xref: /openbmc/linux/fs/ext4/namei.c (revision 6dfcd296)
1 /*
2  *  linux/fs/ext4/namei.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/namei.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  *  Directory entry file type support and forward compatibility hooks
18  *	for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19  *  Hash Tree Directory indexing (c)
20  *	Daniel Phillips, 2001
21  *  Hash Tree Directory indexing porting
22  *	Christopher Li, 2002
23  *  Hash Tree Directory indexing cleanup
24  *	Theodore Ts'o, 2002
25  */
26 
27 #include <linux/fs.h>
28 #include <linux/pagemap.h>
29 #include <linux/time.h>
30 #include <linux/fcntl.h>
31 #include <linux/stat.h>
32 #include <linux/string.h>
33 #include <linux/quotaops.h>
34 #include <linux/buffer_head.h>
35 #include <linux/bio.h>
36 #include "ext4.h"
37 #include "ext4_jbd2.h"
38 
39 #include "xattr.h"
40 #include "acl.h"
41 
42 #include <trace/events/ext4.h>
43 /*
44  * define how far ahead to read directories while searching them.
45  */
46 #define NAMEI_RA_CHUNKS  2
47 #define NAMEI_RA_BLOCKS  4
48 #define NAMEI_RA_SIZE	     (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
49 
50 static struct buffer_head *ext4_append(handle_t *handle,
51 					struct inode *inode,
52 					ext4_lblk_t *block)
53 {
54 	struct buffer_head *bh;
55 	int err;
56 
57 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
58 		     ((inode->i_size >> 10) >=
59 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
60 		return ERR_PTR(-ENOSPC);
61 
62 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
63 
64 	bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
65 	if (IS_ERR(bh))
66 		return bh;
67 	inode->i_size += inode->i_sb->s_blocksize;
68 	EXT4_I(inode)->i_disksize = inode->i_size;
69 	BUFFER_TRACE(bh, "get_write_access");
70 	err = ext4_journal_get_write_access(handle, bh);
71 	if (err) {
72 		brelse(bh);
73 		ext4_std_error(inode->i_sb, err);
74 		return ERR_PTR(err);
75 	}
76 	return bh;
77 }
78 
79 static int ext4_dx_csum_verify(struct inode *inode,
80 			       struct ext4_dir_entry *dirent);
81 
82 typedef enum {
83 	EITHER, INDEX, DIRENT
84 } dirblock_type_t;
85 
86 #define ext4_read_dirblock(inode, block, type) \
87 	__ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
88 
89 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
90 						ext4_lblk_t block,
91 						dirblock_type_t type,
92 						const char *func,
93 						unsigned int line)
94 {
95 	struct buffer_head *bh;
96 	struct ext4_dir_entry *dirent;
97 	int is_dx_block = 0;
98 
99 	bh = ext4_bread(NULL, inode, block, 0);
100 	if (IS_ERR(bh)) {
101 		__ext4_warning(inode->i_sb, func, line,
102 			       "inode #%lu: lblock %lu: comm %s: "
103 			       "error %ld reading directory block",
104 			       inode->i_ino, (unsigned long)block,
105 			       current->comm, PTR_ERR(bh));
106 
107 		return bh;
108 	}
109 	if (!bh) {
110 		ext4_error_inode(inode, func, line, block,
111 				 "Directory hole found");
112 		return ERR_PTR(-EFSCORRUPTED);
113 	}
114 	dirent = (struct ext4_dir_entry *) bh->b_data;
115 	/* Determine whether or not we have an index block */
116 	if (is_dx(inode)) {
117 		if (block == 0)
118 			is_dx_block = 1;
119 		else if (ext4_rec_len_from_disk(dirent->rec_len,
120 						inode->i_sb->s_blocksize) ==
121 			 inode->i_sb->s_blocksize)
122 			is_dx_block = 1;
123 	}
124 	if (!is_dx_block && type == INDEX) {
125 		ext4_error_inode(inode, func, line, block,
126 		       "directory leaf block found instead of index block");
127 		return ERR_PTR(-EFSCORRUPTED);
128 	}
129 	if (!ext4_has_metadata_csum(inode->i_sb) ||
130 	    buffer_verified(bh))
131 		return bh;
132 
133 	/*
134 	 * An empty leaf block can get mistaken for a index block; for
135 	 * this reason, we can only check the index checksum when the
136 	 * caller is sure it should be an index block.
137 	 */
138 	if (is_dx_block && type == INDEX) {
139 		if (ext4_dx_csum_verify(inode, dirent))
140 			set_buffer_verified(bh);
141 		else {
142 			ext4_error_inode(inode, func, line, block,
143 					 "Directory index failed checksum");
144 			brelse(bh);
145 			return ERR_PTR(-EFSBADCRC);
146 		}
147 	}
148 	if (!is_dx_block) {
149 		if (ext4_dirent_csum_verify(inode, dirent))
150 			set_buffer_verified(bh);
151 		else {
152 			ext4_error_inode(inode, func, line, block,
153 					 "Directory block failed checksum");
154 			brelse(bh);
155 			return ERR_PTR(-EFSBADCRC);
156 		}
157 	}
158 	return bh;
159 }
160 
161 #ifndef assert
162 #define assert(test) J_ASSERT(test)
163 #endif
164 
165 #ifdef DX_DEBUG
166 #define dxtrace(command) command
167 #else
168 #define dxtrace(command)
169 #endif
170 
171 struct fake_dirent
172 {
173 	__le32 inode;
174 	__le16 rec_len;
175 	u8 name_len;
176 	u8 file_type;
177 };
178 
179 struct dx_countlimit
180 {
181 	__le16 limit;
182 	__le16 count;
183 };
184 
185 struct dx_entry
186 {
187 	__le32 hash;
188 	__le32 block;
189 };
190 
191 /*
192  * dx_root_info is laid out so that if it should somehow get overlaid by a
193  * dirent the two low bits of the hash version will be zero.  Therefore, the
194  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
195  */
196 
197 struct dx_root
198 {
199 	struct fake_dirent dot;
200 	char dot_name[4];
201 	struct fake_dirent dotdot;
202 	char dotdot_name[4];
203 	struct dx_root_info
204 	{
205 		__le32 reserved_zero;
206 		u8 hash_version;
207 		u8 info_length; /* 8 */
208 		u8 indirect_levels;
209 		u8 unused_flags;
210 	}
211 	info;
212 	struct dx_entry	entries[0];
213 };
214 
215 struct dx_node
216 {
217 	struct fake_dirent fake;
218 	struct dx_entry	entries[0];
219 };
220 
221 
222 struct dx_frame
223 {
224 	struct buffer_head *bh;
225 	struct dx_entry *entries;
226 	struct dx_entry *at;
227 };
228 
229 struct dx_map_entry
230 {
231 	u32 hash;
232 	u16 offs;
233 	u16 size;
234 };
235 
236 /*
237  * This goes at the end of each htree block.
238  */
239 struct dx_tail {
240 	u32 dt_reserved;
241 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
242 };
243 
244 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
245 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
246 static inline unsigned dx_get_hash(struct dx_entry *entry);
247 static void dx_set_hash(struct dx_entry *entry, unsigned value);
248 static unsigned dx_get_count(struct dx_entry *entries);
249 static unsigned dx_get_limit(struct dx_entry *entries);
250 static void dx_set_count(struct dx_entry *entries, unsigned value);
251 static void dx_set_limit(struct dx_entry *entries, unsigned value);
252 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
253 static unsigned dx_node_limit(struct inode *dir);
254 static struct dx_frame *dx_probe(struct ext4_filename *fname,
255 				 struct inode *dir,
256 				 struct dx_hash_info *hinfo,
257 				 struct dx_frame *frame);
258 static void dx_release(struct dx_frame *frames);
259 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
260 		       unsigned blocksize, struct dx_hash_info *hinfo,
261 		       struct dx_map_entry map[]);
262 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
263 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
264 		struct dx_map_entry *offsets, int count, unsigned blocksize);
265 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
266 static void dx_insert_block(struct dx_frame *frame,
267 					u32 hash, ext4_lblk_t block);
268 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
269 				 struct dx_frame *frame,
270 				 struct dx_frame *frames,
271 				 __u32 *start_hash);
272 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
273 		struct ext4_filename *fname,
274 		struct ext4_dir_entry_2 **res_dir);
275 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
276 			     struct inode *dir, struct inode *inode);
277 
278 /* checksumming functions */
279 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
280 			    unsigned int blocksize)
281 {
282 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
283 	t->det_rec_len = ext4_rec_len_to_disk(
284 			sizeof(struct ext4_dir_entry_tail), blocksize);
285 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
286 }
287 
288 /* Walk through a dirent block to find a checksum "dirent" at the tail */
289 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
290 						   struct ext4_dir_entry *de)
291 {
292 	struct ext4_dir_entry_tail *t;
293 
294 #ifdef PARANOID
295 	struct ext4_dir_entry *d, *top;
296 
297 	d = de;
298 	top = (struct ext4_dir_entry *)(((void *)de) +
299 		(EXT4_BLOCK_SIZE(inode->i_sb) -
300 		sizeof(struct ext4_dir_entry_tail)));
301 	while (d < top && d->rec_len)
302 		d = (struct ext4_dir_entry *)(((void *)d) +
303 		    le16_to_cpu(d->rec_len));
304 
305 	if (d != top)
306 		return NULL;
307 
308 	t = (struct ext4_dir_entry_tail *)d;
309 #else
310 	t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
311 #endif
312 
313 	if (t->det_reserved_zero1 ||
314 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
315 	    t->det_reserved_zero2 ||
316 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
317 		return NULL;
318 
319 	return t;
320 }
321 
322 static __le32 ext4_dirent_csum(struct inode *inode,
323 			       struct ext4_dir_entry *dirent, int size)
324 {
325 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
326 	struct ext4_inode_info *ei = EXT4_I(inode);
327 	__u32 csum;
328 
329 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
330 	return cpu_to_le32(csum);
331 }
332 
333 #define warn_no_space_for_csum(inode)					\
334 	__warn_no_space_for_csum((inode), __func__, __LINE__)
335 
336 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
337 				     unsigned int line)
338 {
339 	__ext4_warning_inode(inode, func, line,
340 		"No space for directory leaf checksum. Please run e2fsck -D.");
341 }
342 
343 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
344 {
345 	struct ext4_dir_entry_tail *t;
346 
347 	if (!ext4_has_metadata_csum(inode->i_sb))
348 		return 1;
349 
350 	t = get_dirent_tail(inode, dirent);
351 	if (!t) {
352 		warn_no_space_for_csum(inode);
353 		return 0;
354 	}
355 
356 	if (t->det_checksum != ext4_dirent_csum(inode, dirent,
357 						(void *)t - (void *)dirent))
358 		return 0;
359 
360 	return 1;
361 }
362 
363 static void ext4_dirent_csum_set(struct inode *inode,
364 				 struct ext4_dir_entry *dirent)
365 {
366 	struct ext4_dir_entry_tail *t;
367 
368 	if (!ext4_has_metadata_csum(inode->i_sb))
369 		return;
370 
371 	t = get_dirent_tail(inode, dirent);
372 	if (!t) {
373 		warn_no_space_for_csum(inode);
374 		return;
375 	}
376 
377 	t->det_checksum = ext4_dirent_csum(inode, dirent,
378 					   (void *)t - (void *)dirent);
379 }
380 
381 int ext4_handle_dirty_dirent_node(handle_t *handle,
382 				  struct inode *inode,
383 				  struct buffer_head *bh)
384 {
385 	ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
386 	return ext4_handle_dirty_metadata(handle, inode, bh);
387 }
388 
389 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
390 					       struct ext4_dir_entry *dirent,
391 					       int *offset)
392 {
393 	struct ext4_dir_entry *dp;
394 	struct dx_root_info *root;
395 	int count_offset;
396 
397 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
398 		count_offset = 8;
399 	else if (le16_to_cpu(dirent->rec_len) == 12) {
400 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
401 		if (le16_to_cpu(dp->rec_len) !=
402 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
403 			return NULL;
404 		root = (struct dx_root_info *)(((void *)dp + 12));
405 		if (root->reserved_zero ||
406 		    root->info_length != sizeof(struct dx_root_info))
407 			return NULL;
408 		count_offset = 32;
409 	} else
410 		return NULL;
411 
412 	if (offset)
413 		*offset = count_offset;
414 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
415 }
416 
417 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
418 			   int count_offset, int count, struct dx_tail *t)
419 {
420 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
421 	struct ext4_inode_info *ei = EXT4_I(inode);
422 	__u32 csum;
423 	int size;
424 	__u32 dummy_csum = 0;
425 	int offset = offsetof(struct dx_tail, dt_checksum);
426 
427 	size = count_offset + (count * sizeof(struct dx_entry));
428 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
429 	csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
430 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
431 
432 	return cpu_to_le32(csum);
433 }
434 
435 static int ext4_dx_csum_verify(struct inode *inode,
436 			       struct ext4_dir_entry *dirent)
437 {
438 	struct dx_countlimit *c;
439 	struct dx_tail *t;
440 	int count_offset, limit, count;
441 
442 	if (!ext4_has_metadata_csum(inode->i_sb))
443 		return 1;
444 
445 	c = get_dx_countlimit(inode, dirent, &count_offset);
446 	if (!c) {
447 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
448 		return 0;
449 	}
450 	limit = le16_to_cpu(c->limit);
451 	count = le16_to_cpu(c->count);
452 	if (count_offset + (limit * sizeof(struct dx_entry)) >
453 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
454 		warn_no_space_for_csum(inode);
455 		return 0;
456 	}
457 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
458 
459 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
460 					    count, t))
461 		return 0;
462 	return 1;
463 }
464 
465 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
466 {
467 	struct dx_countlimit *c;
468 	struct dx_tail *t;
469 	int count_offset, limit, count;
470 
471 	if (!ext4_has_metadata_csum(inode->i_sb))
472 		return;
473 
474 	c = get_dx_countlimit(inode, dirent, &count_offset);
475 	if (!c) {
476 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
477 		return;
478 	}
479 	limit = le16_to_cpu(c->limit);
480 	count = le16_to_cpu(c->count);
481 	if (count_offset + (limit * sizeof(struct dx_entry)) >
482 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
483 		warn_no_space_for_csum(inode);
484 		return;
485 	}
486 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
487 
488 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
489 }
490 
491 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
492 					    struct inode *inode,
493 					    struct buffer_head *bh)
494 {
495 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
496 	return ext4_handle_dirty_metadata(handle, inode, bh);
497 }
498 
499 /*
500  * p is at least 6 bytes before the end of page
501  */
502 static inline struct ext4_dir_entry_2 *
503 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
504 {
505 	return (struct ext4_dir_entry_2 *)((char *)p +
506 		ext4_rec_len_from_disk(p->rec_len, blocksize));
507 }
508 
509 /*
510  * Future: use high four bits of block for coalesce-on-delete flags
511  * Mask them off for now.
512  */
513 
514 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
515 {
516 	return le32_to_cpu(entry->block) & 0x00ffffff;
517 }
518 
519 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
520 {
521 	entry->block = cpu_to_le32(value);
522 }
523 
524 static inline unsigned dx_get_hash(struct dx_entry *entry)
525 {
526 	return le32_to_cpu(entry->hash);
527 }
528 
529 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
530 {
531 	entry->hash = cpu_to_le32(value);
532 }
533 
534 static inline unsigned dx_get_count(struct dx_entry *entries)
535 {
536 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
537 }
538 
539 static inline unsigned dx_get_limit(struct dx_entry *entries)
540 {
541 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
542 }
543 
544 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
545 {
546 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
547 }
548 
549 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
550 {
551 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
552 }
553 
554 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
555 {
556 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
557 		EXT4_DIR_REC_LEN(2) - infosize;
558 
559 	if (ext4_has_metadata_csum(dir->i_sb))
560 		entry_space -= sizeof(struct dx_tail);
561 	return entry_space / sizeof(struct dx_entry);
562 }
563 
564 static inline unsigned dx_node_limit(struct inode *dir)
565 {
566 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
567 
568 	if (ext4_has_metadata_csum(dir->i_sb))
569 		entry_space -= sizeof(struct dx_tail);
570 	return entry_space / sizeof(struct dx_entry);
571 }
572 
573 /*
574  * Debug
575  */
576 #ifdef DX_DEBUG
577 static void dx_show_index(char * label, struct dx_entry *entries)
578 {
579 	int i, n = dx_get_count (entries);
580 	printk(KERN_DEBUG "%s index ", label);
581 	for (i = 0; i < n; i++) {
582 		printk("%x->%lu ", i ? dx_get_hash(entries + i) :
583 				0, (unsigned long)dx_get_block(entries + i));
584 	}
585 	printk("\n");
586 }
587 
588 struct stats
589 {
590 	unsigned names;
591 	unsigned space;
592 	unsigned bcount;
593 };
594 
595 static struct stats dx_show_leaf(struct inode *dir,
596 				struct dx_hash_info *hinfo,
597 				struct ext4_dir_entry_2 *de,
598 				int size, int show_names)
599 {
600 	unsigned names = 0, space = 0;
601 	char *base = (char *) de;
602 	struct dx_hash_info h = *hinfo;
603 
604 	printk("names: ");
605 	while ((char *) de < base + size)
606 	{
607 		if (de->inode)
608 		{
609 			if (show_names)
610 			{
611 #ifdef CONFIG_EXT4_FS_ENCRYPTION
612 				int len;
613 				char *name;
614 				struct fscrypt_str fname_crypto_str =
615 					FSTR_INIT(NULL, 0);
616 				int res = 0;
617 
618 				name  = de->name;
619 				len = de->name_len;
620 				if (ext4_encrypted_inode(dir))
621 					res = fscrypt_get_encryption_info(dir);
622 				if (res) {
623 					printk(KERN_WARNING "Error setting up"
624 					       " fname crypto: %d\n", res);
625 				}
626 				if (!fscrypt_has_encryption_key(dir)) {
627 					/* Directory is not encrypted */
628 					ext4fs_dirhash(de->name,
629 						de->name_len, &h);
630 					printk("%*.s:(U)%x.%u ", len,
631 					       name, h.hash,
632 					       (unsigned) ((char *) de
633 							   - base));
634 				} else {
635 					struct fscrypt_str de_name =
636 						FSTR_INIT(name, len);
637 
638 					/* Directory is encrypted */
639 					res = fscrypt_fname_alloc_buffer(
640 						dir, len,
641 						&fname_crypto_str);
642 					if (res)
643 						printk(KERN_WARNING "Error "
644 							"allocating crypto "
645 							"buffer--skipping "
646 							"crypto\n");
647 					res = fscrypt_fname_disk_to_usr(dir,
648 						0, 0, &de_name,
649 						&fname_crypto_str);
650 					if (res) {
651 						printk(KERN_WARNING "Error "
652 							"converting filename "
653 							"from disk to usr"
654 							"\n");
655 						name = "??";
656 						len = 2;
657 					} else {
658 						name = fname_crypto_str.name;
659 						len = fname_crypto_str.len;
660 					}
661 					ext4fs_dirhash(de->name, de->name_len,
662 						       &h);
663 					printk("%*.s:(E)%x.%u ", len, name,
664 					       h.hash, (unsigned) ((char *) de
665 								   - base));
666 					fscrypt_fname_free_buffer(
667 							&fname_crypto_str);
668 				}
669 #else
670 				int len = de->name_len;
671 				char *name = de->name;
672 				ext4fs_dirhash(de->name, de->name_len, &h);
673 				printk("%*.s:%x.%u ", len, name, h.hash,
674 				       (unsigned) ((char *) de - base));
675 #endif
676 			}
677 			space += EXT4_DIR_REC_LEN(de->name_len);
678 			names++;
679 		}
680 		de = ext4_next_entry(de, size);
681 	}
682 	printk("(%i)\n", names);
683 	return (struct stats) { names, space, 1 };
684 }
685 
686 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
687 			     struct dx_entry *entries, int levels)
688 {
689 	unsigned blocksize = dir->i_sb->s_blocksize;
690 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
691 	unsigned bcount = 0;
692 	struct buffer_head *bh;
693 	printk("%i indexed blocks...\n", count);
694 	for (i = 0; i < count; i++, entries++)
695 	{
696 		ext4_lblk_t block = dx_get_block(entries);
697 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
698 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
699 		struct stats stats;
700 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
701 		bh = ext4_bread(NULL,dir, block, 0);
702 		if (!bh || IS_ERR(bh))
703 			continue;
704 		stats = levels?
705 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
706 		   dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
707 			bh->b_data, blocksize, 0);
708 		names += stats.names;
709 		space += stats.space;
710 		bcount += stats.bcount;
711 		brelse(bh);
712 	}
713 	if (bcount)
714 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
715 		       levels ? "" : "   ", names, space/bcount,
716 		       (space/bcount)*100/blocksize);
717 	return (struct stats) { names, space, bcount};
718 }
719 #endif /* DX_DEBUG */
720 
721 /*
722  * Probe for a directory leaf block to search.
723  *
724  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
725  * error in the directory index, and the caller should fall back to
726  * searching the directory normally.  The callers of dx_probe **MUST**
727  * check for this error code, and make sure it never gets reflected
728  * back to userspace.
729  */
730 static struct dx_frame *
731 dx_probe(struct ext4_filename *fname, struct inode *dir,
732 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
733 {
734 	unsigned count, indirect;
735 	struct dx_entry *at, *entries, *p, *q, *m;
736 	struct dx_root *root;
737 	struct dx_frame *frame = frame_in;
738 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
739 	u32 hash;
740 
741 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
742 	if (IS_ERR(frame->bh))
743 		return (struct dx_frame *) frame->bh;
744 
745 	root = (struct dx_root *) frame->bh->b_data;
746 	if (root->info.hash_version != DX_HASH_TEA &&
747 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
748 	    root->info.hash_version != DX_HASH_LEGACY) {
749 		ext4_warning_inode(dir, "Unrecognised inode hash code %u",
750 				   root->info.hash_version);
751 		goto fail;
752 	}
753 	if (fname)
754 		hinfo = &fname->hinfo;
755 	hinfo->hash_version = root->info.hash_version;
756 	if (hinfo->hash_version <= DX_HASH_TEA)
757 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
758 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
759 	if (fname && fname_name(fname))
760 		ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
761 	hash = hinfo->hash;
762 
763 	if (root->info.unused_flags & 1) {
764 		ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
765 				   root->info.unused_flags);
766 		goto fail;
767 	}
768 
769 	indirect = root->info.indirect_levels;
770 	if (indirect > 1) {
771 		ext4_warning_inode(dir, "Unimplemented hash depth: %#06x",
772 				   root->info.indirect_levels);
773 		goto fail;
774 	}
775 
776 	entries = (struct dx_entry *)(((char *)&root->info) +
777 				      root->info.info_length);
778 
779 	if (dx_get_limit(entries) != dx_root_limit(dir,
780 						   root->info.info_length)) {
781 		ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
782 				   dx_get_limit(entries),
783 				   dx_root_limit(dir, root->info.info_length));
784 		goto fail;
785 	}
786 
787 	dxtrace(printk("Look up %x", hash));
788 	while (1) {
789 		count = dx_get_count(entries);
790 		if (!count || count > dx_get_limit(entries)) {
791 			ext4_warning_inode(dir,
792 					   "dx entry: count %u beyond limit %u",
793 					   count, dx_get_limit(entries));
794 			goto fail;
795 		}
796 
797 		p = entries + 1;
798 		q = entries + count - 1;
799 		while (p <= q) {
800 			m = p + (q - p) / 2;
801 			dxtrace(printk("."));
802 			if (dx_get_hash(m) > hash)
803 				q = m - 1;
804 			else
805 				p = m + 1;
806 		}
807 
808 		if (0) { // linear search cross check
809 			unsigned n = count - 1;
810 			at = entries;
811 			while (n--)
812 			{
813 				dxtrace(printk(","));
814 				if (dx_get_hash(++at) > hash)
815 				{
816 					at--;
817 					break;
818 				}
819 			}
820 			assert (at == p - 1);
821 		}
822 
823 		at = p - 1;
824 		dxtrace(printk(" %x->%u\n", at == entries ? 0 : dx_get_hash(at),
825 			       dx_get_block(at)));
826 		frame->entries = entries;
827 		frame->at = at;
828 		if (!indirect--)
829 			return frame;
830 		frame++;
831 		frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
832 		if (IS_ERR(frame->bh)) {
833 			ret_err = (struct dx_frame *) frame->bh;
834 			frame->bh = NULL;
835 			goto fail;
836 		}
837 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
838 
839 		if (dx_get_limit(entries) != dx_node_limit(dir)) {
840 			ext4_warning_inode(dir,
841 				"dx entry: limit %u != node limit %u",
842 				dx_get_limit(entries), dx_node_limit(dir));
843 			goto fail;
844 		}
845 	}
846 fail:
847 	while (frame >= frame_in) {
848 		brelse(frame->bh);
849 		frame--;
850 	}
851 
852 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
853 		ext4_warning_inode(dir,
854 			"Corrupt directory, running e2fsck is recommended");
855 	return ret_err;
856 }
857 
858 static void dx_release(struct dx_frame *frames)
859 {
860 	if (frames[0].bh == NULL)
861 		return;
862 
863 	if (((struct dx_root *)frames[0].bh->b_data)->info.indirect_levels)
864 		brelse(frames[1].bh);
865 	brelse(frames[0].bh);
866 }
867 
868 /*
869  * This function increments the frame pointer to search the next leaf
870  * block, and reads in the necessary intervening nodes if the search
871  * should be necessary.  Whether or not the search is necessary is
872  * controlled by the hash parameter.  If the hash value is even, then
873  * the search is only continued if the next block starts with that
874  * hash value.  This is used if we are searching for a specific file.
875  *
876  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
877  *
878  * This function returns 1 if the caller should continue to search,
879  * or 0 if it should not.  If there is an error reading one of the
880  * index blocks, it will a negative error code.
881  *
882  * If start_hash is non-null, it will be filled in with the starting
883  * hash of the next page.
884  */
885 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
886 				 struct dx_frame *frame,
887 				 struct dx_frame *frames,
888 				 __u32 *start_hash)
889 {
890 	struct dx_frame *p;
891 	struct buffer_head *bh;
892 	int num_frames = 0;
893 	__u32 bhash;
894 
895 	p = frame;
896 	/*
897 	 * Find the next leaf page by incrementing the frame pointer.
898 	 * If we run out of entries in the interior node, loop around and
899 	 * increment pointer in the parent node.  When we break out of
900 	 * this loop, num_frames indicates the number of interior
901 	 * nodes need to be read.
902 	 */
903 	while (1) {
904 		if (++(p->at) < p->entries + dx_get_count(p->entries))
905 			break;
906 		if (p == frames)
907 			return 0;
908 		num_frames++;
909 		p--;
910 	}
911 
912 	/*
913 	 * If the hash is 1, then continue only if the next page has a
914 	 * continuation hash of any value.  This is used for readdir
915 	 * handling.  Otherwise, check to see if the hash matches the
916 	 * desired contiuation hash.  If it doesn't, return since
917 	 * there's no point to read in the successive index pages.
918 	 */
919 	bhash = dx_get_hash(p->at);
920 	if (start_hash)
921 		*start_hash = bhash;
922 	if ((hash & 1) == 0) {
923 		if ((bhash & ~1) != hash)
924 			return 0;
925 	}
926 	/*
927 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
928 	 * block so no check is necessary
929 	 */
930 	while (num_frames--) {
931 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
932 		if (IS_ERR(bh))
933 			return PTR_ERR(bh);
934 		p++;
935 		brelse(p->bh);
936 		p->bh = bh;
937 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
938 	}
939 	return 1;
940 }
941 
942 
943 /*
944  * This function fills a red-black tree with information from a
945  * directory block.  It returns the number directory entries loaded
946  * into the tree.  If there is an error it is returned in err.
947  */
948 static int htree_dirblock_to_tree(struct file *dir_file,
949 				  struct inode *dir, ext4_lblk_t block,
950 				  struct dx_hash_info *hinfo,
951 				  __u32 start_hash, __u32 start_minor_hash)
952 {
953 	struct buffer_head *bh;
954 	struct ext4_dir_entry_2 *de, *top;
955 	int err = 0, count = 0;
956 	struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
957 
958 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
959 							(unsigned long)block));
960 	bh = ext4_read_dirblock(dir, block, DIRENT);
961 	if (IS_ERR(bh))
962 		return PTR_ERR(bh);
963 
964 	de = (struct ext4_dir_entry_2 *) bh->b_data;
965 	top = (struct ext4_dir_entry_2 *) ((char *) de +
966 					   dir->i_sb->s_blocksize -
967 					   EXT4_DIR_REC_LEN(0));
968 #ifdef CONFIG_EXT4_FS_ENCRYPTION
969 	/* Check if the directory is encrypted */
970 	if (ext4_encrypted_inode(dir)) {
971 		err = fscrypt_get_encryption_info(dir);
972 		if (err < 0) {
973 			brelse(bh);
974 			return err;
975 		}
976 		err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
977 						     &fname_crypto_str);
978 		if (err < 0) {
979 			brelse(bh);
980 			return err;
981 		}
982 	}
983 #endif
984 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
985 		if (ext4_check_dir_entry(dir, NULL, de, bh,
986 				bh->b_data, bh->b_size,
987 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
988 					 + ((char *)de - bh->b_data))) {
989 			/* silently ignore the rest of the block */
990 			break;
991 		}
992 		ext4fs_dirhash(de->name, de->name_len, hinfo);
993 		if ((hinfo->hash < start_hash) ||
994 		    ((hinfo->hash == start_hash) &&
995 		     (hinfo->minor_hash < start_minor_hash)))
996 			continue;
997 		if (de->inode == 0)
998 			continue;
999 		if (!ext4_encrypted_inode(dir)) {
1000 			tmp_str.name = de->name;
1001 			tmp_str.len = de->name_len;
1002 			err = ext4_htree_store_dirent(dir_file,
1003 				   hinfo->hash, hinfo->minor_hash, de,
1004 				   &tmp_str);
1005 		} else {
1006 			int save_len = fname_crypto_str.len;
1007 			struct fscrypt_str de_name = FSTR_INIT(de->name,
1008 								de->name_len);
1009 
1010 			/* Directory is encrypted */
1011 			err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1012 					hinfo->minor_hash, &de_name,
1013 					&fname_crypto_str);
1014 			if (err) {
1015 				count = err;
1016 				goto errout;
1017 			}
1018 			err = ext4_htree_store_dirent(dir_file,
1019 				   hinfo->hash, hinfo->minor_hash, de,
1020 					&fname_crypto_str);
1021 			fname_crypto_str.len = save_len;
1022 		}
1023 		if (err != 0) {
1024 			count = err;
1025 			goto errout;
1026 		}
1027 		count++;
1028 	}
1029 errout:
1030 	brelse(bh);
1031 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1032 	fscrypt_fname_free_buffer(&fname_crypto_str);
1033 #endif
1034 	return count;
1035 }
1036 
1037 
1038 /*
1039  * This function fills a red-black tree with information from a
1040  * directory.  We start scanning the directory in hash order, starting
1041  * at start_hash and start_minor_hash.
1042  *
1043  * This function returns the number of entries inserted into the tree,
1044  * or a negative error code.
1045  */
1046 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1047 			 __u32 start_minor_hash, __u32 *next_hash)
1048 {
1049 	struct dx_hash_info hinfo;
1050 	struct ext4_dir_entry_2 *de;
1051 	struct dx_frame frames[2], *frame;
1052 	struct inode *dir;
1053 	ext4_lblk_t block;
1054 	int count = 0;
1055 	int ret, err;
1056 	__u32 hashval;
1057 	struct fscrypt_str tmp_str;
1058 
1059 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1060 		       start_hash, start_minor_hash));
1061 	dir = file_inode(dir_file);
1062 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1063 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1064 		if (hinfo.hash_version <= DX_HASH_TEA)
1065 			hinfo.hash_version +=
1066 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
1067 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1068 		if (ext4_has_inline_data(dir)) {
1069 			int has_inline_data = 1;
1070 			count = htree_inlinedir_to_tree(dir_file, dir, 0,
1071 							&hinfo, start_hash,
1072 							start_minor_hash,
1073 							&has_inline_data);
1074 			if (has_inline_data) {
1075 				*next_hash = ~0;
1076 				return count;
1077 			}
1078 		}
1079 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1080 					       start_hash, start_minor_hash);
1081 		*next_hash = ~0;
1082 		return count;
1083 	}
1084 	hinfo.hash = start_hash;
1085 	hinfo.minor_hash = 0;
1086 	frame = dx_probe(NULL, dir, &hinfo, frames);
1087 	if (IS_ERR(frame))
1088 		return PTR_ERR(frame);
1089 
1090 	/* Add '.' and '..' from the htree header */
1091 	if (!start_hash && !start_minor_hash) {
1092 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1093 		tmp_str.name = de->name;
1094 		tmp_str.len = de->name_len;
1095 		err = ext4_htree_store_dirent(dir_file, 0, 0,
1096 					      de, &tmp_str);
1097 		if (err != 0)
1098 			goto errout;
1099 		count++;
1100 	}
1101 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1102 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1103 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1104 		tmp_str.name = de->name;
1105 		tmp_str.len = de->name_len;
1106 		err = ext4_htree_store_dirent(dir_file, 2, 0,
1107 					      de, &tmp_str);
1108 		if (err != 0)
1109 			goto errout;
1110 		count++;
1111 	}
1112 
1113 	while (1) {
1114 		if (fatal_signal_pending(current)) {
1115 			err = -ERESTARTSYS;
1116 			goto errout;
1117 		}
1118 		cond_resched();
1119 		block = dx_get_block(frame->at);
1120 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1121 					     start_hash, start_minor_hash);
1122 		if (ret < 0) {
1123 			err = ret;
1124 			goto errout;
1125 		}
1126 		count += ret;
1127 		hashval = ~0;
1128 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1129 					    frame, frames, &hashval);
1130 		*next_hash = hashval;
1131 		if (ret < 0) {
1132 			err = ret;
1133 			goto errout;
1134 		}
1135 		/*
1136 		 * Stop if:  (a) there are no more entries, or
1137 		 * (b) we have inserted at least one entry and the
1138 		 * next hash value is not a continuation
1139 		 */
1140 		if ((ret == 0) ||
1141 		    (count && ((hashval & 1) == 0)))
1142 			break;
1143 	}
1144 	dx_release(frames);
1145 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1146 		       "next hash: %x\n", count, *next_hash));
1147 	return count;
1148 errout:
1149 	dx_release(frames);
1150 	return (err);
1151 }
1152 
1153 static inline int search_dirblock(struct buffer_head *bh,
1154 				  struct inode *dir,
1155 				  struct ext4_filename *fname,
1156 				  const struct qstr *d_name,
1157 				  unsigned int offset,
1158 				  struct ext4_dir_entry_2 **res_dir)
1159 {
1160 	return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1161 			       fname, d_name, offset, res_dir);
1162 }
1163 
1164 /*
1165  * Directory block splitting, compacting
1166  */
1167 
1168 /*
1169  * Create map of hash values, offsets, and sizes, stored at end of block.
1170  * Returns number of entries mapped.
1171  */
1172 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1173 		       unsigned blocksize, struct dx_hash_info *hinfo,
1174 		       struct dx_map_entry *map_tail)
1175 {
1176 	int count = 0;
1177 	char *base = (char *) de;
1178 	struct dx_hash_info h = *hinfo;
1179 
1180 	while ((char *) de < base + blocksize) {
1181 		if (de->name_len && de->inode) {
1182 			ext4fs_dirhash(de->name, de->name_len, &h);
1183 			map_tail--;
1184 			map_tail->hash = h.hash;
1185 			map_tail->offs = ((char *) de - base)>>2;
1186 			map_tail->size = le16_to_cpu(de->rec_len);
1187 			count++;
1188 			cond_resched();
1189 		}
1190 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1191 		de = ext4_next_entry(de, blocksize);
1192 	}
1193 	return count;
1194 }
1195 
1196 /* Sort map by hash value */
1197 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1198 {
1199 	struct dx_map_entry *p, *q, *top = map + count - 1;
1200 	int more;
1201 	/* Combsort until bubble sort doesn't suck */
1202 	while (count > 2) {
1203 		count = count*10/13;
1204 		if (count - 9 < 2) /* 9, 10 -> 11 */
1205 			count = 11;
1206 		for (p = top, q = p - count; q >= map; p--, q--)
1207 			if (p->hash < q->hash)
1208 				swap(*p, *q);
1209 	}
1210 	/* Garden variety bubble sort */
1211 	do {
1212 		more = 0;
1213 		q = top;
1214 		while (q-- > map) {
1215 			if (q[1].hash >= q[0].hash)
1216 				continue;
1217 			swap(*(q+1), *q);
1218 			more = 1;
1219 		}
1220 	} while(more);
1221 }
1222 
1223 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1224 {
1225 	struct dx_entry *entries = frame->entries;
1226 	struct dx_entry *old = frame->at, *new = old + 1;
1227 	int count = dx_get_count(entries);
1228 
1229 	assert(count < dx_get_limit(entries));
1230 	assert(old < entries + count);
1231 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1232 	dx_set_hash(new, hash);
1233 	dx_set_block(new, block);
1234 	dx_set_count(entries, count + 1);
1235 }
1236 
1237 /*
1238  * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1239  *
1240  * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1241  * `de != NULL' is guaranteed by caller.
1242  */
1243 static inline int ext4_match(struct ext4_filename *fname,
1244 			     struct ext4_dir_entry_2 *de)
1245 {
1246 	const void *name = fname_name(fname);
1247 	u32 len = fname_len(fname);
1248 
1249 	if (!de->inode)
1250 		return 0;
1251 
1252 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1253 	if (unlikely(!name)) {
1254 		if (fname->usr_fname->name[0] == '_') {
1255 			int ret;
1256 			if (de->name_len < 16)
1257 				return 0;
1258 			ret = memcmp(de->name + de->name_len - 16,
1259 				     fname->crypto_buf.name + 8, 16);
1260 			return (ret == 0) ? 1 : 0;
1261 		}
1262 		name = fname->crypto_buf.name;
1263 		len = fname->crypto_buf.len;
1264 	}
1265 #endif
1266 	if (de->name_len != len)
1267 		return 0;
1268 	return (memcmp(de->name, name, len) == 0) ? 1 : 0;
1269 }
1270 
1271 /*
1272  * Returns 0 if not found, -1 on failure, and 1 on success
1273  */
1274 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1275 		    struct inode *dir, struct ext4_filename *fname,
1276 		    const struct qstr *d_name,
1277 		    unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1278 {
1279 	struct ext4_dir_entry_2 * de;
1280 	char * dlimit;
1281 	int de_len;
1282 	int res;
1283 
1284 	de = (struct ext4_dir_entry_2 *)search_buf;
1285 	dlimit = search_buf + buf_size;
1286 	while ((char *) de < dlimit) {
1287 		/* this code is executed quadratically often */
1288 		/* do minimal checking `by hand' */
1289 		if ((char *) de + de->name_len <= dlimit) {
1290 			res = ext4_match(fname, de);
1291 			if (res < 0) {
1292 				res = -1;
1293 				goto return_result;
1294 			}
1295 			if (res > 0) {
1296 				/* found a match - just to be sure, do
1297 				 * a full check */
1298 				if (ext4_check_dir_entry(dir, NULL, de, bh,
1299 						bh->b_data,
1300 						 bh->b_size, offset)) {
1301 					res = -1;
1302 					goto return_result;
1303 				}
1304 				*res_dir = de;
1305 				res = 1;
1306 				goto return_result;
1307 			}
1308 
1309 		}
1310 		/* prevent looping on a bad block */
1311 		de_len = ext4_rec_len_from_disk(de->rec_len,
1312 						dir->i_sb->s_blocksize);
1313 		if (de_len <= 0) {
1314 			res = -1;
1315 			goto return_result;
1316 		}
1317 		offset += de_len;
1318 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1319 	}
1320 
1321 	res = 0;
1322 return_result:
1323 	return res;
1324 }
1325 
1326 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1327 			       struct ext4_dir_entry *de)
1328 {
1329 	struct super_block *sb = dir->i_sb;
1330 
1331 	if (!is_dx(dir))
1332 		return 0;
1333 	if (block == 0)
1334 		return 1;
1335 	if (de->inode == 0 &&
1336 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1337 			sb->s_blocksize)
1338 		return 1;
1339 	return 0;
1340 }
1341 
1342 /*
1343  *	ext4_find_entry()
1344  *
1345  * finds an entry in the specified directory with the wanted name. It
1346  * returns the cache buffer in which the entry was found, and the entry
1347  * itself (as a parameter - res_dir). It does NOT read the inode of the
1348  * entry - you'll have to do that yourself if you want to.
1349  *
1350  * The returned buffer_head has ->b_count elevated.  The caller is expected
1351  * to brelse() it when appropriate.
1352  */
1353 static struct buffer_head * ext4_find_entry (struct inode *dir,
1354 					const struct qstr *d_name,
1355 					struct ext4_dir_entry_2 **res_dir,
1356 					int *inlined)
1357 {
1358 	struct super_block *sb;
1359 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1360 	struct buffer_head *bh, *ret = NULL;
1361 	ext4_lblk_t start, block, b;
1362 	const u8 *name = d_name->name;
1363 	int ra_max = 0;		/* Number of bh's in the readahead
1364 				   buffer, bh_use[] */
1365 	int ra_ptr = 0;		/* Current index into readahead
1366 				   buffer */
1367 	int num = 0;
1368 	ext4_lblk_t  nblocks;
1369 	int i, namelen, retval;
1370 	struct ext4_filename fname;
1371 
1372 	*res_dir = NULL;
1373 	sb = dir->i_sb;
1374 	namelen = d_name->len;
1375 	if (namelen > EXT4_NAME_LEN)
1376 		return NULL;
1377 
1378 	retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1379 	if (retval)
1380 		return ERR_PTR(retval);
1381 
1382 	if (ext4_has_inline_data(dir)) {
1383 		int has_inline_data = 1;
1384 		ret = ext4_find_inline_entry(dir, &fname, d_name, res_dir,
1385 					     &has_inline_data);
1386 		if (has_inline_data) {
1387 			if (inlined)
1388 				*inlined = 1;
1389 			goto cleanup_and_exit;
1390 		}
1391 	}
1392 
1393 	if ((namelen <= 2) && (name[0] == '.') &&
1394 	    (name[1] == '.' || name[1] == '\0')) {
1395 		/*
1396 		 * "." or ".." will only be in the first block
1397 		 * NFS may look up ".."; "." should be handled by the VFS
1398 		 */
1399 		block = start = 0;
1400 		nblocks = 1;
1401 		goto restart;
1402 	}
1403 	if (is_dx(dir)) {
1404 		ret = ext4_dx_find_entry(dir, &fname, res_dir);
1405 		/*
1406 		 * On success, or if the error was file not found,
1407 		 * return.  Otherwise, fall back to doing a search the
1408 		 * old fashioned way.
1409 		 */
1410 		if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1411 			goto cleanup_and_exit;
1412 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1413 			       "falling back\n"));
1414 	}
1415 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1416 	start = EXT4_I(dir)->i_dir_start_lookup;
1417 	if (start >= nblocks)
1418 		start = 0;
1419 	block = start;
1420 restart:
1421 	do {
1422 		/*
1423 		 * We deal with the read-ahead logic here.
1424 		 */
1425 		if (ra_ptr >= ra_max) {
1426 			/* Refill the readahead buffer */
1427 			ra_ptr = 0;
1428 			b = block;
1429 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1430 				/*
1431 				 * Terminate if we reach the end of the
1432 				 * directory and must wrap, or if our
1433 				 * search has finished at this block.
1434 				 */
1435 				if (b >= nblocks || (num && block == start)) {
1436 					bh_use[ra_max] = NULL;
1437 					break;
1438 				}
1439 				num++;
1440 				bh = ext4_getblk(NULL, dir, b++, 0);
1441 				if (IS_ERR(bh)) {
1442 					if (ra_max == 0) {
1443 						ret = bh;
1444 						goto cleanup_and_exit;
1445 					}
1446 					break;
1447 				}
1448 				bh_use[ra_max] = bh;
1449 				if (bh)
1450 					ll_rw_block(REQ_OP_READ,
1451 						    REQ_META | REQ_PRIO,
1452 						    1, &bh);
1453 			}
1454 		}
1455 		if ((bh = bh_use[ra_ptr++]) == NULL)
1456 			goto next;
1457 		wait_on_buffer(bh);
1458 		if (!buffer_uptodate(bh)) {
1459 			/* read error, skip block & hope for the best */
1460 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1461 					 (unsigned long) block);
1462 			brelse(bh);
1463 			goto next;
1464 		}
1465 		if (!buffer_verified(bh) &&
1466 		    !is_dx_internal_node(dir, block,
1467 					 (struct ext4_dir_entry *)bh->b_data) &&
1468 		    !ext4_dirent_csum_verify(dir,
1469 				(struct ext4_dir_entry *)bh->b_data)) {
1470 			EXT4_ERROR_INODE(dir, "checksumming directory "
1471 					 "block %lu", (unsigned long)block);
1472 			brelse(bh);
1473 			goto next;
1474 		}
1475 		set_buffer_verified(bh);
1476 		i = search_dirblock(bh, dir, &fname, d_name,
1477 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1478 		if (i == 1) {
1479 			EXT4_I(dir)->i_dir_start_lookup = block;
1480 			ret = bh;
1481 			goto cleanup_and_exit;
1482 		} else {
1483 			brelse(bh);
1484 			if (i < 0)
1485 				goto cleanup_and_exit;
1486 		}
1487 	next:
1488 		if (++block >= nblocks)
1489 			block = 0;
1490 	} while (block != start);
1491 
1492 	/*
1493 	 * If the directory has grown while we were searching, then
1494 	 * search the last part of the directory before giving up.
1495 	 */
1496 	block = nblocks;
1497 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1498 	if (block < nblocks) {
1499 		start = 0;
1500 		goto restart;
1501 	}
1502 
1503 cleanup_and_exit:
1504 	/* Clean up the read-ahead blocks */
1505 	for (; ra_ptr < ra_max; ra_ptr++)
1506 		brelse(bh_use[ra_ptr]);
1507 	ext4_fname_free_filename(&fname);
1508 	return ret;
1509 }
1510 
1511 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1512 			struct ext4_filename *fname,
1513 			struct ext4_dir_entry_2 **res_dir)
1514 {
1515 	struct super_block * sb = dir->i_sb;
1516 	struct dx_frame frames[2], *frame;
1517 	const struct qstr *d_name = fname->usr_fname;
1518 	struct buffer_head *bh;
1519 	ext4_lblk_t block;
1520 	int retval;
1521 
1522 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1523 	*res_dir = NULL;
1524 #endif
1525 	frame = dx_probe(fname, dir, NULL, frames);
1526 	if (IS_ERR(frame))
1527 		return (struct buffer_head *) frame;
1528 	do {
1529 		block = dx_get_block(frame->at);
1530 		bh = ext4_read_dirblock(dir, block, DIRENT);
1531 		if (IS_ERR(bh))
1532 			goto errout;
1533 
1534 		retval = search_dirblock(bh, dir, fname, d_name,
1535 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1536 					 res_dir);
1537 		if (retval == 1)
1538 			goto success;
1539 		brelse(bh);
1540 		if (retval == -1) {
1541 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1542 			goto errout;
1543 		}
1544 
1545 		/* Check to see if we should continue to search */
1546 		retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1547 					       frames, NULL);
1548 		if (retval < 0) {
1549 			ext4_warning_inode(dir,
1550 				"error %d reading directory index block",
1551 				retval);
1552 			bh = ERR_PTR(retval);
1553 			goto errout;
1554 		}
1555 	} while (retval == 1);
1556 
1557 	bh = NULL;
1558 errout:
1559 	dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1560 success:
1561 	dx_release(frames);
1562 	return bh;
1563 }
1564 
1565 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1566 {
1567 	struct inode *inode;
1568 	struct ext4_dir_entry_2 *de;
1569 	struct buffer_head *bh;
1570 
1571 	if (ext4_encrypted_inode(dir)) {
1572 		int res = fscrypt_get_encryption_info(dir);
1573 
1574 		/*
1575 		 * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
1576 		 * created while the directory was encrypted and we
1577 		 * have access to the key.
1578 		 */
1579 		if (fscrypt_has_encryption_key(dir))
1580 			fscrypt_set_encrypted_dentry(dentry);
1581 		fscrypt_set_d_op(dentry);
1582 		if (res && res != -ENOKEY)
1583 			return ERR_PTR(res);
1584 	}
1585 
1586        if (dentry->d_name.len > EXT4_NAME_LEN)
1587 	       return ERR_PTR(-ENAMETOOLONG);
1588 
1589 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1590 	if (IS_ERR(bh))
1591 		return (struct dentry *) bh;
1592 	inode = NULL;
1593 	if (bh) {
1594 		__u32 ino = le32_to_cpu(de->inode);
1595 		brelse(bh);
1596 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1597 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1598 			return ERR_PTR(-EFSCORRUPTED);
1599 		}
1600 		if (unlikely(ino == dir->i_ino)) {
1601 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1602 					 dentry);
1603 			return ERR_PTR(-EFSCORRUPTED);
1604 		}
1605 		inode = ext4_iget_normal(dir->i_sb, ino);
1606 		if (inode == ERR_PTR(-ESTALE)) {
1607 			EXT4_ERROR_INODE(dir,
1608 					 "deleted inode referenced: %u",
1609 					 ino);
1610 			return ERR_PTR(-EFSCORRUPTED);
1611 		}
1612 		if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1613 		    (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1614 		    !fscrypt_has_permitted_context(dir, inode)) {
1615 			int nokey = ext4_encrypted_inode(inode) &&
1616 				!fscrypt_has_encryption_key(inode);
1617 			iput(inode);
1618 			if (nokey)
1619 				return ERR_PTR(-ENOKEY);
1620 			ext4_warning(inode->i_sb,
1621 				     "Inconsistent encryption contexts: %lu/%lu",
1622 				     (unsigned long) dir->i_ino,
1623 				     (unsigned long) inode->i_ino);
1624 			return ERR_PTR(-EPERM);
1625 		}
1626 	}
1627 	return d_splice_alias(inode, dentry);
1628 }
1629 
1630 
1631 struct dentry *ext4_get_parent(struct dentry *child)
1632 {
1633 	__u32 ino;
1634 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1635 	struct ext4_dir_entry_2 * de;
1636 	struct buffer_head *bh;
1637 
1638 	bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1639 	if (IS_ERR(bh))
1640 		return (struct dentry *) bh;
1641 	if (!bh)
1642 		return ERR_PTR(-ENOENT);
1643 	ino = le32_to_cpu(de->inode);
1644 	brelse(bh);
1645 
1646 	if (!ext4_valid_inum(child->d_sb, ino)) {
1647 		EXT4_ERROR_INODE(d_inode(child),
1648 				 "bad parent inode number: %u", ino);
1649 		return ERR_PTR(-EFSCORRUPTED);
1650 	}
1651 
1652 	return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1653 }
1654 
1655 /*
1656  * Move count entries from end of map between two memory locations.
1657  * Returns pointer to last entry moved.
1658  */
1659 static struct ext4_dir_entry_2 *
1660 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1661 		unsigned blocksize)
1662 {
1663 	unsigned rec_len = 0;
1664 
1665 	while (count--) {
1666 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1667 						(from + (map->offs<<2));
1668 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1669 		memcpy (to, de, rec_len);
1670 		((struct ext4_dir_entry_2 *) to)->rec_len =
1671 				ext4_rec_len_to_disk(rec_len, blocksize);
1672 		de->inode = 0;
1673 		map++;
1674 		to += rec_len;
1675 	}
1676 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1677 }
1678 
1679 /*
1680  * Compact each dir entry in the range to the minimal rec_len.
1681  * Returns pointer to last entry in range.
1682  */
1683 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1684 {
1685 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1686 	unsigned rec_len = 0;
1687 
1688 	prev = to = de;
1689 	while ((char*)de < base + blocksize) {
1690 		next = ext4_next_entry(de, blocksize);
1691 		if (de->inode && de->name_len) {
1692 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1693 			if (de > to)
1694 				memmove(to, de, rec_len);
1695 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1696 			prev = to;
1697 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1698 		}
1699 		de = next;
1700 	}
1701 	return prev;
1702 }
1703 
1704 /*
1705  * Split a full leaf block to make room for a new dir entry.
1706  * Allocate a new block, and move entries so that they are approx. equally full.
1707  * Returns pointer to de in block into which the new entry will be inserted.
1708  */
1709 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1710 			struct buffer_head **bh,struct dx_frame *frame,
1711 			struct dx_hash_info *hinfo)
1712 {
1713 	unsigned blocksize = dir->i_sb->s_blocksize;
1714 	unsigned count, continued;
1715 	struct buffer_head *bh2;
1716 	ext4_lblk_t newblock;
1717 	u32 hash2;
1718 	struct dx_map_entry *map;
1719 	char *data1 = (*bh)->b_data, *data2;
1720 	unsigned split, move, size;
1721 	struct ext4_dir_entry_2 *de = NULL, *de2;
1722 	struct ext4_dir_entry_tail *t;
1723 	int	csum_size = 0;
1724 	int	err = 0, i;
1725 
1726 	if (ext4_has_metadata_csum(dir->i_sb))
1727 		csum_size = sizeof(struct ext4_dir_entry_tail);
1728 
1729 	bh2 = ext4_append(handle, dir, &newblock);
1730 	if (IS_ERR(bh2)) {
1731 		brelse(*bh);
1732 		*bh = NULL;
1733 		return (struct ext4_dir_entry_2 *) bh2;
1734 	}
1735 
1736 	BUFFER_TRACE(*bh, "get_write_access");
1737 	err = ext4_journal_get_write_access(handle, *bh);
1738 	if (err)
1739 		goto journal_error;
1740 
1741 	BUFFER_TRACE(frame->bh, "get_write_access");
1742 	err = ext4_journal_get_write_access(handle, frame->bh);
1743 	if (err)
1744 		goto journal_error;
1745 
1746 	data2 = bh2->b_data;
1747 
1748 	/* create map in the end of data2 block */
1749 	map = (struct dx_map_entry *) (data2 + blocksize);
1750 	count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1751 			     blocksize, hinfo, map);
1752 	map -= count;
1753 	dx_sort_map(map, count);
1754 	/* Split the existing block in the middle, size-wise */
1755 	size = 0;
1756 	move = 0;
1757 	for (i = count-1; i >= 0; i--) {
1758 		/* is more than half of this entry in 2nd half of the block? */
1759 		if (size + map[i].size/2 > blocksize/2)
1760 			break;
1761 		size += map[i].size;
1762 		move++;
1763 	}
1764 	/* map index at which we will split */
1765 	split = count - move;
1766 	hash2 = map[split].hash;
1767 	continued = hash2 == map[split - 1].hash;
1768 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1769 			(unsigned long)dx_get_block(frame->at),
1770 					hash2, split, count-split));
1771 
1772 	/* Fancy dance to stay within two buffers */
1773 	de2 = dx_move_dirents(data1, data2, map + split, count - split,
1774 			      blocksize);
1775 	de = dx_pack_dirents(data1, blocksize);
1776 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1777 					   (char *) de,
1778 					   blocksize);
1779 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1780 					    (char *) de2,
1781 					    blocksize);
1782 	if (csum_size) {
1783 		t = EXT4_DIRENT_TAIL(data2, blocksize);
1784 		initialize_dirent_tail(t, blocksize);
1785 
1786 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1787 		initialize_dirent_tail(t, blocksize);
1788 	}
1789 
1790 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1791 			blocksize, 1));
1792 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1793 			blocksize, 1));
1794 
1795 	/* Which block gets the new entry? */
1796 	if (hinfo->hash >= hash2) {
1797 		swap(*bh, bh2);
1798 		de = de2;
1799 	}
1800 	dx_insert_block(frame, hash2 + continued, newblock);
1801 	err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1802 	if (err)
1803 		goto journal_error;
1804 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1805 	if (err)
1806 		goto journal_error;
1807 	brelse(bh2);
1808 	dxtrace(dx_show_index("frame", frame->entries));
1809 	return de;
1810 
1811 journal_error:
1812 	brelse(*bh);
1813 	brelse(bh2);
1814 	*bh = NULL;
1815 	ext4_std_error(dir->i_sb, err);
1816 	return ERR_PTR(err);
1817 }
1818 
1819 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1820 		      struct buffer_head *bh,
1821 		      void *buf, int buf_size,
1822 		      struct ext4_filename *fname,
1823 		      struct ext4_dir_entry_2 **dest_de)
1824 {
1825 	struct ext4_dir_entry_2 *de;
1826 	unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1827 	int nlen, rlen;
1828 	unsigned int offset = 0;
1829 	char *top;
1830 	int res;
1831 
1832 	de = (struct ext4_dir_entry_2 *)buf;
1833 	top = buf + buf_size - reclen;
1834 	while ((char *) de <= top) {
1835 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1836 					 buf, buf_size, offset)) {
1837 			res = -EFSCORRUPTED;
1838 			goto return_result;
1839 		}
1840 		/* Provide crypto context and crypto buffer to ext4 match */
1841 		res = ext4_match(fname, de);
1842 		if (res < 0)
1843 			goto return_result;
1844 		if (res > 0) {
1845 			res = -EEXIST;
1846 			goto return_result;
1847 		}
1848 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1849 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1850 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1851 			break;
1852 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1853 		offset += rlen;
1854 	}
1855 
1856 	if ((char *) de > top)
1857 		res = -ENOSPC;
1858 	else {
1859 		*dest_de = de;
1860 		res = 0;
1861 	}
1862 return_result:
1863 	return res;
1864 }
1865 
1866 int ext4_insert_dentry(struct inode *dir,
1867 		       struct inode *inode,
1868 		       struct ext4_dir_entry_2 *de,
1869 		       int buf_size,
1870 		       struct ext4_filename *fname)
1871 {
1872 
1873 	int nlen, rlen;
1874 
1875 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1876 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1877 	if (de->inode) {
1878 		struct ext4_dir_entry_2 *de1 =
1879 			(struct ext4_dir_entry_2 *)((char *)de + nlen);
1880 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1881 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1882 		de = de1;
1883 	}
1884 	de->file_type = EXT4_FT_UNKNOWN;
1885 	de->inode = cpu_to_le32(inode->i_ino);
1886 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1887 	de->name_len = fname_len(fname);
1888 	memcpy(de->name, fname_name(fname), fname_len(fname));
1889 	return 0;
1890 }
1891 
1892 /*
1893  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1894  * it points to a directory entry which is guaranteed to be large
1895  * enough for new directory entry.  If de is NULL, then
1896  * add_dirent_to_buf will attempt search the directory block for
1897  * space.  It will return -ENOSPC if no space is available, and -EIO
1898  * and -EEXIST if directory entry already exists.
1899  */
1900 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1901 			     struct inode *dir,
1902 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1903 			     struct buffer_head *bh)
1904 {
1905 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1906 	int		csum_size = 0;
1907 	int		err;
1908 
1909 	if (ext4_has_metadata_csum(inode->i_sb))
1910 		csum_size = sizeof(struct ext4_dir_entry_tail);
1911 
1912 	if (!de) {
1913 		err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1914 					blocksize - csum_size, fname, &de);
1915 		if (err)
1916 			return err;
1917 	}
1918 	BUFFER_TRACE(bh, "get_write_access");
1919 	err = ext4_journal_get_write_access(handle, bh);
1920 	if (err) {
1921 		ext4_std_error(dir->i_sb, err);
1922 		return err;
1923 	}
1924 
1925 	/* By now the buffer is marked for journaling. Due to crypto operations,
1926 	 * the following function call may fail */
1927 	err = ext4_insert_dentry(dir, inode, de, blocksize, fname);
1928 	if (err < 0)
1929 		return err;
1930 
1931 	/*
1932 	 * XXX shouldn't update any times until successful
1933 	 * completion of syscall, but too many callers depend
1934 	 * on this.
1935 	 *
1936 	 * XXX similarly, too many callers depend on
1937 	 * ext4_new_inode() setting the times, but error
1938 	 * recovery deletes the inode, so the worst that can
1939 	 * happen is that the times are slightly out of date
1940 	 * and/or different from the directory change time.
1941 	 */
1942 	dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1943 	ext4_update_dx_flag(dir);
1944 	dir->i_version++;
1945 	ext4_mark_inode_dirty(handle, dir);
1946 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1947 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1948 	if (err)
1949 		ext4_std_error(dir->i_sb, err);
1950 	return 0;
1951 }
1952 
1953 /*
1954  * This converts a one block unindexed directory to a 3 block indexed
1955  * directory, and adds the dentry to the indexed directory.
1956  */
1957 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1958 			    struct inode *dir,
1959 			    struct inode *inode, struct buffer_head *bh)
1960 {
1961 	struct buffer_head *bh2;
1962 	struct dx_root	*root;
1963 	struct dx_frame	frames[2], *frame;
1964 	struct dx_entry *entries;
1965 	struct ext4_dir_entry_2	*de, *de2;
1966 	struct ext4_dir_entry_tail *t;
1967 	char		*data1, *top;
1968 	unsigned	len;
1969 	int		retval;
1970 	unsigned	blocksize;
1971 	ext4_lblk_t  block;
1972 	struct fake_dirent *fde;
1973 	int csum_size = 0;
1974 
1975 	if (ext4_has_metadata_csum(inode->i_sb))
1976 		csum_size = sizeof(struct ext4_dir_entry_tail);
1977 
1978 	blocksize =  dir->i_sb->s_blocksize;
1979 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1980 	BUFFER_TRACE(bh, "get_write_access");
1981 	retval = ext4_journal_get_write_access(handle, bh);
1982 	if (retval) {
1983 		ext4_std_error(dir->i_sb, retval);
1984 		brelse(bh);
1985 		return retval;
1986 	}
1987 	root = (struct dx_root *) bh->b_data;
1988 
1989 	/* The 0th block becomes the root, move the dirents out */
1990 	fde = &root->dotdot;
1991 	de = (struct ext4_dir_entry_2 *)((char *)fde +
1992 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
1993 	if ((char *) de >= (((char *) root) + blocksize)) {
1994 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1995 		brelse(bh);
1996 		return -EFSCORRUPTED;
1997 	}
1998 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1999 
2000 	/* Allocate new block for the 0th block's dirents */
2001 	bh2 = ext4_append(handle, dir, &block);
2002 	if (IS_ERR(bh2)) {
2003 		brelse(bh);
2004 		return PTR_ERR(bh2);
2005 	}
2006 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
2007 	data1 = bh2->b_data;
2008 
2009 	memcpy (data1, de, len);
2010 	de = (struct ext4_dir_entry_2 *) data1;
2011 	top = data1 + len;
2012 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
2013 		de = de2;
2014 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
2015 					   (char *) de,
2016 					   blocksize);
2017 
2018 	if (csum_size) {
2019 		t = EXT4_DIRENT_TAIL(data1, blocksize);
2020 		initialize_dirent_tail(t, blocksize);
2021 	}
2022 
2023 	/* Initialize the root; the dot dirents already exist */
2024 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
2025 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
2026 					   blocksize);
2027 	memset (&root->info, 0, sizeof(root->info));
2028 	root->info.info_length = sizeof(root->info);
2029 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
2030 	entries = root->entries;
2031 	dx_set_block(entries, 1);
2032 	dx_set_count(entries, 1);
2033 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
2034 
2035 	/* Initialize as for dx_probe */
2036 	fname->hinfo.hash_version = root->info.hash_version;
2037 	if (fname->hinfo.hash_version <= DX_HASH_TEA)
2038 		fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
2039 	fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
2040 	ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
2041 
2042 	memset(frames, 0, sizeof(frames));
2043 	frame = frames;
2044 	frame->entries = entries;
2045 	frame->at = entries;
2046 	frame->bh = bh;
2047 
2048 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2049 	if (retval)
2050 		goto out_frames;
2051 	retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
2052 	if (retval)
2053 		goto out_frames;
2054 
2055 	de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2056 	if (IS_ERR(de)) {
2057 		retval = PTR_ERR(de);
2058 		goto out_frames;
2059 	}
2060 
2061 	retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2062 out_frames:
2063 	/*
2064 	 * Even if the block split failed, we have to properly write
2065 	 * out all the changes we did so far. Otherwise we can end up
2066 	 * with corrupted filesystem.
2067 	 */
2068 	if (retval)
2069 		ext4_mark_inode_dirty(handle, dir);
2070 	dx_release(frames);
2071 	brelse(bh2);
2072 	return retval;
2073 }
2074 
2075 /*
2076  *	ext4_add_entry()
2077  *
2078  * adds a file entry to the specified directory, using the same
2079  * semantics as ext4_find_entry(). It returns NULL if it failed.
2080  *
2081  * NOTE!! The inode part of 'de' is left at 0 - which means you
2082  * may not sleep between calling this and putting something into
2083  * the entry, as someone else might have used it while you slept.
2084  */
2085 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2086 			  struct inode *inode)
2087 {
2088 	struct inode *dir = d_inode(dentry->d_parent);
2089 	struct buffer_head *bh = NULL;
2090 	struct ext4_dir_entry_2 *de;
2091 	struct ext4_dir_entry_tail *t;
2092 	struct super_block *sb;
2093 	struct ext4_filename fname;
2094 	int	retval;
2095 	int	dx_fallback=0;
2096 	unsigned blocksize;
2097 	ext4_lblk_t block, blocks;
2098 	int	csum_size = 0;
2099 
2100 	if (ext4_has_metadata_csum(inode->i_sb))
2101 		csum_size = sizeof(struct ext4_dir_entry_tail);
2102 
2103 	sb = dir->i_sb;
2104 	blocksize = sb->s_blocksize;
2105 	if (!dentry->d_name.len)
2106 		return -EINVAL;
2107 
2108 	retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2109 	if (retval)
2110 		return retval;
2111 
2112 	if (ext4_has_inline_data(dir)) {
2113 		retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2114 		if (retval < 0)
2115 			goto out;
2116 		if (retval == 1) {
2117 			retval = 0;
2118 			goto out;
2119 		}
2120 	}
2121 
2122 	if (is_dx(dir)) {
2123 		retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2124 		if (!retval || (retval != ERR_BAD_DX_DIR))
2125 			goto out;
2126 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2127 		dx_fallback++;
2128 		ext4_mark_inode_dirty(handle, dir);
2129 	}
2130 	blocks = dir->i_size >> sb->s_blocksize_bits;
2131 	for (block = 0; block < blocks; block++) {
2132 		bh = ext4_read_dirblock(dir, block, DIRENT);
2133 		if (IS_ERR(bh)) {
2134 			retval = PTR_ERR(bh);
2135 			bh = NULL;
2136 			goto out;
2137 		}
2138 		retval = add_dirent_to_buf(handle, &fname, dir, inode,
2139 					   NULL, bh);
2140 		if (retval != -ENOSPC)
2141 			goto out;
2142 
2143 		if (blocks == 1 && !dx_fallback &&
2144 		    ext4_has_feature_dir_index(sb)) {
2145 			retval = make_indexed_dir(handle, &fname, dir,
2146 						  inode, bh);
2147 			bh = NULL; /* make_indexed_dir releases bh */
2148 			goto out;
2149 		}
2150 		brelse(bh);
2151 	}
2152 	bh = ext4_append(handle, dir, &block);
2153 	if (IS_ERR(bh)) {
2154 		retval = PTR_ERR(bh);
2155 		bh = NULL;
2156 		goto out;
2157 	}
2158 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2159 	de->inode = 0;
2160 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2161 
2162 	if (csum_size) {
2163 		t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2164 		initialize_dirent_tail(t, blocksize);
2165 	}
2166 
2167 	retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2168 out:
2169 	ext4_fname_free_filename(&fname);
2170 	brelse(bh);
2171 	if (retval == 0)
2172 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2173 	return retval;
2174 }
2175 
2176 /*
2177  * Returns 0 for success, or a negative error value
2178  */
2179 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2180 			     struct inode *dir, struct inode *inode)
2181 {
2182 	struct dx_frame frames[2], *frame;
2183 	struct dx_entry *entries, *at;
2184 	struct buffer_head *bh;
2185 	struct super_block *sb = dir->i_sb;
2186 	struct ext4_dir_entry_2 *de;
2187 	int err;
2188 
2189 	frame = dx_probe(fname, dir, NULL, frames);
2190 	if (IS_ERR(frame))
2191 		return PTR_ERR(frame);
2192 	entries = frame->entries;
2193 	at = frame->at;
2194 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2195 	if (IS_ERR(bh)) {
2196 		err = PTR_ERR(bh);
2197 		bh = NULL;
2198 		goto cleanup;
2199 	}
2200 
2201 	BUFFER_TRACE(bh, "get_write_access");
2202 	err = ext4_journal_get_write_access(handle, bh);
2203 	if (err)
2204 		goto journal_error;
2205 
2206 	err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2207 	if (err != -ENOSPC)
2208 		goto cleanup;
2209 
2210 	/* Block full, should compress but for now just split */
2211 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2212 		       dx_get_count(entries), dx_get_limit(entries)));
2213 	/* Need to split index? */
2214 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2215 		ext4_lblk_t newblock;
2216 		unsigned icount = dx_get_count(entries);
2217 		int levels = frame - frames;
2218 		struct dx_entry *entries2;
2219 		struct dx_node *node2;
2220 		struct buffer_head *bh2;
2221 
2222 		if (levels && (dx_get_count(frames->entries) ==
2223 			       dx_get_limit(frames->entries))) {
2224 			ext4_warning_inode(dir, "Directory index full!");
2225 			err = -ENOSPC;
2226 			goto cleanup;
2227 		}
2228 		bh2 = ext4_append(handle, dir, &newblock);
2229 		if (IS_ERR(bh2)) {
2230 			err = PTR_ERR(bh2);
2231 			goto cleanup;
2232 		}
2233 		node2 = (struct dx_node *)(bh2->b_data);
2234 		entries2 = node2->entries;
2235 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2236 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2237 							   sb->s_blocksize);
2238 		BUFFER_TRACE(frame->bh, "get_write_access");
2239 		err = ext4_journal_get_write_access(handle, frame->bh);
2240 		if (err)
2241 			goto journal_error;
2242 		if (levels) {
2243 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2244 			unsigned hash2 = dx_get_hash(entries + icount1);
2245 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2246 				       icount1, icount2));
2247 
2248 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2249 			err = ext4_journal_get_write_access(handle,
2250 							     frames[0].bh);
2251 			if (err)
2252 				goto journal_error;
2253 
2254 			memcpy((char *) entries2, (char *) (entries + icount1),
2255 			       icount2 * sizeof(struct dx_entry));
2256 			dx_set_count(entries, icount1);
2257 			dx_set_count(entries2, icount2);
2258 			dx_set_limit(entries2, dx_node_limit(dir));
2259 
2260 			/* Which index block gets the new entry? */
2261 			if (at - entries >= icount1) {
2262 				frame->at = at = at - entries - icount1 + entries2;
2263 				frame->entries = entries = entries2;
2264 				swap(frame->bh, bh2);
2265 			}
2266 			dx_insert_block(frames + 0, hash2, newblock);
2267 			dxtrace(dx_show_index("node", frames[1].entries));
2268 			dxtrace(dx_show_index("node",
2269 			       ((struct dx_node *) bh2->b_data)->entries));
2270 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2271 			if (err)
2272 				goto journal_error;
2273 			brelse (bh2);
2274 		} else {
2275 			dxtrace(printk(KERN_DEBUG
2276 				       "Creating second level index...\n"));
2277 			memcpy((char *) entries2, (char *) entries,
2278 			       icount * sizeof(struct dx_entry));
2279 			dx_set_limit(entries2, dx_node_limit(dir));
2280 
2281 			/* Set up root */
2282 			dx_set_count(entries, 1);
2283 			dx_set_block(entries + 0, newblock);
2284 			((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
2285 
2286 			/* Add new access path frame */
2287 			frame = frames + 1;
2288 			frame->at = at = at - entries + entries2;
2289 			frame->entries = entries = entries2;
2290 			frame->bh = bh2;
2291 			err = ext4_journal_get_write_access(handle,
2292 							     frame->bh);
2293 			if (err)
2294 				goto journal_error;
2295 		}
2296 		err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
2297 		if (err) {
2298 			ext4_std_error(inode->i_sb, err);
2299 			goto cleanup;
2300 		}
2301 	}
2302 	de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2303 	if (IS_ERR(de)) {
2304 		err = PTR_ERR(de);
2305 		goto cleanup;
2306 	}
2307 	err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2308 	goto cleanup;
2309 
2310 journal_error:
2311 	ext4_std_error(dir->i_sb, err);
2312 cleanup:
2313 	brelse(bh);
2314 	dx_release(frames);
2315 	return err;
2316 }
2317 
2318 /*
2319  * ext4_generic_delete_entry deletes a directory entry by merging it
2320  * with the previous entry
2321  */
2322 int ext4_generic_delete_entry(handle_t *handle,
2323 			      struct inode *dir,
2324 			      struct ext4_dir_entry_2 *de_del,
2325 			      struct buffer_head *bh,
2326 			      void *entry_buf,
2327 			      int buf_size,
2328 			      int csum_size)
2329 {
2330 	struct ext4_dir_entry_2 *de, *pde;
2331 	unsigned int blocksize = dir->i_sb->s_blocksize;
2332 	int i;
2333 
2334 	i = 0;
2335 	pde = NULL;
2336 	de = (struct ext4_dir_entry_2 *)entry_buf;
2337 	while (i < buf_size - csum_size) {
2338 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2339 					 bh->b_data, bh->b_size, i))
2340 			return -EFSCORRUPTED;
2341 		if (de == de_del)  {
2342 			if (pde)
2343 				pde->rec_len = ext4_rec_len_to_disk(
2344 					ext4_rec_len_from_disk(pde->rec_len,
2345 							       blocksize) +
2346 					ext4_rec_len_from_disk(de->rec_len,
2347 							       blocksize),
2348 					blocksize);
2349 			else
2350 				de->inode = 0;
2351 			dir->i_version++;
2352 			return 0;
2353 		}
2354 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2355 		pde = de;
2356 		de = ext4_next_entry(de, blocksize);
2357 	}
2358 	return -ENOENT;
2359 }
2360 
2361 static int ext4_delete_entry(handle_t *handle,
2362 			     struct inode *dir,
2363 			     struct ext4_dir_entry_2 *de_del,
2364 			     struct buffer_head *bh)
2365 {
2366 	int err, csum_size = 0;
2367 
2368 	if (ext4_has_inline_data(dir)) {
2369 		int has_inline_data = 1;
2370 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2371 					       &has_inline_data);
2372 		if (has_inline_data)
2373 			return err;
2374 	}
2375 
2376 	if (ext4_has_metadata_csum(dir->i_sb))
2377 		csum_size = sizeof(struct ext4_dir_entry_tail);
2378 
2379 	BUFFER_TRACE(bh, "get_write_access");
2380 	err = ext4_journal_get_write_access(handle, bh);
2381 	if (unlikely(err))
2382 		goto out;
2383 
2384 	err = ext4_generic_delete_entry(handle, dir, de_del,
2385 					bh, bh->b_data,
2386 					dir->i_sb->s_blocksize, csum_size);
2387 	if (err)
2388 		goto out;
2389 
2390 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2391 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2392 	if (unlikely(err))
2393 		goto out;
2394 
2395 	return 0;
2396 out:
2397 	if (err != -ENOENT)
2398 		ext4_std_error(dir->i_sb, err);
2399 	return err;
2400 }
2401 
2402 /*
2403  * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2404  * since this indicates that nlinks count was previously 1.
2405  */
2406 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2407 {
2408 	inc_nlink(inode);
2409 	if (is_dx(inode) && inode->i_nlink > 1) {
2410 		/* limit is 16-bit i_links_count */
2411 		if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2412 			set_nlink(inode, 1);
2413 			ext4_set_feature_dir_nlink(inode->i_sb);
2414 		}
2415 	}
2416 }
2417 
2418 /*
2419  * If a directory had nlink == 1, then we should let it be 1. This indicates
2420  * directory has >EXT4_LINK_MAX subdirs.
2421  */
2422 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2423 {
2424 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2425 		drop_nlink(inode);
2426 }
2427 
2428 
2429 static int ext4_add_nondir(handle_t *handle,
2430 		struct dentry *dentry, struct inode *inode)
2431 {
2432 	int err = ext4_add_entry(handle, dentry, inode);
2433 	if (!err) {
2434 		ext4_mark_inode_dirty(handle, inode);
2435 		unlock_new_inode(inode);
2436 		d_instantiate(dentry, inode);
2437 		return 0;
2438 	}
2439 	drop_nlink(inode);
2440 	unlock_new_inode(inode);
2441 	iput(inode);
2442 	return err;
2443 }
2444 
2445 /*
2446  * By the time this is called, we already have created
2447  * the directory cache entry for the new file, but it
2448  * is so far negative - it has no inode.
2449  *
2450  * If the create succeeds, we fill in the inode information
2451  * with d_instantiate().
2452  */
2453 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2454 		       bool excl)
2455 {
2456 	handle_t *handle;
2457 	struct inode *inode;
2458 	int err, credits, retries = 0;
2459 
2460 	err = dquot_initialize(dir);
2461 	if (err)
2462 		return err;
2463 
2464 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2465 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2466 retry:
2467 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2468 					    NULL, EXT4_HT_DIR, credits);
2469 	handle = ext4_journal_current_handle();
2470 	err = PTR_ERR(inode);
2471 	if (!IS_ERR(inode)) {
2472 		inode->i_op = &ext4_file_inode_operations;
2473 		inode->i_fop = &ext4_file_operations;
2474 		ext4_set_aops(inode);
2475 		err = ext4_add_nondir(handle, dentry, inode);
2476 		if (!err && IS_DIRSYNC(dir))
2477 			ext4_handle_sync(handle);
2478 	}
2479 	if (handle)
2480 		ext4_journal_stop(handle);
2481 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2482 		goto retry;
2483 	return err;
2484 }
2485 
2486 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2487 		      umode_t mode, dev_t rdev)
2488 {
2489 	handle_t *handle;
2490 	struct inode *inode;
2491 	int err, credits, retries = 0;
2492 
2493 	err = dquot_initialize(dir);
2494 	if (err)
2495 		return err;
2496 
2497 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2498 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2499 retry:
2500 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2501 					    NULL, EXT4_HT_DIR, credits);
2502 	handle = ext4_journal_current_handle();
2503 	err = PTR_ERR(inode);
2504 	if (!IS_ERR(inode)) {
2505 		init_special_inode(inode, inode->i_mode, rdev);
2506 		inode->i_op = &ext4_special_inode_operations;
2507 		err = ext4_add_nondir(handle, dentry, inode);
2508 		if (!err && IS_DIRSYNC(dir))
2509 			ext4_handle_sync(handle);
2510 	}
2511 	if (handle)
2512 		ext4_journal_stop(handle);
2513 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2514 		goto retry;
2515 	return err;
2516 }
2517 
2518 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2519 {
2520 	handle_t *handle;
2521 	struct inode *inode;
2522 	int err, retries = 0;
2523 
2524 	err = dquot_initialize(dir);
2525 	if (err)
2526 		return err;
2527 
2528 retry:
2529 	inode = ext4_new_inode_start_handle(dir, mode,
2530 					    NULL, 0, NULL,
2531 					    EXT4_HT_DIR,
2532 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2533 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2534 	handle = ext4_journal_current_handle();
2535 	err = PTR_ERR(inode);
2536 	if (!IS_ERR(inode)) {
2537 		inode->i_op = &ext4_file_inode_operations;
2538 		inode->i_fop = &ext4_file_operations;
2539 		ext4_set_aops(inode);
2540 		d_tmpfile(dentry, inode);
2541 		err = ext4_orphan_add(handle, inode);
2542 		if (err)
2543 			goto err_unlock_inode;
2544 		mark_inode_dirty(inode);
2545 		unlock_new_inode(inode);
2546 	}
2547 	if (handle)
2548 		ext4_journal_stop(handle);
2549 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2550 		goto retry;
2551 	return err;
2552 err_unlock_inode:
2553 	ext4_journal_stop(handle);
2554 	unlock_new_inode(inode);
2555 	return err;
2556 }
2557 
2558 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2559 			  struct ext4_dir_entry_2 *de,
2560 			  int blocksize, int csum_size,
2561 			  unsigned int parent_ino, int dotdot_real_len)
2562 {
2563 	de->inode = cpu_to_le32(inode->i_ino);
2564 	de->name_len = 1;
2565 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2566 					   blocksize);
2567 	strcpy(de->name, ".");
2568 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2569 
2570 	de = ext4_next_entry(de, blocksize);
2571 	de->inode = cpu_to_le32(parent_ino);
2572 	de->name_len = 2;
2573 	if (!dotdot_real_len)
2574 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2575 					(csum_size + EXT4_DIR_REC_LEN(1)),
2576 					blocksize);
2577 	else
2578 		de->rec_len = ext4_rec_len_to_disk(
2579 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2580 	strcpy(de->name, "..");
2581 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2582 
2583 	return ext4_next_entry(de, blocksize);
2584 }
2585 
2586 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2587 			     struct inode *inode)
2588 {
2589 	struct buffer_head *dir_block = NULL;
2590 	struct ext4_dir_entry_2 *de;
2591 	struct ext4_dir_entry_tail *t;
2592 	ext4_lblk_t block = 0;
2593 	unsigned int blocksize = dir->i_sb->s_blocksize;
2594 	int csum_size = 0;
2595 	int err;
2596 
2597 	if (ext4_has_metadata_csum(dir->i_sb))
2598 		csum_size = sizeof(struct ext4_dir_entry_tail);
2599 
2600 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2601 		err = ext4_try_create_inline_dir(handle, dir, inode);
2602 		if (err < 0 && err != -ENOSPC)
2603 			goto out;
2604 		if (!err)
2605 			goto out;
2606 	}
2607 
2608 	inode->i_size = 0;
2609 	dir_block = ext4_append(handle, inode, &block);
2610 	if (IS_ERR(dir_block))
2611 		return PTR_ERR(dir_block);
2612 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2613 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2614 	set_nlink(inode, 2);
2615 	if (csum_size) {
2616 		t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2617 		initialize_dirent_tail(t, blocksize);
2618 	}
2619 
2620 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2621 	err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2622 	if (err)
2623 		goto out;
2624 	set_buffer_verified(dir_block);
2625 out:
2626 	brelse(dir_block);
2627 	return err;
2628 }
2629 
2630 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2631 {
2632 	handle_t *handle;
2633 	struct inode *inode;
2634 	int err, credits, retries = 0;
2635 
2636 	if (EXT4_DIR_LINK_MAX(dir))
2637 		return -EMLINK;
2638 
2639 	err = dquot_initialize(dir);
2640 	if (err)
2641 		return err;
2642 
2643 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2644 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2645 retry:
2646 	inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2647 					    &dentry->d_name,
2648 					    0, NULL, EXT4_HT_DIR, credits);
2649 	handle = ext4_journal_current_handle();
2650 	err = PTR_ERR(inode);
2651 	if (IS_ERR(inode))
2652 		goto out_stop;
2653 
2654 	inode->i_op = &ext4_dir_inode_operations;
2655 	inode->i_fop = &ext4_dir_operations;
2656 	err = ext4_init_new_dir(handle, dir, inode);
2657 	if (err)
2658 		goto out_clear_inode;
2659 	err = ext4_mark_inode_dirty(handle, inode);
2660 	if (!err)
2661 		err = ext4_add_entry(handle, dentry, inode);
2662 	if (err) {
2663 out_clear_inode:
2664 		clear_nlink(inode);
2665 		unlock_new_inode(inode);
2666 		ext4_mark_inode_dirty(handle, inode);
2667 		iput(inode);
2668 		goto out_stop;
2669 	}
2670 	ext4_inc_count(handle, dir);
2671 	ext4_update_dx_flag(dir);
2672 	err = ext4_mark_inode_dirty(handle, dir);
2673 	if (err)
2674 		goto out_clear_inode;
2675 	unlock_new_inode(inode);
2676 	d_instantiate(dentry, inode);
2677 	if (IS_DIRSYNC(dir))
2678 		ext4_handle_sync(handle);
2679 
2680 out_stop:
2681 	if (handle)
2682 		ext4_journal_stop(handle);
2683 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2684 		goto retry;
2685 	return err;
2686 }
2687 
2688 /*
2689  * routine to check that the specified directory is empty (for rmdir)
2690  */
2691 bool ext4_empty_dir(struct inode *inode)
2692 {
2693 	unsigned int offset;
2694 	struct buffer_head *bh;
2695 	struct ext4_dir_entry_2 *de, *de1;
2696 	struct super_block *sb;
2697 
2698 	if (ext4_has_inline_data(inode)) {
2699 		int has_inline_data = 1;
2700 		int ret;
2701 
2702 		ret = empty_inline_dir(inode, &has_inline_data);
2703 		if (has_inline_data)
2704 			return ret;
2705 	}
2706 
2707 	sb = inode->i_sb;
2708 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2709 		EXT4_ERROR_INODE(inode, "invalid size");
2710 		return true;
2711 	}
2712 	bh = ext4_read_dirblock(inode, 0, EITHER);
2713 	if (IS_ERR(bh))
2714 		return true;
2715 
2716 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2717 	de1 = ext4_next_entry(de, sb->s_blocksize);
2718 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2719 			le32_to_cpu(de1->inode) == 0 ||
2720 			strcmp(".", de->name) || strcmp("..", de1->name)) {
2721 		ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2722 		brelse(bh);
2723 		return true;
2724 	}
2725 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2726 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2727 	de = ext4_next_entry(de1, sb->s_blocksize);
2728 	while (offset < inode->i_size) {
2729 		if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2730 			unsigned int lblock;
2731 			brelse(bh);
2732 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2733 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2734 			if (IS_ERR(bh))
2735 				return true;
2736 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2737 		}
2738 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2739 					 bh->b_data, bh->b_size, offset)) {
2740 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2741 							 sb->s_blocksize);
2742 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2743 			continue;
2744 		}
2745 		if (le32_to_cpu(de->inode)) {
2746 			brelse(bh);
2747 			return false;
2748 		}
2749 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2750 		de = ext4_next_entry(de, sb->s_blocksize);
2751 	}
2752 	brelse(bh);
2753 	return true;
2754 }
2755 
2756 /*
2757  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2758  * such inodes, starting at the superblock, in case we crash before the
2759  * file is closed/deleted, or in case the inode truncate spans multiple
2760  * transactions and the last transaction is not recovered after a crash.
2761  *
2762  * At filesystem recovery time, we walk this list deleting unlinked
2763  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2764  *
2765  * Orphan list manipulation functions must be called under i_mutex unless
2766  * we are just creating the inode or deleting it.
2767  */
2768 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2769 {
2770 	struct super_block *sb = inode->i_sb;
2771 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2772 	struct ext4_iloc iloc;
2773 	int err = 0, rc;
2774 	bool dirty = false;
2775 
2776 	if (!sbi->s_journal || is_bad_inode(inode))
2777 		return 0;
2778 
2779 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2780 		     !inode_is_locked(inode));
2781 	/*
2782 	 * Exit early if inode already is on orphan list. This is a big speedup
2783 	 * since we don't have to contend on the global s_orphan_lock.
2784 	 */
2785 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2786 		return 0;
2787 
2788 	/*
2789 	 * Orphan handling is only valid for files with data blocks
2790 	 * being truncated, or files being unlinked. Note that we either
2791 	 * hold i_mutex, or the inode can not be referenced from outside,
2792 	 * so i_nlink should not be bumped due to race
2793 	 */
2794 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2795 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2796 
2797 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2798 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2799 	if (err)
2800 		goto out;
2801 
2802 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2803 	if (err)
2804 		goto out;
2805 
2806 	mutex_lock(&sbi->s_orphan_lock);
2807 	/*
2808 	 * Due to previous errors inode may be already a part of on-disk
2809 	 * orphan list. If so skip on-disk list modification.
2810 	 */
2811 	if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2812 	    (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2813 		/* Insert this inode at the head of the on-disk orphan list */
2814 		NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2815 		sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2816 		dirty = true;
2817 	}
2818 	list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2819 	mutex_unlock(&sbi->s_orphan_lock);
2820 
2821 	if (dirty) {
2822 		err = ext4_handle_dirty_super(handle, sb);
2823 		rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2824 		if (!err)
2825 			err = rc;
2826 		if (err) {
2827 			/*
2828 			 * We have to remove inode from in-memory list if
2829 			 * addition to on disk orphan list failed. Stray orphan
2830 			 * list entries can cause panics at unmount time.
2831 			 */
2832 			mutex_lock(&sbi->s_orphan_lock);
2833 			list_del_init(&EXT4_I(inode)->i_orphan);
2834 			mutex_unlock(&sbi->s_orphan_lock);
2835 		}
2836 	}
2837 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2838 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2839 			inode->i_ino, NEXT_ORPHAN(inode));
2840 out:
2841 	ext4_std_error(sb, err);
2842 	return err;
2843 }
2844 
2845 /*
2846  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2847  * of such inodes stored on disk, because it is finally being cleaned up.
2848  */
2849 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2850 {
2851 	struct list_head *prev;
2852 	struct ext4_inode_info *ei = EXT4_I(inode);
2853 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2854 	__u32 ino_next;
2855 	struct ext4_iloc iloc;
2856 	int err = 0;
2857 
2858 	if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2859 		return 0;
2860 
2861 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2862 		     !inode_is_locked(inode));
2863 	/* Do this quick check before taking global s_orphan_lock. */
2864 	if (list_empty(&ei->i_orphan))
2865 		return 0;
2866 
2867 	if (handle) {
2868 		/* Grab inode buffer early before taking global s_orphan_lock */
2869 		err = ext4_reserve_inode_write(handle, inode, &iloc);
2870 	}
2871 
2872 	mutex_lock(&sbi->s_orphan_lock);
2873 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2874 
2875 	prev = ei->i_orphan.prev;
2876 	list_del_init(&ei->i_orphan);
2877 
2878 	/* If we're on an error path, we may not have a valid
2879 	 * transaction handle with which to update the orphan list on
2880 	 * disk, but we still need to remove the inode from the linked
2881 	 * list in memory. */
2882 	if (!handle || err) {
2883 		mutex_unlock(&sbi->s_orphan_lock);
2884 		goto out_err;
2885 	}
2886 
2887 	ino_next = NEXT_ORPHAN(inode);
2888 	if (prev == &sbi->s_orphan) {
2889 		jbd_debug(4, "superblock will point to %u\n", ino_next);
2890 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2891 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2892 		if (err) {
2893 			mutex_unlock(&sbi->s_orphan_lock);
2894 			goto out_brelse;
2895 		}
2896 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2897 		mutex_unlock(&sbi->s_orphan_lock);
2898 		err = ext4_handle_dirty_super(handle, inode->i_sb);
2899 	} else {
2900 		struct ext4_iloc iloc2;
2901 		struct inode *i_prev =
2902 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2903 
2904 		jbd_debug(4, "orphan inode %lu will point to %u\n",
2905 			  i_prev->i_ino, ino_next);
2906 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2907 		if (err) {
2908 			mutex_unlock(&sbi->s_orphan_lock);
2909 			goto out_brelse;
2910 		}
2911 		NEXT_ORPHAN(i_prev) = ino_next;
2912 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2913 		mutex_unlock(&sbi->s_orphan_lock);
2914 	}
2915 	if (err)
2916 		goto out_brelse;
2917 	NEXT_ORPHAN(inode) = 0;
2918 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2919 out_err:
2920 	ext4_std_error(inode->i_sb, err);
2921 	return err;
2922 
2923 out_brelse:
2924 	brelse(iloc.bh);
2925 	goto out_err;
2926 }
2927 
2928 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2929 {
2930 	int retval;
2931 	struct inode *inode;
2932 	struct buffer_head *bh;
2933 	struct ext4_dir_entry_2 *de;
2934 	handle_t *handle = NULL;
2935 
2936 	/* Initialize quotas before so that eventual writes go in
2937 	 * separate transaction */
2938 	retval = dquot_initialize(dir);
2939 	if (retval)
2940 		return retval;
2941 	retval = dquot_initialize(d_inode(dentry));
2942 	if (retval)
2943 		return retval;
2944 
2945 	retval = -ENOENT;
2946 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2947 	if (IS_ERR(bh))
2948 		return PTR_ERR(bh);
2949 	if (!bh)
2950 		goto end_rmdir;
2951 
2952 	inode = d_inode(dentry);
2953 
2954 	retval = -EFSCORRUPTED;
2955 	if (le32_to_cpu(de->inode) != inode->i_ino)
2956 		goto end_rmdir;
2957 
2958 	retval = -ENOTEMPTY;
2959 	if (!ext4_empty_dir(inode))
2960 		goto end_rmdir;
2961 
2962 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2963 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2964 	if (IS_ERR(handle)) {
2965 		retval = PTR_ERR(handle);
2966 		handle = NULL;
2967 		goto end_rmdir;
2968 	}
2969 
2970 	if (IS_DIRSYNC(dir))
2971 		ext4_handle_sync(handle);
2972 
2973 	retval = ext4_delete_entry(handle, dir, de, bh);
2974 	if (retval)
2975 		goto end_rmdir;
2976 	if (!EXT4_DIR_LINK_EMPTY(inode))
2977 		ext4_warning_inode(inode,
2978 			     "empty directory '%.*s' has too many links (%u)",
2979 			     dentry->d_name.len, dentry->d_name.name,
2980 			     inode->i_nlink);
2981 	inode->i_version++;
2982 	clear_nlink(inode);
2983 	/* There's no need to set i_disksize: the fact that i_nlink is
2984 	 * zero will ensure that the right thing happens during any
2985 	 * recovery. */
2986 	inode->i_size = 0;
2987 	ext4_orphan_add(handle, inode);
2988 	inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2989 	ext4_mark_inode_dirty(handle, inode);
2990 	ext4_dec_count(handle, dir);
2991 	ext4_update_dx_flag(dir);
2992 	ext4_mark_inode_dirty(handle, dir);
2993 
2994 end_rmdir:
2995 	brelse(bh);
2996 	if (handle)
2997 		ext4_journal_stop(handle);
2998 	return retval;
2999 }
3000 
3001 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
3002 {
3003 	int retval;
3004 	struct inode *inode;
3005 	struct buffer_head *bh;
3006 	struct ext4_dir_entry_2 *de;
3007 	handle_t *handle = NULL;
3008 
3009 	trace_ext4_unlink_enter(dir, dentry);
3010 	/* Initialize quotas before so that eventual writes go
3011 	 * in separate transaction */
3012 	retval = dquot_initialize(dir);
3013 	if (retval)
3014 		return retval;
3015 	retval = dquot_initialize(d_inode(dentry));
3016 	if (retval)
3017 		return retval;
3018 
3019 	retval = -ENOENT;
3020 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3021 	if (IS_ERR(bh))
3022 		return PTR_ERR(bh);
3023 	if (!bh)
3024 		goto end_unlink;
3025 
3026 	inode = d_inode(dentry);
3027 
3028 	retval = -EFSCORRUPTED;
3029 	if (le32_to_cpu(de->inode) != inode->i_ino)
3030 		goto end_unlink;
3031 
3032 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3033 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3034 	if (IS_ERR(handle)) {
3035 		retval = PTR_ERR(handle);
3036 		handle = NULL;
3037 		goto end_unlink;
3038 	}
3039 
3040 	if (IS_DIRSYNC(dir))
3041 		ext4_handle_sync(handle);
3042 
3043 	if (inode->i_nlink == 0) {
3044 		ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3045 				   dentry->d_name.len, dentry->d_name.name);
3046 		set_nlink(inode, 1);
3047 	}
3048 	retval = ext4_delete_entry(handle, dir, de, bh);
3049 	if (retval)
3050 		goto end_unlink;
3051 	dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
3052 	ext4_update_dx_flag(dir);
3053 	ext4_mark_inode_dirty(handle, dir);
3054 	drop_nlink(inode);
3055 	if (!inode->i_nlink)
3056 		ext4_orphan_add(handle, inode);
3057 	inode->i_ctime = ext4_current_time(inode);
3058 	ext4_mark_inode_dirty(handle, inode);
3059 
3060 end_unlink:
3061 	brelse(bh);
3062 	if (handle)
3063 		ext4_journal_stop(handle);
3064 	trace_ext4_unlink_exit(dentry, retval);
3065 	return retval;
3066 }
3067 
3068 static int ext4_symlink(struct inode *dir,
3069 			struct dentry *dentry, const char *symname)
3070 {
3071 	handle_t *handle;
3072 	struct inode *inode;
3073 	int err, len = strlen(symname);
3074 	int credits;
3075 	bool encryption_required;
3076 	struct fscrypt_str disk_link;
3077 	struct fscrypt_symlink_data *sd = NULL;
3078 
3079 	disk_link.len = len + 1;
3080 	disk_link.name = (char *) symname;
3081 
3082 	encryption_required = (ext4_encrypted_inode(dir) ||
3083 			       DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb)));
3084 	if (encryption_required) {
3085 		err = fscrypt_get_encryption_info(dir);
3086 		if (err)
3087 			return err;
3088 		if (!fscrypt_has_encryption_key(dir))
3089 			return -EPERM;
3090 		disk_link.len = (fscrypt_fname_encrypted_size(dir, len) +
3091 				 sizeof(struct fscrypt_symlink_data));
3092 		sd = kzalloc(disk_link.len, GFP_KERNEL);
3093 		if (!sd)
3094 			return -ENOMEM;
3095 	}
3096 
3097 	if (disk_link.len > dir->i_sb->s_blocksize) {
3098 		err = -ENAMETOOLONG;
3099 		goto err_free_sd;
3100 	}
3101 
3102 	err = dquot_initialize(dir);
3103 	if (err)
3104 		goto err_free_sd;
3105 
3106 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3107 		/*
3108 		 * For non-fast symlinks, we just allocate inode and put it on
3109 		 * orphan list in the first transaction => we need bitmap,
3110 		 * group descriptor, sb, inode block, quota blocks, and
3111 		 * possibly selinux xattr blocks.
3112 		 */
3113 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3114 			  EXT4_XATTR_TRANS_BLOCKS;
3115 	} else {
3116 		/*
3117 		 * Fast symlink. We have to add entry to directory
3118 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3119 		 * allocate new inode (bitmap, group descriptor, inode block,
3120 		 * quota blocks, sb is already counted in previous macros).
3121 		 */
3122 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3123 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3124 	}
3125 
3126 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3127 					    &dentry->d_name, 0, NULL,
3128 					    EXT4_HT_DIR, credits);
3129 	handle = ext4_journal_current_handle();
3130 	if (IS_ERR(inode)) {
3131 		if (handle)
3132 			ext4_journal_stop(handle);
3133 		err = PTR_ERR(inode);
3134 		goto err_free_sd;
3135 	}
3136 
3137 	if (encryption_required) {
3138 		struct qstr istr;
3139 		struct fscrypt_str ostr =
3140 			FSTR_INIT(sd->encrypted_path, disk_link.len);
3141 
3142 		istr.name = (const unsigned char *) symname;
3143 		istr.len = len;
3144 		err = fscrypt_fname_usr_to_disk(inode, &istr, &ostr);
3145 		if (err)
3146 			goto err_drop_inode;
3147 		sd->len = cpu_to_le16(ostr.len);
3148 		disk_link.name = (char *) sd;
3149 		inode->i_op = &ext4_encrypted_symlink_inode_operations;
3150 	}
3151 
3152 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3153 		if (!encryption_required)
3154 			inode->i_op = &ext4_symlink_inode_operations;
3155 		inode_nohighmem(inode);
3156 		ext4_set_aops(inode);
3157 		/*
3158 		 * We cannot call page_symlink() with transaction started
3159 		 * because it calls into ext4_write_begin() which can wait
3160 		 * for transaction commit if we are running out of space
3161 		 * and thus we deadlock. So we have to stop transaction now
3162 		 * and restart it when symlink contents is written.
3163 		 *
3164 		 * To keep fs consistent in case of crash, we have to put inode
3165 		 * to orphan list in the mean time.
3166 		 */
3167 		drop_nlink(inode);
3168 		err = ext4_orphan_add(handle, inode);
3169 		ext4_journal_stop(handle);
3170 		handle = NULL;
3171 		if (err)
3172 			goto err_drop_inode;
3173 		err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3174 		if (err)
3175 			goto err_drop_inode;
3176 		/*
3177 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3178 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3179 		 */
3180 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
3181 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3182 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3183 		if (IS_ERR(handle)) {
3184 			err = PTR_ERR(handle);
3185 			handle = NULL;
3186 			goto err_drop_inode;
3187 		}
3188 		set_nlink(inode, 1);
3189 		err = ext4_orphan_del(handle, inode);
3190 		if (err)
3191 			goto err_drop_inode;
3192 	} else {
3193 		/* clear the extent format for fast symlink */
3194 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3195 		if (!encryption_required) {
3196 			inode->i_op = &ext4_fast_symlink_inode_operations;
3197 			inode->i_link = (char *)&EXT4_I(inode)->i_data;
3198 		}
3199 		memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3200 		       disk_link.len);
3201 		inode->i_size = disk_link.len - 1;
3202 	}
3203 	EXT4_I(inode)->i_disksize = inode->i_size;
3204 	err = ext4_add_nondir(handle, dentry, inode);
3205 	if (!err && IS_DIRSYNC(dir))
3206 		ext4_handle_sync(handle);
3207 
3208 	if (handle)
3209 		ext4_journal_stop(handle);
3210 	kfree(sd);
3211 	return err;
3212 err_drop_inode:
3213 	if (handle)
3214 		ext4_journal_stop(handle);
3215 	clear_nlink(inode);
3216 	unlock_new_inode(inode);
3217 	iput(inode);
3218 err_free_sd:
3219 	kfree(sd);
3220 	return err;
3221 }
3222 
3223 static int ext4_link(struct dentry *old_dentry,
3224 		     struct inode *dir, struct dentry *dentry)
3225 {
3226 	handle_t *handle;
3227 	struct inode *inode = d_inode(old_dentry);
3228 	int err, retries = 0;
3229 
3230 	if (inode->i_nlink >= EXT4_LINK_MAX)
3231 		return -EMLINK;
3232 	if (ext4_encrypted_inode(dir) &&
3233 			!fscrypt_has_permitted_context(dir, inode))
3234 		return -EPERM;
3235 
3236        if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3237 	   (!projid_eq(EXT4_I(dir)->i_projid,
3238 		       EXT4_I(old_dentry->d_inode)->i_projid)))
3239 		return -EXDEV;
3240 
3241 	err = dquot_initialize(dir);
3242 	if (err)
3243 		return err;
3244 
3245 retry:
3246 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3247 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3248 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3249 	if (IS_ERR(handle))
3250 		return PTR_ERR(handle);
3251 
3252 	if (IS_DIRSYNC(dir))
3253 		ext4_handle_sync(handle);
3254 
3255 	inode->i_ctime = ext4_current_time(inode);
3256 	ext4_inc_count(handle, inode);
3257 	ihold(inode);
3258 
3259 	err = ext4_add_entry(handle, dentry, inode);
3260 	if (!err) {
3261 		ext4_mark_inode_dirty(handle, inode);
3262 		/* this can happen only for tmpfile being
3263 		 * linked the first time
3264 		 */
3265 		if (inode->i_nlink == 1)
3266 			ext4_orphan_del(handle, inode);
3267 		d_instantiate(dentry, inode);
3268 	} else {
3269 		drop_nlink(inode);
3270 		iput(inode);
3271 	}
3272 	ext4_journal_stop(handle);
3273 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3274 		goto retry;
3275 	return err;
3276 }
3277 
3278 
3279 /*
3280  * Try to find buffer head where contains the parent block.
3281  * It should be the inode block if it is inlined or the 1st block
3282  * if it is a normal dir.
3283  */
3284 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3285 					struct inode *inode,
3286 					int *retval,
3287 					struct ext4_dir_entry_2 **parent_de,
3288 					int *inlined)
3289 {
3290 	struct buffer_head *bh;
3291 
3292 	if (!ext4_has_inline_data(inode)) {
3293 		bh = ext4_read_dirblock(inode, 0, EITHER);
3294 		if (IS_ERR(bh)) {
3295 			*retval = PTR_ERR(bh);
3296 			return NULL;
3297 		}
3298 		*parent_de = ext4_next_entry(
3299 					(struct ext4_dir_entry_2 *)bh->b_data,
3300 					inode->i_sb->s_blocksize);
3301 		return bh;
3302 	}
3303 
3304 	*inlined = 1;
3305 	return ext4_get_first_inline_block(inode, parent_de, retval);
3306 }
3307 
3308 struct ext4_renament {
3309 	struct inode *dir;
3310 	struct dentry *dentry;
3311 	struct inode *inode;
3312 	bool is_dir;
3313 	int dir_nlink_delta;
3314 
3315 	/* entry for "dentry" */
3316 	struct buffer_head *bh;
3317 	struct ext4_dir_entry_2 *de;
3318 	int inlined;
3319 
3320 	/* entry for ".." in inode if it's a directory */
3321 	struct buffer_head *dir_bh;
3322 	struct ext4_dir_entry_2 *parent_de;
3323 	int dir_inlined;
3324 };
3325 
3326 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3327 {
3328 	int retval;
3329 
3330 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3331 					      &retval, &ent->parent_de,
3332 					      &ent->dir_inlined);
3333 	if (!ent->dir_bh)
3334 		return retval;
3335 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3336 		return -EFSCORRUPTED;
3337 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3338 	return ext4_journal_get_write_access(handle, ent->dir_bh);
3339 }
3340 
3341 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3342 				  unsigned dir_ino)
3343 {
3344 	int retval;
3345 
3346 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3347 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3348 	if (!ent->dir_inlined) {
3349 		if (is_dx(ent->inode)) {
3350 			retval = ext4_handle_dirty_dx_node(handle,
3351 							   ent->inode,
3352 							   ent->dir_bh);
3353 		} else {
3354 			retval = ext4_handle_dirty_dirent_node(handle,
3355 							       ent->inode,
3356 							       ent->dir_bh);
3357 		}
3358 	} else {
3359 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3360 	}
3361 	if (retval) {
3362 		ext4_std_error(ent->dir->i_sb, retval);
3363 		return retval;
3364 	}
3365 	return 0;
3366 }
3367 
3368 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3369 		       unsigned ino, unsigned file_type)
3370 {
3371 	int retval;
3372 
3373 	BUFFER_TRACE(ent->bh, "get write access");
3374 	retval = ext4_journal_get_write_access(handle, ent->bh);
3375 	if (retval)
3376 		return retval;
3377 	ent->de->inode = cpu_to_le32(ino);
3378 	if (ext4_has_feature_filetype(ent->dir->i_sb))
3379 		ent->de->file_type = file_type;
3380 	ent->dir->i_version++;
3381 	ent->dir->i_ctime = ent->dir->i_mtime =
3382 		ext4_current_time(ent->dir);
3383 	ext4_mark_inode_dirty(handle, ent->dir);
3384 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3385 	if (!ent->inlined) {
3386 		retval = ext4_handle_dirty_dirent_node(handle,
3387 						       ent->dir, ent->bh);
3388 		if (unlikely(retval)) {
3389 			ext4_std_error(ent->dir->i_sb, retval);
3390 			return retval;
3391 		}
3392 	}
3393 	brelse(ent->bh);
3394 	ent->bh = NULL;
3395 
3396 	return 0;
3397 }
3398 
3399 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3400 				  const struct qstr *d_name)
3401 {
3402 	int retval = -ENOENT;
3403 	struct buffer_head *bh;
3404 	struct ext4_dir_entry_2 *de;
3405 
3406 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3407 	if (IS_ERR(bh))
3408 		return PTR_ERR(bh);
3409 	if (bh) {
3410 		retval = ext4_delete_entry(handle, dir, de, bh);
3411 		brelse(bh);
3412 	}
3413 	return retval;
3414 }
3415 
3416 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3417 			       int force_reread)
3418 {
3419 	int retval;
3420 	/*
3421 	 * ent->de could have moved from under us during htree split, so make
3422 	 * sure that we are deleting the right entry.  We might also be pointing
3423 	 * to a stale entry in the unused part of ent->bh so just checking inum
3424 	 * and the name isn't enough.
3425 	 */
3426 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3427 	    ent->de->name_len != ent->dentry->d_name.len ||
3428 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3429 		    ent->de->name_len) ||
3430 	    force_reread) {
3431 		retval = ext4_find_delete_entry(handle, ent->dir,
3432 						&ent->dentry->d_name);
3433 	} else {
3434 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3435 		if (retval == -ENOENT) {
3436 			retval = ext4_find_delete_entry(handle, ent->dir,
3437 							&ent->dentry->d_name);
3438 		}
3439 	}
3440 
3441 	if (retval) {
3442 		ext4_warning_inode(ent->dir,
3443 				   "Deleting old file: nlink %d, error=%d",
3444 				   ent->dir->i_nlink, retval);
3445 	}
3446 }
3447 
3448 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3449 {
3450 	if (ent->dir_nlink_delta) {
3451 		if (ent->dir_nlink_delta == -1)
3452 			ext4_dec_count(handle, ent->dir);
3453 		else
3454 			ext4_inc_count(handle, ent->dir);
3455 		ext4_mark_inode_dirty(handle, ent->dir);
3456 	}
3457 }
3458 
3459 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3460 					      int credits, handle_t **h)
3461 {
3462 	struct inode *wh;
3463 	handle_t *handle;
3464 	int retries = 0;
3465 
3466 	/*
3467 	 * for inode block, sb block, group summaries,
3468 	 * and inode bitmap
3469 	 */
3470 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3471 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3472 retry:
3473 	wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3474 					 &ent->dentry->d_name, 0, NULL,
3475 					 EXT4_HT_DIR, credits);
3476 
3477 	handle = ext4_journal_current_handle();
3478 	if (IS_ERR(wh)) {
3479 		if (handle)
3480 			ext4_journal_stop(handle);
3481 		if (PTR_ERR(wh) == -ENOSPC &&
3482 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3483 			goto retry;
3484 	} else {
3485 		*h = handle;
3486 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3487 		wh->i_op = &ext4_special_inode_operations;
3488 	}
3489 	return wh;
3490 }
3491 
3492 /*
3493  * Anybody can rename anything with this: the permission checks are left to the
3494  * higher-level routines.
3495  *
3496  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3497  * while new_{dentry,inode) refers to the destination dentry/inode
3498  * This comes from rename(const char *oldpath, const char *newpath)
3499  */
3500 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3501 		       struct inode *new_dir, struct dentry *new_dentry,
3502 		       unsigned int flags)
3503 {
3504 	handle_t *handle = NULL;
3505 	struct ext4_renament old = {
3506 		.dir = old_dir,
3507 		.dentry = old_dentry,
3508 		.inode = d_inode(old_dentry),
3509 	};
3510 	struct ext4_renament new = {
3511 		.dir = new_dir,
3512 		.dentry = new_dentry,
3513 		.inode = d_inode(new_dentry),
3514 	};
3515 	int force_reread;
3516 	int retval;
3517 	struct inode *whiteout = NULL;
3518 	int credits;
3519 	u8 old_file_type;
3520 
3521 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3522 	    (!projid_eq(EXT4_I(new_dir)->i_projid,
3523 			EXT4_I(old_dentry->d_inode)->i_projid)))
3524 		return -EXDEV;
3525 
3526 	retval = dquot_initialize(old.dir);
3527 	if (retval)
3528 		return retval;
3529 	retval = dquot_initialize(new.dir);
3530 	if (retval)
3531 		return retval;
3532 
3533 	/* Initialize quotas before so that eventual writes go
3534 	 * in separate transaction */
3535 	if (new.inode) {
3536 		retval = dquot_initialize(new.inode);
3537 		if (retval)
3538 			return retval;
3539 	}
3540 
3541 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3542 	if (IS_ERR(old.bh))
3543 		return PTR_ERR(old.bh);
3544 	/*
3545 	 *  Check for inode number is _not_ due to possible IO errors.
3546 	 *  We might rmdir the source, keep it as pwd of some process
3547 	 *  and merrily kill the link to whatever was created under the
3548 	 *  same name. Goodbye sticky bit ;-<
3549 	 */
3550 	retval = -ENOENT;
3551 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3552 		goto end_rename;
3553 
3554 	if ((old.dir != new.dir) &&
3555 	    ext4_encrypted_inode(new.dir) &&
3556 	    !fscrypt_has_permitted_context(new.dir, old.inode)) {
3557 		retval = -EPERM;
3558 		goto end_rename;
3559 	}
3560 
3561 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3562 				 &new.de, &new.inlined);
3563 	if (IS_ERR(new.bh)) {
3564 		retval = PTR_ERR(new.bh);
3565 		new.bh = NULL;
3566 		goto end_rename;
3567 	}
3568 	if (new.bh) {
3569 		if (!new.inode) {
3570 			brelse(new.bh);
3571 			new.bh = NULL;
3572 		}
3573 	}
3574 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3575 		ext4_alloc_da_blocks(old.inode);
3576 
3577 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3578 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3579 	if (!(flags & RENAME_WHITEOUT)) {
3580 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3581 		if (IS_ERR(handle)) {
3582 			retval = PTR_ERR(handle);
3583 			handle = NULL;
3584 			goto end_rename;
3585 		}
3586 	} else {
3587 		whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3588 		if (IS_ERR(whiteout)) {
3589 			retval = PTR_ERR(whiteout);
3590 			whiteout = NULL;
3591 			goto end_rename;
3592 		}
3593 	}
3594 
3595 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3596 		ext4_handle_sync(handle);
3597 
3598 	if (S_ISDIR(old.inode->i_mode)) {
3599 		if (new.inode) {
3600 			retval = -ENOTEMPTY;
3601 			if (!ext4_empty_dir(new.inode))
3602 				goto end_rename;
3603 		} else {
3604 			retval = -EMLINK;
3605 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3606 				goto end_rename;
3607 		}
3608 		retval = ext4_rename_dir_prepare(handle, &old);
3609 		if (retval)
3610 			goto end_rename;
3611 	}
3612 	/*
3613 	 * If we're renaming a file within an inline_data dir and adding or
3614 	 * setting the new dirent causes a conversion from inline_data to
3615 	 * extents/blockmap, we need to force the dirent delete code to
3616 	 * re-read the directory, or else we end up trying to delete a dirent
3617 	 * from what is now the extent tree root (or a block map).
3618 	 */
3619 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3620 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3621 
3622 	old_file_type = old.de->file_type;
3623 	if (whiteout) {
3624 		/*
3625 		 * Do this before adding a new entry, so the old entry is sure
3626 		 * to be still pointing to the valid old entry.
3627 		 */
3628 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3629 				     EXT4_FT_CHRDEV);
3630 		if (retval)
3631 			goto end_rename;
3632 		ext4_mark_inode_dirty(handle, whiteout);
3633 	}
3634 	if (!new.bh) {
3635 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3636 		if (retval)
3637 			goto end_rename;
3638 	} else {
3639 		retval = ext4_setent(handle, &new,
3640 				     old.inode->i_ino, old_file_type);
3641 		if (retval)
3642 			goto end_rename;
3643 	}
3644 	if (force_reread)
3645 		force_reread = !ext4_test_inode_flag(new.dir,
3646 						     EXT4_INODE_INLINE_DATA);
3647 
3648 	/*
3649 	 * Like most other Unix systems, set the ctime for inodes on a
3650 	 * rename.
3651 	 */
3652 	old.inode->i_ctime = ext4_current_time(old.inode);
3653 	ext4_mark_inode_dirty(handle, old.inode);
3654 
3655 	if (!whiteout) {
3656 		/*
3657 		 * ok, that's it
3658 		 */
3659 		ext4_rename_delete(handle, &old, force_reread);
3660 	}
3661 
3662 	if (new.inode) {
3663 		ext4_dec_count(handle, new.inode);
3664 		new.inode->i_ctime = ext4_current_time(new.inode);
3665 	}
3666 	old.dir->i_ctime = old.dir->i_mtime = ext4_current_time(old.dir);
3667 	ext4_update_dx_flag(old.dir);
3668 	if (old.dir_bh) {
3669 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3670 		if (retval)
3671 			goto end_rename;
3672 
3673 		ext4_dec_count(handle, old.dir);
3674 		if (new.inode) {
3675 			/* checked ext4_empty_dir above, can't have another
3676 			 * parent, ext4_dec_count() won't work for many-linked
3677 			 * dirs */
3678 			clear_nlink(new.inode);
3679 		} else {
3680 			ext4_inc_count(handle, new.dir);
3681 			ext4_update_dx_flag(new.dir);
3682 			ext4_mark_inode_dirty(handle, new.dir);
3683 		}
3684 	}
3685 	ext4_mark_inode_dirty(handle, old.dir);
3686 	if (new.inode) {
3687 		ext4_mark_inode_dirty(handle, new.inode);
3688 		if (!new.inode->i_nlink)
3689 			ext4_orphan_add(handle, new.inode);
3690 	}
3691 	retval = 0;
3692 
3693 end_rename:
3694 	brelse(old.dir_bh);
3695 	brelse(old.bh);
3696 	brelse(new.bh);
3697 	if (whiteout) {
3698 		if (retval)
3699 			drop_nlink(whiteout);
3700 		unlock_new_inode(whiteout);
3701 		iput(whiteout);
3702 	}
3703 	if (handle)
3704 		ext4_journal_stop(handle);
3705 	return retval;
3706 }
3707 
3708 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3709 			     struct inode *new_dir, struct dentry *new_dentry)
3710 {
3711 	handle_t *handle = NULL;
3712 	struct ext4_renament old = {
3713 		.dir = old_dir,
3714 		.dentry = old_dentry,
3715 		.inode = d_inode(old_dentry),
3716 	};
3717 	struct ext4_renament new = {
3718 		.dir = new_dir,
3719 		.dentry = new_dentry,
3720 		.inode = d_inode(new_dentry),
3721 	};
3722 	u8 new_file_type;
3723 	int retval;
3724 
3725 	if ((ext4_encrypted_inode(old_dir) ||
3726 	     ext4_encrypted_inode(new_dir)) &&
3727 	    (old_dir != new_dir) &&
3728 	    (!fscrypt_has_permitted_context(new_dir, old.inode) ||
3729 	     !fscrypt_has_permitted_context(old_dir, new.inode)))
3730 		return -EPERM;
3731 
3732 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3733 	     !projid_eq(EXT4_I(new_dir)->i_projid,
3734 			EXT4_I(old_dentry->d_inode)->i_projid)) ||
3735 	    (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3736 	     !projid_eq(EXT4_I(old_dir)->i_projid,
3737 			EXT4_I(new_dentry->d_inode)->i_projid)))
3738 		return -EXDEV;
3739 
3740 	retval = dquot_initialize(old.dir);
3741 	if (retval)
3742 		return retval;
3743 	retval = dquot_initialize(new.dir);
3744 	if (retval)
3745 		return retval;
3746 
3747 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3748 				 &old.de, &old.inlined);
3749 	if (IS_ERR(old.bh))
3750 		return PTR_ERR(old.bh);
3751 	/*
3752 	 *  Check for inode number is _not_ due to possible IO errors.
3753 	 *  We might rmdir the source, keep it as pwd of some process
3754 	 *  and merrily kill the link to whatever was created under the
3755 	 *  same name. Goodbye sticky bit ;-<
3756 	 */
3757 	retval = -ENOENT;
3758 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3759 		goto end_rename;
3760 
3761 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3762 				 &new.de, &new.inlined);
3763 	if (IS_ERR(new.bh)) {
3764 		retval = PTR_ERR(new.bh);
3765 		new.bh = NULL;
3766 		goto end_rename;
3767 	}
3768 
3769 	/* RENAME_EXCHANGE case: old *and* new must both exist */
3770 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3771 		goto end_rename;
3772 
3773 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3774 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3775 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3776 	if (IS_ERR(handle)) {
3777 		retval = PTR_ERR(handle);
3778 		handle = NULL;
3779 		goto end_rename;
3780 	}
3781 
3782 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3783 		ext4_handle_sync(handle);
3784 
3785 	if (S_ISDIR(old.inode->i_mode)) {
3786 		old.is_dir = true;
3787 		retval = ext4_rename_dir_prepare(handle, &old);
3788 		if (retval)
3789 			goto end_rename;
3790 	}
3791 	if (S_ISDIR(new.inode->i_mode)) {
3792 		new.is_dir = true;
3793 		retval = ext4_rename_dir_prepare(handle, &new);
3794 		if (retval)
3795 			goto end_rename;
3796 	}
3797 
3798 	/*
3799 	 * Other than the special case of overwriting a directory, parents'
3800 	 * nlink only needs to be modified if this is a cross directory rename.
3801 	 */
3802 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
3803 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
3804 		new.dir_nlink_delta = -old.dir_nlink_delta;
3805 		retval = -EMLINK;
3806 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3807 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3808 			goto end_rename;
3809 	}
3810 
3811 	new_file_type = new.de->file_type;
3812 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3813 	if (retval)
3814 		goto end_rename;
3815 
3816 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3817 	if (retval)
3818 		goto end_rename;
3819 
3820 	/*
3821 	 * Like most other Unix systems, set the ctime for inodes on a
3822 	 * rename.
3823 	 */
3824 	old.inode->i_ctime = ext4_current_time(old.inode);
3825 	new.inode->i_ctime = ext4_current_time(new.inode);
3826 	ext4_mark_inode_dirty(handle, old.inode);
3827 	ext4_mark_inode_dirty(handle, new.inode);
3828 
3829 	if (old.dir_bh) {
3830 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3831 		if (retval)
3832 			goto end_rename;
3833 	}
3834 	if (new.dir_bh) {
3835 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3836 		if (retval)
3837 			goto end_rename;
3838 	}
3839 	ext4_update_dir_count(handle, &old);
3840 	ext4_update_dir_count(handle, &new);
3841 	retval = 0;
3842 
3843 end_rename:
3844 	brelse(old.dir_bh);
3845 	brelse(new.dir_bh);
3846 	brelse(old.bh);
3847 	brelse(new.bh);
3848 	if (handle)
3849 		ext4_journal_stop(handle);
3850 	return retval;
3851 }
3852 
3853 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3854 			struct inode *new_dir, struct dentry *new_dentry,
3855 			unsigned int flags)
3856 {
3857 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3858 		return -EINVAL;
3859 
3860 	if (flags & RENAME_EXCHANGE) {
3861 		return ext4_cross_rename(old_dir, old_dentry,
3862 					 new_dir, new_dentry);
3863 	}
3864 
3865 	return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3866 }
3867 
3868 /*
3869  * directories can handle most operations...
3870  */
3871 const struct inode_operations ext4_dir_inode_operations = {
3872 	.create		= ext4_create,
3873 	.lookup		= ext4_lookup,
3874 	.link		= ext4_link,
3875 	.unlink		= ext4_unlink,
3876 	.symlink	= ext4_symlink,
3877 	.mkdir		= ext4_mkdir,
3878 	.rmdir		= ext4_rmdir,
3879 	.mknod		= ext4_mknod,
3880 	.tmpfile	= ext4_tmpfile,
3881 	.rename		= ext4_rename2,
3882 	.setattr	= ext4_setattr,
3883 	.listxattr	= ext4_listxattr,
3884 	.get_acl	= ext4_get_acl,
3885 	.set_acl	= ext4_set_acl,
3886 	.fiemap         = ext4_fiemap,
3887 };
3888 
3889 const struct inode_operations ext4_special_inode_operations = {
3890 	.setattr	= ext4_setattr,
3891 	.listxattr	= ext4_listxattr,
3892 	.get_acl	= ext4_get_acl,
3893 	.set_acl	= ext4_set_acl,
3894 };
3895