, blocksize);
1745 	}
1746 
1747 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1748 			blocksize, 1));
1749 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1750 			blocksize, 1));
1751 
1752 	/* Which block gets the new entry? */
1753 	if (hinfo->hash >= hash2) {
1754 		swap(*bh, bh2);
1755 		de = de2;
1756 	}
1757 	dx_insert_block(frame, hash2 + continued, newblock);
1758 	err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1759 	if (err)
1760 		goto journal_error;
1761 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1762 	if (err)
1763 		goto journal_error;
1764 	brelse(bh2);
1765 	dxtrace(dx_show_index("frame", frame->entries));
1766 	return de;
1767 
1768 journal_error:
1769 	brelse(*bh);
1770 	brelse(bh2);
1771 	*bh = NULL;
1772 	ext4_std_error(dir->i_sb, err);
1773 	return ERR_PTR(err);
1774 }
1775 
1776 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1777 		      struct buffer_head *bh,
1778 		      void *buf, int buf_size,
1779 		      struct ext4_filename *fname,
1780 		      struct ext4_dir_entry_2 **dest_de)
1781 {
1782 	struct ext4_dir_entry_2 *de;
1783 	unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1784 	int nlen, rlen;
1785 	unsigned int offset = 0;
1786 	char *top;
1787 
1788 	de = (struct ext4_dir_entry_2 *)buf;
1789 	top = buf + buf_size - reclen;
1790 	while ((char *) de <= top) {
1791 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1792 					 buf, buf_size, offset))
1793 			return -EFSCORRUPTED;
1794 		if (ext4_match(fname, de))
1795 			return -EEXIST;
1796 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1797 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1798 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1799 			break;
1800 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1801 		offset += rlen;
1802 	}
1803 	if ((char *) de > top)
1804 		return -ENOSPC;
1805 
1806 	*dest_de = de;
1807 	return 0;
1808 }
1809 
1810 void ext4_insert_dentry(struct inode *inode,
1811 			struct ext4_dir_entry_2 *de,
1812 			int buf_size,
1813 			struct ext4_filename *fname)
1814 {
1815 
1816 	int nlen, rlen;
1817 
1818 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1819 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1820 	if (de->inode) {
1821 		struct ext4_dir_entry_2 *de1 =
1822 			(struct ext4_dir_entry_2 *)((char *)de + nlen);
1823 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1824 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1825 		de = de1;
1826 	}
1827 	de->file_type = EXT4_FT_UNKNOWN;
1828 	de->inode = cpu_to_le32(inode->i_ino);
1829 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1830 	de->name_len = fname_len(fname);
1831 	memcpy(de->name, fname_name(fname), fname_len(fname));
1832 }
1833 
1834 /*
1835  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1836  * it points to a directory entry which is guaranteed to be large
1837  * enough for new directory entry.  If de is NULL, then
1838  * add_dirent_to_buf will attempt search the directory block for
1839  * space.  It will return -ENOSPC if no space is available, and -EIO
1840  * and -EEXIST if directory entry already exists.
1841  */
1842 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1843 			     struct inode *dir,
1844 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1845 			     struct buffer_head *bh)
1846 {
1847 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1848 	int		csum_size = 0;
1849 	int		err;
1850 
1851 	if (ext4_has_metadata_csum(inode->i_sb))
1852 		csum_size = sizeof(struct ext4_dir_entry_tail);
1853 
1854 	if (!de) {
1855 		err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1856 					blocksize - csum_size, fname, &de);
1857 		if (err)
1858 			return err;
1859 	}
1860 	BUFFER_TRACE(bh, "get_write_access");
1861 	err = ext4_journal_get_write_access(handle, bh);
1862 	if (err) {
1863 		ext4_std_error(dir->i_sb, err);
1864 		return err;
1865 	}
1866 
1867 	/* By now the buffer is marked for journaling */
1868 	ext4_insert_dentry(inode, de, blocksize, fname);
1869 
1870 	/*
1871 	 * XXX shouldn't update any times until successful
1872 	 * completion of syscall, but too many callers depend
1873 	 * on this.
1874 	 *
1875 	 * XXX similarly, too many callers depend on
1876 	 * ext4_new_inode() setting the times, but error
1877 	 * recovery deletes the inode, so the worst that can
1878 	 * happen is that the times are slightly out of date
1879 	 * and/or different from the directory change time.
1880 	 */
1881 	dir->i_mtime = dir->i_ctime = current_time(dir);
1882 	ext4_update_dx_flag(dir);
1883 	inode_inc_iversion(dir);
1884 	ext4_mark_inode_dirty(handle, dir);
1885 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1886 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1887 	if (err)
1888 		ext4_std_error(dir->i_sb, err);
1889 	return 0;
1890 }
1891 
1892 /*
1893  * This converts a one block unindexed directory to a 3 block indexed
1894  * directory, and adds the dentry to the indexed directory.
1895  */
1896 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1897 			    struct inode *dir,
1898 			    struct inode *inode, struct buffer_head *bh)
1899 {
1900 	struct buffer_head *bh2;
1901 	struct dx_root	*root;
1902 	struct dx_frame	frames[EXT4_HTREE_LEVEL], *frame;
1903 	struct dx_entry *entries;
1904 	struct ext4_dir_entry_2	*de, *de2;
1905 	struct ext4_dir_entry_tail *t;
1906 	char		*data1, *top;
1907 	unsigned	len;
1908 	int		retval;
1909 	unsigned	blocksize;
1910 	ext4_lblk_t  block;
1911 	struct fake_dirent *fde;
1912 	int csum_size = 0;
1913 
1914 	if (ext4_has_metadata_csum(inode->i_sb))
1915 		csum_size = sizeof(struct ext4_dir_entry_tail);
1916 
1917 	blocksize =  dir->i_sb->s_blocksize;
1918 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1919 	BUFFER_TRACE(bh, "get_write_access");
1920 	retval = ext4_journal_get_write_access(handle, bh);
1921 	if (retval) {
1922 		ext4_std_error(dir->i_sb, retval);
1923 		brelse(bh);
1924 		return retval;
1925 	}
1926 	root = (struct dx_root *) bh->b_data;
1927 
1928 	/* The 0th block becomes the root, move the dirents out */
1929 	fde = &root->dotdot;
1930 	de = (struct ext4_dir_entry_2 *)((char *)fde +
1931 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
1932 	if ((char *) de >= (((char *) root) + blocksize)) {
1933 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1934 		brelse(bh);
1935 		return -EFSCORRUPTED;
1936 	}
1937 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1938 
1939 	/* Allocate new block for the 0th block's dirents */
1940 	bh2 = ext4_append(handle, dir, &block);
1941 	if (IS_ERR(bh2)) {
1942 		brelse(bh);
1943 		return PTR_ERR(bh2);
1944 	}
1945 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1946 	data1 = bh2->b_data;
1947 
1948 	memcpy (data1, de, len);
1949 	de = (struct ext4_dir_entry_2 *) data1;
1950 	top = data1 + len;
1951 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1952 		de = de2;
1953 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1954 					   (char *) de,
1955 					   blocksize);
1956 
1957 	if (csum_size) {
1958 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1959 		initialize_dirent_tail(t, blocksize);
1960 	}
1961 
1962 	/* Initialize the root; the dot dirents already exist */
1963 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1964 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1965 					   blocksize);
1966 	memset (&root->info, 0, sizeof(root->info));
1967 	root->info.info_length = sizeof(root->info);
1968 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1969 	entries = root->entries;
1970 	dx_set_block(entries, 1);
1971 	dx_set_count(entries, 1);
1972 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1973 
1974 	/* Initialize as for dx_probe */
1975 	fname->hinfo.hash_version = root->info.hash_version;
1976 	if (fname->hinfo.hash_version <= DX_HASH_TEA)
1977 		fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1978 	fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1979 	ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1980 
1981 	memset(frames, 0, sizeof(frames));
1982 	frame = frames;
1983 	frame->entries = entries;
1984 	frame->at = entries;
1985 	frame->bh = bh;
1986 
1987 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1988 	if (retval)
1989 		goto out_frames;
1990 	retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1991 	if (retval)
1992 		goto out_frames;
1993 
1994 	de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
1995 	if (IS_ERR(de)) {
1996 		retval = PTR_ERR(de);
1997 		goto out_frames;
1998 	}
1999 
2000 	retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2001 out_frames:
2002 	/*
2003 	 * Even if the block split failed, we have to properly write
2004 	 * out all the changes we did so far. Otherwise we can end up
2005 	 * with corrupted filesystem.
2006 	 */
2007 	if (retval)
2008 		ext4_mark_inode_dirty(handle, dir);
2009 	dx_release(frames);
2010 	brelse(bh2);
2011 	return retval;
2012 }
2013 
2014 /*
2015  *	ext4_add_entry()
2016  *
2017  * adds a file entry to the specified directory, using the same
2018  * semantics as ext4_find_entry(). It returns NULL if it failed.
2019  *
2020  * NOTE!! The inode part of 'de' is left at 0 - which means you
2021  * may not sleep between calling this and putting something into
2022  * the entry, as someone else might have used it while you slept.
2023  */
2024 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2025 			  struct inode *inode)
2026 {
2027 	struct inode *dir = d_inode(dentry->d_parent);
2028 	struct buffer_head *bh = NULL;
2029 	struct ext4_dir_entry_2 *de;
2030 	struct ext4_dir_entry_tail *t;
2031 	struct super_block *sb;
2032 	struct ext4_filename fname;
2033 	int	retval;
2034 	int	dx_fallback=0;
2035 	unsigned blocksize;
2036 	ext4_lblk_t block, blocks;
2037 	int	csum_size = 0;
2038 
2039 	if (ext4_has_metadata_csum(inode->i_sb))
2040 		csum_size = sizeof(struct ext4_dir_entry_tail);
2041 
2042 	sb = dir->i_sb;
2043 	blocksize = sb->s_blocksize;
2044 	if (!dentry->d_name.len)
2045 		return -EINVAL;
2046 
2047 	retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2048 	if (retval)
2049 		return retval;
2050 
2051 	if (ext4_has_inline_data(dir)) {
2052 		retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2053 		if (retval < 0)
2054 			goto out;
2055 		if (retval == 1) {
2056 			retval = 0;
2057 			goto out;
2058 		}
2059 	}
2060 
2061 	if (is_dx(dir)) {
2062 		retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2063 		if (!retval || (retval != ERR_BAD_DX_DIR))
2064 			goto out;
2065 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2066 		dx_fallback++;
2067 		ext4_mark_inode_dirty(handle, dir);
2068 	}
2069 	blocks = dir->i_size >> sb->s_blocksize_bits;
2070 	for (block = 0; block < blocks; block++) {
2071 		bh = ext4_read_dirblock(dir, block, DIRENT);
2072 		if (IS_ERR(bh)) {
2073 			retval = PTR_ERR(bh);
2074 			bh = NULL;
2075 			goto out;
2076 		}
2077 		retval = add_dirent_to_buf(handle, &fname, dir, inode,
2078 					   NULL, bh);
2079 		if (retval != -ENOSPC)
2080 			goto out;
2081 
2082 		if (blocks == 1 && !dx_fallback &&
2083 		    ext4_has_feature_dir_index(sb)) {
2084 			retval = make_indexed_dir(handle, &fname, dir,
2085 						  inode, bh);
2086 			bh = NULL; /* make_indexed_dir releases bh */
2087 			goto out;
2088 		}
2089 		brelse(bh);
2090 	}
2091 	bh = ext4_append(handle, dir, &block);
2092 	if (IS_ERR(bh)) {
2093 		retval = PTR_ERR(bh);
2094 		bh = NULL;
2095 		goto out;
2096 	}
2097 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2098 	de->inode = 0;
2099 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2100 
2101 	if (csum_size) {
2102 		t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2103 		initialize_dirent_tail(t, blocksize);
2104 	}
2105 
2106 	retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2107 out:
2108 	ext4_fname_free_filename(&fname);
2109 	brelse(bh);
2110 	if (retval == 0)
2111 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2112 	return retval;
2113 }
2114 
2115 /*
2116  * Returns 0 for success, or a negative error value
2117  */
2118 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2119 			     struct inode *dir, struct inode *inode)
2120 {
2121 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2122 	struct dx_entry *entries, *at;
2123 	struct buffer_head *bh;
2124 	struct super_block *sb = dir->i_sb;
2125 	struct ext4_dir_entry_2 *de;
2126 	int restart;
2127 	int err;
2128 
2129 again:
2130 	restart = 0;
2131 	frame = dx_probe(fname, dir, NULL, frames);
2132 	if (IS_ERR(frame))
2133 		return PTR_ERR(frame);
2134 	entries = frame->entries;
2135 	at = frame->at;
2136 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2137 	if (IS_ERR(bh)) {
2138 		err = PTR_ERR(bh);
2139 		bh = NULL;
2140 		goto cleanup;
2141 	}
2142 
2143 	BUFFER_TRACE(bh, "get_write_access");
2144 	err = ext4_journal_get_write_access(handle, bh);
2145 	if (err)
2146 		goto journal_error;
2147 
2148 	err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2149 	if (err != -ENOSPC)
2150 		goto cleanup;
2151 
2152 	err = 0;
2153 	/* Block full, should compress but for now just split */
2154 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2155 		       dx_get_count(entries), dx_get_limit(entries)));
2156 	/* Need to split index? */
2157 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2158 		ext4_lblk_t newblock;
2159 		int levels = frame - frames + 1;
2160 		unsigned int icount;
2161 		int add_level = 1;
2162 		struct dx_entry *entries2;
2163 		struct dx_node *node2;
2164 		struct buffer_head *bh2;
2165 
2166 		while (frame > frames) {
2167 			if (dx_get_count((frame - 1)->entries) <
2168 			    dx_get_limit((frame - 1)->entries)) {
2169 				add_level = 0;
2170 				break;
2171 			}
2172 			frame--; /* split higher index block */
2173 			at = frame->at;
2174 			entries = frame->entries;
2175 			restart = 1;
2176 		}
2177 		if (add_level && levels == ext4_dir_htree_level(sb)) {
2178 			ext4_warning(sb, "Directory (ino: %lu) index full, "
2179 					 "reach max htree level :%d",
2180 					 dir->i_ino, levels);
2181 			if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2182 				ext4_warning(sb, "Large directory feature is "
2183 						 "not enabled on this "
2184 						 "filesystem");
2185 			}
2186 			err = -ENOSPC;
2187 			goto cleanup;
2188 		}
2189 		icount = dx_get_count(entries);
2190 		bh2 = ext4_append(handle, dir, &newblock);
2191 		if (IS_ERR(bh2)) {
2192 			err = PTR_ERR(bh2);
2193 			goto cleanup;
2194 		}
2195 		node2 = (struct dx_node *)(bh2->b_data);
2196 		entries2 = node2->entries;
2197 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2198 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2199 							   sb->s_blocksize);
2200 		BUFFER_TRACE(frame->bh, "get_write_access");
2201 		err = ext4_journal_get_write_access(handle, frame->bh);
2202 		if (err)
2203 			goto journal_error;
2204 		if (!add_level) {
2205 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2206 			unsigned hash2 = dx_get_hash(entries + icount1);
2207 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2208 				       icount1, icount2));
2209 
2210 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2211 			err = ext4_journal_get_write_access(handle,
2212 							     (frame - 1)->bh);
2213 			if (err)
2214 				goto journal_error;
2215 
2216 			memcpy((char *) entries2, (char *) (entries + icount1),
2217 			       icount2 * sizeof(struct dx_entry));
2218 			dx_set_count(entries, icount1);
2219 			dx_set_count(entries2, icount2);
2220 			dx_set_limit(entries2, dx_node_limit(dir));
2221 
2222 			/* Which index block gets the new entry? */
2223 			if (at - entries >= icount1) {
2224 				frame->at = at = at - entries - icount1 + entries2;
2225 				frame->entries = entries = entries2;
2226 				swap(frame->bh, bh2);
2227 			}
2228 			dx_insert_block((frame - 1), hash2, newblock);
2229 			dxtrace(dx_show_index("node", frame->entries));
2230 			dxtrace(dx_show_index("node",
2231 			       ((struct dx_node *) bh2->b_data)->entries));
2232 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2233 			if (err)
2234 				goto journal_error;
2235 			brelse (bh2);
2236 			err = ext4_handle_dirty_dx_node(handle, dir,
2237 						   (frame - 1)->bh);
2238 			if (err)
2239 				goto journal_error;
2240 			if (restart) {
2241 				err = ext4_handle_dirty_dx_node(handle, dir,
2242 							   frame->bh);
2243 				goto journal_error;
2244 			}
2245 		} else {
2246 			struct dx_root *dxroot;
2247 			memcpy((char *) entries2, (char *) entries,
2248 			       icount * sizeof(struct dx_entry));
2249 			dx_set_limit(entries2, dx_node_limit(dir));
2250 
2251 			/* Set up root */
2252 			dx_set_count(entries, 1);
2253 			dx_set_block(entries + 0, newblock);
2254 			dxroot = (struct dx_root *)frames[0].bh->b_data;
2255 			dxroot->info.indirect_levels += 1;
2256 			dxtrace(printk(KERN_DEBUG
2257 				       "Creating %d level index...\n",
2258 				       info->indirect_levels));
2259 			err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2260 			if (err)
2261 				goto journal_error;
2262 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2263 			brelse(bh2);
2264 			restart = 1;
2265 			goto journal_error;
2266 		}
2267 	}
2268 	de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2269 	if (IS_ERR(de)) {
2270 		err = PTR_ERR(de);
2271 		goto cleanup;
2272 	}
2273 	err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2274 	goto cleanup;
2275 
2276 journal_error:
2277 	ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2278 cleanup:
2279 	brelse(bh);
2280 	dx_release(frames);
2281 	/* @restart is true means htree-path has been changed, we need to
2282 	 * repeat dx_probe() to find out valid htree-path
2283 	 */
2284 	if (restart && err == 0)
2285 		goto again;
2286 	return err;
2287 }
2288 
2289 /*
2290  * ext4_generic_delete_entry deletes a directory entry by merging it
2291  * with the previous entry
2292  */
2293 int ext4_generic_delete_entry(handle_t *handle,
2294 			      struct inode *dir,
2295 			      struct ext4_dir_entry_2 *de_del,
2296 			      struct buffer_head *bh,
2297 			      void *entry_buf,
2298 			      int buf_size,
2299 			      int csum_size)
2300 {
2301 	struct ext4_dir_entry_2 *de, *pde;
2302 	unsigned int blocksize = dir->i_sb->s_blocksize;
2303 	int i;
2304 
2305 	i = 0;
2306 	pde = NULL;
2307 	de = (struct ext4_dir_entry_2 *)entry_buf;
2308 	while (i < buf_size - csum_size) {
2309 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2310 					 bh->b_data, bh->b_size, i))
2311 			return -EFSCORRUPTED;
2312 		if (de == de_del)  {
2313 			if (pde)
2314 				pde->rec_len = ext4_rec_len_to_disk(
2315 					ext4_rec_len_from_disk(pde->rec_len,
2316 							       blocksize) +
2317 					ext4_rec_len_from_disk(de->rec_len,
2318 							       blocksize),
2319 					blocksize);
2320 			else
2321 				de->inode = 0;
2322 			inode_inc_iversion(dir);
2323 			return 0;
2324 		}
2325 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2326 		pde = de;
2327 		de = ext4_next_entry(de, blocksize);
2328 	}
2329 	return -ENOENT;
2330 }
2331 
2332 static int ext4_delete_entry(handle_t *handle,
2333 			     struct inode *dir,
2334 			     struct ext4_dir_entry_2 *de_del,
2335 			     struct buffer_head *bh)
2336 {
2337 	int err, csum_size = 0;
2338 
2339 	if (ext4_has_inline_data(dir)) {
2340 		int has_inline_data = 1;
2341 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2342 					       &has_inline_data);
2343 		if (has_inline_data)
2344 			return err;
2345 	}
2346 
2347 	if (ext4_has_metadata_csum(dir->i_sb))
2348 		csum_size = sizeof(struct ext4_dir_entry_tail);
2349 
2350 	BUFFER_TRACE(bh, "get_write_access");
2351 	err = ext4_journal_get_write_access(handle, bh);
2352 	if (unlikely(err))
2353 		goto out;
2354 
2355 	err = ext4_generic_delete_entry(handle, dir, de_del,
2356 					bh, bh->b_data,
2357 					dir->i_sb->s_blocksize, csum_size);
2358 	if (err)
2359 		goto out;
2360 
2361 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2362 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2363 	if (unlikely(err))
2364 		goto out;
2365 
2366 	return 0;
2367 out:
2368 	if (err != -ENOENT)
2369 		ext4_std_error(dir->i_sb, err);
2370 	return err;
2371 }
2372 
2373 /*
2374  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2375  * since this indicates that nlinks count was previously 1 to avoid overflowing
2376  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2377  * that subdirectory link counts are not being maintained accurately.
2378  *
2379  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2380  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2381  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2382  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2383  */
2384 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2385 {
2386 	inc_nlink(inode);
2387 	if (is_dx(inode) &&
2388 	    (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2389 		set_nlink(inode, 1);
2390 }
2391 
2392 /*
2393  * If a directory had nlink == 1, then we should let it be 1. This indicates
2394  * directory has >EXT4_LINK_MAX subdirs.
2395  */
2396 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2397 {
2398 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2399 		drop_nlink(inode);
2400 }
2401 
2402 
2403 static int ext4_add_nondir(handle_t *handle,
2404 		struct dentry *dentry, struct inode *inode)
2405 {
2406 	int err = ext4_add_entry(handle, dentry, inode);
2407 	if (!err) {
2408 		ext4_mark_inode_dirty(handle, inode);
2409 		unlock_new_inode(inode);
2410 		d_instantiate(dentry, inode);
2411 		return 0;
2412 	}
2413 	drop_nlink(inode);
2414 	unlock_new_inode(inode);
2415 	iput(inode);
2416 	return err;
2417 }
2418 
2419 /*
2420  * By the time this is called, we already have created
2421  * the directory cache entry for the new file, but it
2422  * is so far negative - it has no inode.
2423  *
2424  * If the create succeeds, we fill in the inode information
2425  * with d_instantiate().
2426  */
2427 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2428 		       bool excl)
2429 {
2430 	handle_t *handle;
2431 	struct inode *inode;
2432 	int err, credits, retries = 0;
2433 
2434 	err = dquot_initialize(dir);
2435 	if (err)
2436 		return err;
2437 
2438 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2439 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2440 retry:
2441 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2442 					    NULL, EXT4_HT_DIR, credits);
2443 	handle = ext4_journal_current_handle();
2444 	err = PTR_ERR(inode);
2445 	if (!IS_ERR(inode)) {
2446 		inode->i_op = &ext4_file_inode_operations;
2447 		inode->i_fop = &ext4_file_operations;
2448 		ext4_set_aops(inode);
2449 		err = ext4_add_nondir(handle, dentry, inode);
2450 		if (!err && IS_DIRSYNC(dir))
2451 			ext4_handle_sync(handle);
2452 	}
2453 	if (handle)
2454 		ext4_journal_stop(handle);
2455 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2456 		goto retry;
2457 	return err;
2458 }
2459 
2460 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2461 		      umode_t mode, dev_t rdev)
2462 {
2463 	handle_t *handle;
2464 	struct inode *inode;
2465 	int err, credits, retries = 0;
2466 
2467 	err = dquot_initialize(dir);
2468 	if (err)
2469 		return err;
2470 
2471 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2472 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2473 retry:
2474 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2475 					    NULL, EXT4_HT_DIR, credits);
2476 	handle = ext4_journal_current_handle();
2477 	err = PTR_ERR(inode);
2478 	if (!IS_ERR(inode)) {
2479 		init_special_inode(inode, inode->i_mode, rdev);
2480 		inode->i_op = &ext4_special_inode_operations;
2481 		err = ext4_add_nondir(handle, dentry, inode);
2482 		if (!err && IS_DIRSYNC(dir))
2483 			ext4_handle_sync(handle);
2484 	}
2485 	if (handle)
2486 		ext4_journal_stop(handle);
2487 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2488 		goto retry;
2489 	return err;
2490 }
2491 
2492 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2493 {
2494 	handle_t *handle;
2495 	struct inode *inode;
2496 	int err, retries = 0;
2497 
2498 	err = dquot_initialize(dir);
2499 	if (err)
2500 		return err;
2501 
2502 retry:
2503 	inode = ext4_new_inode_start_handle(dir, mode,
2504 					    NULL, 0, NULL,
2505 					    EXT4_HT_DIR,
2506 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2507 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2508 	handle = ext4_journal_current_handle();
2509 	err = PTR_ERR(inode);
2510 	if (!IS_ERR(inode)) {
2511 		inode->i_op = &ext4_file_inode_operations;
2512 		inode->i_fop = &ext4_file_operations;
2513 		ext4_set_aops(inode);
2514 		d_tmpfile(dentry, inode);
2515 		err = ext4_orphan_add(handle, inode);
2516 		if (err)
2517 			goto err_unlock_inode;
2518 		mark_inode_dirty(inode);
2519 		unlock_new_inode(inode);
2520 	}
2521 	if (handle)
2522 		ext4_journal_stop(handle);
2523 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2524 		goto retry;
2525 	return err;
2526 err_unlock_inode:
2527 	ext4_journal_stop(handle);
2528 	unlock_new_inode(inode);
2529 	return err;
2530 }
2531 
2532 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2533 			  struct ext4_dir_entry_2 *de,
2534 			  int blocksize, int csum_size,
2535 			  unsigned int parent_ino, int dotdot_real_len)
2536 {
2537 	de->inode = cpu_to_le32(inode->i_ino);
2538 	de->name_len = 1;
2539 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2540 					   blocksize);
2541 	strcpy(de->name, ".");
2542 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2543 
2544 	de = ext4_next_entry(de, blocksize);
2545 	de->inode = cpu_to_le32(parent_ino);
2546 	de->name_len = 2;
2547 	if (!dotdot_real_len)
2548 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2549 					(csum_size + EXT4_DIR_REC_LEN(1)),
2550 					blocksize);
2551 	else
2552 		de->rec_len = ext4_rec_len_to_disk(
2553 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2554 	strcpy(de->name, "..");
2555 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2556 
2557 	return ext4_next_entry(de, blocksize);
2558 }
2559 
2560 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2561 			     struct inode *inode)
2562 {
2563 	struct buffer_head *dir_block = NULL;
2564 	struct ext4_dir_entry_2 *de;
2565 	struct ext4_dir_entry_tail *t;
2566 	ext4_lblk_t block = 0;
2567 	unsigned int blocksize = dir->i_sb->s_blocksize;
2568 	int csum_size = 0;
2569 	int err;
2570 
2571 	if (ext4_has_metadata_csum(dir->i_sb))
2572 		csum_size = sizeof(struct ext4_dir_entry_tail);
2573 
2574 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2575 		err = ext4_try_create_inline_dir(handle, dir, inode);
2576 		if (err < 0 && err != -ENOSPC)
2577 			goto out;
2578 		if (!err)
2579 			goto out;
2580 	}
2581 
2582 	inode->i_size = 0;
2583 	dir_block = ext4_append(handle, inode, &block);
2584 	if (IS_ERR(dir_block))
2585 		return PTR_ERR(dir_block);
2586 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2587 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2588 	set_nlink(inode, 2);
2589 	if (csum_size) {
2590 		t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2591 		initialize_dirent_tail(t, blocksize);
2592 	}
2593 
2594 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2595 	err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2596 	if (err)
2597 		goto out;
2598 	set_buffer_verified(dir_block);
2599 out:
2600 	brelse(dir_block);
2601 	return err;
2602 }
2603 
2604 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2605 {
2606 	handle_t *handle;
2607 	struct inode *inode;
2608 	int err, credits, retries = 0;
2609 
2610 	if (EXT4_DIR_LINK_MAX(dir))
2611 		return -EMLINK;
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, S_IFDIR | mode,
2621 					    &dentry->d_name,
2622 					    0, NULL, EXT4_HT_DIR, credits);
2623 	handle = ext4_journal_current_handle();
2624 	err = PTR_ERR(inode);
2625 	if (IS_ERR(inode))
2626 		goto out_stop;
2627 
2628 	inode->i_op = &ext4_dir_inode_operations;
2629 	inode->i_fop = &ext4_dir_operations;
2630 	err = ext4_init_new_dir(handle, dir, inode);
2631 	if (err)
2632 		goto out_clear_inode;
2633 	err = ext4_mark_inode_dirty(handle, inode);
2634 	if (!err)
2635 		err = ext4_add_entry(handle, dentry, inode);
2636 	if (err) {
2637 out_clear_inode:
2638 		clear_nlink(inode);
2639 		unlock_new_inode(inode);
2640 		ext4_mark_inode_dirty(handle, inode);
2641 		iput(inode);
2642 		goto out_stop;
2643 	}
2644 	ext4_inc_count(handle, dir);
2645 	ext4_update_dx_flag(dir);
2646 	err = ext4_mark_inode_dirty(handle, dir);
2647 	if (err)
2648 		goto out_clear_inode;
2649 	unlock_new_inode(inode);
2650 	d_instantiate(dentry, inode);
2651 	if (IS_DIRSYNC(dir))
2652 		ext4_handle_sync(handle);
2653 
2654 out_stop:
2655 	if (handle)
2656 		ext4_journal_stop(handle);
2657 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2658 		goto retry;
2659 	return err;
2660 }
2661 
2662 /*
2663  * routine to check that the specified directory is empty (for rmdir)
2664  */
2665 bool ext4_empty_dir(struct inode *inode)
2666 {
2667 	unsigned int offset;
2668 	struct buffer_head *bh;
2669 	struct ext4_dir_entry_2 *de, *de1;
2670 	struct super_block *sb;
2671 
2672 	if (ext4_has_inline_data(inode)) {
2673 		int has_inline_data = 1;
2674 		int ret;
2675 
2676 		ret = empty_inline_dir(inode, &has_inline_data);
2677 		if (has_inline_data)
2678 			return ret;
2679 	}
2680 
2681 	sb = inode->i_sb;
2682 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2683 		EXT4_ERROR_INODE(inode, "invalid size");
2684 		return true;
2685 	}
2686 	bh = ext4_read_dirblock(inode, 0, EITHER);
2687 	if (IS_ERR(bh))
2688 		return true;
2689 
2690 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2691 	de1 = ext4_next_entry(de, sb->s_blocksize);
2692 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2693 			le32_to_cpu(de1->inode) == 0 ||
2694 			strcmp(".", de->name) || strcmp("..", de1->name)) {
2695 		ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2696 		brelse(bh);
2697 		return true;
2698 	}
2699 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2700 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2701 	de = ext4_next_entry(de1, sb->s_blocksize);
2702 	while (offset < inode->i_size) {
2703 		if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2704 			unsigned int lblock;
2705 			brelse(bh);
2706 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2707 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2708 			if (IS_ERR(bh))
2709 				return true;
2710 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2711 		}
2712 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2713 					 bh->b_data, bh->b_size, offset)) {
2714 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2715 							 sb->s_blocksize);
2716 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2717 			continue;
2718 		}
2719 		if (le32_to_cpu(de->inode)) {
2720 			brelse(bh);
2721 			return false;
2722 		}
2723 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2724 		de = ext4_next_entry(de, sb->s_blocksize);
2725 	}
2726 	brelse(bh);
2727 	return true;
2728 }
2729 
2730 /*
2731  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2732  * such inodes, starting at the superblock, in case we crash before the
2733  * file is closed/deleted, or in case the inode truncate spans multiple
2734  * transactions and the last transaction is not recovered after a crash.
2735  *
2736  * At filesystem recovery time, we walk this list deleting unlinked
2737  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2738  *
2739  * Orphan list manipulation functions must be called under i_mutex unless
2740  * we are just creating the inode or deleting it.
2741  */
2742 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2743 {
2744 	struct super_block *sb = inode->i_sb;
2745 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2746 	struct ext4_iloc iloc;
2747 	int err = 0, rc;
2748 	bool dirty = false;
2749 
2750 	if (!sbi->s_journal || is_bad_inode(inode))
2751 		return 0;
2752 
2753 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2754 		     !inode_is_locked(inode));
2755 	/*
2756 	 * Exit early if inode already is on orphan list. This is a big speedup
2757 	 * since we don't have to contend on the global s_orphan_lock.
2758 	 */
2759 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2760 		return 0;
2761 
2762 	/*
2763 	 * Orphan handling is only valid for files with data blocks
2764 	 * being truncated, or files being unlinked. Note that we either
2765 	 * hold i_mutex, or the inode can not be referenced from outside,
2766 	 * so i_nlink should not be bumped due to race
2767 	 */
2768 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2769 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2770 
2771 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2772 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2773 	if (err)
2774 		goto out;
2775 
2776 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2777 	if (err)
2778 		goto out;
2779 
2780 	mutex_lock(&sbi->s_orphan_lock);
2781 	/*
2782 	 * Due to previous errors inode may be already a part of on-disk
2783 	 * orphan list. If so skip on-disk list modification.
2784 	 */
2785 	if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2786 	    (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2787 		/* Insert this inode at the head of the on-disk orphan list */
2788 		NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2789 		sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2790 		dirty = true;
2791 	}
2792 	list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2793 	mutex_unlock(&sbi->s_orphan_lock);
2794 
2795 	if (dirty) {
2796 		err = ext4_handle_dirty_super(handle, sb);
2797 		rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2798 		if (!err)
2799 			err = rc;
2800 		if (err) {
2801 			/*
2802 			 * We have to remove inode from in-memory list if
2803 			 * addition to on disk orphan list failed. Stray orphan
2804 			 * list entries can cause panics at unmount time.
2805 			 */
2806 			mutex_lock(&sbi->s_orphan_lock);
2807 			list_del_init(&EXT4_I(inode)->i_orphan);
2808 			mutex_unlock(&sbi->s_orphan_lock);
2809 		}
2810 	}
2811 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2812 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2813 			inode->i_ino, NEXT_ORPHAN(inode));
2814 out:
2815 	ext4_std_error(sb, err);
2816 	return err;
2817 }
2818 
2819 /*
2820  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2821  * of such inodes stored on disk, because it is finally being cleaned up.
2822  */
2823 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2824 {
2825 	struct list_head *prev;
2826 	struct ext4_inode_info *ei = EXT4_I(inode);
2827 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2828 	__u32 ino_next;
2829 	struct ext4_iloc iloc;
2830 	int err = 0;
2831 
2832 	if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2833 		return 0;
2834 
2835 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2836 		     !inode_is_locked(inode));
2837 	/* Do this quick check before taking global s_orphan_lock. */
2838 	if (list_empty(&ei->i_orphan))
2839 		return 0;
2840 
2841 	if (handle) {
2842 		/* Grab inode buffer early before taking global s_orphan_lock */
2843 		err = ext4_reserve_inode_write(handle, inode, &iloc);
2844 	}
2845 
2846 	mutex_lock(&sbi->s_orphan_lock);
2847 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2848 
2849 	prev = ei->i_orphan.prev;
2850 	list_del_init(&ei->i_orphan);
2851 
2852 	/* If we're on an error path, we may not have a valid
2853 	 * transaction handle with which to update the orphan list on
2854 	 * disk, but we still need to remove the inode from the linked
2855 	 * list in memory. */
2856 	if (!handle || err) {
2857 		mutex_unlock(&sbi->s_orphan_lock);
2858 		goto out_err;
2859 	}
2860 
2861 	ino_next = NEXT_ORPHAN(inode);
2862 	if (prev == &sbi->s_orphan) {
2863 		jbd_debug(4, "superblock will point to %u\n", ino_next);
2864 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2865 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2866 		if (err) {
2867 			mutex_unlock(&sbi->s_orphan_lock);
2868 			goto out_brelse;
2869 		}
2870 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2871 		mutex_unlock(&sbi->s_orphan_lock);
2872 		err = ext4_handle_dirty_super(handle, inode->i_sb);
2873 	} else {
2874 		struct ext4_iloc iloc2;
2875 		struct inode *i_prev =
2876 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2877 
2878 		jbd_debug(4, "orphan inode %lu will point to %u\n",
2879 			  i_prev->i_ino, ino_next);
2880 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2881 		if (err) {
2882 			mutex_unlock(&sbi->s_orphan_lock);
2883 			goto out_brelse;
2884 		}
2885 		NEXT_ORPHAN(i_prev) = ino_next;
2886 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2887 		mutex_unlock(&sbi->s_orphan_lock);
2888 	}
2889 	if (err)
2890 		goto out_brelse;
2891 	NEXT_ORPHAN(inode) = 0;
2892 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2893 out_err:
2894 	ext4_std_error(inode->i_sb, err);
2895 	return err;
2896 
2897 out_brelse:
2898 	brelse(iloc.bh);
2899 	goto out_err;
2900 }
2901 
2902 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2903 {
2904 	int retval;
2905 	struct inode *inode;
2906 	struct buffer_head *bh;
2907 	struct ext4_dir_entry_2 *de;
2908 	handle_t *handle = NULL;
2909 
2910 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2911 		return -EIO;
2912 
2913 	/* Initialize quotas before so that eventual writes go in
2914 	 * separate transaction */
2915 	retval = dquot_initialize(dir);
2916 	if (retval)
2917 		return retval;
2918 	retval = dquot_initialize(d_inode(dentry));
2919 	if (retval)
2920 		return retval;
2921 
2922 	retval = -ENOENT;
2923 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2924 	if (IS_ERR(bh))
2925 		return PTR_ERR(bh);
2926 	if (!bh)
2927 		goto end_rmdir;
2928 
2929 	inode = d_inode(dentry);
2930 
2931 	retval = -EFSCORRUPTED;
2932 	if (le32_to_cpu(de->inode) != inode->i_ino)
2933 		goto end_rmdir;
2934 
2935 	retval = -ENOTEMPTY;
2936 	if (!ext4_empty_dir(inode))
2937 		goto end_rmdir;
2938 
2939 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2940 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2941 	if (IS_ERR(handle)) {
2942 		retval = PTR_ERR(handle);
2943 		handle = NULL;
2944 		goto end_rmdir;
2945 	}
2946 
2947 	if (IS_DIRSYNC(dir))
2948 		ext4_handle_sync(handle);
2949 
2950 	retval = ext4_delete_entry(handle, dir, de, bh);
2951 	if (retval)
2952 		goto end_rmdir;
2953 	if (!EXT4_DIR_LINK_EMPTY(inode))
2954 		ext4_warning_inode(inode,
2955 			     "empty directory '%.*s' has too many links (%u)",
2956 			     dentry->d_name.len, dentry->d_name.name,
2957 			     inode->i_nlink);
2958 	inode->i_version++;
2959 	clear_nlink(inode);
2960 	/* There's no need to set i_disksize: the fact that i_nlink is
2961 	 * zero will ensure that the right thing happens during any
2962 	 * recovery. */
2963 	inode->i_size = 0;
2964 	ext4_orphan_add(handle, inode);
2965 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2966 	ext4_mark_inode_dirty(handle, inode);
2967 	ext4_dec_count(handle, dir);
2968 	ext4_update_dx_flag(dir);
2969 	ext4_mark_inode_dirty(handle, dir);
2970 
2971 end_rmdir:
2972 	brelse(bh);
2973 	if (handle)
2974 		ext4_journal_stop(handle);
2975 	return retval;
2976 }
2977 
2978 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2979 {
2980 	int retval;
2981 	struct inode *inode;
2982 	struct buffer_head *bh;
2983 	struct ext4_dir_entry_2 *de;
2984 	handle_t *handle = NULL;
2985 
2986 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2987 		return -EIO;
2988 
2989 	trace_ext4_unlink_enter(dir, dentry);
2990 	/* Initialize quotas before so that eventual writes go
2991 	 * in separate transaction */
2992 	retval = dquot_initialize(dir);
2993 	if (retval)
2994 		return retval;
2995 	retval = dquot_initialize(d_inode(dentry));
2996 	if (retval)
2997 		return retval;
2998 
2999 	retval = -ENOENT;
3000 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3001 	if (IS_ERR(bh))
3002 		return PTR_ERR(bh);
3003 	if (!bh)
3004 		goto end_unlink;
3005 
3006 	inode = d_inode(dentry);
3007 
3008 	retval = -EFSCORRUPTED;
3009 	if (le32_to_cpu(de->inode) != inode->i_ino)
3010 		goto end_unlink;
3011 
3012 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3013 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3014 	if (IS_ERR(handle)) {
3015 		retval = PTR_ERR(handle);
3016 		handle = NULL;
3017 		goto end_unlink;
3018 	}
3019 
3020 	if (IS_DIRSYNC(dir))
3021 		ext4_handle_sync(handle);
3022 
3023 	if (inode->i_nlink == 0) {
3024 		ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3025 				   dentry->d_name.len, dentry->d_name.name);
3026 		set_nlink(inode, 1);
3027 	}
3028 	retval = ext4_delete_entry(handle, dir, de, bh);
3029 	if (retval)
3030 		goto end_unlink;
3031 	dir->i_ctime = dir->i_mtime = current_time(dir);
3032 	ext4_update_dx_flag(dir);
3033 	ext4_mark_inode_dirty(handle, dir);
3034 	drop_nlink(inode);
3035 	if (!inode->i_nlink)
3036 		ext4_orphan_add(handle, inode);
3037 	inode->i_ctime = current_time(inode);
3038 	ext4_mark_inode_dirty(handle, inode);
3039 
3040 end_unlink:
3041 	brelse(bh);
3042 	if (handle)
3043 		ext4_journal_stop(handle);
3044 	trace_ext4_unlink_exit(dentry, retval);
3045 	return retval;
3046 }
3047 
3048 static int ext4_symlink(struct inode *dir,
3049 			struct dentry *dentry, const char *symname)
3050 {
3051 	handle_t *handle;
3052 	struct inode *inode;
3053 	int err, len = strlen(symname);
3054 	int credits;
3055 	bool encryption_required;
3056 	struct fscrypt_str disk_link;
3057 	struct fscrypt_symlink_data *sd = NULL;
3058 
3059 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3060 		return -EIO;
3061 
3062 	disk_link.len = len + 1;
3063 	disk_link.name = (char *) symname;
3064 
3065 	encryption_required = (ext4_encrypted_inode(dir) ||
3066 			       DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3067 	if (encryption_required) {
3068 		err = fscrypt_get_encryption_info(dir);
3069 		if (err)
3070 			return err;
3071 		if (!fscrypt_has_encryption_key(dir))
3072 			return -ENOKEY;
3073 		disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3074 				 sizeof(struct fscrypt_symlink_data));
3075 		sd = kzalloc(disk_link.len, GFP_KERNEL);
3076 		if (!sd)
3077 			return -ENOMEM;
3078 	}
3079 
3080 	if (disk_link.len > dir->i_sb->s_blocksize) {
3081 		err = -ENAMETOOLONG;
3082 		goto err_free_sd;
3083 	}
3084 
3085 	err = dquot_initialize(dir);
3086 	if (err)
3087 		goto err_free_sd;
3088 
3089 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3090 		/*
3091 		 * For non-fast symlinks, we just allocate inode and put it on
3092 		 * orphan list in the first transaction => we need bitmap,
3093 		 * group descriptor, sb, inode block, quota blocks, and
3094 		 * possibly selinux xattr blocks.
3095 		 */
3096 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3097 			  EXT4_XATTR_TRANS_BLOCKS;
3098 	} else {
3099 		/*
3100 		 * Fast symlink. We have to add entry to directory
3101 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3102 		 * allocate new inode (bitmap, group descriptor, inode block,
3103 		 * quota blocks, sb is already counted in previous macros).
3104 		 */
3105 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3106 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3107 	}
3108 
3109 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3110 					    &dentry->d_name, 0, NULL,
3111 					    EXT4_HT_DIR, credits);
3112 	handle = ext4_journal_current_handle();
3113 	if (IS_ERR(inode)) {
3114 		if (handle)
3115 			ext4_journal_stop(handle);
3116 		err = PTR_ERR(inode);
3117 		goto err_free_sd;
3118 	}
3119 
3120 	if (encryption_required) {
3121 		struct qstr istr;
3122 		struct fscrypt_str ostr =
3123 			FSTR_INIT(sd->encrypted_path, disk_link.len);
3124 
3125 		istr.name = (const unsigned char *) symname;
3126 		istr.len = len;
3127 		err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3128 		if (err)
3129 			goto err_drop_inode;
3130 		sd->len = cpu_to_le16(ostr.len);
3131 		disk_link.name = (char *) sd;
3132 		inode->i_op = &ext4_encrypted_symlink_inode_operations;
3133 	}
3134 
3135 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3136 		if (!encryption_required)
3137 			inode->i_op = &ext4_symlink_inode_operations;
3138 		inode_nohighmem(inode);
3139 		ext4_set_aops(inode);
3140 		/*
3141 		 * We cannot call page_symlink() with transaction started
3142 		 * because it calls into ext4_write_begin() which can wait
3143 		 * for transaction commit if we are running out of space
3144 		 * and thus we deadlock. So we have to stop transaction now
3145 		 * and restart it when symlink contents is written.
3146 		 *
3147 		 * To keep fs consistent in case of crash, we have to put inode
3148 		 * to orphan list in the mean time.
3149 		 */
3150 		drop_nlink(inode);
3151 		err = ext4_orphan_add(handle, inode);
3152 		ext4_journal_stop(handle);
3153 		handle = NULL;
3154 		if (err)
3155 			goto err_drop_inode;
3156 		err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3157 		if (err)
3158 			goto err_drop_inode;
3159 		/*
3160 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3161 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3162 		 */
3163 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
3164 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3165 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3166 		if (IS_ERR(handle)) {
3167 			err = PTR_ERR(handle);
3168 			handle = NULL;
3169 			goto err_drop_inode;
3170 		}
3171 		set_nlink(inode, 1);
3172 		err = ext4_orphan_del(handle, inode);
3173 		if (err)
3174 			goto err_drop_inode;
3175 	} else {
3176 		/* clear the extent format for fast symlink */
3177 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3178 		if (!encryption_required) {
3179 			inode->i_op = &ext4_fast_symlink_inode_operations;
3180 			inode->i_link = (char *)&EXT4_I(inode)->i_data;
3181 		}
3182 		memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3183 		       disk_link.len);
3184 		inode->i_size = disk_link.len - 1;
3185 	}
3186 	EXT4_I(inode)->i_disksize = inode->i_size;
3187 	err = ext4_add_nondir(handle, dentry, inode);
3188 	if (!err && IS_DIRSYNC(dir))
3189 		ext4_handle_sync(handle);
3190 
3191 	if (handle)
3192 		ext4_journal_stop(handle);
3193 	kfree(sd);
3194 	return err;
3195 err_drop_inode:
3196 	if (handle)
3197 		ext4_journal_stop(handle);
3198 	clear_nlink(inode);
3199 	unlock_new_inode(inode);
3200 	iput(inode);
3201 err_free_sd:
3202 	kfree(sd);
3203 	return err;
3204 }
3205 
3206 static int ext4_link(struct dentry *old_dentry,
3207 		     struct inode *dir, struct dentry *dentry)
3208 {
3209 	handle_t *handle;
3210 	struct inode *inode = d_inode(old_dentry);
3211 	int err, retries = 0;
3212 
3213 	if (inode->i_nlink >= EXT4_LINK_MAX)
3214 		return -EMLINK;
3215 
3216 	err = fscrypt_prepare_link(old_dentry, dir, dentry);
3217 	if (err)
3218 		return err;
3219 
3220        if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3221 	   (!projid_eq(EXT4_I(dir)->i_projid,
3222 		       EXT4_I(old_dentry->d_inode)->i_projid)))
3223 		return -EXDEV;
3224 
3225 	err = dquot_initialize(dir);
3226 	if (err)
3227 		return err;
3228 
3229 retry:
3230 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3231 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3232 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3233 	if (IS_ERR(handle))
3234 		return PTR_ERR(handle);
3235 
3236 	if (IS_DIRSYNC(dir))
3237 		ext4_handle_sync(handle);
3238 
3239 	inode->i_ctime = current_time(inode);
3240 	ext4_inc_count(handle, inode);
3241 	ihold(inode);
3242 
3243 	err = ext4_add_entry(handle, dentry, inode);
3244 	if (!err) {
3245 		ext4_mark_inode_dirty(handle, inode);
3246 		/* this can happen only for tmpfile being
3247 		 * linked the first time
3248 		 */
3249 		if (inode->i_nlink == 1)
3250 			ext4_orphan_del(handle, inode);
3251 		d_instantiate(dentry, inode);
3252 	} else {
3253 		drop_nlink(inode);
3254 		iput(inode);
3255 	}
3256 	ext4_journal_stop(handle);
3257 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3258 		goto retry;
3259 	return err;
3260 }
3261 
3262 
3263 /*
3264  * Try to find buffer head where contains the parent block.
3265  * It should be the inode block if it is inlined or the 1st block
3266  * if it is a normal dir.
3267  */
3268 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3269 					struct inode *inode,
3270 					int *retval,
3271 					struct ext4_dir_entry_2 **parent_de,
3272 					int *inlined)
3273 {
3274 	struct buffer_head *bh;
3275 
3276 	if (!ext4_has_inline_data(inode)) {
3277 		bh = ext4_read_dirblock(inode, 0, EITHER);
3278 		if (IS_ERR(bh)) {
3279 			*retval = PTR_ERR(bh);
3280 			return NULL;
3281 		}
3282 		*parent_de = ext4_next_entry(
3283 					(struct ext4_dir_entry_2 *)bh->b_data,
3284 					inode->i_sb->s_blocksize);
3285 		return bh;
3286 	}
3287 
3288 	*inlined = 1;
3289 	return ext4_get_first_inline_block(inode, parent_de, retval);
3290 }
3291 
3292 struct ext4_renament {
3293 	struct inode *dir;
3294 	struct dentry *dentry;
3295 	struct inode *inode;
3296 	bool is_dir;
3297 	int dir_nlink_delta;
3298 
3299 	/* entry for "dentry" */
3300 	struct buffer_head *bh;
3301 	struct ext4_dir_entry_2 *de;
3302 	int inlined;
3303 
3304 	/* entry for ".." in inode if it's a directory */
3305 	struct buffer_head *dir_bh;
3306 	struct ext4_dir_entry_2 *parent_de;
3307 	int dir_inlined;
3308 };
3309 
3310 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3311 {
3312 	int retval;
3313 
3314 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3315 					      &retval, &ent->parent_de,
3316 					      &ent->dir_inlined);
3317 	if (!ent->dir_bh)
3318 		return retval;
3319 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3320 		return -EFSCORRUPTED;
3321 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3322 	return ext4_journal_get_write_access(handle, ent->dir_bh);
3323 }
3324 
3325 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3326 				  unsigned dir_ino)
3327 {
3328 	int retval;
3329 
3330 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3331 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3332 	if (!ent->dir_inlined) {
3333 		if (is_dx(ent->inode)) {
3334 			retval = ext4_handle_dirty_dx_node(handle,
3335 							   ent->inode,
3336 							   ent->dir_bh);
3337 		} else {
3338 			retval = ext4_handle_dirty_dirent_node(handle,
3339 							       ent->inode,
3340 							       ent->dir_bh);
3341 		}
3342 	} else {
3343 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3344 	}
3345 	if (retval) {
3346 		ext4_std_error(ent->dir->i_sb, retval);
3347 		return retval;
3348 	}
3349 	return 0;
3350 }
3351 
3352 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3353 		       unsigned ino, unsigned file_type)
3354 {
3355 	int retval;
3356 
3357 	BUFFER_TRACE(ent->bh, "get write access");
3358 	retval = ext4_journal_get_write_access(handle, ent->bh);
3359 	if (retval)
3360 		return retval;
3361 	ent->de->inode = cpu_to_le32(ino);
3362 	if (ext4_has_feature_filetype(ent->dir->i_sb))
3363 		ent->de->file_type = file_type;
3364 	ent->dir->i_version++;
3365 	ent->dir->i_ctime = ent->dir->i_mtime =
3366 		current_time(ent->dir);
3367 	ext4_mark_inode_dirty(handle, ent->dir);
3368 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3369 	if (!ent->inlined) {
3370 		retval = ext4_handle_dirty_dirent_node(handle,
3371 						       ent->dir, ent->bh);
3372 		if (unlikely(retval)) {
3373 			ext4_std_error(ent->dir->i_sb, retval);
3374 			return retval;
3375 		}
3376 	}
3377 	brelse(ent->bh);
3378 	ent->bh = NULL;
3379 
3380 	return 0;
3381 }
3382 
3383 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3384 				  const struct qstr *d_name)
3385 {
3386 	int retval = -ENOENT;
3387 	struct buffer_head *bh;
3388 	struct ext4_dir_entry_2 *de;
3389 
3390 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3391 	if (IS_ERR(bh))
3392 		return PTR_ERR(bh);
3393 	if (bh) {
3394 		retval = ext4_delete_entry(handle, dir, de, bh);
3395 		brelse(bh);
3396 	}
3397 	return retval;
3398 }
3399 
3400 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3401 			       int force_reread)
3402 {
3403 	int retval;
3404 	/*
3405 	 * ent->de could have moved from under us during htree split, so make
3406 	 * sure that we are deleting the right entry.  We might also be pointing
3407 	 * to a stale entry in the unused part of ent->bh so just checking inum
3408 	 * and the name isn't enough.
3409 	 */
3410 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3411 	    ent->de->name_len != ent->dentry->d_name.len ||
3412 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3413 		    ent->de->name_len) ||
3414 	    force_reread) {
3415 		retval = ext4_find_delete_entry(handle, ent->dir,
3416 						&ent->dentry->d_name);
3417 	} else {
3418 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3419 		if (retval == -ENOENT) {
3420 			retval = ext4_find_delete_entry(handle, ent->dir,
3421 							&ent->dentry->d_name);
3422 		}
3423 	}
3424 
3425 	if (retval) {
3426 		ext4_warning_inode(ent->dir,
3427 				   "Deleting old file: nlink %d, error=%d",
3428 				   ent->dir->i_nlink, retval);
3429 	}
3430 }
3431 
3432 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3433 {
3434 	if (ent->dir_nlink_delta) {
3435 		if (ent->dir_nlink_delta == -1)
3436 			ext4_dec_count(handle, ent->dir);
3437 		else
3438 			ext4_inc_count(handle, ent->dir);
3439 		ext4_mark_inode_dirty(handle, ent->dir);
3440 	}
3441 }
3442 
3443 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3444 					      int credits, handle_t **h)
3445 {
3446 	struct inode *wh;
3447 	handle_t *handle;
3448 	int retries = 0;
3449 
3450 	/*
3451 	 * for inode block, sb block, group summaries,
3452 	 * and inode bitmap
3453 	 */
3454 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3455 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3456 retry:
3457 	wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3458 					 &ent->dentry->d_name, 0, NULL,
3459 					 EXT4_HT_DIR, credits);
3460 
3461 	handle = ext4_journal_current_handle();
3462 	if (IS_ERR(wh)) {
3463 		if (handle)
3464 			ext4_journal_stop(handle);
3465 		if (PTR_ERR(wh) == -ENOSPC &&
3466 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3467 			goto retry;
3468 	} else {
3469 		*h = handle;
3470 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3471 		wh->i_op = &ext4_special_inode_operations;
3472 	}
3473 	return wh;
3474 }
3475 
3476 /*
3477  * Anybody can rename anything with this: the permission checks are left to the
3478  * higher-level routines.
3479  *
3480  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3481  * while new_{dentry,inode) refers to the destination dentry/inode
3482  * This comes from rename(const char *oldpath, const char *newpath)
3483  */
3484 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3485 		       struct inode *new_dir, struct dentry *new_dentry,
3486 		       unsigned int flags)
3487 {
3488 	handle_t *handle = NULL;
3489 	struct ext4_renament old = {
3490 		.dir = old_dir,
3491 		.dentry = old_dentry,
3492 		.inode = d_inode(old_dentry),
3493 	};
3494 	struct ext4_renament new = {
3495 		.dir = new_dir,
3496 		.dentry = new_dentry,
3497 		.inode = d_inode(new_dentry),
3498 	};
3499 	int force_reread;
3500 	int retval;
3501 	struct inode *whiteout = NULL;
3502 	int credits;
3503 	u8 old_file_type;
3504 
3505 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3506 	    (!projid_eq(EXT4_I(new_dir)->i_projid,
3507 			EXT4_I(old_dentry->d_inode)->i_projid)))
3508 		return -EXDEV;
3509 
3510 	retval = dquot_initialize(old.dir);
3511 	if (retval)
3512 		return retval;
3513 	retval = dquot_initialize(new.dir);
3514 	if (retval)
3515 		return retval;
3516 
3517 	/* Initialize quotas before so that eventual writes go
3518 	 * in separate transaction */
3519 	if (new.inode) {
3520 		retval = dquot_initialize(new.inode);
3521 		if (retval)
3522 			return retval;
3523 	}
3524 
3525 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3526 	if (IS_ERR(old.bh))
3527 		return PTR_ERR(old.bh);
3528 	/*
3529 	 *  Check for inode number is _not_ due to possible IO errors.
3530 	 *  We might rmdir the source, keep it as pwd of some process
3531 	 *  and merrily kill the link to whatever was created under the
3532 	 *  same name. Goodbye sticky bit ;-<
3533 	 */
3534 	retval = -ENOENT;
3535 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3536 		goto end_rename;
3537 
3538 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3539 				 &new.de, &new.inlined);
3540 	if (IS_ERR(new.bh)) {
3541 		retval = PTR_ERR(new.bh);
3542 		new.bh = NULL;
3543 		goto end_rename;
3544 	}
3545 	if (new.bh) {
3546 		if (!new.inode) {
3547 			brelse(new.bh);
3548 			new.bh = NULL;
3549 		}
3550 	}
3551 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3552 		ext4_alloc_da_blocks(old.inode);
3553 
3554 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3555 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3556 	if (!(flags & RENAME_WHITEOUT)) {
3557 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3558 		if (IS_ERR(handle)) {
3559 			retval = PTR_ERR(handle);
3560 			handle = NULL;
3561 			goto end_rename;
3562 		}
3563 	} else {
3564 		whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3565 		if (IS_ERR(whiteout)) {
3566 			retval = PTR_ERR(whiteout);
3567 			whiteout = NULL;
3568 			goto end_rename;
3569 		}
3570 	}
3571 
3572 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3573 		ext4_handle_sync(handle);
3574 
3575 	if (S_ISDIR(old.inode->i_mode)) {
3576 		if (new.inode) {
3577 			retval = -ENOTEMPTY;
3578 			if (!ext4_empty_dir(new.inode))
3579 				goto end_rename;
3580 		} else {
3581 			retval = -EMLINK;
3582 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3583 				goto end_rename;
3584 		}
3585 		retval = ext4_rename_dir_prepare(handle, &old);
3586 		if (retval)
3587 			goto end_rename;
3588 	}
3589 	/*
3590 	 * If we're renaming a file within an inline_data dir and adding or
3591 	 * setting the new dirent causes a conversion from inline_data to
3592 	 * extents/blockmap, we need to force the dirent delete code to
3593 	 * re-read the directory, or else we end up trying to delete a dirent
3594 	 * from what is now the extent tree root (or a block map).
3595 	 */
3596 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3597 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3598 
3599 	old_file_type = old.de->file_type;
3600 	if (whiteout) {
3601 		/*
3602 		 * Do this before adding a new entry, so the old entry is sure
3603 		 * to be still pointing to the valid old entry.
3604 		 */
3605 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3606 				     EXT4_FT_CHRDEV);
3607 		if (retval)
3608 			goto end_rename;
3609 		ext4_mark_inode_dirty(handle, whiteout);
3610 	}
3611 	if (!new.bh) {
3612 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3613 		if (retval)
3614 			goto end_rename;
3615 	} else {
3616 		retval = ext4_setent(handle, &new,
3617 				     old.inode->i_ino, old_file_type);
3618 		if (retval)
3619 			goto end_rename;
3620 	}
3621 	if (force_reread)
3622 		force_reread = !ext4_test_inode_flag(new.dir,
3623 						     EXT4_INODE_INLINE_DATA);
3624 
3625 	/*
3626 	 * Like most other Unix systems, set the ctime for inodes on a
3627 	 * rename.
3628 	 */
3629 	old.inode->i_ctime = current_time(old.inode);
3630 	ext4_mark_inode_dirty(handle, old.inode);
3631 
3632 	if (!whiteout) {
3633 		/*
3634 		 * ok, that's it
3635 		 */
3636 		ext4_rename_delete(handle, &old, force_reread);
3637 	}
3638 
3639 	if (new.inode) {
3640 		ext4_dec_count(handle, new.inode);
3641 		new.inode->i_ctime = current_time(new.inode);
3642 	}
3643 	old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3644 	ext4_update_dx_flag(old.dir);
3645 	if (old.dir_bh) {
3646 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3647 		if (retval)
3648 			goto end_rename;
3649 
3650 		ext4_dec_count(handle, old.dir);
3651 		if (new.inode) {
3652 			/* checked ext4_empty_dir above, can't have another
3653 			 * parent, ext4_dec_count() won't work for many-linked
3654 			 * dirs */
3655 			clear_nlink(new.inode);
3656 		} else {
3657 			ext4_inc_count(handle, new.dir);
3658 			ext4_update_dx_flag(new.dir);
3659 			ext4_mark_inode_dirty(handle, new.dir);
3660 		}
3661 	}
3662 	ext4_mark_inode_dirty(handle, old.dir);
3663 	if (new.inode) {
3664 		ext4_mark_inode_dirty(handle, new.inode);
3665 		if (!new.inode->i_nlink)
3666 			ext4_orphan_add(handle, new.inode);
3667 	}
3668 	retval = 0;
3669 
3670 end_rename:
3671 	brelse(old.dir_bh);
3672 	brelse(old.bh);
3673 	brelse(new.bh);
3674 	if (whiteout) {
3675 		if (retval)
3676 			drop_nlink(whiteout);
3677 		unlock_new_inode(whiteout);
3678 		iput(whiteout);
3679 	}
3680 	if (handle)
3681 		ext4_journal_stop(handle);
3682 	return retval;
3683 }
3684 
3685 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3686 			     struct inode *new_dir, struct dentry *new_dentry)
3687 {
3688 	handle_t *handle = NULL;
3689 	struct ext4_renament old = {
3690 		.dir = old_dir,
3691 		.dentry = old_dentry,
3692 		.inode = d_inode(old_dentry),
3693 	};
3694 	struct ext4_renament new = {
3695 		.dir = new_dir,
3696 		.dentry = new_dentry,
3697 		.inode = d_inode(new_dentry),
3698 	};
3699 	u8 new_file_type;
3700 	int retval;
3701 	struct timespec ctime;
3702 
3703 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3704 	     !projid_eq(EXT4_I(new_dir)->i_projid,
3705 			EXT4_I(old_dentry->d_inode)->i_projid)) ||
3706 	    (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3707 	     !projid_eq(EXT4_I(old_dir)->i_projid,
3708 			EXT4_I(new_dentry->d_inode)->i_projid)))
3709 		return -EXDEV;
3710 
3711 	retval = dquot_initialize(old.dir);
3712 	if (retval)
3713 		return retval;
3714 	retval = dquot_initialize(new.dir);
3715 	if (retval)
3716 		return retval;
3717 
3718 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3719 				 &old.de, &old.inlined);
3720 	if (IS_ERR(old.bh))
3721 		return PTR_ERR(old.bh);
3722 	/*
3723 	 *  Check for inode number is _not_ due to possible IO errors.
3724 	 *  We might rmdir the source, keep it as pwd of some process
3725 	 *  and merrily kill the link to whatever was created under the
3726 	 *  same name. Goodbye sticky bit ;-<
3727 	 */
3728 	retval = -ENOENT;
3729 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3730 		goto end_rename;
3731 
3732 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3733 				 &new.de, &new.inlined);
3734 	if (IS_ERR(new.bh)) {
3735 		retval = PTR_ERR(new.bh);
3736 		new.bh = NULL;
3737 		goto end_rename;
3738 	}
3739 
3740 	/* RENAME_EXCHANGE case: old *and* new must both exist */
3741 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3742 		goto end_rename;
3743 
3744 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3745 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3746 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3747 	if (IS_ERR(handle)) {
3748 		retval = PTR_ERR(handle);
3749 		handle = NULL;
3750 		goto end_rename;
3751 	}
3752 
3753 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3754 		ext4_handle_sync(handle);
3755 
3756 	if (S_ISDIR(old.inode->i_mode)) {
3757 		old.is_dir = true;
3758 		retval = ext4_rename_dir_prepare(handle, &old);
3759 		if (retval)
3760 			goto end_rename;
3761 	}
3762 	if (S_ISDIR(new.inode->i_mode)) {
3763 		new.is_dir = true;
3764 		retval = ext4_rename_dir_prepare(handle, &new);
3765 		if (retval)
3766 			goto end_rename;
3767 	}
3768 
3769 	/*
3770 	 * Other than the special case of overwriting a directory, parents'
3771 	 * nlink only needs to be modified if this is a cross directory rename.
3772 	 */
3773 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
3774 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
3775 		new.dir_nlink_delta = -old.dir_nlink_delta;
3776 		retval = -EMLINK;
3777 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3778 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3779 			goto end_rename;
3780 	}
3781 
3782 	new_file_type = new.de->file_type;
3783 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3784 	if (retval)
3785 		goto end_rename;
3786 
3787 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3788 	if (retval)
3789 		goto end_rename;
3790 
3791 	/*
3792 	 * Like most other Unix systems, set the ctime for inodes on a
3793 	 * rename.
3794 	 */
3795 	ctime = current_time(old.inode);
3796 	old.inode->i_ctime = ctime;
3797 	new.inode->i_ctime = ctime;
3798 	ext4_mark_inode_dirty(handle, old.inode);
3799 	ext4_mark_inode_dirty(handle, new.inode);
3800 
3801 	if (old.dir_bh) {
3802 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3803 		if (retval)
3804 			goto end_rename;
3805 	}
3806 	if (new.dir_bh) {
3807 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3808 		if (retval)
3809 			goto end_rename;
3810 	}
3811 	ext4_update_dir_count(handle, &old);
3812 	ext4_update_dir_count(handle, &new);
3813 	retval = 0;
3814 
3815 end_rename:
3816 	brelse(old.dir_bh);
3817 	brelse(new.dir_bh);
3818 	brelse(old.bh);
3819 	brelse(new.bh);
3820 	if (handle)
3821 		ext4_journal_stop(handle);
3822 	return retval;
3823 }
3824 
3825 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3826 			struct inode *new_dir, struct dentry *new_dentry,
3827 			unsigned int flags)
3828 {
3829 	int err;
3830 
3831 	if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3832 		return -EIO;
3833 
3834 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3835 		return -EINVAL;
3836 
3837 	err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
3838 				     flags);
3839 	if (err)
3840 		return err;
3841 
3842 	if (flags & RENAME_EXCHANGE) {
3843 		return ext4_cross_rename(old_dir, old_dentry,
3844 					 new_dir, new_dentry);
3845 	}
3846 
3847 	return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3848 }
3849 
3850 /*
3851  * directories can handle most operations...
3852  */
3853 const struct inode_operations ext4_dir_inode_operations = {
3854 	.create		= ext4_create,
3855 	.lookup		= ext4_lookup,
3856 	.link		= ext4_link,
3857 	.unlink		= ext4_unlink,
3858 	.symlink	= ext4_symlink,
3859 	.mkdir		= ext4_mkdir,
3860 	.rmdir		= ext4_rmdir,
3861 	.mknod		= ext4_mknod,
3862 	.tmpfile	= ext4_tmpfile,
3863 	.rename		= ext4_rename2,
3864 	.setattr	= ext4_setattr,
3865 	.getattr	= ext4_getattr,
3866 	.listxattr	= ext4_listxattr,
3867 	.get_acl	= ext4_get_acl,
3868 	.set_acl	= ext4_set_acl,
3869 	.fiemap         = ext4_fiemap,
3870 };
3871 
3872 const struct inode_operations ext4_special_inode_operations = {
3873 	.setattr	= ext4_setattr,
3874 	.getattr	= ext4_getattr,
3875 	.listxattr	= ext4_listxattr,
3876 	.get_acl	= ext4_get_acl,
3877 	.set_acl	= ext4_set_acl,
3878 };
3879