xref: /openbmc/linux/fs/ext4/namei.c (revision e29dc460)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/ext4/namei.c
4  *
5  * Copyright (C) 1992, 1993, 1994, 1995
6  * Remy Card (card@masi.ibp.fr)
7  * Laboratoire MASI - Institut Blaise Pascal
8  * Universite Pierre et Marie Curie (Paris VI)
9  *
10  *  from
11  *
12  *  linux/fs/minix/namei.c
13  *
14  *  Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  *  Big-endian to little-endian byte-swapping/bitmaps by
17  *        David S. Miller (davem@caip.rutgers.edu), 1995
18  *  Directory entry file type support and forward compatibility hooks
19  *	for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
20  *  Hash Tree Directory indexing (c)
21  *	Daniel Phillips, 2001
22  *  Hash Tree Directory indexing porting
23  *	Christopher Li, 2002
24  *  Hash Tree Directory indexing cleanup
25  *	Theodore Ts'o, 2002
26  */
27 
28 #include <linux/fs.h>
29 #include <linux/pagemap.h>
30 #include <linux/time.h>
31 #include <linux/fcntl.h>
32 #include <linux/stat.h>
33 #include <linux/string.h>
34 #include <linux/quotaops.h>
35 #include <linux/buffer_head.h>
36 #include <linux/bio.h>
37 #include <linux/iversion.h>
38 #include "ext4.h"
39 #include "ext4_jbd2.h"
40 
41 #include "xattr.h"
42 #include "acl.h"
43 
44 #include <trace/events/ext4.h>
45 /*
46  * define how far ahead to read directories while searching them.
47  */
48 #define NAMEI_RA_CHUNKS  2
49 #define NAMEI_RA_BLOCKS  4
50 #define NAMEI_RA_SIZE	     (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
51 
52 static struct buffer_head *ext4_append(handle_t *handle,
53 					struct inode *inode,
54 					ext4_lblk_t *block)
55 {
56 	struct buffer_head *bh;
57 	int err;
58 
59 	if (unlikely(EXT4_SB(inode->i_sb)->s_max_dir_size_kb &&
60 		     ((inode->i_size >> 10) >=
61 		      EXT4_SB(inode->i_sb)->s_max_dir_size_kb)))
62 		return ERR_PTR(-ENOSPC);
63 
64 	*block = inode->i_size >> inode->i_sb->s_blocksize_bits;
65 
66 	bh = ext4_bread(handle, inode, *block, EXT4_GET_BLOCKS_CREATE);
67 	if (IS_ERR(bh))
68 		return bh;
69 	inode->i_size += inode->i_sb->s_blocksize;
70 	EXT4_I(inode)->i_disksize = inode->i_size;
71 	BUFFER_TRACE(bh, "get_write_access");
72 	err = ext4_journal_get_write_access(handle, bh);
73 	if (err) {
74 		brelse(bh);
75 		ext4_std_error(inode->i_sb, err);
76 		return ERR_PTR(err);
77 	}
78 	return bh;
79 }
80 
81 static int ext4_dx_csum_verify(struct inode *inode,
82 			       struct ext4_dir_entry *dirent);
83 
84 typedef enum {
85 	EITHER, INDEX, DIRENT
86 } dirblock_type_t;
87 
88 #define ext4_read_dirblock(inode, block, type) \
89 	__ext4_read_dirblock((inode), (block), (type), __func__, __LINE__)
90 
91 static struct buffer_head *__ext4_read_dirblock(struct inode *inode,
92 						ext4_lblk_t block,
93 						dirblock_type_t type,
94 						const char *func,
95 						unsigned int line)
96 {
97 	struct buffer_head *bh;
98 	struct ext4_dir_entry *dirent;
99 	int is_dx_block = 0;
100 
101 	bh = ext4_bread(NULL, inode, block, 0);
102 	if (IS_ERR(bh)) {
103 		__ext4_warning(inode->i_sb, func, line,
104 			       "inode #%lu: lblock %lu: comm %s: "
105 			       "error %ld reading directory block",
106 			       inode->i_ino, (unsigned long)block,
107 			       current->comm, PTR_ERR(bh));
108 
109 		return bh;
110 	}
111 	if (!bh) {
112 		ext4_error_inode(inode, func, line, block,
113 				 "Directory hole found");
114 		return ERR_PTR(-EFSCORRUPTED);
115 	}
116 	dirent = (struct ext4_dir_entry *) bh->b_data;
117 	/* Determine whether or not we have an index block */
118 	if (is_dx(inode)) {
119 		if (block == 0)
120 			is_dx_block = 1;
121 		else if (ext4_rec_len_from_disk(dirent->rec_len,
122 						inode->i_sb->s_blocksize) ==
123 			 inode->i_sb->s_blocksize)
124 			is_dx_block = 1;
125 	}
126 	if (!is_dx_block && type == INDEX) {
127 		ext4_error_inode(inode, func, line, block,
128 		       "directory leaf block found instead of index block");
129 		return ERR_PTR(-EFSCORRUPTED);
130 	}
131 	if (!ext4_has_metadata_csum(inode->i_sb) ||
132 	    buffer_verified(bh))
133 		return bh;
134 
135 	/*
136 	 * An empty leaf block can get mistaken for a index block; for
137 	 * this reason, we can only check the index checksum when the
138 	 * caller is sure it should be an index block.
139 	 */
140 	if (is_dx_block && type == INDEX) {
141 		if (ext4_dx_csum_verify(inode, dirent))
142 			set_buffer_verified(bh);
143 		else {
144 			ext4_error_inode(inode, func, line, block,
145 					 "Directory index failed checksum");
146 			brelse(bh);
147 			return ERR_PTR(-EFSBADCRC);
148 		}
149 	}
150 	if (!is_dx_block) {
151 		if (ext4_dirent_csum_verify(inode, dirent))
152 			set_buffer_verified(bh);
153 		else {
154 			ext4_error_inode(inode, func, line, block,
155 					 "Directory block failed checksum");
156 			brelse(bh);
157 			return ERR_PTR(-EFSBADCRC);
158 		}
159 	}
160 	return bh;
161 }
162 
163 #ifndef assert
164 #define assert(test) J_ASSERT(test)
165 #endif
166 
167 #ifdef DX_DEBUG
168 #define dxtrace(command) command
169 #else
170 #define dxtrace(command)
171 #endif
172 
173 struct fake_dirent
174 {
175 	__le32 inode;
176 	__le16 rec_len;
177 	u8 name_len;
178 	u8 file_type;
179 };
180 
181 struct dx_countlimit
182 {
183 	__le16 limit;
184 	__le16 count;
185 };
186 
187 struct dx_entry
188 {
189 	__le32 hash;
190 	__le32 block;
191 };
192 
193 /*
194  * dx_root_info is laid out so that if it should somehow get overlaid by a
195  * dirent the two low bits of the hash version will be zero.  Therefore, the
196  * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
197  */
198 
199 struct dx_root
200 {
201 	struct fake_dirent dot;
202 	char dot_name[4];
203 	struct fake_dirent dotdot;
204 	char dotdot_name[4];
205 	struct dx_root_info
206 	{
207 		__le32 reserved_zero;
208 		u8 hash_version;
209 		u8 info_length; /* 8 */
210 		u8 indirect_levels;
211 		u8 unused_flags;
212 	}
213 	info;
214 	struct dx_entry	entries[0];
215 };
216 
217 struct dx_node
218 {
219 	struct fake_dirent fake;
220 	struct dx_entry	entries[0];
221 };
222 
223 
224 struct dx_frame
225 {
226 	struct buffer_head *bh;
227 	struct dx_entry *entries;
228 	struct dx_entry *at;
229 };
230 
231 struct dx_map_entry
232 {
233 	u32 hash;
234 	u16 offs;
235 	u16 size;
236 };
237 
238 /*
239  * This goes at the end of each htree block.
240  */
241 struct dx_tail {
242 	u32 dt_reserved;
243 	__le32 dt_checksum;	/* crc32c(uuid+inum+dirblock) */
244 };
245 
246 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
247 static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
248 static inline unsigned dx_get_hash(struct dx_entry *entry);
249 static void dx_set_hash(struct dx_entry *entry, unsigned value);
250 static unsigned dx_get_count(struct dx_entry *entries);
251 static unsigned dx_get_limit(struct dx_entry *entries);
252 static void dx_set_count(struct dx_entry *entries, unsigned value);
253 static void dx_set_limit(struct dx_entry *entries, unsigned value);
254 static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
255 static unsigned dx_node_limit(struct inode *dir);
256 static struct dx_frame *dx_probe(struct ext4_filename *fname,
257 				 struct inode *dir,
258 				 struct dx_hash_info *hinfo,
259 				 struct dx_frame *frame);
260 static void dx_release(struct dx_frame *frames);
261 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
262 		       unsigned blocksize, struct dx_hash_info *hinfo,
263 		       struct dx_map_entry map[]);
264 static void dx_sort_map(struct dx_map_entry *map, unsigned count);
265 static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
266 		struct dx_map_entry *offsets, int count, unsigned blocksize);
267 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
268 static void dx_insert_block(struct dx_frame *frame,
269 					u32 hash, ext4_lblk_t block);
270 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
271 				 struct dx_frame *frame,
272 				 struct dx_frame *frames,
273 				 __u32 *start_hash);
274 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
275 		struct ext4_filename *fname,
276 		struct ext4_dir_entry_2 **res_dir);
277 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
278 			     struct inode *dir, struct inode *inode);
279 
280 /* checksumming functions */
281 void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
282 			    unsigned int blocksize)
283 {
284 	memset(t, 0, sizeof(struct ext4_dir_entry_tail));
285 	t->det_rec_len = ext4_rec_len_to_disk(
286 			sizeof(struct ext4_dir_entry_tail), blocksize);
287 	t->det_reserved_ft = EXT4_FT_DIR_CSUM;
288 }
289 
290 /* Walk through a dirent block to find a checksum "dirent" at the tail */
291 static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
292 						   struct ext4_dir_entry *de)
293 {
294 	struct ext4_dir_entry_tail *t;
295 
296 #ifdef PARANOID
297 	struct ext4_dir_entry *d, *top;
298 
299 	d = de;
300 	top = (struct ext4_dir_entry *)(((void *)de) +
301 		(EXT4_BLOCK_SIZE(inode->i_sb) -
302 		sizeof(struct ext4_dir_entry_tail)));
303 	while (d < top && d->rec_len)
304 		d = (struct ext4_dir_entry *)(((void *)d) +
305 		    le16_to_cpu(d->rec_len));
306 
307 	if (d != top)
308 		return NULL;
309 
310 	t = (struct ext4_dir_entry_tail *)d;
311 #else
312 	t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
313 #endif
314 
315 	if (t->det_reserved_zero1 ||
316 	    le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
317 	    t->det_reserved_zero2 ||
318 	    t->det_reserved_ft != EXT4_FT_DIR_CSUM)
319 		return NULL;
320 
321 	return t;
322 }
323 
324 static __le32 ext4_dirent_csum(struct inode *inode,
325 			       struct ext4_dir_entry *dirent, int size)
326 {
327 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
328 	struct ext4_inode_info *ei = EXT4_I(inode);
329 	__u32 csum;
330 
331 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
332 	return cpu_to_le32(csum);
333 }
334 
335 #define warn_no_space_for_csum(inode)					\
336 	__warn_no_space_for_csum((inode), __func__, __LINE__)
337 
338 static void __warn_no_space_for_csum(struct inode *inode, const char *func,
339 				     unsigned int line)
340 {
341 	__ext4_warning_inode(inode, func, line,
342 		"No space for directory leaf checksum. Please run e2fsck -D.");
343 }
344 
345 int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
346 {
347 	struct ext4_dir_entry_tail *t;
348 
349 	if (!ext4_has_metadata_csum(inode->i_sb))
350 		return 1;
351 
352 	t = get_dirent_tail(inode, dirent);
353 	if (!t) {
354 		warn_no_space_for_csum(inode);
355 		return 0;
356 	}
357 
358 	if (t->det_checksum != ext4_dirent_csum(inode, dirent,
359 						(void *)t - (void *)dirent))
360 		return 0;
361 
362 	return 1;
363 }
364 
365 static void ext4_dirent_csum_set(struct inode *inode,
366 				 struct ext4_dir_entry *dirent)
367 {
368 	struct ext4_dir_entry_tail *t;
369 
370 	if (!ext4_has_metadata_csum(inode->i_sb))
371 		return;
372 
373 	t = get_dirent_tail(inode, dirent);
374 	if (!t) {
375 		warn_no_space_for_csum(inode);
376 		return;
377 	}
378 
379 	t->det_checksum = ext4_dirent_csum(inode, dirent,
380 					   (void *)t - (void *)dirent);
381 }
382 
383 int ext4_handle_dirty_dirent_node(handle_t *handle,
384 				  struct inode *inode,
385 				  struct buffer_head *bh)
386 {
387 	ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
388 	return ext4_handle_dirty_metadata(handle, inode, bh);
389 }
390 
391 static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
392 					       struct ext4_dir_entry *dirent,
393 					       int *offset)
394 {
395 	struct ext4_dir_entry *dp;
396 	struct dx_root_info *root;
397 	int count_offset;
398 
399 	if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
400 		count_offset = 8;
401 	else if (le16_to_cpu(dirent->rec_len) == 12) {
402 		dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
403 		if (le16_to_cpu(dp->rec_len) !=
404 		    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
405 			return NULL;
406 		root = (struct dx_root_info *)(((void *)dp + 12));
407 		if (root->reserved_zero ||
408 		    root->info_length != sizeof(struct dx_root_info))
409 			return NULL;
410 		count_offset = 32;
411 	} else
412 		return NULL;
413 
414 	if (offset)
415 		*offset = count_offset;
416 	return (struct dx_countlimit *)(((void *)dirent) + count_offset);
417 }
418 
419 static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
420 			   int count_offset, int count, struct dx_tail *t)
421 {
422 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
423 	struct ext4_inode_info *ei = EXT4_I(inode);
424 	__u32 csum;
425 	int size;
426 	__u32 dummy_csum = 0;
427 	int offset = offsetof(struct dx_tail, dt_checksum);
428 
429 	size = count_offset + (count * sizeof(struct dx_entry));
430 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
431 	csum = ext4_chksum(sbi, csum, (__u8 *)t, offset);
432 	csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
433 
434 	return cpu_to_le32(csum);
435 }
436 
437 static int ext4_dx_csum_verify(struct inode *inode,
438 			       struct ext4_dir_entry *dirent)
439 {
440 	struct dx_countlimit *c;
441 	struct dx_tail *t;
442 	int count_offset, limit, count;
443 
444 	if (!ext4_has_metadata_csum(inode->i_sb))
445 		return 1;
446 
447 	c = get_dx_countlimit(inode, dirent, &count_offset);
448 	if (!c) {
449 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
450 		return 0;
451 	}
452 	limit = le16_to_cpu(c->limit);
453 	count = le16_to_cpu(c->count);
454 	if (count_offset + (limit * sizeof(struct dx_entry)) >
455 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
456 		warn_no_space_for_csum(inode);
457 		return 0;
458 	}
459 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
460 
461 	if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
462 					    count, t))
463 		return 0;
464 	return 1;
465 }
466 
467 static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
468 {
469 	struct dx_countlimit *c;
470 	struct dx_tail *t;
471 	int count_offset, limit, count;
472 
473 	if (!ext4_has_metadata_csum(inode->i_sb))
474 		return;
475 
476 	c = get_dx_countlimit(inode, dirent, &count_offset);
477 	if (!c) {
478 		EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
479 		return;
480 	}
481 	limit = le16_to_cpu(c->limit);
482 	count = le16_to_cpu(c->count);
483 	if (count_offset + (limit * sizeof(struct dx_entry)) >
484 	    EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
485 		warn_no_space_for_csum(inode);
486 		return;
487 	}
488 	t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
489 
490 	t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
491 }
492 
493 static inline int ext4_handle_dirty_dx_node(handle_t *handle,
494 					    struct inode *inode,
495 					    struct buffer_head *bh)
496 {
497 	ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
498 	return ext4_handle_dirty_metadata(handle, inode, bh);
499 }
500 
501 /*
502  * p is at least 6 bytes before the end of page
503  */
504 static inline struct ext4_dir_entry_2 *
505 ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
506 {
507 	return (struct ext4_dir_entry_2 *)((char *)p +
508 		ext4_rec_len_from_disk(p->rec_len, blocksize));
509 }
510 
511 /*
512  * Future: use high four bits of block for coalesce-on-delete flags
513  * Mask them off for now.
514  */
515 
516 static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
517 {
518 	return le32_to_cpu(entry->block) & 0x0fffffff;
519 }
520 
521 static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
522 {
523 	entry->block = cpu_to_le32(value);
524 }
525 
526 static inline unsigned dx_get_hash(struct dx_entry *entry)
527 {
528 	return le32_to_cpu(entry->hash);
529 }
530 
531 static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
532 {
533 	entry->hash = cpu_to_le32(value);
534 }
535 
536 static inline unsigned dx_get_count(struct dx_entry *entries)
537 {
538 	return le16_to_cpu(((struct dx_countlimit *) entries)->count);
539 }
540 
541 static inline unsigned dx_get_limit(struct dx_entry *entries)
542 {
543 	return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
544 }
545 
546 static inline void dx_set_count(struct dx_entry *entries, unsigned value)
547 {
548 	((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
549 }
550 
551 static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
552 {
553 	((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
554 }
555 
556 static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
557 {
558 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
559 		EXT4_DIR_REC_LEN(2) - infosize;
560 
561 	if (ext4_has_metadata_csum(dir->i_sb))
562 		entry_space -= sizeof(struct dx_tail);
563 	return entry_space / sizeof(struct dx_entry);
564 }
565 
566 static inline unsigned dx_node_limit(struct inode *dir)
567 {
568 	unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
569 
570 	if (ext4_has_metadata_csum(dir->i_sb))
571 		entry_space -= sizeof(struct dx_tail);
572 	return entry_space / sizeof(struct dx_entry);
573 }
574 
575 /*
576  * Debug
577  */
578 #ifdef DX_DEBUG
579 static void dx_show_index(char * label, struct dx_entry *entries)
580 {
581 	int i, n = dx_get_count (entries);
582 	printk(KERN_DEBUG "%s index", label);
583 	for (i = 0; i < n; i++) {
584 		printk(KERN_CONT " %x->%lu",
585 		       i ? dx_get_hash(entries + i) : 0,
586 		       (unsigned long)dx_get_block(entries + i));
587 	}
588 	printk(KERN_CONT "\n");
589 }
590 
591 struct stats
592 {
593 	unsigned names;
594 	unsigned space;
595 	unsigned bcount;
596 };
597 
598 static struct stats dx_show_leaf(struct inode *dir,
599 				struct dx_hash_info *hinfo,
600 				struct ext4_dir_entry_2 *de,
601 				int size, int show_names)
602 {
603 	unsigned names = 0, space = 0;
604 	char *base = (char *) de;
605 	struct dx_hash_info h = *hinfo;
606 
607 	printk("names: ");
608 	while ((char *) de < base + size)
609 	{
610 		if (de->inode)
611 		{
612 			if (show_names)
613 			{
614 #ifdef CONFIG_EXT4_FS_ENCRYPTION
615 				int len;
616 				char *name;
617 				struct fscrypt_str fname_crypto_str =
618 					FSTR_INIT(NULL, 0);
619 				int res = 0;
620 
621 				name  = de->name;
622 				len = de->name_len;
623 				if (ext4_encrypted_inode(dir))
624 					res = fscrypt_get_encryption_info(dir);
625 				if (res) {
626 					printk(KERN_WARNING "Error setting up"
627 					       " fname crypto: %d\n", res);
628 				}
629 				if (!fscrypt_has_encryption_key(dir)) {
630 					/* Directory is not encrypted */
631 					ext4fs_dirhash(de->name,
632 						de->name_len, &h);
633 					printk("%*.s:(U)%x.%u ", len,
634 					       name, h.hash,
635 					       (unsigned) ((char *) de
636 							   - base));
637 				} else {
638 					struct fscrypt_str de_name =
639 						FSTR_INIT(name, len);
640 
641 					/* Directory is encrypted */
642 					res = fscrypt_fname_alloc_buffer(
643 						dir, len,
644 						&fname_crypto_str);
645 					if (res)
646 						printk(KERN_WARNING "Error "
647 							"allocating crypto "
648 							"buffer--skipping "
649 							"crypto\n");
650 					res = fscrypt_fname_disk_to_usr(dir,
651 						0, 0, &de_name,
652 						&fname_crypto_str);
653 					if (res) {
654 						printk(KERN_WARNING "Error "
655 							"converting filename "
656 							"from disk to usr"
657 							"\n");
658 						name = "??";
659 						len = 2;
660 					} else {
661 						name = fname_crypto_str.name;
662 						len = fname_crypto_str.len;
663 					}
664 					ext4fs_dirhash(de->name, de->name_len,
665 						       &h);
666 					printk("%*.s:(E)%x.%u ", len, name,
667 					       h.hash, (unsigned) ((char *) de
668 								   - base));
669 					fscrypt_fname_free_buffer(
670 							&fname_crypto_str);
671 				}
672 #else
673 				int len = de->name_len;
674 				char *name = de->name;
675 				ext4fs_dirhash(de->name, de->name_len, &h);
676 				printk("%*.s:%x.%u ", len, name, h.hash,
677 				       (unsigned) ((char *) de - base));
678 #endif
679 			}
680 			space += EXT4_DIR_REC_LEN(de->name_len);
681 			names++;
682 		}
683 		de = ext4_next_entry(de, size);
684 	}
685 	printk(KERN_CONT "(%i)\n", names);
686 	return (struct stats) { names, space, 1 };
687 }
688 
689 struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
690 			     struct dx_entry *entries, int levels)
691 {
692 	unsigned blocksize = dir->i_sb->s_blocksize;
693 	unsigned count = dx_get_count(entries), names = 0, space = 0, i;
694 	unsigned bcount = 0;
695 	struct buffer_head *bh;
696 	printk("%i indexed blocks...\n", count);
697 	for (i = 0; i < count; i++, entries++)
698 	{
699 		ext4_lblk_t block = dx_get_block(entries);
700 		ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
701 		u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
702 		struct stats stats;
703 		printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
704 		bh = ext4_bread(NULL,dir, block, 0);
705 		if (!bh || IS_ERR(bh))
706 			continue;
707 		stats = levels?
708 		   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
709 		   dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *)
710 			bh->b_data, blocksize, 0);
711 		names += stats.names;
712 		space += stats.space;
713 		bcount += stats.bcount;
714 		brelse(bh);
715 	}
716 	if (bcount)
717 		printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
718 		       levels ? "" : "   ", names, space/bcount,
719 		       (space/bcount)*100/blocksize);
720 	return (struct stats) { names, space, bcount};
721 }
722 #endif /* DX_DEBUG */
723 
724 /*
725  * Probe for a directory leaf block to search.
726  *
727  * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
728  * error in the directory index, and the caller should fall back to
729  * searching the directory normally.  The callers of dx_probe **MUST**
730  * check for this error code, and make sure it never gets reflected
731  * back to userspace.
732  */
733 static struct dx_frame *
734 dx_probe(struct ext4_filename *fname, struct inode *dir,
735 	 struct dx_hash_info *hinfo, struct dx_frame *frame_in)
736 {
737 	unsigned count, indirect;
738 	struct dx_entry *at, *entries, *p, *q, *m;
739 	struct dx_root *root;
740 	struct dx_frame *frame = frame_in;
741 	struct dx_frame *ret_err = ERR_PTR(ERR_BAD_DX_DIR);
742 	u32 hash;
743 
744 	memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
745 	frame->bh = ext4_read_dirblock(dir, 0, INDEX);
746 	if (IS_ERR(frame->bh))
747 		return (struct dx_frame *) frame->bh;
748 
749 	root = (struct dx_root *) frame->bh->b_data;
750 	if (root->info.hash_version != DX_HASH_TEA &&
751 	    root->info.hash_version != DX_HASH_HALF_MD4 &&
752 	    root->info.hash_version != DX_HASH_LEGACY) {
753 		ext4_warning_inode(dir, "Unrecognised inode hash code %u",
754 				   root->info.hash_version);
755 		goto fail;
756 	}
757 	if (fname)
758 		hinfo = &fname->hinfo;
759 	hinfo->hash_version = root->info.hash_version;
760 	if (hinfo->hash_version <= DX_HASH_TEA)
761 		hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
762 	hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
763 	if (fname && fname_name(fname))
764 		ext4fs_dirhash(fname_name(fname), fname_len(fname), hinfo);
765 	hash = hinfo->hash;
766 
767 	if (root->info.unused_flags & 1) {
768 		ext4_warning_inode(dir, "Unimplemented hash flags: %#06x",
769 				   root->info.unused_flags);
770 		goto fail;
771 	}
772 
773 	indirect = root->info.indirect_levels;
774 	if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
775 		ext4_warning(dir->i_sb,
776 			     "Directory (ino: %lu) htree depth %#06x exceed"
777 			     "supported value", dir->i_ino,
778 			     ext4_dir_htree_level(dir->i_sb));
779 		if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
780 			ext4_warning(dir->i_sb, "Enable large directory "
781 						"feature to access it");
782 		}
783 		goto fail;
784 	}
785 
786 	entries = (struct dx_entry *)(((char *)&root->info) +
787 				      root->info.info_length);
788 
789 	if (dx_get_limit(entries) != dx_root_limit(dir,
790 						   root->info.info_length)) {
791 		ext4_warning_inode(dir, "dx entry: limit %u != root limit %u",
792 				   dx_get_limit(entries),
793 				   dx_root_limit(dir, root->info.info_length));
794 		goto fail;
795 	}
796 
797 	dxtrace(printk("Look up %x", hash));
798 	while (1) {
799 		count = dx_get_count(entries);
800 		if (!count || count > dx_get_limit(entries)) {
801 			ext4_warning_inode(dir,
802 					   "dx entry: count %u beyond limit %u",
803 					   count, dx_get_limit(entries));
804 			goto fail;
805 		}
806 
807 		p = entries + 1;
808 		q = entries + count - 1;
809 		while (p <= q) {
810 			m = p + (q - p) / 2;
811 			dxtrace(printk(KERN_CONT "."));
812 			if (dx_get_hash(m) > hash)
813 				q = m - 1;
814 			else
815 				p = m + 1;
816 		}
817 
818 		if (0) { // linear search cross check
819 			unsigned n = count - 1;
820 			at = entries;
821 			while (n--)
822 			{
823 				dxtrace(printk(KERN_CONT ","));
824 				if (dx_get_hash(++at) > hash)
825 				{
826 					at--;
827 					break;
828 				}
829 			}
830 			assert (at == p - 1);
831 		}
832 
833 		at = p - 1;
834 		dxtrace(printk(KERN_CONT " %x->%u\n",
835 			       at == entries ? 0 : dx_get_hash(at),
836 			       dx_get_block(at)));
837 		frame->entries = entries;
838 		frame->at = at;
839 		if (!indirect--)
840 			return frame;
841 		frame++;
842 		frame->bh = ext4_read_dirblock(dir, dx_get_block(at), INDEX);
843 		if (IS_ERR(frame->bh)) {
844 			ret_err = (struct dx_frame *) frame->bh;
845 			frame->bh = NULL;
846 			goto fail;
847 		}
848 		entries = ((struct dx_node *) frame->bh->b_data)->entries;
849 
850 		if (dx_get_limit(entries) != dx_node_limit(dir)) {
851 			ext4_warning_inode(dir,
852 				"dx entry: limit %u != node limit %u",
853 				dx_get_limit(entries), dx_node_limit(dir));
854 			goto fail;
855 		}
856 	}
857 fail:
858 	while (frame >= frame_in) {
859 		brelse(frame->bh);
860 		frame--;
861 	}
862 
863 	if (ret_err == ERR_PTR(ERR_BAD_DX_DIR))
864 		ext4_warning_inode(dir,
865 			"Corrupt directory, running e2fsck is recommended");
866 	return ret_err;
867 }
868 
869 static void dx_release(struct dx_frame *frames)
870 {
871 	struct dx_root_info *info;
872 	int i;
873 
874 	if (frames[0].bh == NULL)
875 		return;
876 
877 	info = &((struct dx_root *)frames[0].bh->b_data)->info;
878 	for (i = 0; i <= info->indirect_levels; i++) {
879 		if (frames[i].bh == NULL)
880 			break;
881 		brelse(frames[i].bh);
882 		frames[i].bh = NULL;
883 	}
884 }
885 
886 /*
887  * This function increments the frame pointer to search the next leaf
888  * block, and reads in the necessary intervening nodes if the search
889  * should be necessary.  Whether or not the search is necessary is
890  * controlled by the hash parameter.  If the hash value is even, then
891  * the search is only continued if the next block starts with that
892  * hash value.  This is used if we are searching for a specific file.
893  *
894  * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
895  *
896  * This function returns 1 if the caller should continue to search,
897  * or 0 if it should not.  If there is an error reading one of the
898  * index blocks, it will a negative error code.
899  *
900  * If start_hash is non-null, it will be filled in with the starting
901  * hash of the next page.
902  */
903 static int ext4_htree_next_block(struct inode *dir, __u32 hash,
904 				 struct dx_frame *frame,
905 				 struct dx_frame *frames,
906 				 __u32 *start_hash)
907 {
908 	struct dx_frame *p;
909 	struct buffer_head *bh;
910 	int num_frames = 0;
911 	__u32 bhash;
912 
913 	p = frame;
914 	/*
915 	 * Find the next leaf page by incrementing the frame pointer.
916 	 * If we run out of entries in the interior node, loop around and
917 	 * increment pointer in the parent node.  When we break out of
918 	 * this loop, num_frames indicates the number of interior
919 	 * nodes need to be read.
920 	 */
921 	while (1) {
922 		if (++(p->at) < p->entries + dx_get_count(p->entries))
923 			break;
924 		if (p == frames)
925 			return 0;
926 		num_frames++;
927 		p--;
928 	}
929 
930 	/*
931 	 * If the hash is 1, then continue only if the next page has a
932 	 * continuation hash of any value.  This is used for readdir
933 	 * handling.  Otherwise, check to see if the hash matches the
934 	 * desired contiuation hash.  If it doesn't, return since
935 	 * there's no point to read in the successive index pages.
936 	 */
937 	bhash = dx_get_hash(p->at);
938 	if (start_hash)
939 		*start_hash = bhash;
940 	if ((hash & 1) == 0) {
941 		if ((bhash & ~1) != hash)
942 			return 0;
943 	}
944 	/*
945 	 * If the hash is HASH_NB_ALWAYS, we always go to the next
946 	 * block so no check is necessary
947 	 */
948 	while (num_frames--) {
949 		bh = ext4_read_dirblock(dir, dx_get_block(p->at), INDEX);
950 		if (IS_ERR(bh))
951 			return PTR_ERR(bh);
952 		p++;
953 		brelse(p->bh);
954 		p->bh = bh;
955 		p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
956 	}
957 	return 1;
958 }
959 
960 
961 /*
962  * This function fills a red-black tree with information from a
963  * directory block.  It returns the number directory entries loaded
964  * into the tree.  If there is an error it is returned in err.
965  */
966 static int htree_dirblock_to_tree(struct file *dir_file,
967 				  struct inode *dir, ext4_lblk_t block,
968 				  struct dx_hash_info *hinfo,
969 				  __u32 start_hash, __u32 start_minor_hash)
970 {
971 	struct buffer_head *bh;
972 	struct ext4_dir_entry_2 *de, *top;
973 	int err = 0, count = 0;
974 	struct fscrypt_str fname_crypto_str = FSTR_INIT(NULL, 0), tmp_str;
975 
976 	dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
977 							(unsigned long)block));
978 	bh = ext4_read_dirblock(dir, block, DIRENT);
979 	if (IS_ERR(bh))
980 		return PTR_ERR(bh);
981 
982 	de = (struct ext4_dir_entry_2 *) bh->b_data;
983 	top = (struct ext4_dir_entry_2 *) ((char *) de +
984 					   dir->i_sb->s_blocksize -
985 					   EXT4_DIR_REC_LEN(0));
986 #ifdef CONFIG_EXT4_FS_ENCRYPTION
987 	/* Check if the directory is encrypted */
988 	if (ext4_encrypted_inode(dir)) {
989 		err = fscrypt_get_encryption_info(dir);
990 		if (err < 0) {
991 			brelse(bh);
992 			return err;
993 		}
994 		err = fscrypt_fname_alloc_buffer(dir, EXT4_NAME_LEN,
995 						     &fname_crypto_str);
996 		if (err < 0) {
997 			brelse(bh);
998 			return err;
999 		}
1000 	}
1001 #endif
1002 	for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
1003 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1004 				bh->b_data, bh->b_size,
1005 				(block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
1006 					 + ((char *)de - bh->b_data))) {
1007 			/* silently ignore the rest of the block */
1008 			break;
1009 		}
1010 		ext4fs_dirhash(de->name, de->name_len, hinfo);
1011 		if ((hinfo->hash < start_hash) ||
1012 		    ((hinfo->hash == start_hash) &&
1013 		     (hinfo->minor_hash < start_minor_hash)))
1014 			continue;
1015 		if (de->inode == 0)
1016 			continue;
1017 		if (!ext4_encrypted_inode(dir)) {
1018 			tmp_str.name = de->name;
1019 			tmp_str.len = de->name_len;
1020 			err = ext4_htree_store_dirent(dir_file,
1021 				   hinfo->hash, hinfo->minor_hash, de,
1022 				   &tmp_str);
1023 		} else {
1024 			int save_len = fname_crypto_str.len;
1025 			struct fscrypt_str de_name = FSTR_INIT(de->name,
1026 								de->name_len);
1027 
1028 			/* Directory is encrypted */
1029 			err = fscrypt_fname_disk_to_usr(dir, hinfo->hash,
1030 					hinfo->minor_hash, &de_name,
1031 					&fname_crypto_str);
1032 			if (err) {
1033 				count = err;
1034 				goto errout;
1035 			}
1036 			err = ext4_htree_store_dirent(dir_file,
1037 				   hinfo->hash, hinfo->minor_hash, de,
1038 					&fname_crypto_str);
1039 			fname_crypto_str.len = save_len;
1040 		}
1041 		if (err != 0) {
1042 			count = err;
1043 			goto errout;
1044 		}
1045 		count++;
1046 	}
1047 errout:
1048 	brelse(bh);
1049 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1050 	fscrypt_fname_free_buffer(&fname_crypto_str);
1051 #endif
1052 	return count;
1053 }
1054 
1055 
1056 /*
1057  * This function fills a red-black tree with information from a
1058  * directory.  We start scanning the directory in hash order, starting
1059  * at start_hash and start_minor_hash.
1060  *
1061  * This function returns the number of entries inserted into the tree,
1062  * or a negative error code.
1063  */
1064 int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
1065 			 __u32 start_minor_hash, __u32 *next_hash)
1066 {
1067 	struct dx_hash_info hinfo;
1068 	struct ext4_dir_entry_2 *de;
1069 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1070 	struct inode *dir;
1071 	ext4_lblk_t block;
1072 	int count = 0;
1073 	int ret, err;
1074 	__u32 hashval;
1075 	struct fscrypt_str tmp_str;
1076 
1077 	dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
1078 		       start_hash, start_minor_hash));
1079 	dir = file_inode(dir_file);
1080 	if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
1081 		hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1082 		if (hinfo.hash_version <= DX_HASH_TEA)
1083 			hinfo.hash_version +=
1084 				EXT4_SB(dir->i_sb)->s_hash_unsigned;
1085 		hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1086 		if (ext4_has_inline_data(dir)) {
1087 			int has_inline_data = 1;
1088 			count = htree_inlinedir_to_tree(dir_file, dir, 0,
1089 							&hinfo, start_hash,
1090 							start_minor_hash,
1091 							&has_inline_data);
1092 			if (has_inline_data) {
1093 				*next_hash = ~0;
1094 				return count;
1095 			}
1096 		}
1097 		count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
1098 					       start_hash, start_minor_hash);
1099 		*next_hash = ~0;
1100 		return count;
1101 	}
1102 	hinfo.hash = start_hash;
1103 	hinfo.minor_hash = 0;
1104 	frame = dx_probe(NULL, dir, &hinfo, frames);
1105 	if (IS_ERR(frame))
1106 		return PTR_ERR(frame);
1107 
1108 	/* Add '.' and '..' from the htree header */
1109 	if (!start_hash && !start_minor_hash) {
1110 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1111 		tmp_str.name = de->name;
1112 		tmp_str.len = de->name_len;
1113 		err = ext4_htree_store_dirent(dir_file, 0, 0,
1114 					      de, &tmp_str);
1115 		if (err != 0)
1116 			goto errout;
1117 		count++;
1118 	}
1119 	if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
1120 		de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
1121 		de = ext4_next_entry(de, dir->i_sb->s_blocksize);
1122 		tmp_str.name = de->name;
1123 		tmp_str.len = de->name_len;
1124 		err = ext4_htree_store_dirent(dir_file, 2, 0,
1125 					      de, &tmp_str);
1126 		if (err != 0)
1127 			goto errout;
1128 		count++;
1129 	}
1130 
1131 	while (1) {
1132 		if (fatal_signal_pending(current)) {
1133 			err = -ERESTARTSYS;
1134 			goto errout;
1135 		}
1136 		cond_resched();
1137 		block = dx_get_block(frame->at);
1138 		ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
1139 					     start_hash, start_minor_hash);
1140 		if (ret < 0) {
1141 			err = ret;
1142 			goto errout;
1143 		}
1144 		count += ret;
1145 		hashval = ~0;
1146 		ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
1147 					    frame, frames, &hashval);
1148 		*next_hash = hashval;
1149 		if (ret < 0) {
1150 			err = ret;
1151 			goto errout;
1152 		}
1153 		/*
1154 		 * Stop if:  (a) there are no more entries, or
1155 		 * (b) we have inserted at least one entry and the
1156 		 * next hash value is not a continuation
1157 		 */
1158 		if ((ret == 0) ||
1159 		    (count && ((hashval & 1) == 0)))
1160 			break;
1161 	}
1162 	dx_release(frames);
1163 	dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
1164 		       "next hash: %x\n", count, *next_hash));
1165 	return count;
1166 errout:
1167 	dx_release(frames);
1168 	return (err);
1169 }
1170 
1171 static inline int search_dirblock(struct buffer_head *bh,
1172 				  struct inode *dir,
1173 				  struct ext4_filename *fname,
1174 				  unsigned int offset,
1175 				  struct ext4_dir_entry_2 **res_dir)
1176 {
1177 	return ext4_search_dir(bh, bh->b_data, dir->i_sb->s_blocksize, dir,
1178 			       fname, offset, res_dir);
1179 }
1180 
1181 /*
1182  * Directory block splitting, compacting
1183  */
1184 
1185 /*
1186  * Create map of hash values, offsets, and sizes, stored at end of block.
1187  * Returns number of entries mapped.
1188  */
1189 static int dx_make_map(struct inode *dir, struct ext4_dir_entry_2 *de,
1190 		       unsigned blocksize, struct dx_hash_info *hinfo,
1191 		       struct dx_map_entry *map_tail)
1192 {
1193 	int count = 0;
1194 	char *base = (char *) de;
1195 	struct dx_hash_info h = *hinfo;
1196 
1197 	while ((char *) de < base + blocksize) {
1198 		if (de->name_len && de->inode) {
1199 			ext4fs_dirhash(de->name, de->name_len, &h);
1200 			map_tail--;
1201 			map_tail->hash = h.hash;
1202 			map_tail->offs = ((char *) de - base)>>2;
1203 			map_tail->size = le16_to_cpu(de->rec_len);
1204 			count++;
1205 			cond_resched();
1206 		}
1207 		/* XXX: do we need to check rec_len == 0 case? -Chris */
1208 		de = ext4_next_entry(de, blocksize);
1209 	}
1210 	return count;
1211 }
1212 
1213 /* Sort map by hash value */
1214 static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1215 {
1216 	struct dx_map_entry *p, *q, *top = map + count - 1;
1217 	int more;
1218 	/* Combsort until bubble sort doesn't suck */
1219 	while (count > 2) {
1220 		count = count*10/13;
1221 		if (count - 9 < 2) /* 9, 10 -> 11 */
1222 			count = 11;
1223 		for (p = top, q = p - count; q >= map; p--, q--)
1224 			if (p->hash < q->hash)
1225 				swap(*p, *q);
1226 	}
1227 	/* Garden variety bubble sort */
1228 	do {
1229 		more = 0;
1230 		q = top;
1231 		while (q-- > map) {
1232 			if (q[1].hash >= q[0].hash)
1233 				continue;
1234 			swap(*(q+1), *q);
1235 			more = 1;
1236 		}
1237 	} while(more);
1238 }
1239 
1240 static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1241 {
1242 	struct dx_entry *entries = frame->entries;
1243 	struct dx_entry *old = frame->at, *new = old + 1;
1244 	int count = dx_get_count(entries);
1245 
1246 	assert(count < dx_get_limit(entries));
1247 	assert(old < entries + count);
1248 	memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1249 	dx_set_hash(new, hash);
1250 	dx_set_block(new, block);
1251 	dx_set_count(entries, count + 1);
1252 }
1253 
1254 /*
1255  * Test whether a directory entry matches the filename being searched for.
1256  *
1257  * Return: %true if the directory entry matches, otherwise %false.
1258  */
1259 static inline bool ext4_match(const struct ext4_filename *fname,
1260 			      const struct ext4_dir_entry_2 *de)
1261 {
1262 	struct fscrypt_name f;
1263 
1264 	if (!de->inode)
1265 		return false;
1266 
1267 	f.usr_fname = fname->usr_fname;
1268 	f.disk_name = fname->disk_name;
1269 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1270 	f.crypto_buf = fname->crypto_buf;
1271 #endif
1272 	return fscrypt_match_name(&f, de->name, de->name_len);
1273 }
1274 
1275 /*
1276  * Returns 0 if not found, -1 on failure, and 1 on success
1277  */
1278 int ext4_search_dir(struct buffer_head *bh, char *search_buf, int buf_size,
1279 		    struct inode *dir, struct ext4_filename *fname,
1280 		    unsigned int offset, struct ext4_dir_entry_2 **res_dir)
1281 {
1282 	struct ext4_dir_entry_2 * de;
1283 	char * dlimit;
1284 	int de_len;
1285 
1286 	de = (struct ext4_dir_entry_2 *)search_buf;
1287 	dlimit = search_buf + buf_size;
1288 	while ((char *) de < dlimit) {
1289 		/* this code is executed quadratically often */
1290 		/* do minimal checking `by hand' */
1291 		if ((char *) de + de->name_len <= dlimit &&
1292 		    ext4_match(fname, de)) {
1293 			/* found a match - just to be sure, do
1294 			 * a full check */
1295 			if (ext4_check_dir_entry(dir, NULL, de, bh, bh->b_data,
1296 						 bh->b_size, offset))
1297 				return -1;
1298 			*res_dir = de;
1299 			return 1;
1300 		}
1301 		/* prevent looping on a bad block */
1302 		de_len = ext4_rec_len_from_disk(de->rec_len,
1303 						dir->i_sb->s_blocksize);
1304 		if (de_len <= 0)
1305 			return -1;
1306 		offset += de_len;
1307 		de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1308 	}
1309 	return 0;
1310 }
1311 
1312 static int is_dx_internal_node(struct inode *dir, ext4_lblk_t block,
1313 			       struct ext4_dir_entry *de)
1314 {
1315 	struct super_block *sb = dir->i_sb;
1316 
1317 	if (!is_dx(dir))
1318 		return 0;
1319 	if (block == 0)
1320 		return 1;
1321 	if (de->inode == 0 &&
1322 	    ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) ==
1323 			sb->s_blocksize)
1324 		return 1;
1325 	return 0;
1326 }
1327 
1328 /*
1329  *	ext4_find_entry()
1330  *
1331  * finds an entry in the specified directory with the wanted name. It
1332  * returns the cache buffer in which the entry was found, and the entry
1333  * itself (as a parameter - res_dir). It does NOT read the inode of the
1334  * entry - you'll have to do that yourself if you want to.
1335  *
1336  * The returned buffer_head has ->b_count elevated.  The caller is expected
1337  * to brelse() it when appropriate.
1338  */
1339 static struct buffer_head * ext4_find_entry (struct inode *dir,
1340 					const struct qstr *d_name,
1341 					struct ext4_dir_entry_2 **res_dir,
1342 					int *inlined)
1343 {
1344 	struct super_block *sb;
1345 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1346 	struct buffer_head *bh, *ret = NULL;
1347 	ext4_lblk_t start, block;
1348 	const u8 *name = d_name->name;
1349 	size_t ra_max = 0;	/* Number of bh's in the readahead
1350 				   buffer, bh_use[] */
1351 	size_t ra_ptr = 0;	/* Current index into readahead
1352 				   buffer */
1353 	ext4_lblk_t  nblocks;
1354 	int i, namelen, retval;
1355 	struct ext4_filename fname;
1356 
1357 	*res_dir = NULL;
1358 	sb = dir->i_sb;
1359 	namelen = d_name->len;
1360 	if (namelen > EXT4_NAME_LEN)
1361 		return NULL;
1362 
1363 	retval = ext4_fname_setup_filename(dir, d_name, 1, &fname);
1364 	if (retval == -ENOENT)
1365 		return NULL;
1366 	if (retval)
1367 		return ERR_PTR(retval);
1368 
1369 	if (ext4_has_inline_data(dir)) {
1370 		int has_inline_data = 1;
1371 		ret = ext4_find_inline_entry(dir, &fname, res_dir,
1372 					     &has_inline_data);
1373 		if (has_inline_data) {
1374 			if (inlined)
1375 				*inlined = 1;
1376 			goto cleanup_and_exit;
1377 		}
1378 	}
1379 
1380 	if ((namelen <= 2) && (name[0] == '.') &&
1381 	    (name[1] == '.' || name[1] == '\0')) {
1382 		/*
1383 		 * "." or ".." will only be in the first block
1384 		 * NFS may look up ".."; "." should be handled by the VFS
1385 		 */
1386 		block = start = 0;
1387 		nblocks = 1;
1388 		goto restart;
1389 	}
1390 	if (is_dx(dir)) {
1391 		ret = ext4_dx_find_entry(dir, &fname, res_dir);
1392 		/*
1393 		 * On success, or if the error was file not found,
1394 		 * return.  Otherwise, fall back to doing a search the
1395 		 * old fashioned way.
1396 		 */
1397 		if (!IS_ERR(ret) || PTR_ERR(ret) != ERR_BAD_DX_DIR)
1398 			goto cleanup_and_exit;
1399 		dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1400 			       "falling back\n"));
1401 	}
1402 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1403 	if (!nblocks) {
1404 		ret = NULL;
1405 		goto cleanup_and_exit;
1406 	}
1407 	start = EXT4_I(dir)->i_dir_start_lookup;
1408 	if (start >= nblocks)
1409 		start = 0;
1410 	block = start;
1411 restart:
1412 	do {
1413 		/*
1414 		 * We deal with the read-ahead logic here.
1415 		 */
1416 		if (ra_ptr >= ra_max) {
1417 			/* Refill the readahead buffer */
1418 			ra_ptr = 0;
1419 			if (block < start)
1420 				ra_max = start - block;
1421 			else
1422 				ra_max = nblocks - block;
1423 			ra_max = min(ra_max, ARRAY_SIZE(bh_use));
1424 			retval = ext4_bread_batch(dir, block, ra_max,
1425 						  false /* wait */, bh_use);
1426 			if (retval) {
1427 				ret = ERR_PTR(retval);
1428 				ra_max = 0;
1429 				goto cleanup_and_exit;
1430 			}
1431 		}
1432 		if ((bh = bh_use[ra_ptr++]) == NULL)
1433 			goto next;
1434 		wait_on_buffer(bh);
1435 		if (!buffer_uptodate(bh)) {
1436 			EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1437 					 (unsigned long) block);
1438 			brelse(bh);
1439 			ret = ERR_PTR(-EIO);
1440 			goto cleanup_and_exit;
1441 		}
1442 		if (!buffer_verified(bh) &&
1443 		    !is_dx_internal_node(dir, block,
1444 					 (struct ext4_dir_entry *)bh->b_data) &&
1445 		    !ext4_dirent_csum_verify(dir,
1446 				(struct ext4_dir_entry *)bh->b_data)) {
1447 			EXT4_ERROR_INODE(dir, "checksumming directory "
1448 					 "block %lu", (unsigned long)block);
1449 			brelse(bh);
1450 			ret = ERR_PTR(-EFSBADCRC);
1451 			goto cleanup_and_exit;
1452 		}
1453 		set_buffer_verified(bh);
1454 		i = search_dirblock(bh, dir, &fname,
1455 			    block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1456 		if (i == 1) {
1457 			EXT4_I(dir)->i_dir_start_lookup = block;
1458 			ret = bh;
1459 			goto cleanup_and_exit;
1460 		} else {
1461 			brelse(bh);
1462 			if (i < 0)
1463 				goto cleanup_and_exit;
1464 		}
1465 	next:
1466 		if (++block >= nblocks)
1467 			block = 0;
1468 	} while (block != start);
1469 
1470 	/*
1471 	 * If the directory has grown while we were searching, then
1472 	 * search the last part of the directory before giving up.
1473 	 */
1474 	block = nblocks;
1475 	nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1476 	if (block < nblocks) {
1477 		start = 0;
1478 		goto restart;
1479 	}
1480 
1481 cleanup_and_exit:
1482 	/* Clean up the read-ahead blocks */
1483 	for (; ra_ptr < ra_max; ra_ptr++)
1484 		brelse(bh_use[ra_ptr]);
1485 	ext4_fname_free_filename(&fname);
1486 	return ret;
1487 }
1488 
1489 static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
1490 			struct ext4_filename *fname,
1491 			struct ext4_dir_entry_2 **res_dir)
1492 {
1493 	struct super_block * sb = dir->i_sb;
1494 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
1495 	struct buffer_head *bh;
1496 	ext4_lblk_t block;
1497 	int retval;
1498 
1499 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1500 	*res_dir = NULL;
1501 #endif
1502 	frame = dx_probe(fname, dir, NULL, frames);
1503 	if (IS_ERR(frame))
1504 		return (struct buffer_head *) frame;
1505 	do {
1506 		block = dx_get_block(frame->at);
1507 		bh = ext4_read_dirblock(dir, block, DIRENT);
1508 		if (IS_ERR(bh))
1509 			goto errout;
1510 
1511 		retval = search_dirblock(bh, dir, fname,
1512 					 block << EXT4_BLOCK_SIZE_BITS(sb),
1513 					 res_dir);
1514 		if (retval == 1)
1515 			goto success;
1516 		brelse(bh);
1517 		if (retval == -1) {
1518 			bh = ERR_PTR(ERR_BAD_DX_DIR);
1519 			goto errout;
1520 		}
1521 
1522 		/* Check to see if we should continue to search */
1523 		retval = ext4_htree_next_block(dir, fname->hinfo.hash, frame,
1524 					       frames, NULL);
1525 		if (retval < 0) {
1526 			ext4_warning_inode(dir,
1527 				"error %d reading directory index block",
1528 				retval);
1529 			bh = ERR_PTR(retval);
1530 			goto errout;
1531 		}
1532 	} while (retval == 1);
1533 
1534 	bh = NULL;
1535 errout:
1536 	dxtrace(printk(KERN_DEBUG "%s not found\n", fname->usr_fname->name));
1537 success:
1538 	dx_release(frames);
1539 	return bh;
1540 }
1541 
1542 static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1543 {
1544 	struct inode *inode;
1545 	struct ext4_dir_entry_2 *de;
1546 	struct buffer_head *bh;
1547 	int err;
1548 
1549 	err = fscrypt_prepare_lookup(dir, dentry, flags);
1550 	if (err)
1551 		return ERR_PTR(err);
1552 
1553 	if (dentry->d_name.len > EXT4_NAME_LEN)
1554 		return ERR_PTR(-ENAMETOOLONG);
1555 
1556 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
1557 	if (IS_ERR(bh))
1558 		return (struct dentry *) bh;
1559 	inode = NULL;
1560 	if (bh) {
1561 		__u32 ino = le32_to_cpu(de->inode);
1562 		brelse(bh);
1563 		if (!ext4_valid_inum(dir->i_sb, ino)) {
1564 			EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1565 			return ERR_PTR(-EFSCORRUPTED);
1566 		}
1567 		if (unlikely(ino == dir->i_ino)) {
1568 			EXT4_ERROR_INODE(dir, "'%pd' linked to parent dir",
1569 					 dentry);
1570 			return ERR_PTR(-EFSCORRUPTED);
1571 		}
1572 		inode = ext4_iget_normal(dir->i_sb, ino);
1573 		if (inode == ERR_PTR(-ESTALE)) {
1574 			EXT4_ERROR_INODE(dir,
1575 					 "deleted inode referenced: %u",
1576 					 ino);
1577 			return ERR_PTR(-EFSCORRUPTED);
1578 		}
1579 		if (!IS_ERR(inode) && ext4_encrypted_inode(dir) &&
1580 		    (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) &&
1581 		    !fscrypt_has_permitted_context(dir, inode)) {
1582 			ext4_warning(inode->i_sb,
1583 				     "Inconsistent encryption contexts: %lu/%lu",
1584 				     dir->i_ino, inode->i_ino);
1585 			iput(inode);
1586 			return ERR_PTR(-EPERM);
1587 		}
1588 	}
1589 	return d_splice_alias(inode, dentry);
1590 }
1591 
1592 
1593 struct dentry *ext4_get_parent(struct dentry *child)
1594 {
1595 	__u32 ino;
1596 	static const struct qstr dotdot = QSTR_INIT("..", 2);
1597 	struct ext4_dir_entry_2 * de;
1598 	struct buffer_head *bh;
1599 
1600 	bh = ext4_find_entry(d_inode(child), &dotdot, &de, NULL);
1601 	if (IS_ERR(bh))
1602 		return (struct dentry *) bh;
1603 	if (!bh)
1604 		return ERR_PTR(-ENOENT);
1605 	ino = le32_to_cpu(de->inode);
1606 	brelse(bh);
1607 
1608 	if (!ext4_valid_inum(child->d_sb, ino)) {
1609 		EXT4_ERROR_INODE(d_inode(child),
1610 				 "bad parent inode number: %u", ino);
1611 		return ERR_PTR(-EFSCORRUPTED);
1612 	}
1613 
1614 	return d_obtain_alias(ext4_iget_normal(child->d_sb, ino));
1615 }
1616 
1617 /*
1618  * Move count entries from end of map between two memory locations.
1619  * Returns pointer to last entry moved.
1620  */
1621 static struct ext4_dir_entry_2 *
1622 dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1623 		unsigned blocksize)
1624 {
1625 	unsigned rec_len = 0;
1626 
1627 	while (count--) {
1628 		struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1629 						(from + (map->offs<<2));
1630 		rec_len = EXT4_DIR_REC_LEN(de->name_len);
1631 		memcpy (to, de, rec_len);
1632 		((struct ext4_dir_entry_2 *) to)->rec_len =
1633 				ext4_rec_len_to_disk(rec_len, blocksize);
1634 		de->inode = 0;
1635 		map++;
1636 		to += rec_len;
1637 	}
1638 	return (struct ext4_dir_entry_2 *) (to - rec_len);
1639 }
1640 
1641 /*
1642  * Compact each dir entry in the range to the minimal rec_len.
1643  * Returns pointer to last entry in range.
1644  */
1645 static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1646 {
1647 	struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1648 	unsigned rec_len = 0;
1649 
1650 	prev = to = de;
1651 	while ((char*)de < base + blocksize) {
1652 		next = ext4_next_entry(de, blocksize);
1653 		if (de->inode && de->name_len) {
1654 			rec_len = EXT4_DIR_REC_LEN(de->name_len);
1655 			if (de > to)
1656 				memmove(to, de, rec_len);
1657 			to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1658 			prev = to;
1659 			to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1660 		}
1661 		de = next;
1662 	}
1663 	return prev;
1664 }
1665 
1666 /*
1667  * Split a full leaf block to make room for a new dir entry.
1668  * Allocate a new block, and move entries so that they are approx. equally full.
1669  * Returns pointer to de in block into which the new entry will be inserted.
1670  */
1671 static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1672 			struct buffer_head **bh,struct dx_frame *frame,
1673 			struct dx_hash_info *hinfo)
1674 {
1675 	unsigned blocksize = dir->i_sb->s_blocksize;
1676 	unsigned count, continued;
1677 	struct buffer_head *bh2;
1678 	ext4_lblk_t newblock;
1679 	u32 hash2;
1680 	struct dx_map_entry *map;
1681 	char *data1 = (*bh)->b_data, *data2;
1682 	unsigned split, move, size;
1683 	struct ext4_dir_entry_2 *de = NULL, *de2;
1684 	struct ext4_dir_entry_tail *t;
1685 	int	csum_size = 0;
1686 	int	err = 0, i;
1687 
1688 	if (ext4_has_metadata_csum(dir->i_sb))
1689 		csum_size = sizeof(struct ext4_dir_entry_tail);
1690 
1691 	bh2 = ext4_append(handle, dir, &newblock);
1692 	if (IS_ERR(bh2)) {
1693 		brelse(*bh);
1694 		*bh = NULL;
1695 		return (struct ext4_dir_entry_2 *) bh2;
1696 	}
1697 
1698 	BUFFER_TRACE(*bh, "get_write_access");
1699 	err = ext4_journal_get_write_access(handle, *bh);
1700 	if (err)
1701 		goto journal_error;
1702 
1703 	BUFFER_TRACE(frame->bh, "get_write_access");
1704 	err = ext4_journal_get_write_access(handle, frame->bh);
1705 	if (err)
1706 		goto journal_error;
1707 
1708 	data2 = bh2->b_data;
1709 
1710 	/* create map in the end of data2 block */
1711 	map = (struct dx_map_entry *) (data2 + blocksize);
1712 	count = dx_make_map(dir, (struct ext4_dir_entry_2 *) data1,
1713 			     blocksize, hinfo, map);
1714 	map -= count;
1715 	dx_sort_map(map, count);
1716 	/* Split the existing block in the middle, size-wise */
1717 	size = 0;
1718 	move = 0;
1719 	for (i = count-1; i >= 0; i--) {
1720 		/* is more than half of this entry in 2nd half of the block? */
1721 		if (size + map[i].size/2 > blocksize/2)
1722 			break;
1723 		size += map[i].size;
1724 		move++;
1725 	}
1726 	/* map index at which we will split */
1727 	split = count - move;
1728 	hash2 = map[split].hash;
1729 	continued = hash2 == map[split - 1].hash;
1730 	dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1731 			(unsigned long)dx_get_block(frame->at),
1732 					hash2, split, count-split));
1733 
1734 	/* Fancy dance to stay within two buffers */
1735 	de2 = dx_move_dirents(data1, data2, map + split, count - split,
1736 			      blocksize);
1737 	de = dx_pack_dirents(data1, blocksize);
1738 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1739 					   (char *) de,
1740 					   blocksize);
1741 	de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1742 					    (char *) de2,
1743 					    blocksize);
1744 	if (csum_size) {
1745 		t = EXT4_DIRENT_TAIL(data2, blocksize);
1746 		initialize_dirent_tail(t, blocksize);
1747 
1748 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1749 		initialize_dirent_tail(t, blocksize);
1750 	}
1751 
1752 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data1,
1753 			blocksize, 1));
1754 	dxtrace(dx_show_leaf(dir, hinfo, (struct ext4_dir_entry_2 *) data2,
1755 			blocksize, 1));
1756 
1757 	/* Which block gets the new entry? */
1758 	if (hinfo->hash >= hash2) {
1759 		swap(*bh, bh2);
1760 		de = de2;
1761 	}
1762 	dx_insert_block(frame, hash2 + continued, newblock);
1763 	err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1764 	if (err)
1765 		goto journal_error;
1766 	err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1767 	if (err)
1768 		goto journal_error;
1769 	brelse(bh2);
1770 	dxtrace(dx_show_index("frame", frame->entries));
1771 	return de;
1772 
1773 journal_error:
1774 	brelse(*bh);
1775 	brelse(bh2);
1776 	*bh = NULL;
1777 	ext4_std_error(dir->i_sb, err);
1778 	return ERR_PTR(err);
1779 }
1780 
1781 int ext4_find_dest_de(struct inode *dir, struct inode *inode,
1782 		      struct buffer_head *bh,
1783 		      void *buf, int buf_size,
1784 		      struct ext4_filename *fname,
1785 		      struct ext4_dir_entry_2 **dest_de)
1786 {
1787 	struct ext4_dir_entry_2 *de;
1788 	unsigned short reclen = EXT4_DIR_REC_LEN(fname_len(fname));
1789 	int nlen, rlen;
1790 	unsigned int offset = 0;
1791 	char *top;
1792 
1793 	de = (struct ext4_dir_entry_2 *)buf;
1794 	top = buf + buf_size - reclen;
1795 	while ((char *) de <= top) {
1796 		if (ext4_check_dir_entry(dir, NULL, de, bh,
1797 					 buf, buf_size, offset))
1798 			return -EFSCORRUPTED;
1799 		if (ext4_match(fname, de))
1800 			return -EEXIST;
1801 		nlen = EXT4_DIR_REC_LEN(de->name_len);
1802 		rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1803 		if ((de->inode ? rlen - nlen : rlen) >= reclen)
1804 			break;
1805 		de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1806 		offset += rlen;
1807 	}
1808 	if ((char *) de > top)
1809 		return -ENOSPC;
1810 
1811 	*dest_de = de;
1812 	return 0;
1813 }
1814 
1815 void ext4_insert_dentry(struct inode *inode,
1816 			struct ext4_dir_entry_2 *de,
1817 			int buf_size,
1818 			struct ext4_filename *fname)
1819 {
1820 
1821 	int nlen, rlen;
1822 
1823 	nlen = EXT4_DIR_REC_LEN(de->name_len);
1824 	rlen = ext4_rec_len_from_disk(de->rec_len, buf_size);
1825 	if (de->inode) {
1826 		struct ext4_dir_entry_2 *de1 =
1827 			(struct ext4_dir_entry_2 *)((char *)de + nlen);
1828 		de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, buf_size);
1829 		de->rec_len = ext4_rec_len_to_disk(nlen, buf_size);
1830 		de = de1;
1831 	}
1832 	de->file_type = EXT4_FT_UNKNOWN;
1833 	de->inode = cpu_to_le32(inode->i_ino);
1834 	ext4_set_de_type(inode->i_sb, de, inode->i_mode);
1835 	de->name_len = fname_len(fname);
1836 	memcpy(de->name, fname_name(fname), fname_len(fname));
1837 }
1838 
1839 /*
1840  * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1841  * it points to a directory entry which is guaranteed to be large
1842  * enough for new directory entry.  If de is NULL, then
1843  * add_dirent_to_buf will attempt search the directory block for
1844  * space.  It will return -ENOSPC if no space is available, and -EIO
1845  * and -EEXIST if directory entry already exists.
1846  */
1847 static int add_dirent_to_buf(handle_t *handle, struct ext4_filename *fname,
1848 			     struct inode *dir,
1849 			     struct inode *inode, struct ext4_dir_entry_2 *de,
1850 			     struct buffer_head *bh)
1851 {
1852 	unsigned int	blocksize = dir->i_sb->s_blocksize;
1853 	int		csum_size = 0;
1854 	int		err;
1855 
1856 	if (ext4_has_metadata_csum(inode->i_sb))
1857 		csum_size = sizeof(struct ext4_dir_entry_tail);
1858 
1859 	if (!de) {
1860 		err = ext4_find_dest_de(dir, inode, bh, bh->b_data,
1861 					blocksize - csum_size, fname, &de);
1862 		if (err)
1863 			return err;
1864 	}
1865 	BUFFER_TRACE(bh, "get_write_access");
1866 	err = ext4_journal_get_write_access(handle, bh);
1867 	if (err) {
1868 		ext4_std_error(dir->i_sb, err);
1869 		return err;
1870 	}
1871 
1872 	/* By now the buffer is marked for journaling */
1873 	ext4_insert_dentry(inode, de, blocksize, fname);
1874 
1875 	/*
1876 	 * XXX shouldn't update any times until successful
1877 	 * completion of syscall, but too many callers depend
1878 	 * on this.
1879 	 *
1880 	 * XXX similarly, too many callers depend on
1881 	 * ext4_new_inode() setting the times, but error
1882 	 * recovery deletes the inode, so the worst that can
1883 	 * happen is that the times are slightly out of date
1884 	 * and/or different from the directory change time.
1885 	 */
1886 	dir->i_mtime = dir->i_ctime = current_time(dir);
1887 	ext4_update_dx_flag(dir);
1888 	inode_inc_iversion(dir);
1889 	ext4_mark_inode_dirty(handle, dir);
1890 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1891 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1892 	if (err)
1893 		ext4_std_error(dir->i_sb, err);
1894 	return 0;
1895 }
1896 
1897 /*
1898  * This converts a one block unindexed directory to a 3 block indexed
1899  * directory, and adds the dentry to the indexed directory.
1900  */
1901 static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
1902 			    struct inode *dir,
1903 			    struct inode *inode, struct buffer_head *bh)
1904 {
1905 	struct buffer_head *bh2;
1906 	struct dx_root	*root;
1907 	struct dx_frame	frames[EXT4_HTREE_LEVEL], *frame;
1908 	struct dx_entry *entries;
1909 	struct ext4_dir_entry_2	*de, *de2;
1910 	struct ext4_dir_entry_tail *t;
1911 	char		*data1, *top;
1912 	unsigned	len;
1913 	int		retval;
1914 	unsigned	blocksize;
1915 	ext4_lblk_t  block;
1916 	struct fake_dirent *fde;
1917 	int csum_size = 0;
1918 
1919 	if (ext4_has_metadata_csum(inode->i_sb))
1920 		csum_size = sizeof(struct ext4_dir_entry_tail);
1921 
1922 	blocksize =  dir->i_sb->s_blocksize;
1923 	dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1924 	BUFFER_TRACE(bh, "get_write_access");
1925 	retval = ext4_journal_get_write_access(handle, bh);
1926 	if (retval) {
1927 		ext4_std_error(dir->i_sb, retval);
1928 		brelse(bh);
1929 		return retval;
1930 	}
1931 	root = (struct dx_root *) bh->b_data;
1932 
1933 	/* The 0th block becomes the root, move the dirents out */
1934 	fde = &root->dotdot;
1935 	de = (struct ext4_dir_entry_2 *)((char *)fde +
1936 		ext4_rec_len_from_disk(fde->rec_len, blocksize));
1937 	if ((char *) de >= (((char *) root) + blocksize)) {
1938 		EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1939 		brelse(bh);
1940 		return -EFSCORRUPTED;
1941 	}
1942 	len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1943 
1944 	/* Allocate new block for the 0th block's dirents */
1945 	bh2 = ext4_append(handle, dir, &block);
1946 	if (IS_ERR(bh2)) {
1947 		brelse(bh);
1948 		return PTR_ERR(bh2);
1949 	}
1950 	ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1951 	data1 = bh2->b_data;
1952 
1953 	memcpy (data1, de, len);
1954 	de = (struct ext4_dir_entry_2 *) data1;
1955 	top = data1 + len;
1956 	while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1957 		de = de2;
1958 	de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1959 					   (char *) de,
1960 					   blocksize);
1961 
1962 	if (csum_size) {
1963 		t = EXT4_DIRENT_TAIL(data1, blocksize);
1964 		initialize_dirent_tail(t, blocksize);
1965 	}
1966 
1967 	/* Initialize the root; the dot dirents already exist */
1968 	de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1969 	de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1970 					   blocksize);
1971 	memset (&root->info, 0, sizeof(root->info));
1972 	root->info.info_length = sizeof(root->info);
1973 	root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1974 	entries = root->entries;
1975 	dx_set_block(entries, 1);
1976 	dx_set_count(entries, 1);
1977 	dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1978 
1979 	/* Initialize as for dx_probe */
1980 	fname->hinfo.hash_version = root->info.hash_version;
1981 	if (fname->hinfo.hash_version <= DX_HASH_TEA)
1982 		fname->hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1983 	fname->hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1984 	ext4fs_dirhash(fname_name(fname), fname_len(fname), &fname->hinfo);
1985 
1986 	memset(frames, 0, sizeof(frames));
1987 	frame = frames;
1988 	frame->entries = entries;
1989 	frame->at = entries;
1990 	frame->bh = bh;
1991 
1992 	retval = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1993 	if (retval)
1994 		goto out_frames;
1995 	retval = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1996 	if (retval)
1997 		goto out_frames;
1998 
1999 	de = do_split(handle,dir, &bh2, frame, &fname->hinfo);
2000 	if (IS_ERR(de)) {
2001 		retval = PTR_ERR(de);
2002 		goto out_frames;
2003 	}
2004 
2005 	retval = add_dirent_to_buf(handle, fname, dir, inode, de, bh2);
2006 out_frames:
2007 	/*
2008 	 * Even if the block split failed, we have to properly write
2009 	 * out all the changes we did so far. Otherwise we can end up
2010 	 * with corrupted filesystem.
2011 	 */
2012 	if (retval)
2013 		ext4_mark_inode_dirty(handle, dir);
2014 	dx_release(frames);
2015 	brelse(bh2);
2016 	return retval;
2017 }
2018 
2019 /*
2020  *	ext4_add_entry()
2021  *
2022  * adds a file entry to the specified directory, using the same
2023  * semantics as ext4_find_entry(). It returns NULL if it failed.
2024  *
2025  * NOTE!! The inode part of 'de' is left at 0 - which means you
2026  * may not sleep between calling this and putting something into
2027  * the entry, as someone else might have used it while you slept.
2028  */
2029 static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
2030 			  struct inode *inode)
2031 {
2032 	struct inode *dir = d_inode(dentry->d_parent);
2033 	struct buffer_head *bh = NULL;
2034 	struct ext4_dir_entry_2 *de;
2035 	struct ext4_dir_entry_tail *t;
2036 	struct super_block *sb;
2037 	struct ext4_filename fname;
2038 	int	retval;
2039 	int	dx_fallback=0;
2040 	unsigned blocksize;
2041 	ext4_lblk_t block, blocks;
2042 	int	csum_size = 0;
2043 
2044 	if (ext4_has_metadata_csum(inode->i_sb))
2045 		csum_size = sizeof(struct ext4_dir_entry_tail);
2046 
2047 	sb = dir->i_sb;
2048 	blocksize = sb->s_blocksize;
2049 	if (!dentry->d_name.len)
2050 		return -EINVAL;
2051 
2052 	retval = ext4_fname_setup_filename(dir, &dentry->d_name, 0, &fname);
2053 	if (retval)
2054 		return retval;
2055 
2056 	if (ext4_has_inline_data(dir)) {
2057 		retval = ext4_try_add_inline_entry(handle, &fname, dir, inode);
2058 		if (retval < 0)
2059 			goto out;
2060 		if (retval == 1) {
2061 			retval = 0;
2062 			goto out;
2063 		}
2064 	}
2065 
2066 	if (is_dx(dir)) {
2067 		retval = ext4_dx_add_entry(handle, &fname, dir, inode);
2068 		if (!retval || (retval != ERR_BAD_DX_DIR))
2069 			goto out;
2070 		ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
2071 		dx_fallback++;
2072 		ext4_mark_inode_dirty(handle, dir);
2073 	}
2074 	blocks = dir->i_size >> sb->s_blocksize_bits;
2075 	for (block = 0; block < blocks; block++) {
2076 		bh = ext4_read_dirblock(dir, block, DIRENT);
2077 		if (IS_ERR(bh)) {
2078 			retval = PTR_ERR(bh);
2079 			bh = NULL;
2080 			goto out;
2081 		}
2082 		retval = add_dirent_to_buf(handle, &fname, dir, inode,
2083 					   NULL, bh);
2084 		if (retval != -ENOSPC)
2085 			goto out;
2086 
2087 		if (blocks == 1 && !dx_fallback &&
2088 		    ext4_has_feature_dir_index(sb)) {
2089 			retval = make_indexed_dir(handle, &fname, dir,
2090 						  inode, bh);
2091 			bh = NULL; /* make_indexed_dir releases bh */
2092 			goto out;
2093 		}
2094 		brelse(bh);
2095 	}
2096 	bh = ext4_append(handle, dir, &block);
2097 	if (IS_ERR(bh)) {
2098 		retval = PTR_ERR(bh);
2099 		bh = NULL;
2100 		goto out;
2101 	}
2102 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2103 	de->inode = 0;
2104 	de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
2105 
2106 	if (csum_size) {
2107 		t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
2108 		initialize_dirent_tail(t, blocksize);
2109 	}
2110 
2111 	retval = add_dirent_to_buf(handle, &fname, dir, inode, de, bh);
2112 out:
2113 	ext4_fname_free_filename(&fname);
2114 	brelse(bh);
2115 	if (retval == 0)
2116 		ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
2117 	return retval;
2118 }
2119 
2120 /*
2121  * Returns 0 for success, or a negative error value
2122  */
2123 static int ext4_dx_add_entry(handle_t *handle, struct ext4_filename *fname,
2124 			     struct inode *dir, struct inode *inode)
2125 {
2126 	struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
2127 	struct dx_entry *entries, *at;
2128 	struct buffer_head *bh;
2129 	struct super_block *sb = dir->i_sb;
2130 	struct ext4_dir_entry_2 *de;
2131 	int restart;
2132 	int err;
2133 
2134 again:
2135 	restart = 0;
2136 	frame = dx_probe(fname, dir, NULL, frames);
2137 	if (IS_ERR(frame))
2138 		return PTR_ERR(frame);
2139 	entries = frame->entries;
2140 	at = frame->at;
2141 	bh = ext4_read_dirblock(dir, dx_get_block(frame->at), DIRENT);
2142 	if (IS_ERR(bh)) {
2143 		err = PTR_ERR(bh);
2144 		bh = NULL;
2145 		goto cleanup;
2146 	}
2147 
2148 	BUFFER_TRACE(bh, "get_write_access");
2149 	err = ext4_journal_get_write_access(handle, bh);
2150 	if (err)
2151 		goto journal_error;
2152 
2153 	err = add_dirent_to_buf(handle, fname, dir, inode, NULL, bh);
2154 	if (err != -ENOSPC)
2155 		goto cleanup;
2156 
2157 	err = 0;
2158 	/* Block full, should compress but for now just split */
2159 	dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
2160 		       dx_get_count(entries), dx_get_limit(entries)));
2161 	/* Need to split index? */
2162 	if (dx_get_count(entries) == dx_get_limit(entries)) {
2163 		ext4_lblk_t newblock;
2164 		int levels = frame - frames + 1;
2165 		unsigned int icount;
2166 		int add_level = 1;
2167 		struct dx_entry *entries2;
2168 		struct dx_node *node2;
2169 		struct buffer_head *bh2;
2170 
2171 		while (frame > frames) {
2172 			if (dx_get_count((frame - 1)->entries) <
2173 			    dx_get_limit((frame - 1)->entries)) {
2174 				add_level = 0;
2175 				break;
2176 			}
2177 			frame--; /* split higher index block */
2178 			at = frame->at;
2179 			entries = frame->entries;
2180 			restart = 1;
2181 		}
2182 		if (add_level && levels == ext4_dir_htree_level(sb)) {
2183 			ext4_warning(sb, "Directory (ino: %lu) index full, "
2184 					 "reach max htree level :%d",
2185 					 dir->i_ino, levels);
2186 			if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
2187 				ext4_warning(sb, "Large directory feature is "
2188 						 "not enabled on this "
2189 						 "filesystem");
2190 			}
2191 			err = -ENOSPC;
2192 			goto cleanup;
2193 		}
2194 		icount = dx_get_count(entries);
2195 		bh2 = ext4_append(handle, dir, &newblock);
2196 		if (IS_ERR(bh2)) {
2197 			err = PTR_ERR(bh2);
2198 			goto cleanup;
2199 		}
2200 		node2 = (struct dx_node *)(bh2->b_data);
2201 		entries2 = node2->entries;
2202 		memset(&node2->fake, 0, sizeof(struct fake_dirent));
2203 		node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
2204 							   sb->s_blocksize);
2205 		BUFFER_TRACE(frame->bh, "get_write_access");
2206 		err = ext4_journal_get_write_access(handle, frame->bh);
2207 		if (err)
2208 			goto journal_error;
2209 		if (!add_level) {
2210 			unsigned icount1 = icount/2, icount2 = icount - icount1;
2211 			unsigned hash2 = dx_get_hash(entries + icount1);
2212 			dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
2213 				       icount1, icount2));
2214 
2215 			BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
2216 			err = ext4_journal_get_write_access(handle,
2217 							     (frame - 1)->bh);
2218 			if (err)
2219 				goto journal_error;
2220 
2221 			memcpy((char *) entries2, (char *) (entries + icount1),
2222 			       icount2 * sizeof(struct dx_entry));
2223 			dx_set_count(entries, icount1);
2224 			dx_set_count(entries2, icount2);
2225 			dx_set_limit(entries2, dx_node_limit(dir));
2226 
2227 			/* Which index block gets the new entry? */
2228 			if (at - entries >= icount1) {
2229 				frame->at = at = at - entries - icount1 + entries2;
2230 				frame->entries = entries = entries2;
2231 				swap(frame->bh, bh2);
2232 			}
2233 			dx_insert_block((frame - 1), hash2, newblock);
2234 			dxtrace(dx_show_index("node", frame->entries));
2235 			dxtrace(dx_show_index("node",
2236 			       ((struct dx_node *) bh2->b_data)->entries));
2237 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2238 			if (err)
2239 				goto journal_error;
2240 			brelse (bh2);
2241 			err = ext4_handle_dirty_dx_node(handle, dir,
2242 						   (frame - 1)->bh);
2243 			if (err)
2244 				goto journal_error;
2245 			if (restart) {
2246 				err = ext4_handle_dirty_dx_node(handle, dir,
2247 							   frame->bh);
2248 				goto journal_error;
2249 			}
2250 		} else {
2251 			struct dx_root *dxroot;
2252 			memcpy((char *) entries2, (char *) entries,
2253 			       icount * sizeof(struct dx_entry));
2254 			dx_set_limit(entries2, dx_node_limit(dir));
2255 
2256 			/* Set up root */
2257 			dx_set_count(entries, 1);
2258 			dx_set_block(entries + 0, newblock);
2259 			dxroot = (struct dx_root *)frames[0].bh->b_data;
2260 			dxroot->info.indirect_levels += 1;
2261 			dxtrace(printk(KERN_DEBUG
2262 				       "Creating %d level index...\n",
2263 				       info->indirect_levels));
2264 			err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
2265 			if (err)
2266 				goto journal_error;
2267 			err = ext4_handle_dirty_dx_node(handle, dir, bh2);
2268 			brelse(bh2);
2269 			restart = 1;
2270 			goto journal_error;
2271 		}
2272 	}
2273 	de = do_split(handle, dir, &bh, frame, &fname->hinfo);
2274 	if (IS_ERR(de)) {
2275 		err = PTR_ERR(de);
2276 		goto cleanup;
2277 	}
2278 	err = add_dirent_to_buf(handle, fname, dir, inode, de, bh);
2279 	goto cleanup;
2280 
2281 journal_error:
2282 	ext4_std_error(dir->i_sb, err); /* this is a no-op if err == 0 */
2283 cleanup:
2284 	brelse(bh);
2285 	dx_release(frames);
2286 	/* @restart is true means htree-path has been changed, we need to
2287 	 * repeat dx_probe() to find out valid htree-path
2288 	 */
2289 	if (restart && err == 0)
2290 		goto again;
2291 	return err;
2292 }
2293 
2294 /*
2295  * ext4_generic_delete_entry deletes a directory entry by merging it
2296  * with the previous entry
2297  */
2298 int ext4_generic_delete_entry(handle_t *handle,
2299 			      struct inode *dir,
2300 			      struct ext4_dir_entry_2 *de_del,
2301 			      struct buffer_head *bh,
2302 			      void *entry_buf,
2303 			      int buf_size,
2304 			      int csum_size)
2305 {
2306 	struct ext4_dir_entry_2 *de, *pde;
2307 	unsigned int blocksize = dir->i_sb->s_blocksize;
2308 	int i;
2309 
2310 	i = 0;
2311 	pde = NULL;
2312 	de = (struct ext4_dir_entry_2 *)entry_buf;
2313 	while (i < buf_size - csum_size) {
2314 		if (ext4_check_dir_entry(dir, NULL, de, bh,
2315 					 bh->b_data, bh->b_size, i))
2316 			return -EFSCORRUPTED;
2317 		if (de == de_del)  {
2318 			if (pde)
2319 				pde->rec_len = ext4_rec_len_to_disk(
2320 					ext4_rec_len_from_disk(pde->rec_len,
2321 							       blocksize) +
2322 					ext4_rec_len_from_disk(de->rec_len,
2323 							       blocksize),
2324 					blocksize);
2325 			else
2326 				de->inode = 0;
2327 			inode_inc_iversion(dir);
2328 			return 0;
2329 		}
2330 		i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2331 		pde = de;
2332 		de = ext4_next_entry(de, blocksize);
2333 	}
2334 	return -ENOENT;
2335 }
2336 
2337 static int ext4_delete_entry(handle_t *handle,
2338 			     struct inode *dir,
2339 			     struct ext4_dir_entry_2 *de_del,
2340 			     struct buffer_head *bh)
2341 {
2342 	int err, csum_size = 0;
2343 
2344 	if (ext4_has_inline_data(dir)) {
2345 		int has_inline_data = 1;
2346 		err = ext4_delete_inline_entry(handle, dir, de_del, bh,
2347 					       &has_inline_data);
2348 		if (has_inline_data)
2349 			return err;
2350 	}
2351 
2352 	if (ext4_has_metadata_csum(dir->i_sb))
2353 		csum_size = sizeof(struct ext4_dir_entry_tail);
2354 
2355 	BUFFER_TRACE(bh, "get_write_access");
2356 	err = ext4_journal_get_write_access(handle, bh);
2357 	if (unlikely(err))
2358 		goto out;
2359 
2360 	err = ext4_generic_delete_entry(handle, dir, de_del,
2361 					bh, bh->b_data,
2362 					dir->i_sb->s_blocksize, csum_size);
2363 	if (err)
2364 		goto out;
2365 
2366 	BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2367 	err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2368 	if (unlikely(err))
2369 		goto out;
2370 
2371 	return 0;
2372 out:
2373 	if (err != -ENOENT)
2374 		ext4_std_error(dir->i_sb, err);
2375 	return err;
2376 }
2377 
2378 /*
2379  * Set directory link count to 1 if nlinks > EXT4_LINK_MAX, or if nlinks == 2
2380  * since this indicates that nlinks count was previously 1 to avoid overflowing
2381  * the 16-bit i_links_count field on disk.  Directories with i_nlink == 1 mean
2382  * that subdirectory link counts are not being maintained accurately.
2383  *
2384  * The caller has already checked for i_nlink overflow in case the DIR_LINK
2385  * feature is not enabled and returned -EMLINK.  The is_dx() check is a proxy
2386  * for checking S_ISDIR(inode) (since the INODE_INDEX feature will not be set
2387  * on regular files) and to avoid creating huge/slow non-HTREE directories.
2388  */
2389 static void ext4_inc_count(handle_t *handle, struct inode *inode)
2390 {
2391 	inc_nlink(inode);
2392 	if (is_dx(inode) &&
2393 	    (inode->i_nlink > EXT4_LINK_MAX || inode->i_nlink == 2))
2394 		set_nlink(inode, 1);
2395 }
2396 
2397 /*
2398  * If a directory had nlink == 1, then we should let it be 1. This indicates
2399  * directory has >EXT4_LINK_MAX subdirs.
2400  */
2401 static void ext4_dec_count(handle_t *handle, struct inode *inode)
2402 {
2403 	if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2404 		drop_nlink(inode);
2405 }
2406 
2407 
2408 static int ext4_add_nondir(handle_t *handle,
2409 		struct dentry *dentry, struct inode *inode)
2410 {
2411 	int err = ext4_add_entry(handle, dentry, inode);
2412 	if (!err) {
2413 		ext4_mark_inode_dirty(handle, inode);
2414 		d_instantiate_new(dentry, inode);
2415 		return 0;
2416 	}
2417 	drop_nlink(inode);
2418 	unlock_new_inode(inode);
2419 	iput(inode);
2420 	return err;
2421 }
2422 
2423 /*
2424  * By the time this is called, we already have created
2425  * the directory cache entry for the new file, but it
2426  * is so far negative - it has no inode.
2427  *
2428  * If the create succeeds, we fill in the inode information
2429  * with d_instantiate().
2430  */
2431 static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2432 		       bool excl)
2433 {
2434 	handle_t *handle;
2435 	struct inode *inode;
2436 	int err, credits, retries = 0;
2437 
2438 	err = dquot_initialize(dir);
2439 	if (err)
2440 		return err;
2441 
2442 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2443 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2444 retry:
2445 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2446 					    NULL, EXT4_HT_DIR, credits);
2447 	handle = ext4_journal_current_handle();
2448 	err = PTR_ERR(inode);
2449 	if (!IS_ERR(inode)) {
2450 		inode->i_op = &ext4_file_inode_operations;
2451 		inode->i_fop = &ext4_file_operations;
2452 		ext4_set_aops(inode);
2453 		err = ext4_add_nondir(handle, dentry, inode);
2454 		if (!err && IS_DIRSYNC(dir))
2455 			ext4_handle_sync(handle);
2456 	}
2457 	if (handle)
2458 		ext4_journal_stop(handle);
2459 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2460 		goto retry;
2461 	return err;
2462 }
2463 
2464 static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2465 		      umode_t mode, dev_t rdev)
2466 {
2467 	handle_t *handle;
2468 	struct inode *inode;
2469 	int err, credits, retries = 0;
2470 
2471 	err = dquot_initialize(dir);
2472 	if (err)
2473 		return err;
2474 
2475 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2476 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2477 retry:
2478 	inode = ext4_new_inode_start_handle(dir, mode, &dentry->d_name, 0,
2479 					    NULL, EXT4_HT_DIR, credits);
2480 	handle = ext4_journal_current_handle();
2481 	err = PTR_ERR(inode);
2482 	if (!IS_ERR(inode)) {
2483 		init_special_inode(inode, inode->i_mode, rdev);
2484 		inode->i_op = &ext4_special_inode_operations;
2485 		err = ext4_add_nondir(handle, dentry, inode);
2486 		if (!err && IS_DIRSYNC(dir))
2487 			ext4_handle_sync(handle);
2488 	}
2489 	if (handle)
2490 		ext4_journal_stop(handle);
2491 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2492 		goto retry;
2493 	return err;
2494 }
2495 
2496 static int ext4_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2497 {
2498 	handle_t *handle;
2499 	struct inode *inode;
2500 	int err, retries = 0;
2501 
2502 	err = dquot_initialize(dir);
2503 	if (err)
2504 		return err;
2505 
2506 retry:
2507 	inode = ext4_new_inode_start_handle(dir, mode,
2508 					    NULL, 0, NULL,
2509 					    EXT4_HT_DIR,
2510 			EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2511 			  4 + EXT4_XATTR_TRANS_BLOCKS);
2512 	handle = ext4_journal_current_handle();
2513 	err = PTR_ERR(inode);
2514 	if (!IS_ERR(inode)) {
2515 		inode->i_op = &ext4_file_inode_operations;
2516 		inode->i_fop = &ext4_file_operations;
2517 		ext4_set_aops(inode);
2518 		d_tmpfile(dentry, inode);
2519 		err = ext4_orphan_add(handle, inode);
2520 		if (err)
2521 			goto err_unlock_inode;
2522 		mark_inode_dirty(inode);
2523 		unlock_new_inode(inode);
2524 	}
2525 	if (handle)
2526 		ext4_journal_stop(handle);
2527 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2528 		goto retry;
2529 	return err;
2530 err_unlock_inode:
2531 	ext4_journal_stop(handle);
2532 	unlock_new_inode(inode);
2533 	return err;
2534 }
2535 
2536 struct ext4_dir_entry_2 *ext4_init_dot_dotdot(struct inode *inode,
2537 			  struct ext4_dir_entry_2 *de,
2538 			  int blocksize, int csum_size,
2539 			  unsigned int parent_ino, int dotdot_real_len)
2540 {
2541 	de->inode = cpu_to_le32(inode->i_ino);
2542 	de->name_len = 1;
2543 	de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2544 					   blocksize);
2545 	strcpy(de->name, ".");
2546 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2547 
2548 	de = ext4_next_entry(de, blocksize);
2549 	de->inode = cpu_to_le32(parent_ino);
2550 	de->name_len = 2;
2551 	if (!dotdot_real_len)
2552 		de->rec_len = ext4_rec_len_to_disk(blocksize -
2553 					(csum_size + EXT4_DIR_REC_LEN(1)),
2554 					blocksize);
2555 	else
2556 		de->rec_len = ext4_rec_len_to_disk(
2557 				EXT4_DIR_REC_LEN(de->name_len), blocksize);
2558 	strcpy(de->name, "..");
2559 	ext4_set_de_type(inode->i_sb, de, S_IFDIR);
2560 
2561 	return ext4_next_entry(de, blocksize);
2562 }
2563 
2564 static int ext4_init_new_dir(handle_t *handle, struct inode *dir,
2565 			     struct inode *inode)
2566 {
2567 	struct buffer_head *dir_block = NULL;
2568 	struct ext4_dir_entry_2 *de;
2569 	struct ext4_dir_entry_tail *t;
2570 	ext4_lblk_t block = 0;
2571 	unsigned int blocksize = dir->i_sb->s_blocksize;
2572 	int csum_size = 0;
2573 	int err;
2574 
2575 	if (ext4_has_metadata_csum(dir->i_sb))
2576 		csum_size = sizeof(struct ext4_dir_entry_tail);
2577 
2578 	if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2579 		err = ext4_try_create_inline_dir(handle, dir, inode);
2580 		if (err < 0 && err != -ENOSPC)
2581 			goto out;
2582 		if (!err)
2583 			goto out;
2584 	}
2585 
2586 	inode->i_size = 0;
2587 	dir_block = ext4_append(handle, inode, &block);
2588 	if (IS_ERR(dir_block))
2589 		return PTR_ERR(dir_block);
2590 	de = (struct ext4_dir_entry_2 *)dir_block->b_data;
2591 	ext4_init_dot_dotdot(inode, de, blocksize, csum_size, dir->i_ino, 0);
2592 	set_nlink(inode, 2);
2593 	if (csum_size) {
2594 		t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2595 		initialize_dirent_tail(t, blocksize);
2596 	}
2597 
2598 	BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2599 	err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2600 	if (err)
2601 		goto out;
2602 	set_buffer_verified(dir_block);
2603 out:
2604 	brelse(dir_block);
2605 	return err;
2606 }
2607 
2608 static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2609 {
2610 	handle_t *handle;
2611 	struct inode *inode;
2612 	int err, credits, retries = 0;
2613 
2614 	if (EXT4_DIR_LINK_MAX(dir))
2615 		return -EMLINK;
2616 
2617 	err = dquot_initialize(dir);
2618 	if (err)
2619 		return err;
2620 
2621 	credits = (EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2622 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3);
2623 retry:
2624 	inode = ext4_new_inode_start_handle(dir, S_IFDIR | mode,
2625 					    &dentry->d_name,
2626 					    0, NULL, EXT4_HT_DIR, credits);
2627 	handle = ext4_journal_current_handle();
2628 	err = PTR_ERR(inode);
2629 	if (IS_ERR(inode))
2630 		goto out_stop;
2631 
2632 	inode->i_op = &ext4_dir_inode_operations;
2633 	inode->i_fop = &ext4_dir_operations;
2634 	err = ext4_init_new_dir(handle, dir, inode);
2635 	if (err)
2636 		goto out_clear_inode;
2637 	err = ext4_mark_inode_dirty(handle, inode);
2638 	if (!err)
2639 		err = ext4_add_entry(handle, dentry, inode);
2640 	if (err) {
2641 out_clear_inode:
2642 		clear_nlink(inode);
2643 		unlock_new_inode(inode);
2644 		ext4_mark_inode_dirty(handle, inode);
2645 		iput(inode);
2646 		goto out_stop;
2647 	}
2648 	ext4_inc_count(handle, dir);
2649 	ext4_update_dx_flag(dir);
2650 	err = ext4_mark_inode_dirty(handle, dir);
2651 	if (err)
2652 		goto out_clear_inode;
2653 	d_instantiate_new(dentry, inode);
2654 	if (IS_DIRSYNC(dir))
2655 		ext4_handle_sync(handle);
2656 
2657 out_stop:
2658 	if (handle)
2659 		ext4_journal_stop(handle);
2660 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2661 		goto retry;
2662 	return err;
2663 }
2664 
2665 /*
2666  * routine to check that the specified directory is empty (for rmdir)
2667  */
2668 bool ext4_empty_dir(struct inode *inode)
2669 {
2670 	unsigned int offset;
2671 	struct buffer_head *bh;
2672 	struct ext4_dir_entry_2 *de, *de1;
2673 	struct super_block *sb;
2674 
2675 	if (ext4_has_inline_data(inode)) {
2676 		int has_inline_data = 1;
2677 		int ret;
2678 
2679 		ret = empty_inline_dir(inode, &has_inline_data);
2680 		if (has_inline_data)
2681 			return ret;
2682 	}
2683 
2684 	sb = inode->i_sb;
2685 	if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2)) {
2686 		EXT4_ERROR_INODE(inode, "invalid size");
2687 		return true;
2688 	}
2689 	bh = ext4_read_dirblock(inode, 0, EITHER);
2690 	if (IS_ERR(bh))
2691 		return true;
2692 
2693 	de = (struct ext4_dir_entry_2 *) bh->b_data;
2694 	de1 = ext4_next_entry(de, sb->s_blocksize);
2695 	if (le32_to_cpu(de->inode) != inode->i_ino ||
2696 			le32_to_cpu(de1->inode) == 0 ||
2697 			strcmp(".", de->name) || strcmp("..", de1->name)) {
2698 		ext4_warning_inode(inode, "directory missing '.' and/or '..'");
2699 		brelse(bh);
2700 		return true;
2701 	}
2702 	offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2703 		 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2704 	de = ext4_next_entry(de1, sb->s_blocksize);
2705 	while (offset < inode->i_size) {
2706 		if ((void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2707 			unsigned int lblock;
2708 			brelse(bh);
2709 			lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2710 			bh = ext4_read_dirblock(inode, lblock, EITHER);
2711 			if (IS_ERR(bh))
2712 				return true;
2713 			de = (struct ext4_dir_entry_2 *) bh->b_data;
2714 		}
2715 		if (ext4_check_dir_entry(inode, NULL, de, bh,
2716 					 bh->b_data, bh->b_size, offset)) {
2717 			de = (struct ext4_dir_entry_2 *)(bh->b_data +
2718 							 sb->s_blocksize);
2719 			offset = (offset | (sb->s_blocksize - 1)) + 1;
2720 			continue;
2721 		}
2722 		if (le32_to_cpu(de->inode)) {
2723 			brelse(bh);
2724 			return false;
2725 		}
2726 		offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2727 		de = ext4_next_entry(de, sb->s_blocksize);
2728 	}
2729 	brelse(bh);
2730 	return true;
2731 }
2732 
2733 /*
2734  * ext4_orphan_add() links an unlinked or truncated inode into a list of
2735  * such inodes, starting at the superblock, in case we crash before the
2736  * file is closed/deleted, or in case the inode truncate spans multiple
2737  * transactions and the last transaction is not recovered after a crash.
2738  *
2739  * At filesystem recovery time, we walk this list deleting unlinked
2740  * inodes and truncating linked inodes in ext4_orphan_cleanup().
2741  *
2742  * Orphan list manipulation functions must be called under i_mutex unless
2743  * we are just creating the inode or deleting it.
2744  */
2745 int ext4_orphan_add(handle_t *handle, struct inode *inode)
2746 {
2747 	struct super_block *sb = inode->i_sb;
2748 	struct ext4_sb_info *sbi = EXT4_SB(sb);
2749 	struct ext4_iloc iloc;
2750 	int err = 0, rc;
2751 	bool dirty = false;
2752 
2753 	if (!sbi->s_journal || is_bad_inode(inode))
2754 		return 0;
2755 
2756 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2757 		     !inode_is_locked(inode));
2758 	/*
2759 	 * Exit early if inode already is on orphan list. This is a big speedup
2760 	 * since we don't have to contend on the global s_orphan_lock.
2761 	 */
2762 	if (!list_empty(&EXT4_I(inode)->i_orphan))
2763 		return 0;
2764 
2765 	/*
2766 	 * Orphan handling is only valid for files with data blocks
2767 	 * being truncated, or files being unlinked. Note that we either
2768 	 * hold i_mutex, or the inode can not be referenced from outside,
2769 	 * so i_nlink should not be bumped due to race
2770 	 */
2771 	J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2772 		  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2773 
2774 	BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2775 	err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2776 	if (err)
2777 		goto out;
2778 
2779 	err = ext4_reserve_inode_write(handle, inode, &iloc);
2780 	if (err)
2781 		goto out;
2782 
2783 	mutex_lock(&sbi->s_orphan_lock);
2784 	/*
2785 	 * Due to previous errors inode may be already a part of on-disk
2786 	 * orphan list. If so skip on-disk list modification.
2787 	 */
2788 	if (!NEXT_ORPHAN(inode) || NEXT_ORPHAN(inode) >
2789 	    (le32_to_cpu(sbi->s_es->s_inodes_count))) {
2790 		/* Insert this inode at the head of the on-disk orphan list */
2791 		NEXT_ORPHAN(inode) = le32_to_cpu(sbi->s_es->s_last_orphan);
2792 		sbi->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2793 		dirty = true;
2794 	}
2795 	list_add(&EXT4_I(inode)->i_orphan, &sbi->s_orphan);
2796 	mutex_unlock(&sbi->s_orphan_lock);
2797 
2798 	if (dirty) {
2799 		err = ext4_handle_dirty_super(handle, sb);
2800 		rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2801 		if (!err)
2802 			err = rc;
2803 		if (err) {
2804 			/*
2805 			 * We have to remove inode from in-memory list if
2806 			 * addition to on disk orphan list failed. Stray orphan
2807 			 * list entries can cause panics at unmount time.
2808 			 */
2809 			mutex_lock(&sbi->s_orphan_lock);
2810 			list_del_init(&EXT4_I(inode)->i_orphan);
2811 			mutex_unlock(&sbi->s_orphan_lock);
2812 		}
2813 	}
2814 	jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2815 	jbd_debug(4, "orphan inode %lu will point to %d\n",
2816 			inode->i_ino, NEXT_ORPHAN(inode));
2817 out:
2818 	ext4_std_error(sb, err);
2819 	return err;
2820 }
2821 
2822 /*
2823  * ext4_orphan_del() removes an unlinked or truncated inode from the list
2824  * of such inodes stored on disk, because it is finally being cleaned up.
2825  */
2826 int ext4_orphan_del(handle_t *handle, struct inode *inode)
2827 {
2828 	struct list_head *prev;
2829 	struct ext4_inode_info *ei = EXT4_I(inode);
2830 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2831 	__u32 ino_next;
2832 	struct ext4_iloc iloc;
2833 	int err = 0;
2834 
2835 	if (!sbi->s_journal && !(sbi->s_mount_state & EXT4_ORPHAN_FS))
2836 		return 0;
2837 
2838 	WARN_ON_ONCE(!(inode->i_state & (I_NEW | I_FREEING)) &&
2839 		     !inode_is_locked(inode));
2840 	/* Do this quick check before taking global s_orphan_lock. */
2841 	if (list_empty(&ei->i_orphan))
2842 		return 0;
2843 
2844 	if (handle) {
2845 		/* Grab inode buffer early before taking global s_orphan_lock */
2846 		err = ext4_reserve_inode_write(handle, inode, &iloc);
2847 	}
2848 
2849 	mutex_lock(&sbi->s_orphan_lock);
2850 	jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2851 
2852 	prev = ei->i_orphan.prev;
2853 	list_del_init(&ei->i_orphan);
2854 
2855 	/* If we're on an error path, we may not have a valid
2856 	 * transaction handle with which to update the orphan list on
2857 	 * disk, but we still need to remove the inode from the linked
2858 	 * list in memory. */
2859 	if (!handle || err) {
2860 		mutex_unlock(&sbi->s_orphan_lock);
2861 		goto out_err;
2862 	}
2863 
2864 	ino_next = NEXT_ORPHAN(inode);
2865 	if (prev == &sbi->s_orphan) {
2866 		jbd_debug(4, "superblock will point to %u\n", ino_next);
2867 		BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2868 		err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2869 		if (err) {
2870 			mutex_unlock(&sbi->s_orphan_lock);
2871 			goto out_brelse;
2872 		}
2873 		sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2874 		mutex_unlock(&sbi->s_orphan_lock);
2875 		err = ext4_handle_dirty_super(handle, inode->i_sb);
2876 	} else {
2877 		struct ext4_iloc iloc2;
2878 		struct inode *i_prev =
2879 			&list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2880 
2881 		jbd_debug(4, "orphan inode %lu will point to %u\n",
2882 			  i_prev->i_ino, ino_next);
2883 		err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2884 		if (err) {
2885 			mutex_unlock(&sbi->s_orphan_lock);
2886 			goto out_brelse;
2887 		}
2888 		NEXT_ORPHAN(i_prev) = ino_next;
2889 		err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2890 		mutex_unlock(&sbi->s_orphan_lock);
2891 	}
2892 	if (err)
2893 		goto out_brelse;
2894 	NEXT_ORPHAN(inode) = 0;
2895 	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2896 out_err:
2897 	ext4_std_error(inode->i_sb, err);
2898 	return err;
2899 
2900 out_brelse:
2901 	brelse(iloc.bh);
2902 	goto out_err;
2903 }
2904 
2905 static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2906 {
2907 	int retval;
2908 	struct inode *inode;
2909 	struct buffer_head *bh;
2910 	struct ext4_dir_entry_2 *de;
2911 	handle_t *handle = NULL;
2912 
2913 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2914 		return -EIO;
2915 
2916 	/* Initialize quotas before so that eventual writes go in
2917 	 * separate transaction */
2918 	retval = dquot_initialize(dir);
2919 	if (retval)
2920 		return retval;
2921 	retval = dquot_initialize(d_inode(dentry));
2922 	if (retval)
2923 		return retval;
2924 
2925 	retval = -ENOENT;
2926 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
2927 	if (IS_ERR(bh))
2928 		return PTR_ERR(bh);
2929 	if (!bh)
2930 		goto end_rmdir;
2931 
2932 	inode = d_inode(dentry);
2933 
2934 	retval = -EFSCORRUPTED;
2935 	if (le32_to_cpu(de->inode) != inode->i_ino)
2936 		goto end_rmdir;
2937 
2938 	retval = -ENOTEMPTY;
2939 	if (!ext4_empty_dir(inode))
2940 		goto end_rmdir;
2941 
2942 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
2943 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
2944 	if (IS_ERR(handle)) {
2945 		retval = PTR_ERR(handle);
2946 		handle = NULL;
2947 		goto end_rmdir;
2948 	}
2949 
2950 	if (IS_DIRSYNC(dir))
2951 		ext4_handle_sync(handle);
2952 
2953 	retval = ext4_delete_entry(handle, dir, de, bh);
2954 	if (retval)
2955 		goto end_rmdir;
2956 	if (!EXT4_DIR_LINK_EMPTY(inode))
2957 		ext4_warning_inode(inode,
2958 			     "empty directory '%.*s' has too many links (%u)",
2959 			     dentry->d_name.len, dentry->d_name.name,
2960 			     inode->i_nlink);
2961 	inode_inc_iversion(inode);
2962 	clear_nlink(inode);
2963 	/* There's no need to set i_disksize: the fact that i_nlink is
2964 	 * zero will ensure that the right thing happens during any
2965 	 * recovery. */
2966 	inode->i_size = 0;
2967 	ext4_orphan_add(handle, inode);
2968 	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2969 	ext4_mark_inode_dirty(handle, inode);
2970 	ext4_dec_count(handle, dir);
2971 	ext4_update_dx_flag(dir);
2972 	ext4_mark_inode_dirty(handle, dir);
2973 
2974 end_rmdir:
2975 	brelse(bh);
2976 	if (handle)
2977 		ext4_journal_stop(handle);
2978 	return retval;
2979 }
2980 
2981 static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2982 {
2983 	int retval;
2984 	struct inode *inode;
2985 	struct buffer_head *bh;
2986 	struct ext4_dir_entry_2 *de;
2987 	handle_t *handle = NULL;
2988 
2989 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
2990 		return -EIO;
2991 
2992 	trace_ext4_unlink_enter(dir, dentry);
2993 	/* Initialize quotas before so that eventual writes go
2994 	 * in separate transaction */
2995 	retval = dquot_initialize(dir);
2996 	if (retval)
2997 		return retval;
2998 	retval = dquot_initialize(d_inode(dentry));
2999 	if (retval)
3000 		return retval;
3001 
3002 	retval = -ENOENT;
3003 	bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
3004 	if (IS_ERR(bh))
3005 		return PTR_ERR(bh);
3006 	if (!bh)
3007 		goto end_unlink;
3008 
3009 	inode = d_inode(dentry);
3010 
3011 	retval = -EFSCORRUPTED;
3012 	if (le32_to_cpu(de->inode) != inode->i_ino)
3013 		goto end_unlink;
3014 
3015 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3016 				    EXT4_DATA_TRANS_BLOCKS(dir->i_sb));
3017 	if (IS_ERR(handle)) {
3018 		retval = PTR_ERR(handle);
3019 		handle = NULL;
3020 		goto end_unlink;
3021 	}
3022 
3023 	if (IS_DIRSYNC(dir))
3024 		ext4_handle_sync(handle);
3025 
3026 	if (inode->i_nlink == 0) {
3027 		ext4_warning_inode(inode, "Deleting file '%.*s' with no links",
3028 				   dentry->d_name.len, dentry->d_name.name);
3029 		set_nlink(inode, 1);
3030 	}
3031 	retval = ext4_delete_entry(handle, dir, de, bh);
3032 	if (retval)
3033 		goto end_unlink;
3034 	dir->i_ctime = dir->i_mtime = current_time(dir);
3035 	ext4_update_dx_flag(dir);
3036 	ext4_mark_inode_dirty(handle, dir);
3037 	drop_nlink(inode);
3038 	if (!inode->i_nlink)
3039 		ext4_orphan_add(handle, inode);
3040 	inode->i_ctime = current_time(inode);
3041 	ext4_mark_inode_dirty(handle, inode);
3042 
3043 end_unlink:
3044 	brelse(bh);
3045 	if (handle)
3046 		ext4_journal_stop(handle);
3047 	trace_ext4_unlink_exit(dentry, retval);
3048 	return retval;
3049 }
3050 
3051 static int ext4_symlink(struct inode *dir,
3052 			struct dentry *dentry, const char *symname)
3053 {
3054 	handle_t *handle;
3055 	struct inode *inode;
3056 	int err, len = strlen(symname);
3057 	int credits;
3058 	struct fscrypt_str disk_link;
3059 
3060 	if (unlikely(ext4_forced_shutdown(EXT4_SB(dir->i_sb))))
3061 		return -EIO;
3062 
3063 	err = fscrypt_prepare_symlink(dir, symname, len, dir->i_sb->s_blocksize,
3064 				      &disk_link);
3065 	if (err)
3066 		return err;
3067 
3068 	err = dquot_initialize(dir);
3069 	if (err)
3070 		return err;
3071 
3072 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3073 		/*
3074 		 * For non-fast symlinks, we just allocate inode and put it on
3075 		 * orphan list in the first transaction => we need bitmap,
3076 		 * group descriptor, sb, inode block, quota blocks, and
3077 		 * possibly selinux xattr blocks.
3078 		 */
3079 		credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
3080 			  EXT4_XATTR_TRANS_BLOCKS;
3081 	} else {
3082 		/*
3083 		 * Fast symlink. We have to add entry to directory
3084 		 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
3085 		 * allocate new inode (bitmap, group descriptor, inode block,
3086 		 * quota blocks, sb is already counted in previous macros).
3087 		 */
3088 		credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3089 			  EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3;
3090 	}
3091 
3092 	inode = ext4_new_inode_start_handle(dir, S_IFLNK|S_IRWXUGO,
3093 					    &dentry->d_name, 0, NULL,
3094 					    EXT4_HT_DIR, credits);
3095 	handle = ext4_journal_current_handle();
3096 	if (IS_ERR(inode)) {
3097 		if (handle)
3098 			ext4_journal_stop(handle);
3099 		return PTR_ERR(inode);
3100 	}
3101 
3102 	if (IS_ENCRYPTED(inode)) {
3103 		err = fscrypt_encrypt_symlink(inode, symname, len, &disk_link);
3104 		if (err)
3105 			goto err_drop_inode;
3106 		inode->i_op = &ext4_encrypted_symlink_inode_operations;
3107 	}
3108 
3109 	if ((disk_link.len > EXT4_N_BLOCKS * 4)) {
3110 		if (!IS_ENCRYPTED(inode))
3111 			inode->i_op = &ext4_symlink_inode_operations;
3112 		inode_nohighmem(inode);
3113 		ext4_set_aops(inode);
3114 		/*
3115 		 * We cannot call page_symlink() with transaction started
3116 		 * because it calls into ext4_write_begin() which can wait
3117 		 * for transaction commit if we are running out of space
3118 		 * and thus we deadlock. So we have to stop transaction now
3119 		 * and restart it when symlink contents is written.
3120 		 *
3121 		 * To keep fs consistent in case of crash, we have to put inode
3122 		 * to orphan list in the mean time.
3123 		 */
3124 		drop_nlink(inode);
3125 		err = ext4_orphan_add(handle, inode);
3126 		ext4_journal_stop(handle);
3127 		handle = NULL;
3128 		if (err)
3129 			goto err_drop_inode;
3130 		err = __page_symlink(inode, disk_link.name, disk_link.len, 1);
3131 		if (err)
3132 			goto err_drop_inode;
3133 		/*
3134 		 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
3135 		 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
3136 		 */
3137 		handle = ext4_journal_start(dir, EXT4_HT_DIR,
3138 				EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3139 				EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
3140 		if (IS_ERR(handle)) {
3141 			err = PTR_ERR(handle);
3142 			handle = NULL;
3143 			goto err_drop_inode;
3144 		}
3145 		set_nlink(inode, 1);
3146 		err = ext4_orphan_del(handle, inode);
3147 		if (err)
3148 			goto err_drop_inode;
3149 	} else {
3150 		/* clear the extent format for fast symlink */
3151 		ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
3152 		if (!IS_ENCRYPTED(inode)) {
3153 			inode->i_op = &ext4_fast_symlink_inode_operations;
3154 			inode->i_link = (char *)&EXT4_I(inode)->i_data;
3155 		}
3156 		memcpy((char *)&EXT4_I(inode)->i_data, disk_link.name,
3157 		       disk_link.len);
3158 		inode->i_size = disk_link.len - 1;
3159 	}
3160 	EXT4_I(inode)->i_disksize = inode->i_size;
3161 	err = ext4_add_nondir(handle, dentry, inode);
3162 	if (!err && IS_DIRSYNC(dir))
3163 		ext4_handle_sync(handle);
3164 
3165 	if (handle)
3166 		ext4_journal_stop(handle);
3167 	goto out_free_encrypted_link;
3168 
3169 err_drop_inode:
3170 	if (handle)
3171 		ext4_journal_stop(handle);
3172 	clear_nlink(inode);
3173 	unlock_new_inode(inode);
3174 	iput(inode);
3175 out_free_encrypted_link:
3176 	if (disk_link.name != (unsigned char *)symname)
3177 		kfree(disk_link.name);
3178 	return err;
3179 }
3180 
3181 static int ext4_link(struct dentry *old_dentry,
3182 		     struct inode *dir, struct dentry *dentry)
3183 {
3184 	handle_t *handle;
3185 	struct inode *inode = d_inode(old_dentry);
3186 	int err, retries = 0;
3187 
3188 	if (inode->i_nlink >= EXT4_LINK_MAX)
3189 		return -EMLINK;
3190 
3191 	err = fscrypt_prepare_link(old_dentry, dir, dentry);
3192 	if (err)
3193 		return err;
3194 
3195 	if ((ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) &&
3196 	    (!projid_eq(EXT4_I(dir)->i_projid,
3197 			EXT4_I(old_dentry->d_inode)->i_projid)))
3198 		return -EXDEV;
3199 
3200 	err = dquot_initialize(dir);
3201 	if (err)
3202 		return err;
3203 
3204 retry:
3205 	handle = ext4_journal_start(dir, EXT4_HT_DIR,
3206 		(EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
3207 		 EXT4_INDEX_EXTRA_TRANS_BLOCKS) + 1);
3208 	if (IS_ERR(handle))
3209 		return PTR_ERR(handle);
3210 
3211 	if (IS_DIRSYNC(dir))
3212 		ext4_handle_sync(handle);
3213 
3214 	inode->i_ctime = current_time(inode);
3215 	ext4_inc_count(handle, inode);
3216 	ihold(inode);
3217 
3218 	err = ext4_add_entry(handle, dentry, inode);
3219 	if (!err) {
3220 		ext4_mark_inode_dirty(handle, inode);
3221 		/* this can happen only for tmpfile being
3222 		 * linked the first time
3223 		 */
3224 		if (inode->i_nlink == 1)
3225 			ext4_orphan_del(handle, inode);
3226 		d_instantiate(dentry, inode);
3227 	} else {
3228 		drop_nlink(inode);
3229 		iput(inode);
3230 	}
3231 	ext4_journal_stop(handle);
3232 	if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
3233 		goto retry;
3234 	return err;
3235 }
3236 
3237 
3238 /*
3239  * Try to find buffer head where contains the parent block.
3240  * It should be the inode block if it is inlined or the 1st block
3241  * if it is a normal dir.
3242  */
3243 static struct buffer_head *ext4_get_first_dir_block(handle_t *handle,
3244 					struct inode *inode,
3245 					int *retval,
3246 					struct ext4_dir_entry_2 **parent_de,
3247 					int *inlined)
3248 {
3249 	struct buffer_head *bh;
3250 
3251 	if (!ext4_has_inline_data(inode)) {
3252 		bh = ext4_read_dirblock(inode, 0, EITHER);
3253 		if (IS_ERR(bh)) {
3254 			*retval = PTR_ERR(bh);
3255 			return NULL;
3256 		}
3257 		*parent_de = ext4_next_entry(
3258 					(struct ext4_dir_entry_2 *)bh->b_data,
3259 					inode->i_sb->s_blocksize);
3260 		return bh;
3261 	}
3262 
3263 	*inlined = 1;
3264 	return ext4_get_first_inline_block(inode, parent_de, retval);
3265 }
3266 
3267 struct ext4_renament {
3268 	struct inode *dir;
3269 	struct dentry *dentry;
3270 	struct inode *inode;
3271 	bool is_dir;
3272 	int dir_nlink_delta;
3273 
3274 	/* entry for "dentry" */
3275 	struct buffer_head *bh;
3276 	struct ext4_dir_entry_2 *de;
3277 	int inlined;
3278 
3279 	/* entry for ".." in inode if it's a directory */
3280 	struct buffer_head *dir_bh;
3281 	struct ext4_dir_entry_2 *parent_de;
3282 	int dir_inlined;
3283 };
3284 
3285 static int ext4_rename_dir_prepare(handle_t *handle, struct ext4_renament *ent)
3286 {
3287 	int retval;
3288 
3289 	ent->dir_bh = ext4_get_first_dir_block(handle, ent->inode,
3290 					      &retval, &ent->parent_de,
3291 					      &ent->dir_inlined);
3292 	if (!ent->dir_bh)
3293 		return retval;
3294 	if (le32_to_cpu(ent->parent_de->inode) != ent->dir->i_ino)
3295 		return -EFSCORRUPTED;
3296 	BUFFER_TRACE(ent->dir_bh, "get_write_access");
3297 	return ext4_journal_get_write_access(handle, ent->dir_bh);
3298 }
3299 
3300 static int ext4_rename_dir_finish(handle_t *handle, struct ext4_renament *ent,
3301 				  unsigned dir_ino)
3302 {
3303 	int retval;
3304 
3305 	ent->parent_de->inode = cpu_to_le32(dir_ino);
3306 	BUFFER_TRACE(ent->dir_bh, "call ext4_handle_dirty_metadata");
3307 	if (!ent->dir_inlined) {
3308 		if (is_dx(ent->inode)) {
3309 			retval = ext4_handle_dirty_dx_node(handle,
3310 							   ent->inode,
3311 							   ent->dir_bh);
3312 		} else {
3313 			retval = ext4_handle_dirty_dirent_node(handle,
3314 							       ent->inode,
3315 							       ent->dir_bh);
3316 		}
3317 	} else {
3318 		retval = ext4_mark_inode_dirty(handle, ent->inode);
3319 	}
3320 	if (retval) {
3321 		ext4_std_error(ent->dir->i_sb, retval);
3322 		return retval;
3323 	}
3324 	return 0;
3325 }
3326 
3327 static int ext4_setent(handle_t *handle, struct ext4_renament *ent,
3328 		       unsigned ino, unsigned file_type)
3329 {
3330 	int retval;
3331 
3332 	BUFFER_TRACE(ent->bh, "get write access");
3333 	retval = ext4_journal_get_write_access(handle, ent->bh);
3334 	if (retval)
3335 		return retval;
3336 	ent->de->inode = cpu_to_le32(ino);
3337 	if (ext4_has_feature_filetype(ent->dir->i_sb))
3338 		ent->de->file_type = file_type;
3339 	inode_inc_iversion(ent->dir);
3340 	ent->dir->i_ctime = ent->dir->i_mtime =
3341 		current_time(ent->dir);
3342 	ext4_mark_inode_dirty(handle, ent->dir);
3343 	BUFFER_TRACE(ent->bh, "call ext4_handle_dirty_metadata");
3344 	if (!ent->inlined) {
3345 		retval = ext4_handle_dirty_dirent_node(handle,
3346 						       ent->dir, ent->bh);
3347 		if (unlikely(retval)) {
3348 			ext4_std_error(ent->dir->i_sb, retval);
3349 			return retval;
3350 		}
3351 	}
3352 	brelse(ent->bh);
3353 	ent->bh = NULL;
3354 
3355 	return 0;
3356 }
3357 
3358 static int ext4_find_delete_entry(handle_t *handle, struct inode *dir,
3359 				  const struct qstr *d_name)
3360 {
3361 	int retval = -ENOENT;
3362 	struct buffer_head *bh;
3363 	struct ext4_dir_entry_2 *de;
3364 
3365 	bh = ext4_find_entry(dir, d_name, &de, NULL);
3366 	if (IS_ERR(bh))
3367 		return PTR_ERR(bh);
3368 	if (bh) {
3369 		retval = ext4_delete_entry(handle, dir, de, bh);
3370 		brelse(bh);
3371 	}
3372 	return retval;
3373 }
3374 
3375 static void ext4_rename_delete(handle_t *handle, struct ext4_renament *ent,
3376 			       int force_reread)
3377 {
3378 	int retval;
3379 	/*
3380 	 * ent->de could have moved from under us during htree split, so make
3381 	 * sure that we are deleting the right entry.  We might also be pointing
3382 	 * to a stale entry in the unused part of ent->bh so just checking inum
3383 	 * and the name isn't enough.
3384 	 */
3385 	if (le32_to_cpu(ent->de->inode) != ent->inode->i_ino ||
3386 	    ent->de->name_len != ent->dentry->d_name.len ||
3387 	    strncmp(ent->de->name, ent->dentry->d_name.name,
3388 		    ent->de->name_len) ||
3389 	    force_reread) {
3390 		retval = ext4_find_delete_entry(handle, ent->dir,
3391 						&ent->dentry->d_name);
3392 	} else {
3393 		retval = ext4_delete_entry(handle, ent->dir, ent->de, ent->bh);
3394 		if (retval == -ENOENT) {
3395 			retval = ext4_find_delete_entry(handle, ent->dir,
3396 							&ent->dentry->d_name);
3397 		}
3398 	}
3399 
3400 	if (retval) {
3401 		ext4_warning_inode(ent->dir,
3402 				   "Deleting old file: nlink %d, error=%d",
3403 				   ent->dir->i_nlink, retval);
3404 	}
3405 }
3406 
3407 static void ext4_update_dir_count(handle_t *handle, struct ext4_renament *ent)
3408 {
3409 	if (ent->dir_nlink_delta) {
3410 		if (ent->dir_nlink_delta == -1)
3411 			ext4_dec_count(handle, ent->dir);
3412 		else
3413 			ext4_inc_count(handle, ent->dir);
3414 		ext4_mark_inode_dirty(handle, ent->dir);
3415 	}
3416 }
3417 
3418 static struct inode *ext4_whiteout_for_rename(struct ext4_renament *ent,
3419 					      int credits, handle_t **h)
3420 {
3421 	struct inode *wh;
3422 	handle_t *handle;
3423 	int retries = 0;
3424 
3425 	/*
3426 	 * for inode block, sb block, group summaries,
3427 	 * and inode bitmap
3428 	 */
3429 	credits += (EXT4_MAXQUOTAS_TRANS_BLOCKS(ent->dir->i_sb) +
3430 		    EXT4_XATTR_TRANS_BLOCKS + 4);
3431 retry:
3432 	wh = ext4_new_inode_start_handle(ent->dir, S_IFCHR | WHITEOUT_MODE,
3433 					 &ent->dentry->d_name, 0, NULL,
3434 					 EXT4_HT_DIR, credits);
3435 
3436 	handle = ext4_journal_current_handle();
3437 	if (IS_ERR(wh)) {
3438 		if (handle)
3439 			ext4_journal_stop(handle);
3440 		if (PTR_ERR(wh) == -ENOSPC &&
3441 		    ext4_should_retry_alloc(ent->dir->i_sb, &retries))
3442 			goto retry;
3443 	} else {
3444 		*h = handle;
3445 		init_special_inode(wh, wh->i_mode, WHITEOUT_DEV);
3446 		wh->i_op = &ext4_special_inode_operations;
3447 	}
3448 	return wh;
3449 }
3450 
3451 /*
3452  * Anybody can rename anything with this: the permission checks are left to the
3453  * higher-level routines.
3454  *
3455  * n.b.  old_{dentry,inode) refers to the source dentry/inode
3456  * while new_{dentry,inode) refers to the destination dentry/inode
3457  * This comes from rename(const char *oldpath, const char *newpath)
3458  */
3459 static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
3460 		       struct inode *new_dir, struct dentry *new_dentry,
3461 		       unsigned int flags)
3462 {
3463 	handle_t *handle = NULL;
3464 	struct ext4_renament old = {
3465 		.dir = old_dir,
3466 		.dentry = old_dentry,
3467 		.inode = d_inode(old_dentry),
3468 	};
3469 	struct ext4_renament new = {
3470 		.dir = new_dir,
3471 		.dentry = new_dentry,
3472 		.inode = d_inode(new_dentry),
3473 	};
3474 	int force_reread;
3475 	int retval;
3476 	struct inode *whiteout = NULL;
3477 	int credits;
3478 	u8 old_file_type;
3479 
3480 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT)) &&
3481 	    (!projid_eq(EXT4_I(new_dir)->i_projid,
3482 			EXT4_I(old_dentry->d_inode)->i_projid)))
3483 		return -EXDEV;
3484 
3485 	retval = dquot_initialize(old.dir);
3486 	if (retval)
3487 		return retval;
3488 	retval = dquot_initialize(new.dir);
3489 	if (retval)
3490 		return retval;
3491 
3492 	/* Initialize quotas before so that eventual writes go
3493 	 * in separate transaction */
3494 	if (new.inode) {
3495 		retval = dquot_initialize(new.inode);
3496 		if (retval)
3497 			return retval;
3498 	}
3499 
3500 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name, &old.de, NULL);
3501 	if (IS_ERR(old.bh))
3502 		return PTR_ERR(old.bh);
3503 	/*
3504 	 *  Check for inode number is _not_ due to possible IO errors.
3505 	 *  We might rmdir the source, keep it as pwd of some process
3506 	 *  and merrily kill the link to whatever was created under the
3507 	 *  same name. Goodbye sticky bit ;-<
3508 	 */
3509 	retval = -ENOENT;
3510 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3511 		goto end_rename;
3512 
3513 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3514 				 &new.de, &new.inlined);
3515 	if (IS_ERR(new.bh)) {
3516 		retval = PTR_ERR(new.bh);
3517 		new.bh = NULL;
3518 		goto end_rename;
3519 	}
3520 	if (new.bh) {
3521 		if (!new.inode) {
3522 			brelse(new.bh);
3523 			new.bh = NULL;
3524 		}
3525 	}
3526 	if (new.inode && !test_opt(new.dir->i_sb, NO_AUTO_DA_ALLOC))
3527 		ext4_alloc_da_blocks(old.inode);
3528 
3529 	credits = (2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3530 		   EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
3531 	if (!(flags & RENAME_WHITEOUT)) {
3532 		handle = ext4_journal_start(old.dir, EXT4_HT_DIR, credits);
3533 		if (IS_ERR(handle)) {
3534 			retval = PTR_ERR(handle);
3535 			handle = NULL;
3536 			goto end_rename;
3537 		}
3538 	} else {
3539 		whiteout = ext4_whiteout_for_rename(&old, credits, &handle);
3540 		if (IS_ERR(whiteout)) {
3541 			retval = PTR_ERR(whiteout);
3542 			whiteout = NULL;
3543 			goto end_rename;
3544 		}
3545 	}
3546 
3547 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3548 		ext4_handle_sync(handle);
3549 
3550 	if (S_ISDIR(old.inode->i_mode)) {
3551 		if (new.inode) {
3552 			retval = -ENOTEMPTY;
3553 			if (!ext4_empty_dir(new.inode))
3554 				goto end_rename;
3555 		} else {
3556 			retval = -EMLINK;
3557 			if (new.dir != old.dir && EXT4_DIR_LINK_MAX(new.dir))
3558 				goto end_rename;
3559 		}
3560 		retval = ext4_rename_dir_prepare(handle, &old);
3561 		if (retval)
3562 			goto end_rename;
3563 	}
3564 	/*
3565 	 * If we're renaming a file within an inline_data dir and adding or
3566 	 * setting the new dirent causes a conversion from inline_data to
3567 	 * extents/blockmap, we need to force the dirent delete code to
3568 	 * re-read the directory, or else we end up trying to delete a dirent
3569 	 * from what is now the extent tree root (or a block map).
3570 	 */
3571 	force_reread = (new.dir->i_ino == old.dir->i_ino &&
3572 			ext4_test_inode_flag(new.dir, EXT4_INODE_INLINE_DATA));
3573 
3574 	old_file_type = old.de->file_type;
3575 	if (whiteout) {
3576 		/*
3577 		 * Do this before adding a new entry, so the old entry is sure
3578 		 * to be still pointing to the valid old entry.
3579 		 */
3580 		retval = ext4_setent(handle, &old, whiteout->i_ino,
3581 				     EXT4_FT_CHRDEV);
3582 		if (retval)
3583 			goto end_rename;
3584 		ext4_mark_inode_dirty(handle, whiteout);
3585 	}
3586 	if (!new.bh) {
3587 		retval = ext4_add_entry(handle, new.dentry, old.inode);
3588 		if (retval)
3589 			goto end_rename;
3590 	} else {
3591 		retval = ext4_setent(handle, &new,
3592 				     old.inode->i_ino, old_file_type);
3593 		if (retval)
3594 			goto end_rename;
3595 	}
3596 	if (force_reread)
3597 		force_reread = !ext4_test_inode_flag(new.dir,
3598 						     EXT4_INODE_INLINE_DATA);
3599 
3600 	/*
3601 	 * Like most other Unix systems, set the ctime for inodes on a
3602 	 * rename.
3603 	 */
3604 	old.inode->i_ctime = current_time(old.inode);
3605 	ext4_mark_inode_dirty(handle, old.inode);
3606 
3607 	if (!whiteout) {
3608 		/*
3609 		 * ok, that's it
3610 		 */
3611 		ext4_rename_delete(handle, &old, force_reread);
3612 	}
3613 
3614 	if (new.inode) {
3615 		ext4_dec_count(handle, new.inode);
3616 		new.inode->i_ctime = current_time(new.inode);
3617 	}
3618 	old.dir->i_ctime = old.dir->i_mtime = current_time(old.dir);
3619 	ext4_update_dx_flag(old.dir);
3620 	if (old.dir_bh) {
3621 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3622 		if (retval)
3623 			goto end_rename;
3624 
3625 		ext4_dec_count(handle, old.dir);
3626 		if (new.inode) {
3627 			/* checked ext4_empty_dir above, can't have another
3628 			 * parent, ext4_dec_count() won't work for many-linked
3629 			 * dirs */
3630 			clear_nlink(new.inode);
3631 		} else {
3632 			ext4_inc_count(handle, new.dir);
3633 			ext4_update_dx_flag(new.dir);
3634 			ext4_mark_inode_dirty(handle, new.dir);
3635 		}
3636 	}
3637 	ext4_mark_inode_dirty(handle, old.dir);
3638 	if (new.inode) {
3639 		ext4_mark_inode_dirty(handle, new.inode);
3640 		if (!new.inode->i_nlink)
3641 			ext4_orphan_add(handle, new.inode);
3642 	}
3643 	retval = 0;
3644 
3645 end_rename:
3646 	brelse(old.dir_bh);
3647 	brelse(old.bh);
3648 	brelse(new.bh);
3649 	if (whiteout) {
3650 		if (retval)
3651 			drop_nlink(whiteout);
3652 		unlock_new_inode(whiteout);
3653 		iput(whiteout);
3654 	}
3655 	if (handle)
3656 		ext4_journal_stop(handle);
3657 	return retval;
3658 }
3659 
3660 static int ext4_cross_rename(struct inode *old_dir, struct dentry *old_dentry,
3661 			     struct inode *new_dir, struct dentry *new_dentry)
3662 {
3663 	handle_t *handle = NULL;
3664 	struct ext4_renament old = {
3665 		.dir = old_dir,
3666 		.dentry = old_dentry,
3667 		.inode = d_inode(old_dentry),
3668 	};
3669 	struct ext4_renament new = {
3670 		.dir = new_dir,
3671 		.dentry = new_dentry,
3672 		.inode = d_inode(new_dentry),
3673 	};
3674 	u8 new_file_type;
3675 	int retval;
3676 	struct timespec ctime;
3677 
3678 	if ((ext4_test_inode_flag(new_dir, EXT4_INODE_PROJINHERIT) &&
3679 	     !projid_eq(EXT4_I(new_dir)->i_projid,
3680 			EXT4_I(old_dentry->d_inode)->i_projid)) ||
3681 	    (ext4_test_inode_flag(old_dir, EXT4_INODE_PROJINHERIT) &&
3682 	     !projid_eq(EXT4_I(old_dir)->i_projid,
3683 			EXT4_I(new_dentry->d_inode)->i_projid)))
3684 		return -EXDEV;
3685 
3686 	retval = dquot_initialize(old.dir);
3687 	if (retval)
3688 		return retval;
3689 	retval = dquot_initialize(new.dir);
3690 	if (retval)
3691 		return retval;
3692 
3693 	old.bh = ext4_find_entry(old.dir, &old.dentry->d_name,
3694 				 &old.de, &old.inlined);
3695 	if (IS_ERR(old.bh))
3696 		return PTR_ERR(old.bh);
3697 	/*
3698 	 *  Check for inode number is _not_ due to possible IO errors.
3699 	 *  We might rmdir the source, keep it as pwd of some process
3700 	 *  and merrily kill the link to whatever was created under the
3701 	 *  same name. Goodbye sticky bit ;-<
3702 	 */
3703 	retval = -ENOENT;
3704 	if (!old.bh || le32_to_cpu(old.de->inode) != old.inode->i_ino)
3705 		goto end_rename;
3706 
3707 	new.bh = ext4_find_entry(new.dir, &new.dentry->d_name,
3708 				 &new.de, &new.inlined);
3709 	if (IS_ERR(new.bh)) {
3710 		retval = PTR_ERR(new.bh);
3711 		new.bh = NULL;
3712 		goto end_rename;
3713 	}
3714 
3715 	/* RENAME_EXCHANGE case: old *and* new must both exist */
3716 	if (!new.bh || le32_to_cpu(new.de->inode) != new.inode->i_ino)
3717 		goto end_rename;
3718 
3719 	handle = ext4_journal_start(old.dir, EXT4_HT_DIR,
3720 		(2 * EXT4_DATA_TRANS_BLOCKS(old.dir->i_sb) +
3721 		 2 * EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2));
3722 	if (IS_ERR(handle)) {
3723 		retval = PTR_ERR(handle);
3724 		handle = NULL;
3725 		goto end_rename;
3726 	}
3727 
3728 	if (IS_DIRSYNC(old.dir) || IS_DIRSYNC(new.dir))
3729 		ext4_handle_sync(handle);
3730 
3731 	if (S_ISDIR(old.inode->i_mode)) {
3732 		old.is_dir = true;
3733 		retval = ext4_rename_dir_prepare(handle, &old);
3734 		if (retval)
3735 			goto end_rename;
3736 	}
3737 	if (S_ISDIR(new.inode->i_mode)) {
3738 		new.is_dir = true;
3739 		retval = ext4_rename_dir_prepare(handle, &new);
3740 		if (retval)
3741 			goto end_rename;
3742 	}
3743 
3744 	/*
3745 	 * Other than the special case of overwriting a directory, parents'
3746 	 * nlink only needs to be modified if this is a cross directory rename.
3747 	 */
3748 	if (old.dir != new.dir && old.is_dir != new.is_dir) {
3749 		old.dir_nlink_delta = old.is_dir ? -1 : 1;
3750 		new.dir_nlink_delta = -old.dir_nlink_delta;
3751 		retval = -EMLINK;
3752 		if ((old.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(old.dir)) ||
3753 		    (new.dir_nlink_delta > 0 && EXT4_DIR_LINK_MAX(new.dir)))
3754 			goto end_rename;
3755 	}
3756 
3757 	new_file_type = new.de->file_type;
3758 	retval = ext4_setent(handle, &new, old.inode->i_ino, old.de->file_type);
3759 	if (retval)
3760 		goto end_rename;
3761 
3762 	retval = ext4_setent(handle, &old, new.inode->i_ino, new_file_type);
3763 	if (retval)
3764 		goto end_rename;
3765 
3766 	/*
3767 	 * Like most other Unix systems, set the ctime for inodes on a
3768 	 * rename.
3769 	 */
3770 	ctime = current_time(old.inode);
3771 	old.inode->i_ctime = ctime;
3772 	new.inode->i_ctime = ctime;
3773 	ext4_mark_inode_dirty(handle, old.inode);
3774 	ext4_mark_inode_dirty(handle, new.inode);
3775 
3776 	if (old.dir_bh) {
3777 		retval = ext4_rename_dir_finish(handle, &old, new.dir->i_ino);
3778 		if (retval)
3779 			goto end_rename;
3780 	}
3781 	if (new.dir_bh) {
3782 		retval = ext4_rename_dir_finish(handle, &new, old.dir->i_ino);
3783 		if (retval)
3784 			goto end_rename;
3785 	}
3786 	ext4_update_dir_count(handle, &old);
3787 	ext4_update_dir_count(handle, &new);
3788 	retval = 0;
3789 
3790 end_rename:
3791 	brelse(old.dir_bh);
3792 	brelse(new.dir_bh);
3793 	brelse(old.bh);
3794 	brelse(new.bh);
3795 	if (handle)
3796 		ext4_journal_stop(handle);
3797 	return retval;
3798 }
3799 
3800 static int ext4_rename2(struct inode *old_dir, struct dentry *old_dentry,
3801 			struct inode *new_dir, struct dentry *new_dentry,
3802 			unsigned int flags)
3803 {
3804 	int err;
3805 
3806 	if (unlikely(ext4_forced_shutdown(EXT4_SB(old_dir->i_sb))))
3807 		return -EIO;
3808 
3809 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3810 		return -EINVAL;
3811 
3812 	err = fscrypt_prepare_rename(old_dir, old_dentry, new_dir, new_dentry,
3813 				     flags);
3814 	if (err)
3815 		return err;
3816 
3817 	if (flags & RENAME_EXCHANGE) {
3818 		return ext4_cross_rename(old_dir, old_dentry,
3819 					 new_dir, new_dentry);
3820 	}
3821 
3822 	return ext4_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
3823 }
3824 
3825 /*
3826  * directories can handle most operations...
3827  */
3828 const struct inode_operations ext4_dir_inode_operations = {
3829 	.create		= ext4_create,
3830 	.lookup		= ext4_lookup,
3831 	.link		= ext4_link,
3832 	.unlink		= ext4_unlink,
3833 	.symlink	= ext4_symlink,
3834 	.mkdir		= ext4_mkdir,
3835 	.rmdir		= ext4_rmdir,
3836 	.mknod		= ext4_mknod,
3837 	.tmpfile	= ext4_tmpfile,
3838 	.rename		= ext4_rename2,
3839 	.setattr	= ext4_setattr,
3840 	.getattr	= ext4_getattr,
3841 	.listxattr	= ext4_listxattr,
3842 	.get_acl	= ext4_get_acl,
3843 	.set_acl	= ext4_set_acl,
3844 	.fiemap         = ext4_fiemap,
3845 };
3846 
3847 const struct inode_operations ext4_special_inode_operations = {
3848 	.setattr	= ext4_setattr,
3849 	.getattr	= ext4_getattr,
3850 	.listxattr	= ext4_listxattr,
3851 	.get_acl	= ext4_get_acl,
3852 	.set_acl	= ext4_set_acl,
3853 };
3854